# Paradigms regained

Theoretical and empirical arguments for the reassessment of the notion of paradigm

Edited by Gabriele Diewald Katja Politt

Empirically Oriented Theoretical Morphology and Syntax 10

## Empirically Oriented Theoretical Morphology and Syntax

Chief Editor: Stefan Müller Consulting Editors: Berthold Crysmann, Ana R. Luís

In this series:


ISSN: 2366-3529

# Paradigms regained

Theoretical and empirical arguments for the reassessment of the notion of paradigm

Edited by Gabriele Diewald Katja Politt

Gabriele Diewald & Katja Politt (eds.). 2022. *Paradigms regained: Theoretical and empirical arguments for the reassessment of the notion of paradigm* (Empirically Oriented Theoretical Morphology and Syntax 10). Berlin: Language Science Press.

This title can be downloaded at: http://langsci-press.org/catalog/book/318 © 2022, the authors Published under the Creative Commons Attribution 4.0 Licence (CC BY 4.0): http://creativecommons.org/licenses/by/4.0/ ISBN: 978-3-96110-326-3 (Digital) 978-3-98554-022-8 (Hardcover)

ISSN: 2366-3529 DOI: 10.5281/zenodo.5506578 Source code available from www.github.com/langsci/318 Collaborative reading: paperhive.org/documents/remote?type=langsci&id=318

Cover and concept of design: Ulrike Harbort Proofreading: Amir Ghorbanpour, Aniefon Daniel, Eduard S. Lukasiewicz, Jeroen van de Weijer, Marten Stelling, Jean Nitzke, Tom Bossuyt, Troy E. Spier, Vadim Kimmelman Fonts: Libertinus, Arimo, DejaVu Sans Mono Typesetting software: XƎLATEX

Language Science Press xHain Grünberger Str. 16 10243 Berlin, Germany langsci-press.org

Storage and cataloguing done by FU Berlin

## **Contents**


### Contents


## **Chapter 1**

## **Paradigms regained**

## Gabriele Diewald Leibniz Universität Hannover

## Katja Politt

Leibniz Universität Hannover

The contributions in this volume are anchored in the notions of paradigm and the paradigmatic organisation of linguistic items. The papers united here are substantial elaborations and enhancements of concepts as well as case studies that were presented at a workshop "Paradigms regained" held at the 52nd SLE Annual Meeting (SLE 2019), which took place from 21st–24th August 2019 at Leipzig University.

Its background is a long-lasting project aiming at assessing the cognitive reality of (grammatical) paradigms throughout various linguistic domains, thereby testing this notion for its ability to allow for "graceful integration" (Jackendoff 2011). A notion like this should be able to account for empirical findings and general cognitive mechanisms. In this volume, different domains of grammatical phenomena are investigated to illustrate what the concept of grammatical paradigms can and cannot – yet – explain. The theoretical and conceptual foundations of this project are grammaticalisation theory, implicational morphology, usage-based constructional approaches, cognitive semantics, as well as corpus-based and experimental approaches to grammatical structures in diachronic and synchronic phenomena.

## **Definitions and positions**

The notion of paradigm is primarily discussed in morphological theories, where it plays a central role as a tool for describing the structures in which inflectional forms are organised. The members of inflectional paradigms are primarily identified by their formal properties (cf. Fabri 1998: 7). Each member of a paradigm corresponds to a cell, which can be either filled by a form or by a form-feature pair

Gabriele Diewald & Katja Politt. 2022. Paradigms regained. In Gabriele Diewald & Katja Politt (eds.), *Paradigms regained: Theoretical and empirical arguments for the reassessment of the notion of paradigm*, 1–10. Berlin: Language Science Press. DOI: 10.5281/zenodo.5675839

### Gabriele Diewald & Katja Politt

(Lieb 2005, Werner 1994, Wurzel 1984). Lately, work on relational structures in morphological paradigms (Ackerman et al. 2009, Blevins 2015, 2016) has shown that this purely instrumental conception of paradigms as nothing but a useful descriptive convention clearly underestimates its cognitive foundation and functional importance. Paradigms in this sense are structures which provide "cohesive wholes" (Blevins 2015: 94) for the paradigm members. These structures consist of relations and associations between the individual cells within and in between paradigms. As such, they are part of speaker knowledge, because they provide necessary generalisations that allow speakers to infer previously unencountered forms of lexical items (Ackerman et al. 2009: 54). Knowing the overall organisational structure of the forms allows for inferring forms and their functions from one another, i.e., putting them into relation to one another.

This inferential nature of paradigms is what can be generalised as a structuring principle to other areas of grammar (Nørgård-Sørensen et al. 2011: xi). For this it is necessary to expand the notion of paradigms from a purely inflectional notion to a broader sense: It is assumed that grammatical items are structured in *grammatical paradigms*. Grammatical paradigms in this sense are functional sets (Andersen 2008, Diewald 2020, Nørgård-Sørensen et al. 2011, Politt 2021). They are holistic semiotic structures, consisting of ordered bundles of oppositions between all marked and unmarked members of the category in question (which in grammaticalisation are modified in various ways). Take the grammatical category tense as an example: The members of the tense paradigm share a common categorical function, namely situating events relative to the speech time. The unmarked zero in tense is the present. The specific function of all other members of the category – like past and future forms – can be described relative to that unmarked zero, i.e., in opposition to it and of course also in opposition to one another. Those oppositions serve a similar function as the aforementioned relations within inflectional paradigms; they (i) relate the members of a grammatical paradigm to each other and (ii) define the specific categorical function of each member relative to the categorical function of the other members In short: The oppositions and relations between the members of a grammatical paradigm are the very essence of grammatical structures (cf. Politt 2021). They "cannot be described without reference to the paradigmatic organisation that lies behind the syntagmatic realisations" (Nørgård-Sørensen et al. 2011: 71).

It is because of this internal relational structure that grammatical paradigms are an invaluable tool for describing the target structures of grammaticalisation processes and assessing the status of a grammaticalising element. These elements acquire a place in such a structure or change their place within it. By entering grammatical paradigms, elements form new oppositional pairs with other grammatical elements that are members of the same superordinate category, e.g.

### 1 Paradigms regained

tense (Diewald & Smirnova 2010: 4). By developing this opposition, the newly grammaticalised item becomes a member of a grammatical paradigm (Bybee et al. 1994, Lehmann 2015).

While it is undisputed that "grammar" is the target domain of grammaticalisation processes, and that "paradigms" play a role in the development of Indo-European languages, the exact extent of the notion of paradigm and grammatical paradigm and its usefulness for languages with little or no inflectional morphology has been under dispute for some time now. For once, there is fundamental criticism concerning the lack of an exact definition of "grammar", as put forth by Himmelmann: "[w]ork in grammaticalisation […] hardly ever makes explicit the concept of grammar underlying a given investigation" (Himmelmann 1992: 2). Furthermore, there is a lively discussion about (i) whether the notion of paradigm should be extended to syntagmatic linguistic structures beyond bound morphology and periphrastic forms, as for example in Construction Morphology (Booij 2010, 2016, 2018), and include, for example, grammatical oppositions on the level of the whole clause, like the opposition between sentences particles and modal particles. In constructional approaches, paradigms are often "marginalized or even lost" (Diewald 2020: 277). Another hotly disputed issue is (ii) what the benefit of such an extension might be (Bisang 2014, Diewald 2020, Haspelmath 2000, Wiemer & Bisang 2004).

This discussion, which arose in typological research and grammaticalisation studies, meets with current questions and challenges in construction grammar. The latter aims at describing grammatical structures in their entirety. If grammatical paradigms are indeed structures of the internal organisation of grammatical categories, it must be possible to describe them in constructional terms as well. The goal is therefore to find an integrative approach that combines both construction grammar and paradigms as organisational structures of grammar (Diewald 2009, 2015, 2020, Diewald & Smirnova 2010, Politt 2021). In such an approach, paradigms are not only the aforementioned generalisations of associative structures but they can be seen as constructions "whose function and meaning is defined by the specific number and constellation of [their] components", which "mutually define each other's values" through their inherent indexical structure (Diewald 2020: 303).

The basic assumptions derived from this background are:


These basic assumptions are to be tested and refined based on the case studies and theoretical reflections offered by the contributions of this volume.

## **The papers**

The contributions in this volume explore and test these assumptions, raising questions like the following ones:


The contributions range from diachronic and synchronic case studies to broadly scaled surveys of different types of paradigmatic organisation to theoretical reflections of relevant notions within this field of research. This allows for an arrangement of the contributions to this volume into three sections: The first section, containing two papers, deals with general terminological, definitional, and theoretical issues. A broad survey on large-scale diachronic mechanisms and drifts building up inflectional paradigms of various types (Andersen) is followed by a theoretical reflection on the status of paradigms as metaconstructions in the construction grammatical approach (Leino). The second section consists of two papers paying close attention to the details of particular mechanisms and (diachronic) processes steering the organisation of morphological paradigms and more extended constructions, with one of them investigating the interplay of inflection and derivation in Slavic languages (Wiemer), and the other one dealing with the morphological process of recursion in relation to composition, mainly drawing on examples from Turkish (Reiner). The third section consists of six contributions offering detailed language specific case studies, taking up linguistic phenomena of Danish (Hansen, Heltoft), Dutch (Nuyts, Caers & Goelen), German (Hartmann & Neels), Norwegian (Kurek-Przybilski) and French (Kragh), most of them addressing or focusing on diachronic issues. The following are brief outlines of the contributions in the order in which they appear.

### 1 Paradigms regained

In "Paradigms of paradigms" Henning Andersen provides a broad overview of the types of organisational structures of paradigms in inflectional paradigms (selectional sets) arising in the course of diachronic processes. The article provides a classification of formal and functional complexities of paradigms of verbal and nominal categories (case, number, person, tense, aspect, mood, voice) that are due to hierarchical nesting or embedding of their morphological exponents. Attention is given to the interaction of different techniques within one grammatical paradigm in a particular language, e.g., the "typological gradation" in the paradigms of verbal categories in English, French, and Latin conjugations, which includes the phenomena of auxiliarisation, agglutination, irregular forms, fusion, ablaut and suppletion. Contending that morphological systems are typologically diverse (due to diachronic processes), it is argued that morphological theory – also in synchronic analysis – must take care of the fundamental feature of typological diversity in its theoretical and methodological layout from the start.

In "Formalizing paradigms in Construction Grammar" Jaakko Leino discusses the question of how constructions in a language are organised. He draws on both Construction and Cognitive Grammar to explore similarities and differences – relations and oppositions – between constructions and introduces the notion of *metaconstructions* (Leino 2003, Leino & Östman 2005) as a generalisation of constructions on a more schematic level. leino contrasts the two notions of metaconstructions and grammatical paradigms with each other and explores how metaconstructions can serve as a means to describe the internal organisation of grammar and as a base for the formation and integration of new constructions in(to) the system.

The comprehensive contribution by Björn Wiemer "No paradigms without classification: How stem-derivation develops into grammatical aspect" develops detailed suggestions on the subclassification of complex paradigms of verbal aspect in Slavic languages. Based on data from several Slavic languages, it proposes a layered conception of the notion of paradigm. The first layer rests on the principal binary distinction between perfective and imperfective aspect, which is realised by the inherent aspectual features of the verb stems and their associated derivational pattern. The second layer is constituted by subparadigms, which are triggered by specific, mutually exclusive bundles of regular usage conditions and contextual features. Drawing on notions from Construction Grammar approaches and Word-and-Paradigm models, it is suggested that these bundles of features can be conceived as constructional templates for individual aspectual values within subparadigms, which operate on an underlying binary system of

aspectual distinctions based on verb stems. Thus, the paper puts forward farreaching suggestions for analysis of grammatical distinctions that integrate lexical, constructional, and contextual features.

In "Recursion and paradigms" Tabea Reiner discusses morphological paradigms from both a constructional and compositional perspective. By comparing the status of inflectional paradigms in Constructional, Distributed, and Autonomous Morphology, she raises the question what they can contribute to a model of inflectional recursion. Namely, paradigms could serve as a means of modelling restrictions on recursion patterns in inflectional morphology and therefore constitute a fundamental unit of morphological description.

In "Redundant indexicality and paradigmatic reorganisations in the Middle Danish case system" Bjarne Simmelkjær Sandgaard Hansen investigates the fundamental changes of the Danish case system during the Middle Danish period as an instance of grammaticalisation. The central issue is the increasing topological fixation inside the noun phrase and its interaction with phrase internal agreement and case marking. The newly established system of noun phrase marking is shown to be an instance of grammatical change proper (instead of e.g. phonologically induced change) and provides a prime example of the claim that grammaticalisation is inextricably linked with paradigmaticisation.

The contribution by Lars Heltoft "The semantic reorganisation of case paradigms and word order paradigms in the history of Danish" investigates the interaction of word order change, namely the topological fixation of the subject position and serialisation rules in the middle field as well as change in the case system in the history of Danish. Assuming that inflectional and (pluri-item) constructional representations of grammatical information alike are organised in paradigms, this investigation highlights the shifts in the "co-organisation" of the expression of grammatical content. It is suggested that this type of complex collaboration between morphological and topological marking techniques can be called second-order paradigms or hyperparadigms.

In raising the question "The Dutch modals, a paradigm?" Jan Nuyts, Wim Caers and Henri-Joseph Goelen depart from morphology-based definitions of "paradigm" and adopt a cognitive perspective, whereby a paradigm is defined as a "cognitively real phenomenon". Relevant criteria for a paradigm, more precisely the gradual rise of paradigms, are seen in the gradual accumulation of shared grammatical and semantic features, and an increasingly pronounced "divisions of labor", i.e. a stricter internal functional organisation, among the entities involved. In presenting a "meta study" of several diachronic investigations of the development of the Dutch modal verbs *kunnen* 'can', *mogen* 'may', *moeten* 'must', and *hoeven* 'need', the broad lines of change and convergence in structural and

### 1 Paradigms regained

semantic features of this group are taken as an instance of this type of paradigmaticisation. Furthermore, the authors raise the question whether grammaticalisation should always be seen as a necessary correlate of paradigmhood, thus, offering arguments for further discussion on the theoretical issue concerning the distinction between paradigmatic relations on the one hand and narrowly defined paradigms on the other hand.

Stefan Hartmann and Jakob Neels analyze the grammaticalisation of a family of constructions in "Grammaticalisation, schematisation and paradigmaticisation: How they intersect in the development of German degree modifiers". Drawing on both synchronic and diachronic corpus data, they explore the gradual context expansion of German degree modifier-constructions such as [*ein wenig* X] ('a little'), [*ein bissche*n X] ('a bit'), [*ein Quäntchen* X] (lit. 'a quantum'), [*ein Tick* X] (lit. 'a tick') and [eine Idee X] (lit. 'an idea'). They aim to show that paradigmaticisation leads to multiple interrelated paradigms with varying levels of schematicity, similar to the differences of higher and lower level constructions discussed by Traugott (2007).

In "Generics as a paradigm: A corpus-based study of Norwegian" Anna Kurek-Przybilski investigates how the notion of grammatical paradigms can help in modelling language specific grammatical categories. Looking at encyclopedic texts from Nynorsk, she develops a genericity paradigm that can serve as a baseline for investigating co-existing varieties of a language and helps in understanding the grammaticalisation process of generic contexts and expressions.

Kirsten Jeppesen Kragh draws on French diachronic data to illustrate "The importance of paradigmatic analyses: From one lexical input into multiple grammatical paradigms". By following the grammaticalisation path of the French verb *voir* 'to see' and its polygrammaticalisation into multiple grammatical categories, Kragh shows that the target structures of grammaticalisation do not necessarily have to belong to the same grammatical areas. For *voir*, she illustrates grammaticalisation paths into tense, aspect, mood, voice, as well as discourse markers and prepositions. Synchronic paradigms serve as the target structure of these grammaticalisation paths and allow for precise descriptions of the newly acquired grammatical functions of the grammaticalised elements due to their inherent relational structure.

Gabriele Diewald & Katja Politt

## **References**


### 1 Paradigms regained


*its fringes and its components* (Trends in Linguistics. Studies and Monographs 158), 3–20. Berlin, Boston: De Gruyter Mouton.

Wurzel, Wolfgang U. 1984. *Flexionsmorphologie und Natürlichkeit: Ein Beitrag zur morphologischen Theoriebildung*. Berlin: Akademie-Verlag.

## **Chapter 2**

## **Paradigms of paradigms**

## Henning Andersen

University of California, Los Angeles

This paper is concerned with inflectional morphology. Its point of departure is the old insight that paradigms of content categories are typically nested inside other paradigms forming hierarchical structures, e.g., case paradigms are included in number paradigms: case] number]. Similarly, say, in the Latin verb, person] number] tense] aspect] mood] voice] (Section 1). As for exponence, paradigms (selectional sets) of inflectional classes (declensions, conjugations) are more often flat (linear) and asymmetrical with respect to one or more characteristics, e.g., meaning, stem shape, or productivity. But they may include selectional sets of allomorphs at a lower level, say, in individual cases or tenses, presenting paradigms in paradigms within paradigms (...) (Section 1.1). A third dimension in inflectional systems is the typological paradigm of morphological techniques commonly reflected in synchronic variation; English verb morphology, for instance, comprises analytic (*will call, got arrested*), agglutinative (*waded, jogged*), fusional (*kept, built*), and introflective (*sang, hung*) formations (Section 2.1). As additional examples of typological paradigms the conjugations of French, Latin, and Russian are examined (Sections 2.1–2.4). It is shown that such typological variation may reflect the historicity of an inflectional system. Since this variation is part of speakers' competence, it should be recognized as an inomissible part of synchronic description (Section 3).

## **1 Introduction**

## **1.1 Content paradigm**

The purpose of this paper is to draw attention to paradigmatic relations in inflectional morphology that have traditionally been overlooked. The background for this examination is the distinction between content paradigms and exponent paradigms. We begin with paradigms of grammatical categories.

Henning Andersen. 2022. Paradigms of paradigms. In Gabriele Diewald & Katja Politt (eds.), *Paradigms regained: Theoretical and empirical arguments for the reassessment of the notion of paradigm*, 11–36. Berlin: Language Science Press. DOI: 10.5281/zenodo. 5675841

### Henning Andersen

Quite naturally the notion of paradigm – what de Saussure (1959 [1916]) called associative relations – has historically implied a focus on grammatical content categories and their members. Saussure illustrated the notion with members of inflectional categories and derivational categories and, to include the exponent level, sets of homonyms. When Jakobson (1956b) characterized speech as a product of the dual processes of selecting and combining linguistic entities, the selectional sets he referred to were evidently paradigms of lexical and grammatical categories (see also Bloomfield 1933: 164; Hamp 1966, s.v. *selection*). Such categories serve well to illustrate the diversity of selectional sets.

Scalar paradigms perhaps best illustrate Saussure's understanding that linguistic categories are imposed on the world of experience and thereby shape our conceptual categories. Consider the colors in the world around us, which in the world of languages are represented by anywhere from two to a dozen simple color terms. The modern understanding of the hues of the color wheel has facilitated the identification of a multitude of colors. Their multifarious exploitation in design and fashion, as in interior decorating and in clothing, suggests that this perceptual dimension has a potential for infinitely differentiated paradigms of hues and subparadigms of shades, tints, and tones – through the imposition of color names (or code numbers). Many other scalar dimensions of experience are expressed with paradigms of contrary opposites, e.g., *large* vs *small*, *wide* vs *narrow*, *dark* vs *light*, *loud* vs *soft*, which are employed relative to explicit or implicit standards in both literal and metaphorical senses.

Graded (step-wise) categories are as common. The number words from *one* to *ten*, which recursively name units, tens, hundreds, thousands, and so on, illustrate how paradigms can be nested inside paradigms, in this instance ad infinitum; see (1). Decimal fractions of each whole number likewise form nested paradigms stretching to the infinitely small, a conceptual counterpart to the potentially infinite differentiation of the realm of colors.

(1) units ] tens ] hundreds ] thousands ] ten-thousands ] ...

Grammatical categories typically form paradigms of contradictory opposites. These paradigms do not have infinitely many members, but a hierarchical ordering of paradigms is commonly in evidence. Typically, for instance, in languages with grammatical cases, the case paradigm is subordinated to a number paradigm; see Table 1.

This relation of subordination, which can be summed up as case ] number ], becomes manifest in historical change when case is lost while number remains, as in many European languages.

### 2 Paradigms of paradigms


Table 1: Latin First declension

(2) person ] number ] tense ] aspect ] mood ] voice ]

Similarly in verbs, the category of person is often subordinated to that of number. In Latin this relationship is realized in each of the three tenses Present, Imperfect, Future, in both aspects, Infective and Perfective (where the tenses are called Perfect, Pluperfect, and Future Perfect), in both moods, Indicative and Subjunctive, and in both voices, Active and Passive, yielding a structure that is easily represented by a tree-diagram, but can be summed up as in (2).

## **1.2 Exponent paradigms**

When we turn to the expression side of languages, we find paradigms of less consistently hierarchical structure. A language may have extensive allomorphy in its inflectional categories; see, for example the Latin case allomorphy in Table 2.

Table 2: Allomorphy in Latin noun declension. Singular


The allomorphs are organized into several partly overlapping classes, the "declensions". The grammatical tradition recognizes five declensions (Table 3), that is, a paradigm of five noun classes forming a flat structure But several of the declensions have cases with notable internal allomorphy, that is, allomorph paradigms within case paradigms within the paradigm of declensions; see Table 3.

### Henning Andersen


Table 3: Allomorphy in Latin case desinences

In Decl. 2, for instance, feminines and masculines have syncretic desinences in all cases; e.g., *fīcus*.f.nom.sg, *fīcum*.f.acc.sg, *fīcī*.f.gen.sg 'fig', *servus*.m.nom.sg, *servum*.m.acc.sg, *servī*.m.gen.sg 'slave'. Both masculines and neuters have additional, lexically conditioned allomorphs: Some Decl. 2 masculines have a -∅ nom.sg desinence (*vir* 'man', *liber* 'book'), and a few Decl. 2 neuters have *-us*.nom/acc.sg (*vulgus* 'crowd', *vīrus* 'poison'). This allomorphy is conditioned by the gender of lexical stems; hence each of the endings *-um*.nom.sg and *-us*.acc.sg points to (or indicates) the neuter gender of its noun stem; see Table 4.

Table 4: Allomorphy paradigms in Latin Second-declension desinences


Table 5: Allomorphy paradigms in Russian First-declension desinences


A similar case paradigm in Russian (Table 5) includes paradigms of allomorphs that indicate the gender of the stems: nom.sg *-o* (*okn-o* 'window') indicates neuter, nom.sg -∅ (*gorod-∅* 'town') indicates masculine. But additional allomorphs within the masculine point to referential features such as animacy

### 2 Paradigms of paradigms

(acc.sg *syn-a* 'son' vs *gorod-∅*) or mass vs countable (gen.sg *čaj-u* 'tea' vs *goroda*) or material vs nonmaterial referent (loc.sg *v sneg-ú* 'in the snow' vs *v sneg-e* 'in snow (as concept or word)').

In terms of the theory of semiotics of Charles S. Peirce, the semiotic value of this low level allomorphy is identified as indexical (Shapiro 1969; Anttila 1972, 1989; Andersen 1980, 2020): individual allomorphs point to, indicate, or are indexes of phonological, grammatical, or lexical features of given noun stems or features of their referents or combinations of such subsidiary information.

A systematic investigation of the contribution of this secondary level of signaling in communication is a task for the future. But the index values of allomorphs or morphological processes such as mutation or truncation will be relevant repeatedly below.

## **2 Typological paradigms**

Here we turn to yet another paradigmatic relation in inflection, one that opens up a neglected perspective on morphological systems.

For the purposes of this exposition it is useful to be able to refer to Sapir's (1921: 120–146) typology of morphological techniques. It is presented in Table 6 with minimal characterizations of the individual types, derived from Sapir's text; I use *introflection* for Sapir's *symbolism*.


Table 6: Basic morphological techniques

### Henning Andersen

The major types (A, B, C) form a paradigm, and within the synthetic macrotype, agglutination, fusion, and introflection form another paradigm. In the following pages we will see some examples of how these paradigms are exploited in morphological systems. The examples are taken from English (Section 2.1), French (Section 2.2), Latin (Section 2.3), and Russian (Section 2.4).

## **2.1 English conjugation**

It is convenient to begin with the conjugation of the English verb; see the major verbal categories in Table 7. Infinitive and imperative are uninflected. They have identical lexical content; their distinct grammatical content is expressed solely by their syntactic properties. In the Present Indicative, the 3sg is suffixed /-ǝz/ (|| /iz/) after sibilants, /-s/ after voiceless stops, and otherwise /-z/, e.g., *pitch-es*, *bat-s*, *jog-s*, *run-s*. The content of this desinence is debatable; some would consider it cumulative ('3sg Present Indicative'), but it might be just 'Indicative', specifying a predicate as asserted, in the least marked Person–Number–Tense environment. Assuming that cumulative exponence is characteristic of fusion (cf. Table 6), this noncumulative interpretation would be compatible with agglutination, as are the phonotactic adjustment after sibilant stems and the phonological adjustment after voiceless stops. The Present participle in /-iŋ/ is agglutinative.


Table 7: English verb morphology: Categories

The Past tense and Past participle are expressed by three formations that differ in morphological technique. They form a selectional set, a paradigm; data from Bloch (1947), Palmer (1987: 249–257).

Preterite 1, the productive pattern characteristic of thousands of verbs and applied to all new verbs, is agglutinative. It has a phonotactic adjustment /-ǝd/ (|| /id/) after stem-final dental plosives (here written ...T), a voiceless allomorph /-t/ after stem-final voiceless consonants, and otherwise /-d/; see Table 8.

Preterite 2 is fusional. It comprises verbs with the regular /-t/ and /-d/ allomorphs and some that have a lexically conditioned /-t/. Some of these verbs have

### 2 Paradigms of paradigms

a vowel mutation in the Preterite (*keep–kept*, *flee–fled, mean–meant*), some have both a vowel and a coda mutation (*lose–lost*), some have a vowel mutation and coda truncation (*can–could*, *catch–caught*); a couple have just coda truncation (*make–made*); see Table 9.

Table 8: English Preterite 1: Agglutinative


Table 9: English Preterite 2: Fusional


The mutations in the stems of these verbs are conditioned by the Past-tense category. Since Past tense is separately expressed by the distinct Past tense marker in all these verbs, the mutations are indexical, and the morphological type of these forms is fusional; cf. Table 6. Some of these verbs have an agglutinative Preterite 1 variant in which the mutation is omitted, e.g., *leaped*, *dreamed*, *kneeled*, *cleaved*. There are also basically agglutinative verbs with a variant 'Irregular *-t*' Preterite 2 and no mutation, e.g. *spell–spelt*, *spill–spilt*.

There are no verbs in Preterite 2 that have the phonotactic vowel epenthesis (-ǝd) found in Preterite 1. Instead, Preterite 2 verbs in stem-final dental plosive (here written ...T) truncate the ...t or ...d after the appropriate Preterite allomorph has been selected, that is, ...t-t → *-t*, ...d-d → *-d*, ...d-t → *-t*; see Table 10. Some English experts view the ending in these verbs as -∅ (Palmer 1987). The verbs with regular *-t* or -*d* might be described that way; but the verbs with 'Irregular *-t*' suggest that it is indeed the stem-final dental plosive that is truncated.

### Henning Andersen

Table 10: English Preterite 2′ : Fusional


Preterite 3 comprises verbs with unsuffixed and with suffixed pst.ptcp. Besides this distinction there are verbs with three alternating vowels (*sing–sang– sung*, *drive–drove–driv-en*), with two vowels, identical in prs and pst.ptcp (*run– ran–run*, *know–knew–know-n*) or identical in pst and pst.ptcp (*hang–hung– hung*, *speak–spoke–spok-en*), and with a single vowel throughout (*beat–beat– beat-en*); see Table 11.

Preterite 3 is typologically diverse. In the unsuffixed lexemes past tense and past-participle function are expressed through introflection (cells 1.1–1.3). In the suffixed lexemes, past tense is expressed through introflection (cells 2.1–2.4), and past-participle functions by the suffix alone (cells 2.2, 2.4; agglutination) or by the suffix accompanied by stem-vowel mutation (cells 2.1, 2.3); this is fusion.

Casual speaking styles show a strong tendency to extend the introflective pasttense forms to past-participle function, e.g. *I would've did it differently* (cell 2.1), *you could've came earlier* (cell 1.2), *they should've took the other one* (cell 2.2).

The different formations of the English Preterite form a typological paradigm as in Table 12. As a selectional set they are evidenced in synchronic (stylistic, social) variation. Preterite 1 ~ Preterite 2: *dreamed* ~ *dreamt*, *kneeled* ~ *knelt*; Preterite 1 ~ Preterite 3: *strived* ~ *strove*, *thrived* ~ *throve*, (metaphorical) *weaved* (through traffic) ~ *wove*, *slayed* (an audience) ~ *slew*; Preterite 2´ ~ Preterite 3: *(for)bid* ~ *(for)bade*; note Preterite 1 ~ 2 ~ 3 in *cleave*–*cleaved* ~ *cleave*–*cleft* ~ *cleave*–*clove–cloven*; all variants cited from *The American Heritage Dictionary*.

Table 12 displays the paradigm of morphological techniques of English conjugation.<sup>1</sup>

<sup>1</sup>English noun plurals exemplify the same variation. Analytic: *heads of cattle*, *... cabbage*; *pairs of scissors*, *... trousers*, *cloves of garlic*. Agglutination: *horse-s*, *cat-s*, *dog-s*, *cow-s*; *ox-en*. Fusion: *calf–calv-es*, *house–hous-es*, *youth–youth-s*; *child–childr-en*. Introflection: *woman–women*, *man–men*; *foot–feet*, also *tooth*, *goose*; *mouse–mice*, also *louse*; *crisis–crises*, *alumna–alumnae*; *sheep–sheep*, also *deer*, *grouse*, *trout*, *fish*; *sail*, *cannon*.





### Henning Andersen

Here it is worth noting that the vowel alternations in Preterite 3 verbs reflect those of the Indo-European Present vs. Perfect formations. These apophonic alternations, which Germanic shares with all the other Indo-European languages, may be at least 7000 years old. The agglutinative Preterite 1, by contrast, is Common Germanic heritage, perhaps less than 2500 years old. With the exception of the 'be' passive, the analytic, auxiliated formations are much younger.

It appears that this typological perspective reflects what one can call the historicity of the system of verbal morphology.

Now, we know enough of the history of English to recognize that this is not the same as reflecting the history of the language. A historical account will acknowledge (i) that Preterite 2 developed from Preterite 1 thanks to a variety of conditioned sound changes, so it is younger; (ii) that in Old English our apophonic Preterite 3 verbs had a separate set of past-tense desinences, that is, they were fusional; they became introflective only when this 'strong' past-tense inflection was lost; but also (iii) that more Preterite 3 verbs have changed to Preterite 1 or 2 since Old English, than vice versa, and (iv) that there is a similar predominance of Preterite 2 verbs transitioning to Preterite 1.

None of these details can be read off the overview in Table 12. Still the typological paradigm undeniably suggests a generalized historical perspective on this synchronic system.

In the following pages I will look at a few other languages whose history is known, to see to what extent such a historical perspective may be a common feature of morphological systems.

## **2.2 French conjugation**

French has two regular conjugations, one productive, exemplified by *chanter* 'sing', the other practically unproductive, typified by *finir* 'finish'. In addition there is a number of irregular verbs. The system of categories can be seen in Table 13.

An important feature of French conjugation is that finite verbs are obligatorily accompanied by subject clitics; consequently the (contingent) suffixal Participant (Person, Number) marking is strictly an agreement feature.

Conj. 1 is agglutinative: Suffixes for tense, mood, person, and the nonfinite categories simply follow the stem. There is some stem allomorphy: Some verb stems with mid vowels have a regular alternation between pretonic and tonic (final, closed) syllable: /e/, /ǝ/ ~ /ɛ/ (*céder–cède* 'cede', *jeter–jette* 'throw', *appeler– appelle* 'name, call'), /ø/ ~ /œ/ (*beurrer–beurre* 'butter'), and /o/ ~ /ɔ/ (*coller–colle*

### 2 Paradigms of paradigms


Table 13: French tense system

'glue'). The alternation has no apparent synchronic motivation; contrast *aider– aide* 'help' with /ɛ/, *sauver–sauve* 'save' with /o/; but it is phonologically (prosodically) conditioned, and it is irrelevant to the stem–desinence boundary. Thus it is compatible with agglutination.

Conj. 2 is similarly agglutinative. But it is characterized by two truncations that produce distinct stems for (i) Prs.ind.123sg and (ii) the Preterite and Infinitive systems; see (3.a–b), where superscript 0, 00 represent coda and rhyme truncation, respectively; the basic stem ends in /s/, e.g., /finis-/. The truncations produce stem allomorphs with specific grammatical meaning: This is a fusional feature.

	- b. st.ind: fin00-i-∅.123.sg, fin00-i-m.1pl, fin00-i-t.2pl, fin00-i-r.3pl; Pst.sbj: fin00-i-s-∅.12sg.3pl, fin00-i-s-j-õ.1pl, fin00-i.pst.ptcp, fin00-ir.inf

The contrast between the stem-final ...s- in finis-∅.prs.sbj.123sg.3pl (*finisse(nt)*) (3) and the Pst.sbj morpheme -s- in fin00-i-s-∅.pst.sbj.12sg.3pl (*finisse(nt)*) (3b) is recognized by French grammarians (see Grevisse 1961: 588–589). These grammatical forms are systematically homophonous in Conj. 2 verbs, but since they have different morpheme constituency they are not homonymous.

Irregular verbs have largely the same agglutinative suffixations as Conj. 1 and 2 verbs; but they have different allomorphs in the Preterite and Future systems; and they are characterized by stem mutations and truncations, as well as by stem suppletion. The lexical distribution of these features is irregular. A systematic presentation of the whole picture would exceed the space available here. The

### Henning Andersen

following subregularities and a few illustrations in (4) will suffice for the present purpose.

	- b. Coda truncation in Prs.ind.sg; e.g., *dormir* 'sleep': dɔrm-õ.1pl, dɔrm-∅.3pl, but dɔr<sup>0</sup> -∅.3sg (*dormons*, *dorment*, *dort*); *bouillir* 'boil': buj-õ, buj-∅, bu<sup>0</sup> -∅ (*bouillons*, *bouillent*, *bout*); *lire* 'read': liz-õ, liz-∅, li0 -∅ (*lisons*, *lisent*, *lit*).
	- c. Coda truncation in Prs.ind.sg also occurs in some verbs with vowel mutation (but not in Prs.sbj.1–3sg): *recevoir* 'receive': rǝsǝv-õ.1pl, rǝswav-∅.3pl, rǝswa<sup>0</sup> -∅.3sg (*recevons, reçoivent, reçoit*); *devoir* 'ought; owe': dǝv-õ, dwav-∅, dwa<sup>0</sup> -∅ (*devons*, *doivent*, *doit*); *boire* 'drink': byv-õ, bwav-∅, bwa<sup>0</sup> -∅ (*buvons*, *boivent*, *boit*); *mouvoir* 'move': muv-õ, mœv-∅, mø<sup>0</sup> -∅ (*mouvons*. *meuvent*, *meut*).
	- d. Rhyme truncation in pst.ind.sbj, as in Conj. 2; e.g., *acquérir*: aker-õ.1pl, ak00-i-∅.3sg (*acquérons*, *acqui*); *voir* 'see': vwaj-õ, v00-i-∅ (*voyons*, *vi*); *recevoir*: rǝsǝv-õ, rǝs00-y-∅ (*recevons*, *reçu*); *devoir*: dǝv-õ, d <sup>00</sup>-y-1∅1 (*devons*, *du*); *boire:* byv-õ, b00-y-∅ (*buvons*, *bu*); *lire*: liz-õ, l <sup>00</sup>-y-∅ (*lisons*, *lu*); *savoir* 'know': sav-õ, s00-y-∅ (*savons, su*).
	- e. Preterite allomorphy. -i- ~ -y-; e.g., (i) -i- in both pst.ind.sbj and pst.ptcp: e.g., *dormir*: dɔrm-i-∅, dɔrm-i (*dormi*, *dormi*); *bouillir*: buj-i-∅, buj-i (*bouilli*, *bouilli*); (ii) -i- in pst.ind-sbj, -y- in pst.ptcp: *voir*: v00i-∅ and v00-y (*vi*, *vu*), *rompre* 'break': rõp-i-∅, rõp-y (*rompi*, *rompu*), *battre* 'beat': bat-i-∅, bat-y (*batti*, *battu*). (iii) -y- in both pst.ind-sbj and pst.ptcp: *lire*: l00-y-∅ and l00-y (*lu*, *lu*), *courir* 'run': kur-y-∅ and kur-y (*couru*, *couru*).
	- f. Interfixed consonants in Inf and/or Fut.Cond; e.g., *connaître* 'know': kɔnɛs-õ, kɔnɛ<sup>0</sup> -t-r.inf (*connaissons*); *coudre* 'sew': kuz-õ, ku<sup>0</sup> -d-r (*cousons*); *moudre* 'grind': mul-õ, mu<sup>0</sup> -d-r (*moulons*); *tenir* 'hold': tǝn-õ, tjẽ-d-r-ɛ.cond.3sg (*tenons*, *tiendrait*); *vouloir* 'will, want': vul-õ, vu<sup>0</sup> -d-r-ɛ (*voulons*, *voudrait*).

### 2 Paradigms of paradigms

The features in (4a–4f) are relevant to many irregular verbs. They define some stem allomorphy indicating desinential grammatical content and some desinence allomorphy indicating the lexical content of the stem: This is fusion.

The following examples in (5) illustrate wordforms that combine lexical and grammatical content: Introflection.

	- b. A specific stem for Fut and Cond. *tenir*: tǝn-õ, tjẽ-d-r-e.fut.1sg (*tenons*, *tiendrai*), *aller*: al-õ, ir-e (*allons*, *irai*), *voir*: vwaj-õ, vɛr-e (*voyons*, *verrai*).
	- c. A specific wordform for pst.ptcp (feminine endings in parentheses). *Offrir* 'offer': ɔfr-õ, ɔfɛr(-t) (*offrons*, *offert*); *ouvrir* 'open': uvr-õ, uvɛr(-t) (*ouvrons*, *ouvert*); *mourir*: mur-õ, mɔr(-t) (*mourons*, *mort*); combined with rhyme truncation: *acquérir*: aker-õ, ak00-i(-z) (*acqérons*, *acquis*); *mettre* 'put': mɛt-õ, m00-i(-z) (*mettons*, *mis*); *écrire* 'write': ekriv-õ, ekr00-i(-t) (*écrivons*, *écrit*).
	- d. Other grammatically specific stems or wordforms. (i) *haïr* 'hate': ais-õ, ɛ.prs.123sg (*haïssons*, *hait*); (ii) *pouvoir* 'can, be able': pɥi.prs.ind.1sg, pɥis–.prs.sbj (*puis*, *puisse*(*nt*)); (iii) *savoir*: sav–.prs.ind.pl/inf, se.prs.ind.sg, saš–prs.sbj/ptcp, s00–.pst.ind/sbj/ptcp, so-r–.fut/cond (*savons*, *savais*, *sais*, *sache*(*nt*), *sachant*, *su*s, *saurai*); (iv) *aller*: vɛ.prs.ind.1sg, va.23sg, võ.3pl, aj–.prs.sbj, ir–.fut/cond (*vais*, *va*, *allons*, *vont*, *aille*, *ir-ai*); (v) *avoir* 'have': e.prs.ind.1sg, a.23sg, õ.3pl, ɛ.prs.sbj.123sg.3pl, y-.pst/pst.ptcp, or–.fut/cond (*avons*, *ai*, *a*, *ont*, *aie*, *eus*, *aurai*); (vi) *être* 'be': sɥi.prs.ind.1sg, ɛ.23sg, som.1pl, ɛt.2pl, sõ.3pl, swa–.prs.sbj, fy–.pst, ɛt–.inf, sǝr–.fut/cond, ete.pst.ptcp (*suis*, *est*, *sommes*, *sont*, *sois*, *fus*, *êt-re*, *se-r-ai*, *été*); (vii) *faire* 'do, make': fǝz–.prs.ind.1pl/ptcp/impf, fɛ.prs.123sg/inf, fɛt.prs.ind.2pl, fõ.3pl fas–.prs.sbj, fi.pst, fǝr–.fut/cond, fɛ(-t).pst.ptcp (*faisons*, *fait*, *faites*, *font*, *fasse(nt)*, *fi*, *ferai*, *fait*).

The irregular lexical distribution of the many subregularities in the morphology of these verbs makes for some complexity; in Stump & Finkel's (2017) approach, French has 72 conjugations. Still, it is clear that features in (4a–4f) produce allomorphy, in stems or suffixes, that amounts to cross-boundary indexes;

### Henning Andersen

they exemplify the technique of fusion. The examples in (5) are stems or wordforms that combine lexical and grammatical content; this is introflection. Many irregular verbs have no introflective forms at all, but several have a handful or more.


Table 14: French conjugation in typological perspective

In Table 14, only the Passive is old; the other analytic formations have developed since the early Middle Ages. The productive, agglutinative pattern is the descendant of the Latin productive Conj. 1. Among the fusional verbs only the regular but unproductive *finir* conjugation continues a productive Latin formation (Late Lat. *finīscō–finīvī–finīre*). The other fusional patterns as well as all the introflective ones go back to pre-Latin formations that had ceased to be productive in classical Latin; this is true also of some post-Latin suppletive verbs, e.g., Fr. *aller* (< *ambulāre*, *vadere*, *īre*) and *être* (< *esse*(*re*), *stare*).

The analytic formations are of different age. But the synthetic part of the paradigm largely reflects the historicity of the system.

2 Paradigms of paradigms

## **2.3 Latin conjugation**

The hierarchy of Latin verbal categories was briefly summarized in (2), repeated here as (6).

(6) person ] number ] tense ] aspect ] mood ] voice ]

In Latin, verbal inflection is organized in a paradigm of four conjugation classes, traditionally numbered 1 to 4. Conj. 1 is fully productive, Conj. 4 less so; Conj. 2 and 3 are unproductive, the latter, with the exception of inchoative verbs formed with the suffix *-sc-* (*senēscō* 'age', Late Lat. *finīscō* 'finish').

Basic-stem formation. Verb stems are derived from lexical morphemes with interfixes (stem formatives), in Conj. 1 with *-ā-*, in Conj. 4 with *-ī-*, in Conj. 2 with *-ē-* (~ -∅- ~ *-i-*; see below). In Conj. 3, some basic (Infective) stems are derived with several lexically conditioned affixes, other stems are bare, e.g., interfixed (*cap-i-ō* 'catch'), infixed (*ru-m-p-ō* 'break'), and bare stem (*ag-ō* 'lead'). The interfixation, being lexically conditioned, is fusional derivation. Verb stems serve as bases for inflection for Voice, Mood, Aspect, Tense, Number and Person as well as the derivation of a roster of deverbal nominal (Infinitive, Gerund), adjectival (Gerundive, Present and Past Participle), and adverbial (Supines 1 and 2) derivatives, as well as deverbal verbs (e.g., frequentative, desiderative). Here we focus on the obligatory verbal categories.


In Conj. 3 Perfective stems are related to Infective stems by (i) deaffixation (*cap-i-ō–cēp-ī*, *ru-m-p-ō–rūp-ī*), (ii) vowel mutation and/or (iii) quantity change (*ag-ō–ēg-ī*), (iv) reduplication (*curr-ō–cu-curr-ī* 'run'), (v) a combination of some

### Henning Andersen

of these (*ta-n-g-ō–te-tig-ī* 'touch'), (vi) affixation (*scrīb-ō–scrīp-s-ī* 'write', *ping-ō– pinx-ī* 'paint'), or (vii) invariant-stem inflection (*scand-ō–scand-ī* 'ascend'). Types (i–v) are fusional: the stem allomorph points to the Perfect desinences; (vi) and (vii) are agglutinative, (vi) with an overt *-s-*, (vii) with zero Perfect suffix.

	- b. Past. Conj. 1 *amā-b-a-m*.pst.infv, *amā-v-er-a-m*.pst.pfv; Conj. 2 *monē-b-a-m*, *mon-u-er-a-m*; Conj. 3 *capi-ē-b-a-m*, *cēp-er-a-m*; *ag-ē-b-a-m*, *ēg-er-a-m*; Conj. 4 *audi-ē-b-a-m*, *audī-v-er-a-m*.
	- c. Future. Conj. 1, 2 *amā-b-ō*.fut.inf, *amā-v-er-ō*.fut.pfv; *monē-b-ō*, *mon-u-er-ō*; *vs.* Conj. 3, 4 *capi-a-m*, *...-e-s, ...-e-nt*, *cēp-er-ō*; *ag-a-m*, *ēg-er-ō*; *audi-a-m, audī-v-er-ō*.
	- d. Prs.subj *am<sup>0</sup> -e-m, mone-a-m*, *capi-a-m*, *audi-a-m*; Impf.subj *amā-r-e-m, monē-r-e-m*, *cap-er-e-m*, *audī-r-e-m*; Prf.subj *amā-v-er-i-m, mon-u-er-i-m*, *cēp-er-i-m*, *audī-v-er-i-m*; Plup.subj *amā-v-iss-e-m*, *mon-u-iss-e-m*, *cēp-iss-e-m*, *audī-v-iss-e-m*.
	- e. Conj. 1: *amā-re*.inf. but *am<sup>0</sup> -ō*.prs.ind.1sg, *am*<sup>0</sup> *-em*.prs.sbj.1.sg.
	- f. Prs *amā-s*.2sg, *ama-t*.3sg, *amā-mus*.1pl, *amā-tis*.2pl, *ama-nt*.3pl.
	- g. Conj. 1: *amā-re*.inf, *amā-v-ī*.prf.1.sg, *amā-t-us*.pst.pass.ptcp; Conj. 2: *monē-re*, *mon-u-ī*, *moni-t-us*; Conj. 3 *cap-e-re*, *cēp-ī*, *cap-t-us*; Conj. 4: *finī-re*, *finī-v-ī*, *finī-t-us*.
	- h. *am<sup>0</sup> -o-r*, *amā-r-is*, *amā-t-ur*, *amā-m-ur*, *amā-mini*, *ama-nt-ur*.

In the Subjunctive, Present is expressed by <sup>0</sup> *-e-* in Conj. 1, otherwise by *-a-*; Imperfect is expressed by *-r-e-* in Conj. 1, 2, 4 and by *-er-e-* in Conj. 3: phonological

### 2 Paradigms of paradigms

conditioning. Perfect is *-er-i-* and Pluperfect, *-iss-e-* (7d). In sum, Tense is largely expressed agglutinatively, but indicates Mood and Aspect: Fusion. Person and Number allomorphy is conditioned by Voice, Aspect and Tense (7e).

In the Active Participant desinences, separate plural exponents can be recognized in the final *-s* of *-mu-s*.1pl, *-ti-s*.2pl, and in the longer desinence *-nt*.3pl vs *-t-*.3sg (7f). Passive suffixes are partly fused with Participant desinences: *-or*.1sg.pass, *-r-is*.pass.2sg, *-t-ur*.3sg.pass, and *-nt-ur*.3pl.pass are agglutinative; in *-m-ur*.1pl.pass, Person and Number are cumulative; in *-mini*.pass.2pl, similarly, Voice, Person, and Number are cumulative (7g): Fusion.

In the Perfective Present (the Perfect), *-ī*.1sg is a covariant of the Infective Present and Future *-ō*.1sg and the default *-m*.1sg. The desinences *-is-tī*.2sg and *-is-ti-s*.2pl contain the Perfect suffix *-is-* ~ *-er-*; *-ēre*.3pl is cumulative: Fusion.

Phonotactic and phonological adjustments. In the Infective, Conj. 1 interfix *-ā*is truncated before vocalic endings (7e) and exemplifies a general alternation in quantity phonologically conditioned by following desinences (7f). The long interfix vowels of Conj. 2, 4 shorten in hiatus, e.g., *monē-re*.inf *mone-ō*.prs.infv.1sg, *finī-re*, *fini-ō*. Inflection within each of the aspects is fairly transparent.

Among the irregular verbs there are instances of (i) stem suppletion yielding wordforms that combine lexical and grammatical content, (ii) stems with ambiguous aspect (*pluit*.prs/prf.3sg 'rains/rained'), and (iii) a few stems with aspect or voice meaning that overrides that of their inflection, e.g., *ōd-ī* 'hate', *me-min-ī* 'remember' (Infective meaning despite Perfective form), *ūt-or* 'use', *fru-or* 'enjoy' (Active meaning despite Passive form): Introflection.

The typological paradigm of Latin verb inflection is in Table 15. In the typological paradigm (Table 15) I leave aside basic-stem formation to focus on obligatory grammatical categories. The analytic formations, the mainly agglutinative productive conjugations, the fusional Conj. 3 and the introflective suppletive verbs make for an apparent historical perspective. The unproductive patterns reflect (original aorist and perfect) formations from the distant past of the language. The productive formations may have ancient ancestors too (Sihler 2010), as may the Perfective Passive. But in the synchronic view, the transparent productive patterns of inflection and the unproductive ones form a clear reflection of the historicity of the system of conjugation.

## **2.4 Russian conjugation**

The hierarchy of obligatory verbal categories in Russian (7) (Jakobson 1956a) is similar to that of Latin:

(8) person ] number ] tense ] aspect ] mood ] voice ]

### Henning Andersen

Table 15: Latin conjugation in typological perspective



In a discussion of morphological techniques a first distinction must be made between the non-obligatory semantic categories of essentially monoaspectual

### 2 Paradigms of paradigms

"procedural" verbs<sup>2</sup> , which will not be discussed here. We will focus on the "basic" non-procedural verbs. As Imperfectives these represent states or activities, and as Perfectives they represent results of activities.

The foundation of this lexico-grammatical category is a large and open class of simplex Imperfective verbs (9a) and a few dozen simplex Perfective verbs (9b) (Isačenko 1962: 352–355, 381–385). These two groups of primary simplicia, in which each lexeme combines lexical and aspectual meaning, can be considered introflective. Secondary Perfective verbs are formed from simplex Imperfectives by prefixation (Janda et al. 2013). In (9c) perfectivization is agglutinative. Secondary Imperfective verbs are formed from both primary and secondary Perfectives by suffixation (*-a, -va, -iva*), regularly accompanied by mutation of stem vowel and/or stem-final consonant and stress displacement. In (9d) imperfectivization is fusional. It must be acknowledged that the vast majority of Russian verbs are old and replete with codified semantic extensions. But the processes of Perfectivization (9c) and Imperfectivization (9d) are perfectly productive and apply to neologisms.

	- b. Primary Perfective. E.g., *liší-t'* 'deprive', *da-t'* 'give';
	- c. Secondary Perfective. E.g., *na-pisá-t'* 'write', *s-pisá-t'* 'copy', *pod-pisá-t'* 'sign', *za-rabóta-t'* 'earn (lit.: work in)', *pro-rabóta-t'* 'study, analyze (lit.: work through)';
	- d. Secondary Imperfective. E.g. *liš-á-t'* 'deprive', *da-vá-t'* 'give', *s-pís-yva-t'* 'copy', *pod-pís-yva-t'* 'sign', *za-rabát-yva-t'* 'earn', *pro-rabát-yva-t'* 'study, analyse'.

To these can be added the deverbal (nominal) Infinitive, (adjectival) participles, and (adverbial) gerund. Their expression of Aspect and Tense (or Taxis; Jakobson 1956a) is fusional.

<sup>2</sup>R *sposoby glagol'nogo dejstvija*; Isačenko (1962: 385–418); https://russkiyyazik.ru/889/

### Henning Andersen

A key morphophonemic fact of Russian verb inflection is that Infinitive and Past endings begin with consonants, whereas Present and Imperative endings begin with vowels (in the Imperative, alternating with -∅). The endings entail different phonotactic adjustments of stems that end in a consonant and stems that end in a vowel.

There are four productive inflection classes (10a–10d). The majority of Russian verbs, which includes most secondary Perfectives, most primary, and all secondary Imperfectives (10a), have a stem in ...*a-* or ...*e-* in Infinitive and Past and a stem in ...*aj-* or ...*ej-* in Present and Imperative. There is evidence from historical morphology and from child speech that synchronically the /j/ is inserted between a stem-final vowel and a vocalic ending (cf. Andersen 1980). The /j/ epenthesis counts as a phonotactic adjustment, compatible with agglutination. The three other productive patterns are fusional: A basic vowel-final stem is modified in Present and Imperative by a suffix-allomorph replacement (10b), by truncation (10c), or by truncation and a mutation of the Prs.1sg stem-final consonant (10d). The mutation is one of several consonant mutations that reduce the number of phonological distinctions in specific derivational and inflectional environments (Andersen 1995), and which are regular in verb inflection.

	- b. *tolk-ová-t'*–*tolk-új-ut* prs.3pl 'interpret';
	- c. *mók-nu-t'*–*mók-n<sup>0</sup> -ut.*prs.3pl 'get wet';
	- d. *prosí-t'*–*proš<sup>0</sup> -ú*.prs.1sg 'ask'

There are (synchronically) underived verbs that pattern with the productive formations (10a) and (10d). In addition to these, there are some two dozen groups of additional simplex verbs with unproductive subregularities. In those, stem allomorphy indicates the grammatical content of endings, that is, they exemplify fusion. A few of them have suppletive stems, e.g., *sést'* 'sit down': *s'é-*.inf/pst, *s'ád-*.prs/impv; (*po*)*nját'* 'understand': *-n'a-*.inf/pst, *-jm-*.prs-impv.

Besides these suppletive verbs there are some that differ from the regular alternation of Infinitive–Past stem and Present–Imperative stem: *éxat'* 'ride': *jéxa–* .inf/pst, *jéd-*.prs, *pojezžáj-*.impv; *idtí* 'go, walk': *i-*.inf, *id-*.prs/impv, *šol-*.pst; *léč'* 'lie down': *l'é-*.inf, *l'ág-*.prs/impv, *l'og-*.pst; *ést* 'eat': *jé-* default, *jed'-*.prs.pl; *dát'* 'give': *da-* default, *dad-*.prs.pl, *daj-*.impv; *být'* 'be': *jést'*.prs.3sg,*sút'*.prs.3pl (bookish), *búd-*.fut.

These suppletive stems cooccur with grammatical morphemes in concatenations that can be viewed as fusional. But some of their alternant stems are limited to, and hence are indexes of, specific grammatical content. In practical terms,

### 2 Paradigms of paradigms

then, they can be considered introflective; compare the similarly ambiguous status of the English Preterite 2 verbs (Section 3, note 3).


Table 16: Russian verb inflection in typological perspective

The synchronic overview of Russian verb inflection in Table 16 only partly corresponds to the historical perspective it suggests.

The analytic Futures are quite young, first attested in the 1300s, though as a category, with different auxiliaries, the analytic Future must have originated before the 1000s (Andersen 2006). Also, the productive agglutinative verbs with /j/ epenthesis in the Present are likely younger than the unproductive fusional patterns. At the other extreme, the introflective verbs do include some of the

### Henning Andersen

oldest, and long since unproductive, formations, originally athematic (*ést'* 'eat', *dát'* 'give') and infixed presents (*léč'–ljágut* 'lie down', *sést'–sjádut* 'sit down').

But the agglutinative prefixation (serving Perfectivization) is as old as preverbs in other Indo-European languages; they were grammatized as Perfectivizers in recent prehistory. The fusional Imperfectivizing suffixation is specific to Slavic and cannot be much younger. Again, the agglutinative Russian Past tense developed as a Perfect in the Middle Ages, being regrammatized as a simple past no later than the 1200s, whereas the analytic pst.pass with participles in *-en-* and *-t*has ancient origins.

We are reminded of the English verb system, in which the fusional Preterite 2 is younger than the agglutinative Preterite 1, and the apparent historicity of the system to some extent is at odds with its known history.

## **3 Conclusion**

In this study of paradigms of paradigms I have drawn attention to notable differences between the hierarchical structures constituted by paradigms of morphological categories (Section 1.1) and the mainly flat or mixed flat–hierarchical paradigms of exponent allomorphs (Section 1.2). Such language-particular paradigms of paradigms will long occupy students of morphology, and although they are language-particular a detailed study of them can be expected to yield insights into common and perhaps universal principles that underlie their organization.

Against this background I have looked at the presumably universal paradigm of morphological techniques that defines synchronic typological variation (Section 2). I have limited the definitions of these techniques to the bare bones (Table 1); a discussion of details will be offered elsewhere.

My aim in this regard has been twofold:

First of all, I wished to highlight the fact that synchronic typological paradigms may afford an extraordinary perspective on the historicity of inflectional systems. In each of the systems sketched here the typological paradigm points up the contrast between the less restricted, productive formations and the more restricted, unproductive ones, between the systems' younger and older parts. This imprint of history by and large reflects the well-known Morphological Cycle (Hodge 1970). True enough, as emphasized in Section 2.1, a synchronic paradigm of morphological techniques is unlikely to directly reflect the historical development of the given system. For languages whose history we know it is clear that besides a main-stream development from analysis to agglutination to fusion to introflection, there are many renewals that loop back from each of the synthetic types to structurally simpler techniques (Werner 1987, Igartua 2015).

### 2 Paradigms of paradigms

Still, the perceived historicity of any typological paradigm similar to Table 12 and Tables 14–16 implies hypotheses about past developments. In any language that lacks a historical record, this perspective extends an invitation to historical linguists to uncover the actual historical past of the given language through internal reconstruction.

But more importantly, I wished to advocate for an approach to morphological description that acknowledges typological paradigms. In simple, practical terms, the paradigms of morphological techniques that can be observed in language after language show us that only an approach to morphological analysis that captures this synchronic variation can attain descriptive adequacy.

A theory of morphology that presumed all inflection to be agglutinative would be artificial and inadequate (Hockett 1954), not to say useless. A Word-and-Paradigm approach that operates with unanalysed wordforms as if all inflectional systems were introflective is no better. The recent advance into the dead end of Word-and-Paradigm theory by Stump & Finkel (2017) divides the wordforms of inflectional paradigms into stable and alternating fragments (termed "themes" and "plats") that are divorced from both lexical and grammatical content; e.g., Fr. *mouvoir* 'move': m- + -uvwar, -ø, -uv, -œv, ...; *mourir* 'die': m- + urir, -urõ, -œr, -ɔr; or *moudre* 'grind': mu- + -dr, -l, -ly...). This approach achieves descriptions that are truly meaningless.

Due attention to synchronic variation in inflectional morphology – stylistic, inflection-class, allomorphic, and typological – and to the innovations that give rise to such variation will convince the interested linguist of the priority – in the minds of speakers – of productive patterns over unproductive ones, of regular patterns over irregular ones, and of speakers' concern with ultimate elements of exponence and their correlations with elements of meaning (symbolic as well as indexical).

Since all such variation reflects the speakers' competence, it must be acknowledged in any theory of morphology, and recognized as an inomissible part of any adequate synchronic description.<sup>3</sup>

## **Abbreviations**


3 I am grateful to Lars Heltoft and Lene Schøsler for their insightful comments on an early draft of this paper.


## **References**


Bloomfield, Leonard. 1933. *Language*. New York: Henry Holt & Co.


Palmer, F. R. 1987. *The English verb*. 2nd edn. London: Longman.


### Henning Andersen

Wiemer, Björn. 2022. No paradigms without classification: How stem-derivation develops into grammatical aspect. In Gabriele Diewald & Katja Politt (eds.), *Paradigms regained: Theoretical and empirical arguments for the reassessment of the notion of paradigm*, 67–125. Berlin: Language Science Press. DOI: 10.5281/ zenodo.5675845.

## **Chapter 3**

## **Formalizing paradigms in Construction Grammar**

## Jaakko Leino

University of Helsinki

Construction Grammar sees the language system as consisting solely of conventionalized pairings of form and meaning, i.e. constructions. Constructions may be of any size and complexity, and they may be abstract (or schematic) to any degree. They may be templates for sentences, lexical items, inflectional morphemes, discourse patterns (Östman 2005) that organize whole texts or even genres, etc. However, the notion of constructions seems incapable of capturing patterns found within the grammar: systematic similarities between constructions and, notably, paradigms of different sorts. For instance, inflection paradigms consist of sets of constructions, but nothing in common varieties of Construction Grammar explains how those constructions join together to form a paradigm. The paper argues that in addition to constructions, the language system must also include specifiable relations which hold between the constructions of a language and which organize them into a functional system. Crucially, such relations are necessary for the organization of paradigms, be they of morphological, syntactic or other nature. Relations between constructions within the grammar can be – and have previously been – described in terms of inheritance (e.g. Goldberg 1995), taxonomic and meronomic links (Croft 2001), and the like. However, such very abstract links can only capture simple relations between constructions. Yet, more complex relations, notably of an analogical nature, exist widely within the grammar of apparently all human languages. To capture such analogical relations, the paper uses the notion of *metaconstruction*, briefly introduced in Leino & Östman (2005). Metaconstructions may be thought of as generalizations of constructions, partly in the same sense as constructions may be seen as generalizations of actual expressions. It will be argued that such analogical relations, formalizable as metaconstructions, hold paradigms together and also facilitate both producing and interpreting complex expressions.

Jaakko Leino. 2022. Formalizing paradigms in Construction Grammar. In Gabriele Diewald & Katja Politt (eds.), *Paradigms regained: Theoretical and empirical arguments for the reassessment of the notion of paradigm*, 37–65. Berlin: Language Science Press. DOI: 10.5281/zenodo.5675843

### Jaakko Leino

## **1 Introduction**

Construction Grammar sees the language system as consisting solely of conventional pairings of form and meaning, i.e. constructions. Constructions may be of any size and complexity, and they may be abstract (or schematic) to any degree. However, the notion of construction seems incapable to capture patterns found within the grammar: systematic similarities between constructions and, notably, paradigms of different sorts.

Besides holding paradigms together, analogical relations also facilitate both the production and interpretation of complex expressions and serve as a notable source of linguistic creativity and innovation. Systematic analogical structures often both show existing gaps in the language system and provide means of coining novel but instantly comprehensible ways of filling such gaps. The mechanism is ubiquitous in language, but it seems to lead to (conceived) paradigms only in certain parts of grammar. This, in turn, may be revelatory of the nature of paradigms.

If one takes as a starting point the claim, often made in Construction Grammar (e.g. Fillmore & Kay 1995: 1·15–16, Goldberg 1995: 1–5, Croft 2005: 273–275), that grammar consists of constructions, an obvious question arises: How are the constructions of a given language organized? Is the grammar of a given language merely a "warehouse" or an inventory of constructions, from which a language user picks out whatever is necessary to produce an utterance?

Cognitive Grammar, a close relative of Construction Grammar,<sup>1</sup> conceptualizes grammar as a *structured inventory of linguistic units* (Langacker 1987: 73). In this view, grammar is organized mainly in terms of *categorization:* the inventory of linguistic units is structured into schematic networks. The relation of *symbolization* also structures the inventory by establishing correspondencies between particular semantic structures and phonological structures: a symbolic relation is necessarily present in every linguistic unit of a language, as these units are taken to be inherently bipolar, i.e. to represent a conventionalized correspondence of form and meaning. (For a more detailed discussion, see Langacker 1987: 73–76.)

Cognitive Grammar greatly resembles Construction Grammar in many important aspects (cf. Leino 2005a, Croft 2001: 6–7), and Langacker's characterization of grammar could therefore conceivably be rephrased as a *structured inventory of grammatical constructions.* And, indeed, Tomasello (2006: 258) does so (with regard to "Cognitive-Functional Linguistics" in general, but explicitly including

<sup>1</sup> Some, e.g. Goldberg (2006), go as far as to consider Cognitive Grammar a variant of Construction Grammar.

### 3 Formalizing paradigms in Construction Grammar

Construction Grammar in this notion): "In this approach, mature linguistic competence is conceived as a structured inventory of meaningful linguistic constructions – including both the more regular and the more idiomatic structures in a given language (and all structures in between)."

Thus, besides constructions, the language system must include specifiable relations which hold between the constructions of a language and which organize them into a functional system. Such relations have been described in terms of inheritance (e.g. Goldberg 1995), or taxonomic and meronomic links (e.g. Croft 2001). However, such very abstract links can only capture simple relations between constructions. More complex relations, notably analogical ones, exist widely within the grammar of apparently all human languages. Analogical relations may be captured by the notion of metaconstruction (briefly introduced in Leino & Östman 2005).

It is not clear, however, how either Cognitive Grammar or Construction Grammar would express relations between expression types which show obvious similarities but which cannot be said to be in a schematic relation to each other. For example, English active and passive sentences are obviously related expression types, but neither is schematic with regard to the other. Similarly, English assertive sentences and questions are related in a very similar manner, but this relationship cannot be captured in terms of categorization or schematicity either:

### (1) English

a. John built the house.

b. The house was built by John.

	- a. Lisa has met my wife
	- b. Has Lisa met my wife?

In both cases, it is clear that the (a) and (b) sentences are related to each other. This relatedness is not incidental but systematic: it is not only the sentences that are related but the sentence *types,* i.e. the constructions – or, in Cognitive Grammar terms, constructional schemas – behind the sentences that are related. However, this relatedness cannot be captured as an organizing feature of the grammatical system with the tools provided by Cognitive Grammar.<sup>2</sup>

<sup>2</sup>This is, of course, not to say that Cognitive Grammar is unable to analyze these sentences or even address the essential similarities between them. The problem that I wish to point out concerns the internal organization of the grammar and the lack of tools in both Cognitive Grammar and Construction Grammar to describe this internal organization in sufficient detail and systematicity.

### Jaakko Leino

Essentially the same situation holds for Construction Grammar as well. Relations between constructions within the grammar can be – and have been – described in terms of inheritance links (e.g. Goldberg 1995: 73–81, Michaelis & Lambrecht 1996: 235–245, Croft 2001: 53–57). However, such very abstract links only capture certain rather simple relations between constructions in terms of what is made of what and what is a part of what. In addition to the aforementioned inheritance links, a complementary mechanism should be taken into use to capture such relations which cannot be described by means of simple networks of schematicity and part–whole relations.

Recently Diessel (2019) has presented a greatly improved way of representing inter-constructional relations within the grammar. His book specifically devotes to the analysis of grammar as a complex network of interconnected constructions, and greatly improve our understanding of that challenging topic. He also describes (e.g. pp. 18–19 & chap. 11) paradigms in terms of emerging networks of constructions, a view which I shall also adopt in the following. Yet, I feel, there is still more to be said about different types of relations between constructions, and the complex architecture of grammar.

As we saw above, there are other kinds of similarities between the constructions of a language as well. Notably, relationships of an *analogical* nature exist widely within the grammar of apparently all human languages. If the theoretical machinery that we use isn't sufficient for capturing these similarities, then the description of the language in question will miss possible generalizations, and will thereby not conform to the requirement of full coverage spelled out by Kay (1995).

To capture such generalizations, I shall make use of the notion of *metaconstruction,* briefly introduced in Leino (2003) and Leino & Östman (2005). Metaconstructions may be thought of as generalizations of constructions, partly in the same sense as constructions may be seen as generalizations of actual expressions. Any given construction may be related to other constructions in the language by means of such metaconstructions. Ultimately, the language system will not appear as an unstructured list of constructions, but rather as a structured system in which a certain kind of order prevails.

One notable point of relevance for metaconstructions are paradigms of different sorts within the language system. For the purposes of this paper, a paradigm is seen as a set of constructions which has the following properties:

1. The members of the paradigm, i.e. the constructions which make up the paradigm, are alternatives to one another in a given linguistic context.

### 3 Formalizing paradigms in Construction Grammar


Metaconstructions have a notable role not only in the internal organization of grammar but also in the production of novel types of utterances. In other words, they do more than merely organize the system in a synchronic sense statically: they also serve as dynamic and diachronically relevant instructions of how to form new constructions. More specifically, metaconstructions capture analogical relationships, which have been shown to be of great importance for the internal organization of the language system and the creative use of constructions (e.g. Leino 2003: 260–284), as well as syntactic creativity in general (e.g. Tabor 1994: 202–205) – not to mention the importance of analogy in language and linguistic description more generally (cf. e.g. Bloomfield 1933 for a linguistic classic which emphasizes the importance of analogy, and Anttila 1977 and Itkonen 2005 for more general accounts of analogy in language).

## **2 Some cases in point**

In what follows, I shall discuss two cases where analogical relations between constructions have a crucial role in the functioning of the language in question. The first case (also presented in Leino & Östman 2005), in Section 2.1, concerns an arising variation in Finnish subject and object case marking. The second one, in Section 2.2, is the relationship between assertions and questions referred to above, but in the context of the Swedish language. In addition, these sentence types will be discussed with regard to active and passive voice.

## **2.1 Metaconstructions and Finnish case marking**

Finnish subject and object case marking provides a clear-cut example of how analogical relations between constructions affect the functioning of the language system. Briefly stated, Finnish is an accusative language, but it also has a peculiar ergative-like subsystem marked with the partitive case and related to the boundedness of the object, as well as the resultativity – or, more precisely, telicity – of the activity denoted by the predicate (for details, see e.g. P. Leino 1991, Karlsson 1999, Heinämäki 1984).

### Jaakko Leino

For the purposes of this paper, we may state the following simplified rule of thumb: an object in Finnish is marked with the accusative case if the sentence is not negated and the activity denoted by the verb is telic, i.e. if the activity is carried out completely. If the sentence is negated or the activity is atelic, the object is marked with the partitive case.

To further complicate the Finnish case marking system, the subject – which is normally in the nominative case – may also be marked with the partitive case. However, this is traditionally said to require that the following conditions be met:


In other words, transitive sentences in Finnish cannot, traditionally speaking, have a partitive subject. However, transitive sentences with partitive subjects do in fact show up sporadically, although normative grammars do not allow them and the vast majority of Finnish speakers find them ungrammatical, or awkward at best.

In terms of Construction Grammar, the Finnish language may be said to have a transitive sentence construction which licenses sentences like those in (3):<sup>3</sup>

	- a. Lapset child-pl.nom rikkoivat break-pst.3pl ikkunan. window-acc 'The children broke the/a window.'
	- b. Miehet man-pl.nom kaatoivat cut-down-pst.3pl puut. tree-pl.acc 'The men cut down the trees.'

This construction may be characterized schematically as in Figure 1 or, in the traditional boxes-within-boxes notation as in Figure 2.

In other words, this construction licenses transitive sentences with a nominative case subject and an accusative case object.

The Finnish language also has an intransitive sentence construction which licenses sentences like the following:

<sup>3</sup> Finnish has a rich case inflection morphology. A list of abbreviations which indicate morphological case is printed at the end of this chapter. Ablative, allative, essive and comitative do not occur in the examples.

### 3 Formalizing paradigms in Construction Grammar

```
[Snom V Oacc]
```
Figure 1: Transitive sentence

Figure 2: Transitive sentence, box notation

	- a. Lapset child-pl.nom leikkivät play-3pl pihalla. yard-ade 'The children are playing in the yard.'
	- b. Puut tree-pl.nom kaatuivat fall-down-past.3pl myrskyssä. storm-ine 'The trees fell down in the storm.'

Example (5a) is a basic intransitive sentence with a nominative subject, whereas (5b) is an intransitive sentence with a partitive subject:

(5) Finnish


The similarity between these two sentence types is all the greater due to the fact that, while Finnish allegedly has "free" word order (i.e. one that mostly expresses information structure rather than grammatical relations, see e.g. Vilkuna 1989), in a neutral context the partitive subject of the intransitive sentence follows the verb like the partitive object of the transitive sentence.

### Jaakko Leino

The constructions which license examples (4a) and (4b) are connected together by a metaconstruction which may be characterized by Figure 3.

$$\{\left[\mathbf{S}\_{\text{NOM}} \lor \mathbf{X}\right] \leftrightarrow \left[\mathbf{X} \lor \mathbf{S}\_{\text{PAR}}\right]\}$$

Figure 3: Metaconstruction between nominative and partitive subject

In Figure 3, *S* stands for the subject, nom and par for the nominative and partitive case, *V* for the predicate verb, and *X* for a potential other argument. The same information may be expressed in the boxes-within-boxes notation (greatly simplified) as in Figure 4.

Figure 4: Metaconstruction between nominative and partitive subject, box notation

However, the Finnish language does not have a construction that is connected with this metaconstruction to the construction exemplified by the sentences (3a– b); i.e., as pointed out above, transitive sentences in Finnish cannot have partitive subjects. Yet, it is very easy to note, on the basis of that construction and this metaconstruction, that such a construction would have the form expressed in Figures 5 and 6.

\* [Oacc V Spar]

Figure 5: Transitive sentence with partitive subject

Figure 6: Transitive sentence with partitive subject, box notation

This construction would license such sentences as those in (6a–b):

(6) Finnish


### 3 Formalizing paradigms in Construction Grammar

In (6), the sentences have been marked as ungrammatical. However, a language user might well wish to express a transitive event with an unbounded subject – which would rather naturally be coded as a transitive sentence with a partitive subject, if only this were grammatical in Finnish. In other words, the construction sketched out in Figures 5 and 6 would be a natural tool for expressing such a meaning.

Since the language already has, so to speak, all the "ingredients" for such a construction, it would not be difficult to coin such a construction and start using it. And, in fact, this is happening in the Finnish language at the moment. Sentences like in (6) do in fact show up not only in colloquial language but also in newspaper headlines and practically all registers of the Finnish language, though only very sporadically:

	- a. \* Tuhansia thousand-pl.par Soneran Sonera-gen piensijoittajia minor.investor-pl.par jätti leave-past-3sg käyttämättä use-inf3-abe merkintäoikeutensa right.to.subscribe.for.shares-acc.ps3sg/pl Soneran Sonera-gen annissa.

rights.offering-ine

'Thousands of Sonera's minor investors left their share subscription right unused in the Sonera stock rights offering.' (*Helsingin Sanomat,* 11/24/2001)

b. Minkä what-gen maan country-gen jalkapalloilijoita footboller-pl.par haki apply-past.3sg viime last viikolla week turvapaikkaa asylum-par Suomesta? Finland-ela

'What country where the football players from who sought asylum in Finland last week?' (*Uutislehti 100* 8/25/2003)

In other words, what is happening in the Finnish language in this respect is essentially that the existing constructions, and generalizations based on them, are coupled in such a way that a new construction is taken into use. Actually, this is a rather ordinary case of *analogy*, and the metaconstruction I sketched out serves as an *analogy model* here.

Metaconstructions may thus have a role in diachronic change in that they motivate new constructions through several existing constructions and their systematic similarities and differences. However, two points of clarification are in place with regard to the role of metaconstructions in this process. First, they are not

### Jaakko Leino

the direct cause of the change: the emergence of a new grammatical construction stems primarily either from the need to express a new kind of meaning or from the tendency towards systematicity and simplicity in grammar. Secondly, metaconstructions do not serve as the goal, or the target structure, in such a change. Rather, they provide a structured analogy which serves to motivate (perhaps initially single ad hoc utterances which may then give rise to) the target construction.

## **2.2 Assertions, questions, voice, and metaconstructions**

As we saw with example (2) at the beginning of this paper, there is an obvious similarity between assertive sentences and questions in English. Swedish shows a very similar relationship between assertions and questions. This relationship is systematic rather than incidental; that is, the same similarity holds for each question and a corresponding assertion. Therefore, it is plausible to say that assertions and questions as sentence types, i.e. constructions, are related in some manner.

More generally, not only different kinds of interrogative sentences but sentence types in general form a wide network of different but interrelated constructions. This network includes assertive sentences, several types of question sentences, and a number of other sentence types as well. In fact, at a yet more general level, the entire grammar of a language may be represented as a network of interrelated constructions (perhaps much in the same manner as suggested in Diessel 2019) made up of individual constructions and relationships between them which organize the network. Often, as in the case of sentence types, there are parts of the network which may be seen as "subsystems" or, indeed, paradigms.

For the sake of clarity, I shall only refer to yes/no questions here. Of course, the discussion here holds (mutatis mutandis) for other types of questions as well, provided that we take each question type to be a separate construction.

Let us consider the following pairs of sentences:

	- a. Du you läste read-pst boken. book-def 'You read the book.'

b. Läste read-pst du you boken? book-def 'Did you read the book?'

	- a. Kalle Charlie har have ätit eat-pcp soppan. soup-def 'Charlie has eaten the soup.'

### 3 Formalizing paradigms in Construction Grammar

b. Har Have Kalle Charlie ätit eat-pcp soppan? soup-def 'Has Charlie eaten the soup?'

In ordinary terms, forming a yes/no question involves subject and object inversion (or, differently stated, verb-initial word order). This may be stated very simply with the following metaconstruction.

[[S V X] ↔ [V S X]]

Figure 7: Metaconstruction between assertive sentence and yes/no question in Swedish

Figure 7 only shows a very schematic structural association of the associated constructions (i.e. the assertive sentence construction and the yes/no question construction). A more detailed description of this metaconstruction would, of course, include information on the discourse functions of these constructions, on more specific structural properties of the constructions, etc.<sup>4</sup>

To take a broader perspective, and to further illustrate the role of metaconstructions in organizing the grammatical system, let us add to this the active vs. passive alternation. In Swedish, the regular way of forming a passive is adding the suffix *-s* to the main verb and marking its object argument as the grammatical subject:

(10) Swedish


soup-def have eat-pst.pass 'The soup has been eaten.'

b. Har Have soppan soup-def ätits? eat-pst.pass 'Has the soup been eaten?'

<sup>4</sup>There are no a priori limits to what, and how much, information a metaconstruction may include, just as there are no such limits for the information content of a construction. Constructions, as usage-based generalizations of expressions, may include any amount of observed and generalized linguistic information. Similarly, metaconstructions may, at least in principle, include any amount of information relevant to the constructions that they relate to one another and the relationship between those constructions. In practice, however, the core of a metaconstruction is a relatively simple analogical relation, and the rest of the information included in it is a selection of features of the constructions involved. Even so, of course, metaconstructions tend to be rather complex knowledge structures.

### Jaakko Leino

Thus, the relationship between active and passive in Swedish may be characterized as the following metaconstruction.<sup>5</sup>

$$\left[ \left[ \text{S}\_{\text{S}} \text{ V O}\_{\text{O}} \right] \leftrightarrow \left[ \text{S}\_{\text{O}} \text{ V-s} \right] \right]$$

Figure 8: Metaconstruction between active and passive sentences in Swedish

We may relate the metaconstructions shown in Figures 7 and 8, and thereby present a limited subsystem of the Swedish grammar organized by these metaconstructions, in the following manner.

Figure 9: Assertive sentences, yes/no questions and active and passive voice as a subsystem in Swedish

In Figure 9, there are two instances of both of the metaconstructions shown in Figures 7 and 8. The one shown in Figure 7, i.e. [[S V X] ↔ [V S X]], connects the sentence types [S<sup>S</sup> V OO] (i.e. active assertive sentence) and [V S<sup>S</sup> OO] (active yes/no question) on the one hand, and the sentence types [S<sup>O</sup> V-*s*] (passive assertive sentence) and [V-*s* SO] (passive yes/no question) on the other. Correspondingly, the metaconstruction shown in Figure 8, i.e. [[S<sup>S</sup> V OO] ↔ [S<sup>O</sup> V-*s*]], connects the sentence types [S<sup>S</sup> V OO] (active assertive sentence) and [S<sup>O</sup> V-*s*] (passive assertive sentence) on the one hand, and [V S<sup>S</sup> OO] (active yes/no question) and [V-*s* SO] (passive yes/no question) on the other.

This example shows a notably different aspect of metaconstructions than that discussed in Section 2. While the example of Finnish subject and object case marking was a case of diachronic change taking place, and metaconstructions serving as a vehicle of such change, the case of the Swedish sentence types is purely synchronical. The synchronic role of metaconstructions may be argued to include

<sup>5</sup> In Figure 8, the notion S<sup>S</sup> stands for a grammatical subject which expresses the subject argument of the verb, S<sup>O</sup> for a grammatical subject which expresses the object argument, etc.

### 3 Formalizing paradigms in Construction Grammar

such tasks as finding the right construction for a given discourse function, indicating correspondencies between different expression types, and the like. The synchronic and diachronic aspects of metaconstructions will be further discussed in Sections 4 and 5. It should be noted, however, that metaconstructions do have both aspects to them: they serve both as synchronic devices which organize the grammar of a language and as diachronic analogy models for re-organizing the grammar.

## **3 Some related theoretical notions**

The notion of metaconstruction is quite obviously related to other previously suggested notions. In the following, I shall point out some similarities between metaconstructions and Kay's *patterns of coining,* on the one hand, and Chomskyan transformations, on the other.

## **3.1 Metaconstructions and patterns of coining**

The concept of metaconstruction, in particular in its use as a basis for novel expressions and expression types, shows great resemblace to Kay's *patterns of coining* (Kay 2013).<sup>6</sup> Kay's first example is the word *underwhelm,* which is the result of analogy along the following lines:

(12) English (Kay 2013: 33)

*over : overwhelm :: under : underwhelm*

Above, I have not extended the concept of metaconstruction to morphological phenomena, simply because Construction Grammar does not yet have conventionalized ways of representing morphology and morphological phenomena. However, we may rather comprehensibly – albeit pre-theoretically – represent (12) in terms of metaconstructions as follows.

[[P] ↔ [P*whelm*]]

Figure 10: A metaconstructional account of underwhelm

What the metaconstruction in Figure 10 states is essentially that there is a relationship between the combination of a preposition and another word which

<sup>6</sup>Kay, in fact, attributes the notion to Charles Fillmore. Apparently, Fillmore has presented the notion in a lecture, the text of which is available online (Fillmore 1997). However, since Fillmore does not elaborate on the notion, whereas Kay does, it seems justified to refer to the notion as *Kay's* patterns of coining.

### Jaakko Leino

consists of that preposition and the affix (or affix-like element) *-whelm.* The pair of words related to one another by this metaconstruction share semantic features in a systematic way. In other words, for the two pairs of words, *over & overwhelm* and *under & underwhelm,* the metaconstruction essentially expresses the same information as Kay's proportional analogy shown in example (12).

The metaconstruction in Figure 10 obviously allows for such hypothetical words as *upwhelm, downwhelm, throughwhelm, inwhelm, outwhelm* etc. This could be used as an argument against this formulation, claiming that the metaconstruction in Figure 10 *overgenerates* such expression. However, as I shall discuss below, metaconstructions are not intended as generative entities. Rather, they express observed analogies. Thus, the metaconstruction in Figure 10 does not state that we should expect such words as *upwhelm* and *inwhelm*. What it does state is that if we were to encounter such words, then *upwhelm* would be to *overwhelm* what *up* is to *over*; i.e. that the relation between *upwhelm* and *overwhelm* is analogous to that between *up* and *over*.

The same goes, mutatis mutandis, for other patterns of coining discussed by Kay as well. While metaconstructions merely capture similarities between observed expressions and, notably, *types* of expressions, patterns of coining are used (as the name implies) to coin new expressions. In other words, patterns of coining are *productive* to some extent, whereas metaconstructions are not (except for some rare cases such as the one discussed in Section 2.1).

This point leads us to an interesting continuum from metaconstructions via patterns of coining to constructions. According to Kay (2013: 38), the formula *A as NP* 'extremely A' is not a construction but, rather, a pattern of coining. He states two reasons for this (ibid.):

First, knowledge of formula (12) [A as NP 'extremely A'] plus knowledge of the constituent words is not sufficient [to license examples of this formula]. If a young, foreign or sheltered speaker of English knew what *easy* meant, and knew what *pie, duck,* and *soup* meant and knew all the expressions in [Kay's examples] plus many more built on the same pattern, they would still not know that *easy as pie* and *easy as a duck soup* are ways of saying *very easy*. Secondly, one can't freely use the pattern to coin new expressions.

The central point of Kay's argument – and, indeed, his whole paper – is that patterns of coining are less productive than grammatical constructions. As noted above, metaconstructions are less productive than patterns of coining. Thus, we may think of these three as a cline from less to more productive generalizations.

### 3 Formalizing paradigms in Construction Grammar


Figure 11: Constructions, patterns of coining, and metaconstructions on a productivity scale

While this is by no means the only difference between these three theoretical notions (notably, metaconstructions are generalizations of expression *types,* or more accurately of relations which hold between them, while the other two are generalizations of actual expressions), this aspect of the notions is useful in pinning down the essential nature of each of these notions. And, as we saw in the case of *underwhelm,* metaconstructions can actually be postulated as generalizations of actual words and utterances as well.

## **3.2 Metaconstructions and transformations**

A rather different way to conceive of the notion of metaconstruction has to do with a somewhat different branch of linguistics. As the observant reader may well have noticed, such metaconstructions resemble, to a great degree, the *grammatical transformations* used in the tradition started by Chomsky (1957). And indeed, such a tool might well be used to revitalize the transformational school of thought. For example, Chomsky's (1957: 43) example:

If S1 is a grammatical sentence of the form *NP1 – Aux – V – NP2,* then the corresponding string of the form *NP2 – Aux + be + en – V – by + NP1* is also a grammatical sentence

can be expressed, with a notation which greatly resembles the characterizations of metaconstructions, in the following form:

[[NP1 – Aux – V – NP2] ↔ [NP2 – Aux + be + en – V – by + NP1]]

However, this is by no means what metaconstructions are intended to do. The nature of metaconstructions is deeply different from that of transformations. Although the differences between metaconstructions and transformations may not appear to be as obvious as the similarities, they are all the more noteworthy from a theoretical perspective.

### Jaakko Leino

First of all, it is clear that metaconstructions are not nearly as productive as transformations. Metaconstructions are generalizations which a language user may or may not make, and their central function is to keep up analogical relationships among different sets of constructions.

Secondly, metaconstructions are not used to turn linguistic material into some other linguistic material, or deep structure into surface structure, the way that transformations are: Metaconstructions are not generative in nature. Where transformations may be said to describe alternations, metaconstructions describe correspondencies. In the construction grammar world view, no material *changes* into other material; rather, everything is described in terms of correspondencies and compatibility.<sup>7</sup> In accordance with this tradition, metaconstructions do not involve transforming an expression into some more suitable expression.

And thirdly, metaconstructions have, as we saw above, a more or less diachronic nature. They are generalizations over *types* of expressions, not over actual expressions. If they do have a generative nature, that nature must be diachronic in the sense that metaconstructions are used to create new types of expressions, new constructions, rather than just new expressions.

The synchronic vs. diachronic nature of new constructions being conventionalized, and the role of metaconstructions in that, naturally is a broad topic and falls outside of the scope of this paper. As a general rule, however, I see the conventionalization of new constructions as a diachronic process (cf. e.g. Rostila 2006), and certainly more so than the process of generating utterances based on already existing constructions.

## **4 Metaconstructions and paradigms**

The examples in Section 2 show that metaconstructions capture both static and dynamic relations between groups of constructions. They both show systematic groups of constructions within the grammar and may even serve as patterns for coining new constructions. Crucially with regard to the theme of the present volume, they may also be used to capture paradigmatic relations, unlike any other device yet postulated for Construction Grammar that I am aware of. In what follows, I present an analysis of the person inflection paradigm in Finnish as exemplified in different contexts, or groups of constructions.

<sup>7</sup>Cf. Kay (1995) and the *non-derivational* and *usage-based* properties of Construction Grammar. While metaconstructions may sometimes be used to coin new expressions or expression types, as shown in Section 2.1, they are not part of the standard mechanism of generating sentences in the same manner as transformations are (or were) in transformational grammar.

### 3 Formalizing paradigms in Construction Grammar

## **4.1 Finnish personal pronouns and the verb inflection paradigm**

The Finnish person paradigm rather straightforwardly consists of the set of 1st , 2 nd, and 3rd person in singular vs. plural. This paradigm is found both in verb inflection and in personal pronouns (Table 1).


Table 1: Finnish person paradigm

A local peculiarity of Finnish, so to speak, is the form conventionally known as the passive, perhaps more accurately an impersonal form. It is sometimes referred to as "the 4th person" (originally by Tuomikoski 1971) to reflect the fact that it is effectively a part of the person inflection paradigm, despite the fact that it is used, to some extent, in constructions distinct from the ones used for the active forms. It does, however, resemble the active forms in usage significantly more than e.g. most Indo-European passives. For instance, it can be formed of practically all Finnish verbs, including intransitive verbs and even the copula.

While the passive morphologically has, in addition to the voice marker, a person suffix of its own, there is no separate personal pronoun for the passive. This is naturally in accordance with its primary use for agent demotion or impersonalization. Therefore, it might be argued that the passive is not a full-fledged "4th person" but rather a set of constructions which are used in contexts where the person inflection paradigm is not relevant. As will become apparent in the following sections, personal pronouns are not the only context to speak in favor of such a view. On the other hand, as will also become apparent, the finite verb inflection paradigm is by no means the only context to speak in favor of the opposing view, i.e. of interpreting the passive as a part of the person inflection paradigm.

### Jaakko Leino

The precise nature of the Finnish passive need not concern us here. The interested reader is referred to sources like Shore (1988), Manninen & Nelson (2004) and Helasvuo (2006) for further details. For the purposes of this paper, the crucial thing is that the passive is a conventional part of the person paradigm but not present in all of its manifestations.

## **4.2 Possessive suffixes**

Possessive relations in Finnish are typically marked, redundantly, with two devices simultaneously: the genitive form of a personal pronoun on the one hand, and a possessive suffix on the other.


Table 2: Finnish possessive paradigm

As shown above, the passive is left out of the paradigm again. Not only has it no personal pronoun of its own, but it also has no corresponding possessive suffix. What exactly a "passive ownership" would mean is left an open question here. For the sake of the argument, we may assume that a bare noun with no possessive marker is the closest match to "impersonal ownership". As will become apparent, however, there are other contexts in which the possessive suffixes are used where a passive possessive suffix would be useful to complete the paradigm. As is also shown above, the possessive suffix paradigm does not distinguish between 3rd person singular and plural. While this is an interesting observation per se, it need not concern us here.

It seems perfectly natural that the same distinctions among 1st, 2nd, and 3rd person and singular vs. plural are found both in verb inflection and in nominal possessive marking. It seems equally natural that the same personal pronouns occur with verbs inflected for person and nouns marked for possession. And yet, there is no a priori reason why this should be the case, and, more importantly for

### 3 Formalizing paradigms in Construction Grammar

present purposes, there is nothing in the architecture of Construction Grammar that could capture, let alone explain, the fact that it is the case.

While there has, so far, been no notable attempt at capturing Finnish morphology in a construction-based formalism, it is obviously possible to represent the morphological structures as constructions, e.g. with the formalism proposed in Booij (2010). Such a formalization is beyond the scope of this paper, but for the present case the crucial point is that the paradigms shown above can be represented as organized groups of constructions. Furthermore, it is important to observe that the organization of the group of constructions discussed in this section (possessive suffixes) is essentially identical to the organization of the groups of constructions discussed in the previous section (personal pronouns and verb inflection).

Given the discussion in Section 2 of this paper, it seems justified to claim that it is possible to capture the systematic, and, in fact, analogous organization of these groups of constructions in terms of a (somewhat complex) metaconstruction. And, as will become apparent in following sections, the same metaconstruction is also necessary for capturing and explaining other phenomena in the Finnish grammar.

## **4.3 The Finnish infinitive system**

In Finnish, as in languages more generally, person inflection is essentially a property of finite verb forms. However, as will be shown in the following sections, Finnish infinitives also show features at least reminiscent of person inflection in some contexts. In order to properly understand that phenomenon, a brief introduction of the complex of Finnish infinitive forms is in place.

According to the traditional view, predominant since the 19th Century, Finnish is said, on morphological grounds, to have either four or five distinct infinitives, each of which has a different morphological marker. Each of the infinitives shows some case inflection, but none of them has a full case inflection paradigm. The forms are traditionally referred to with numbers, but since Hakulinen et al. (2004), they are more commonly referred to by their morphological marker. Hakulinen et al. only treat the first three as true infinitives for reasons that need not concern us here. The forms are briefly introduced in the following.

	- i. morphological marker -*TA*<sup>8</sup> (i.e. *-a, -tä, -da, -dä,* and assimilated variants *-lA, -rA*, etc.)

<sup>8</sup>The vowel quality in Finnish affixes is dependent on vowel harmony. In front vowel contexts, the archephoneme /A/ is realized as the frontal vowel *ä* [a], and in back vowel contexts as the

### Jaakko Leino

	- i. morphological marker *-den, -ten*
	- ii. inessive: teh-de-ssä (do-inf2.ine)
	- iii. instructive: teh-de-n (do-inf2.ins)
	- i. morphological marker -*mA* (-ma, -mä)
	- ii. inessive: *teke-mä-ssä* (do-inf3.ine)
	- iii. elative: *teke-mä-stä* (do-inf3.ela)
	- iv. illative: *teke-mä-än* (do-inf3.ill)
	- v. adessive: *teke-mä-llä* (do-inf3.ade)
	- vi. instructive: *teke-mä-n* (do-inf3.ins)
	- vii. abessive: *teke-mä-ttä* (do-inf3.abe)
	- i. morphological marker *-minen, -mis-*
	- ii. nominative: *teke-minen* (do-inf4)
	- iii. partitive: *teke-mis-tä* (do-inf4.par)
	- i. morphological marker –-maisi-, -mäisi-
	- ii. adessive: *tekemäisillään* (do-inf5.ade.3sg/pl)

In other words, according to the traditional view, Finnish has several separate infinitives, each of which has a defective case inflection paradigm. A radically different view, originally presented already by Lönnrot (1841: 44) and rediscovered by Siro (1964), treats the different forms as variants of the same infinitive, with the infinitive marker varying in different case forms. As Siro points out, the different infinitive markers are in fact nearly in a complementary distribution with regard to case inflection.

back vowel [ɑ]. Similarly, the archephoneme /O/ is realized either as the front vowel ö [ø] or the back vowel o [o] depending on the phonemic context. To further confuse the uninitiated reader, the verb stem varies according to consonant gradation (which also differentiates between the realization of the /T/ of the infinitive marker as either [t] or [d]) and other sound changes triggered by the following affix. Fortunately, for the purposes of this paper, these peculiarities are beside the point.

### 3 Formalizing paradigms in Construction Grammar


Table 3: Finnish infinitive forms sorted by morphological case

*a* (or basic form)

To complete the complementary distribution, each of the forms is only used in a limited set of non-finite expression types, or constructions, and those few instances where there are two different infinitive forms corresponding to the same case form (nominative, inessive, adessive, and instructive), the two forms are never mutually interchangeable in any of the constructions in which they are used.

## **4.4 Finnish infinitives with possessive suffixes**

Infinitives are often described as verb forms which resemble nouns. This characterization typically refers to their syntactic behavior, but in Finnish it is true also of their morphology. As was shown in Section 4.3, Finnish infinitives are inflected for case like nouns. In addition, some of them also take the possessive suffix in some contexts – the translative form of the 1st infinitive and the adessive form of the 5th infinitive in fact never occur without a possessive suffix.

Not all Finnish infinitive forms occur with possessive suffixes however. Of the thirteen different infinitive forms (combinations of an infinitive marker and a case ending) listed in Table 3, only five occur with possessive suffixes.

In all of the occurrences of these forms, the possessive suffix corresponds to the actor of the event denoted by the infinitive. In other words, for all intents

### Jaakko Leino


Table 4: Finnish infinitive forms with possessive suffixes

and purposes, the possessive suffix in these forms semantically corresponds to person inflection. Yet, the morphemes used for that purpose in the infinitives are not those used as person affixes in finite verb forms but rather those used for marking possession with nouns as shown in Section 4.2.

## **4.5 Passive infinitive forms**

As was pointed out in Section 4.1, the Finnish passive is effectively a part of the person inflection paradigm. Given that, and the observation made in Section 4.4 that possessive suffixes in infinitives resemble person inflection, the question arises whether there are passive forms to complete the person paradigm in those infinitive forms which do take the possessive suffix. And, indeed, there are passive infinitive forms in Finnish, even though they are few in terms of both types and tokens. In fact, there are only three passive infinitive forms in Finnish.

As has been pointed out by Leino (2005b), the following observation holds in the Finnish infinitive system: if a given infinitive (i.e. a combination of one of the four or five infinitive markers and a specific case suffix) has a passive variant, then it also has a variant with a possessive suffix. In other words, there are no


Table 5: Finnish passive infinitive forms

passive infinitive forms with no corresponding possessive suffixed forms. This seems to suggest that the few existing passive infinitive forms are motivated by the forms with the possessive suffixes, and have arisen in order to complete what looks like their person inflection paradigm.

## **4.6 Emerging non-finite person inflection?**

As an interim summary, Finnish shows the same paradigm of 1st, 2nd, and 3rd person in singular vs. plural in a number of different contexts, including personal pronouns, finite verb inflection, possession marking on nouns, and the usage of possessive suffixes on infinitives, closely resembling person inflection. Furthermore, the paradigm is supplemented in several, but not all, instances by the passive. Thus, what we observe is a paradigm consisting of seven parts, organized in terms of voice, person, and number.

It seems rather obvious that some of the instances of this pattern, or paradigm, are more fundamental than others. Notably, the apparent person inflection of infinitives seems secondary in comparison to both its clear model, finite verb inflection, and the much more common use of the possessive suffixes as markers of possession. This impression is backed up by the fact that other Baltic Finnic languages make even less, if any, use of passive and possessive suffixed infinitives than Finnish does, suggesting that they are the result of a relatively late development.

While the apparently emerging non-finite person inflection is a relatively young development, and one that has not been fully developed (at least yet),

### Jaakko Leino

it is strongly motivated by related phenomena in the Finnish language. Notably, of course, person marking on finite verbs serves as the single most important semantic model. Possessive marking, on the other hand, provides suitable morphological means to realize the marking, in the form of affixes that naturally attach to noun-like forms such as infinitives.<sup>9</sup>

To illustrate the similarieties, consider the following:

```
(14) Finnish
```

The resemblance of the possessive suffixed infinitival expression in (14b) to the finite verb expression in (14a), on the one hand, and the NP with possessive marking and the same case inflection in (14c), on the other, is clear.

More importantly with regard to the topic of the present paper, however, all of the expressions in (14a) may be altered to represent the 1st, 2nd, and 3rd person in singular vs. plural, and both of the verbal expression, i.e. (14a) and (14b), may also take the passive morphology. Even (14c) may be claimed to have a counterpart for the passive in the form of a simple noun with no possessive marking. Thus, the correspondence of these expression types is systematic with regard to the person paradigm.

The following figure shows the set of phenomena described above, with their interconnections, and the person paradigm which is present in all of them.

Unfortunately, a proper formalization of the various (sets of) constructions and interconnections shown in Figure 12 is beyond the scope of this paper – not least because it would not be possible without a lot of basic groundwork for describing Finnish morphology in a constructionist framework. Importantly, however, it is possible to represent all the relevant expression types as constructions and to

<sup>9</sup>A further motivation is the genitive subject which occurs in many non-finite constructions in Finnish and which motivates the use of the possessive suffix. For the sake of simplicity, I will omit this part of the complex here. For details, see Leino (2015).

### 3 Formalizing paradigms in Construction Grammar

Figure 12: Motivational structure of the possibly emerging non-finite person inflection

formalize their analogical interconnections in the form of metaconstructions, in the spirit of the analyses and formalizations in Section 2 of this paper.

Understanding the complex Figure 12 as an analogically interconnected group of constructions is revelatory to the nature of paradigms. It serves as an example of what was already observed in Section 2: that language promotes systematicity and recurring patterns by extending observed analogical relations to other contexts, in other words by "copying" ways to organize grammatical information from one part of the grammar to others. This naturally leads to paradigms which hold to more than one single set of constructions.

## **5 Implications and conclusions**

Based on the phenomena discussed above, and the observations made, we may conclude that analogy plays a central role in the internal organization of grammar. In order to capture analogical relations within grammar, a concept like metaconstruction, or at least something similar capable of capturing those relations, is necessary. Acknowledging, and also formalizing, systematic analogical relations between sets of constructions lets us see paradigms as emergent categories of grammar, based on generalizations over repeatedly observed analogies. Paradigms often involve very complex sets of relations between constructions, but they may nonetheless be described in terms of systematic interconnectedness of rather simple constructions. This becomes particularly clear in a case like the one described in Section 4 where a new instantiation of a paradigm is emerging.

### Jaakko Leino

I hope to have shown that metaconstructions have a role in both the synchronic organization and the diachronic reorganization of grammatical constructions. They participate both in statically structuring the inventory of constructions in a given language and in dynamically restructuring that inventory. This restructuring may occur by reanalysis of existing constructions and relationships between them or by coining new constructions through analogy based on existing ones.

Metaconstructions are *generalizations* of constructions, and their central function is to keep up analogical relationships among different sets of constructions. Thus, metaconstructions have to do with networking certain structures and meanings into a coherent system, and with choosing a construction that fits into a given communication setting. This has to do with the topic of synchronic organization of constructions.

Metaconstructions also have, as we have seen, a diachronic nature. They are generalizations of *types* of expressions, not of actual expressions. If they are to be seen as *generative* parts of the grammar, then they must be generative only in the sense that they are not used to create new expressions but, rather, new *types* of expressions, new *constructions*. Consequently, they also have a major role in the diachronic re-organization of constructions.

Furthermore, metaconstructions are a formalization of a phenomenon which is of great importance with regard to language acquisition. Metaconstructions may, in a number of cases, be thought of as *proto*constructions: they are observed analogies, and when they reach a sufficient level of generality, the language learner may use them to abstract a new construction. In other words, we may assume that a child observes an analogical relation between two linguistic expressions and the situations that they represent, and abstracts a construction based on these expressions which is associated to an abstraction of these situations.<sup>10</sup> Similarly, a language learner may observe a similarity of an analogical nature among a group constructions, and abstract a more general construction.

Goldberg (1995: 75) states that inheritance links are "objects in our system", they are an essential part of the language and the grammar, to the extent that a grammar consists not only of constructions but also of different kinds of links which express different kinds of relations between those constructions. As I have argued in this paper, grammar also includes metaconstructions, which further structure and organize the inventory of constructions and which also express relations between constructions – albeit more complicated ones than those expressed with inheritance links – and also provide a dynamic, re-organizational aspect to the grammatical system.

<sup>10</sup>For a more detailed account of language acquisition based on this type of reasoning, see Kauppinen (1998, 1999).

3 Formalizing paradigms in Construction Grammar

## **Abbreviations**


## **References**


Leino, Pentti. 1991. *Lauseet ja tilanteet: Suomen objektin ongelmia*. Helsinki: SKS. *Bidrag till Finska Språkets Grammatik*. 1841.

Manninen, Satu & Diane Nelson. 2004. What is a passive? The case of Finnish. *Studia Linguistica* 58(3). 212–251.


## **Chapter 4**

## **No paradigms without classification: How stem-derivation develops into grammatical aspect**

## Björn Wiemer

Johannes Gutenberg Universität Mainz

The Slavic aspect (pfv.:ipfv.) system is based on a binary opposition of stems connected by derivational patterns. The choice between pfv. and ipfv. stem is unavoidable, since it affects virtually any finite and non-finite form. Simultaneously, this opposition is characterized by classificatory properties resulting from complementary inventories of heterogeneous functions and constraints assigned to pfv. and ipfv. stems, respectively. This situation raises questions as for the paradigmatic organization of lexical units represented by verb stems, and an argument is developed for a two-layer structure of paradigms which integrates the crucial role played by complementary inventories of functions and constraints associated to ipfv. vs pfv. stems. Concomitantly, a case is made for a moderate paradigm-based model of morphology in which stems are the basic units. On this background also Construction Grammar approaches to grammatical categories are evaluated and useful parallels to Word-and-Paradigm models are elaborated on to show which hitherto unnoticed profit can be gained for a theory of grammatical oppositions by a due account of the properties of the Slavic aspect system.

## **1 Introduction**

The opposition of perfective (pfv.) : imperfective (ipfv.) aspect is a kind of "proprietary label" of Slavic languages as a whole. The fundamental architecture of this system is the same for all varieties of Slavic, both concerning the morphological patterns and the basic functional distinctions (see Section 2). The pfv.:ipfv.

Björn Wiemer. 2022. No paradigms without classification: How stem-derivation develops into grammatical aspect. In Gabriele Diewald & Katja Politt (eds.), *Paradigms regained: Theoretical and empirical arguments for the reassessment of the notion of paradigm*, 67–125. Berlin: Language Science Press. DOI: 10.5281/zenodo.5675845

opposition is binary, and the backbone of the system are regular and predictable patterns of stem-derivation (Section 2.1). As a consequence, practically every finite and non-finite form of a verb belongs to either pfv. or ipfv. aspect. This makes aspect choice almost unavoidable, and since verbs play a central role in syntax and participate in many categorial distinctions, aspect choice bears severe consequences for the grammatical system of any Slavic language. Simultaneously, the distribution of pfv. and ipfv. stems over contexts that can be defined by grammatical or pragmatic conditions raises questions about the lexical units on which aspect operates, and about the paradigmatic organization of verb stems representing such units (Section 2.2). This is why the history of Slavic aspectology abounds in discussions concerning the morphological status of the pfv.:ipfv. opposition and its interaction with the lexicon. Central to this discussion are countless debates as for which ipfv. and pfv. stems represent the same lexical meaning (so that they can be considered pairs, partners or groups) and how these relations should be captured in systematic lexicography, but also how the paradigmatic structure of morphologically related verb stems of opposite aspect may be captured (see Section 3).

These traditional debates, in particular the issue of paradigmatic organization, are here re-evaluated: by choosing a stem, one chooses aspect, and this choice correlates with various grammatical constraints and functional oppositions, largely depending on different types of construction. The choice therefore depends on sets of functions and constraints, and these sets lean toward complementary distribution over two classes of stems, called perfective and imperfective. Aspectual distinctions belong to the core functions influencing this choice, but many more oppositions and constraints have appeared since the onset of a development by which diverse kinds of contextual implicatures have been strengthened and entrenched via the choice of stems. These stems organize into two classes (class of pfv. stems vs class of ipfv. stems) and thus make up a binary classificatory system.

Plungjan (2000: 125) defines classificatory categories as follows:

[1] (…) a given amount of lexemes of a language distributes over non-overlapping subclasses without any remainder; every subclass is characterized by its meaning for a certain grammatical category. Therefore, this category is assigned not to word forms, but to lexemes and determines the grammatical *classification* of the [given section of the] lexicon. This is why it is called classificatory.

(my translation, emphasis original; see also Plungjan 2011: 53–54)

### 4 No paradigms without classification

Plungjan considers Slavic aspect a prominent example of a classificatory category. Accepting this, I further argue that the units underlying grammatical classification are stems and that, first of all, the grammatical classes (inventories of pfv. stems vs ipfv. stems) are themselves established and strengthened by an increasing amount of very heterogeneous functions and constraints in complementary distribution. Thus, complementary inventories and the binary opposition of two increasingly abstract classes condition each other mutually (see Section 2.2).

From these premises, I arrive at two interrelated claims which will be argued for in this article, namely


Concomitantly, these properties explain why the rise of the Slavic aspect opposition is no good example of grammaticalization. Among standard parameters, as those in Lehmann (1995), Heine & Kuteva (2002), Himmelmann (2004), only paradigmatic tightening and, to a much smaller extent, syntagmatic tightening apply, in a sense context expansion can be considered as well, but other parameters are practically irrelevant, or useless (Wiemer 2002: Ch. 6, 2008, 2020a: 267– 270, Wiemer & Seržant 2017). This said, I have to specify in which sense paradigmatic and syntagmatic tightening as well as context expansion apply and why the rise of tighter paradigmatic relations is crucial for an understanding of this grammatical opposition. The question whether the related processes and their results may be subsumed under grammaticalization is tangential to (and unrevealing for) a proper explanation of how this opposition works and how its paradigmatic organization looks like. More so, I dare claim that neither Construction Grammar nor Word-and-Paradigm approaches, as known so far, are able to provide a fully adequate description (let alone explanation), although it is certainly among these approaches where we have to look for an adequate model.

This sets the stage for the following. I will start with the essentials about the Slavic pfv.:ipfv. opposition; this presentation glosses over many details, is far

from comprehensive and refrains from giving a diachronic background. It is restricted to what is necessary to justify why aspect in contemporary Slavic should be considered a classificatory system based on stem derivation (Section 2), and what is indispensible for a proper understanding of the problematic relation between paradigm and lexeme, which will be addressed in Section 3. Subsequently, I will discuss the merits and limits of Construction Grammar approaches and Word-and-Paradigm models for an account of the architecture of aspect in Slavic (Section 4), before I address my own proposal of an extended notion of paradigm applied to aspect choice (Section 5). The article ends with a summary and an outlook (Section 6).

Two "technical" remarks should be added: if no elaborate glossing is needed, I will indicate the aspect of the stem by upper case small caps (ipfv , pfv), and examples given without an indication of their source are constructed.

## **2 The basic architecture of the Slavic pfv.:ipfv. opposition**

Essentially, the pfv.:ipfv. opposition of Slavic languages rests on two pillars, one consists in patterns of morphological derivation (Section 2.1), the other in the distribution of verb stems over (a) function inventories and (b) sets of combinatorial restrictions (Section 2.2). Both pillars provide the foundation for paradigmatic oppositions and render them grammatical, but while the first pillar assigns some regularity of form to this opposition, it is the distributional properties of the involved stems which turn it into a classificatory category. These two characteristics – patterns of stem derivation and grammatical classification – are not only compatible, by capturing different properties of Slavic aspect (Wiemer 2006), but depend on one another; moreover, they justify the combination of different models of morphological analysis. Among these, Word-and-Paradigm approaches are more important.

## **2.1 Stem-derivational patterns**

In Slavic, aspect is not indicated by unequivocal, monofunctional morphemes. Instead, the morphology of aspect builds on the functional reinterpretation of patterns of stem derivation which uses both prefixes and suffixes (Breu 2000, Wiemer 2008, Wiemer & Seržant 2017, among many others). Apart from minor, often lexically restricted and obsolete patterns, two patterns predominate and are productive across contemporary Slavic; see Figure 1 with infinitives.

4 No paradigms without classification

$$\begin{array}{rcl} \text{a. } \mathit{simplex} & \Rightarrow & \mathtt{pREFx} + \text{simplex} \\ \text{b. } & & \text{prefixed stem} \Rightarrow [\mathit{prefixed stem}] + \text{surFFx} \\ & & & \begin{cases} \mathit{reinterpolation} \text{:} (\mathit{a}) \,\mathit{identical} \,\mathit{exical\,\,\text{comcept}} \\ \text{ } & \text{(b)} \,\mathit{different\,\,\text{gram\,\,\text{`}} \text{stem} \\ \text{(b)} \,\mathit{different\,\,\text{gram\,\,\text{`}} \text{stem} \end{cases} \\ \text{c. } & \text{simplex}^{\text{prev}} \,\mathit{\,\,\text{`}} \text{stem}^{\text{prev}} \\ \text{d. } & \text{prefixed stem}^{\text{prev}} \,\mathit{\,\,\text{`}} \text{\,\,\text{`}} \text{stem} \,\mathit{\,\,\text{`}} \text{stem} \,\mathit{\,\,\text{`}} \text{stem} \,\mathit{\,\,\text{`}} \text{\,\,\text{`}} \\ & & \text{(e.g.,\,\text{P} \,\text{\$} \mathit{\,\,\text{\$}} \text{\,\,\text{\$}} \text{\,\,\text{\$}} \text{\,\,\text{\$}} \text{\,\,\text{\$}} \text{\,\,\text{\$}} \text{\,\,\text{\$}} \text{\,\,\text{\$}} \text{\,\,\text{\$}} \text{\,\,\text{\$}} \text{\,\,\text{\$}} \text{\,\,\text{\$}} \text{\,\,\text{\$}} \text{\,\,\text{\$}} \text{\,\,\text{\$}} \text{\,\,\text{\$}} \text{\,\,\text{\$}} \text{\,\,\text{\$}} \text{\,\,\text{\$}} \text{\,\,\,\text{\$$$

Figure 1: Predominant patterns of stem derivation and their grammatical reinterpretation

Importantly, markers of tense, mood, agreement (person/number) and of nonfinite forms all occur outside of these stems. Although such ending sets are associated with systematic morphonological alternations, e.g., between present (resp. non-past) and infinitive stems, these alternations apply to any stem regardless of its aspect. This predictably leads to multiple exponence, e.g. the non-past and the infinitive stem each imply particular sets of endings (see 1a–b), so do allomorphs of imperfectivizing suffixes, like -*ywa*- vs -*uj*- in Polish (see 2–c). Thus, the patterns of stem changes that distinguish aspect are entirely dissociated from alternations between non-past and infinitive stems, i.e. from what Brown (1998: 199) dubs "allostems". These two types of stem change are not related diachronically, and despite regular morphonological alternations at the end of stems (or: before endings), the word-form structure shows clear morpheme boundaries;<sup>1</sup> in this respect, it follows concatenative principles: typically, portmanteau morphemes are known for material added on stems, not of stems themselves. See (1– 2) from Polish, where square brackets indicate the part of the word forms which constitutes the stem relevant for determining aspect. They show the word form structure and non-aspect related stem-alternations in the derivation of perfective and imperfective stems with finite past and non-past forms.

<sup>1</sup>There also occur language-specific regular changes in the root vowel (apophony), e.g. Pol. *przepowiedzieć*pfv *– przepowiadać*ipfv 'forecast.inf', but these non-concatenative alternations are likewise irrelevant for aspect membership. They can supply additional cues for its recognition, but never independently from the patterns given in Figure 1, and from multiple prefixation if it applies (see below).

(1) [1] Simplex imperfective stem ⇒ perfective stem by prefixation


(2) [2] Perfective stem by prefixation ⇒ secondary imperfective stem by suffixation


This standard segmentation of word forms into morphemes, with an account of allomorphy, looks like an application of an Item-and-Process (IP) analysis, inasmuch as morphonological changes are accounted for, or even an Item-and-Arrangement (IA) analysis, inasmuch as merely concatenative principles are considered. However, such analyses miss the crucial point that these allomorphic alternations are accidental, and thus irrelevant for determining the membership of the stem to pfv. or ipfv. aspect. It is not the affixes themselves which determine aspect but patterns standing in opposition to another stem which is formally distinguished by presence vs lack of a prefix or suffix.<sup>2</sup> Since these stem distinctions apply "before" markers of tense and agreement are added, each stem already distinguished as pfv. or ipfv. can have all kinds of finite and non-finite forms, or conversely: aspect is in principle distinguished for all finite and non-finite forms (infinitive, participles, action noun); cf. also Andersen (2022 [this volume]).

As the patterns in Figure 1 and (1–2) show, the locus of the affix is also crucial: most simplex stems are ipfv., while prefixation almost always yields a pfv. stem,

<sup>2</sup>Here I am neglecting differences between addition and change (or alternation) of suffixes, for reasons that become clear in Sections 2.2 and 3.

### 4 No paradigms without classification

and the addition (or change) of another suffix makes an already prefixed stem ipfv. While the latter, i.e. so-called secondary imperfectivization, almost never changes the lexical meaning of the stem, prefixes normally change it. In those cases, the prefix serves both a lexical function (it changes lexical meaning and thereby partakes in creating a new lexical unit) and a grammatical function (it changes an ipfv. stem into a pfv. one).

In contrast to prefixes, the number of suffixes used in secondary imperfectivization (see 2b) is much smaller, in fact contemporary Slavic languages tend to have only one productive suffix, such as Russ./Pol./Cr. {iva}, Czech {ova}, Bulg./ Bel. {va}. It has as such become a salient sign of imperfectivization which, above all, does not change lexical meaning. For this reason, many have regarded it as a grammatical marker of ipfv. aspect par excellence. Moreover, because of its productivity secondary imperfectivization has also been viewed as inflection, while other patterns of stem derivation would be ascribed a different grammatical status – with the consequence that Russian aspect has sometimes been treated as a morphologically mixed category. Without going too deeply into this recently revived debate (cf. Gorbova 2015), I here only want to point out that nothing in the morphological structure of verb stems forces us to assume that the addition or alternation of suffixes which precede morphemes marking agreement categories (person/number) or tense should be considered as inflection, even if they do not affect lexical meaning and are very productive. Ultimately, the discussion boils down to the question of what counts as a lexical unit and whether these units can be integrated in paradigms (see Section 3).

Importantly, aspect membership (i.e. the grammatical function) cannot be explained by derivation as such. Instead, it results from distributional restrictions which do not hinge on pairs of pfv. and ipfv. stems united by an identical lexical meaning, although such pairs constitute a central and necessary part of the system (see Section 2.2). By a similar token, aspect assignment does not depend on (a)telicity or actionality features. These tenets will be briefly explained in the remainder of this subsection.

To start with, although the dual – lexical and grammatical – function is characteristic for prefixes added to simplex (predominantly ipfv.) stems, there is a considerable number of cases in which the prefix only marks change of aspect, but does not alter the lexical meaning of the initial ipfv. stem (henceforth: simplex ipfv. = IPFV1). These cases are called Natural Perfectives, following Janda et al. (2013): The perfectivizing prefix has a meaning profile which makes it compatible with a component of the lexical meaning of the simplex stem. This results in semantic overlap, so that the lexical meaning remains unaltered. Standard examples are Russ. *pisat'*ipfv ⇒ *na-pisat'*pfv 'write', *stroit'*ipfv ⇒ *po-stroit'*pfv 'build',

*pit'*ipfv ⇒ *vy-pit'*pfv 'drink', *čitat'*ipfv ⇒ *pro-čitat'*pfv 'read'. Notably, there are no prefixes "specializing" as such in Natural Perfectives (NatPerfs).<sup>3</sup> NatPerfs are only claimed for telic stems (resp. their telic uses); in atelic contexts, usually the prefix *po*- is encountered, which while also perfectivizing, does not mark an action as accomplished, but only as limited in time. Thus, one also gets *pisat'*ipfv ⇒ *po-pisat'*pfv 'write', *pit'*ipfv ⇒ *po-pit'*pfv 'drink',*čitat'*ipfv ⇒ *po-čitat'*pfv 'read' (together with inherently atelic stems like *sidet'*ipfv ⇒ *po-sidet'*pfv 'sit', *kričat'*ipfv ⇒ *po-kričat*ipfv 'shout'), while *po-stroit'*pfv 'build' is telic simply because *stroit'*ipfv is incapable of atelic readings. This evidences that, regardless of whether telicity is analyzed as a feature of verb (better: stem) semantics or of clause semantics, it is not a defining property of aspect.<sup>4</sup> Instead, the stem-derivational patterns discussed here are grammatical, among other things, because they are not restricted by telicity. Pairs of atelic simplexipfv—prefixedpfv stems are numerous and can be productively derived (at least in Russian and many other Slavic languages).

Moreover, there are many verbs which denote punctual events and are, in this respect, related to limited events. These demonstrate systematic pairings of pfv. and ipfv. stems (in either direction of morphological derivation) in which the ipfv. stem simply denotes the same event as the pfv. one without any shifts in actionality. Consider, for instance, Russ. *spotk-nu-t'-sja*pfv – *spotyk-a-t'-sja*ipfv 'stumble', *lop-nu-t'*pfv – *lop-a-t'-sja*ipfv 'pop', verbs denoting illocutionary acts (Russ. *prosit'*ipfv *– po-prosit'*pfv 'request', *zajav-i-t'*pfv *– zajavlj-a-t'*ipfv 'declare'), verbs denoting mental or social events (Russ. *zamet-i-t'*pfv *– zameč-a-t'*ipfv 'notice, spot', *prost-i-t'*pfv *– prošč-a-t'*ipfv 'forgive'). The ipfv. stems only "copy" the event meaning of the pfv. stem, but their function in the grammatical system is important, as they serve to replace their pfv. counterparts in contexts for which the latter are avoided or altogether inadmissible (see Section 2.2). In sum, regardless of the actionality type and differences in actional behavior on the clause level, stem derivation changing the grammatical behavior is pervasive and able to "overwrite" differences of actionality types and actional shifts between the related stems.

<sup>3</sup>The number of NatPerfs can only be estimated roughly, but they are not a marginal phenomenon. Łaziński (2020) counted about 1,670 NatPerfs in Polish, which amounts to approx. 36% of all aspect pairs acknowledged in authoritative dictionaries. For similar figures concerning Russian cf. Janda et al. (2013), who emphasize that, on average, the token frequency of NatPerfs exceeds that of pfv. stems with prefixes that modify the lexical meaning of the simplex about 10 times.

<sup>4</sup>By the same token, perfective aspect only has the function of limiting the situation denoted by the verb; it per se does not mark completion (or similar notions). The latter is possible only with telic predicates.

### 4 No paradigms without classification

Two more phenomena are to be mentioned here because they are widespread and, though complicating the system on first sight, on closer inspection they substantially support the generalization that aspect membership is indicated not by particular prefixes (or suffixes), but based on derivational patterns of verb stems which lead to a binary division of stems into grammatical classes. The first "complication" arises with prefixed stems that are able to occur in a double relationship with ipfv. stems. For instance, Russ. *raz-delit'*pfv can be considered the pfv. counterpart to *delit'*ipfv1 in the sense of 'divide, separate' (= NatPerf), but it also occurs as the pfv. stem to the secondary imperfective (= IPFV2) *razdelj-a-t*ipfv2 in the sense of 'share (somebody's opinion)'. The same applies to the Polish cognates *dzielić*ipfv1 ⇒ *po-dzielić*pfv 'divide, separate' (= NatPerf) vs *po-dzielić*pfv ⇒ *podziel-a-ć*ipfv2 'share (somebody's opinion, fate)'. In these cases the two different pairs (IPFV1—PFV vs PFV—IPFV2) represent different lexical meanings. However, in many other cases the relation of IPFV1—PFV—IPFV2 establishes so-called aspect triplets: both ipfv. stems can provide "copies" of the lexical meaning of the pfv. stem, while each of them is subject to different constraints and function preferences assigned to ipfv. stems in their entirety (see Section 2.2). The pfv. stem, in turn, functions like a pivot, since it connects the two ipfv. stems in a derivational chain; compare:

(3) Russian


Such triplets depend, of course, on NatPerfs, and they are even more difficult to count than NatPerfs (see fn. 3). Preliminarily, even in conservative counts<sup>5</sup> the number of triplets amounts to considerably more than 500 in contemporary Russian, Polish and Czech (Wiemer et al. 2020: §3.2).

Another pervasive kind of triplet is based on atelic IPFV1 stems denoting unbounded activities consisting of rapid cyclic acts (e.g., 'wave', 'shiver', 'blinker', 'knock'). They derive two different pfv. stems, one with the prefix *po*- adding a temporal limit, the other with a nasal suffix (e.g. Russ. -*nu*-) denoting a single act out of the cyclic repetition (semelfactives). Compare:

<sup>5</sup> For a survey of the criteria which have been employed to subdivide aspect triplets cf. Wiemer (2019: 51–56).

(4) Russian

```
a. IPFV PFV
  mig-a-t' ⇒ po-mig-a-t'
      ⇓
  PFV
  mig-nu-t'
  'blinker'
b. IPFV PFV
  vilj-a-t' ⇒ po-vilj-a-t'
      ⇓
  PFV
  vil'-nu-t'
  'wag'
c. IPFV PFV
  krič-a-t' ⇒ po-krič-a-t'
      ⇓
  PFV
  krik-nu-t'
  'shout'
```
In addition, in some Slavic languages, IPFV1 stems denoting atelic activities (among them the same as in 4) can have prefixes marking the ingressive phase (e.g., Russ. *za-kričat'*pfv 'start shouting', *za-volnovat'sja*pfv 'become nervous, start worrying'). Both kinds of triplets provide further evidence that the morphologically related stems distribute over complementary inventories of constraints and functions regardless (i) of the type and tightness of their lexical relation, and (ii) of telicity.

The same holds true for the second "complication": multiple prefixation. Certain prefixes can be used "on top" of already prefixed, or even secondarily suffixed, stems. They do not modify the lexical concept, but add a pluractional<sup>6</sup> or some other quantifying feature, such as repetition (Russ. *pere-za-pisa-t'*pfv *lekciju* '*again* record the lesson'), cumulativity (Russ. *na-so-bira-t'*pfv *ruxljadi* 'collect *a certain/larger amount of lumber*') or distributivity (Russ. *po-ot-davat'*pfv *kvartiry bezdomnym* 'give apartments (*one after the other*) to homeless people'). Prefixes with such functions are called outer or supralexical prefixes;<sup>7</sup> they can also attach to IPFV1 stems (e.g., Russ. *pere-čitat'*pfv 'reread', *na-rvat'*pfv *(cvetov)*

<sup>6</sup> For a classification of pluractionality types cf. Šluinskij (2006) and Wood (2007).

<sup>7</sup>Cf. Tatevosov (2009) for a comprehensive analysis on Russian and further references.

### 4 No paradigms without classification

'pick some amount (of flowers)'), also the numerous "delimitative" pfv. atelic stems prefixed with *po*- belong here (e.g., Russ. *po-sporit'*pfv 'quarrel (over some time)'). The combination of (inner and outer) prefixes, possibly with an "interspersed" suffix, creates chains of derivation, which in an IA-fashion can be analyzed like the successive addition of morphemes; see (5). Importantly, the aspect of the stem depends on whether the last morpheme added is a prefix or the suffix. Thus, the entire stem *po-do-za-pis-yva*- in (5) is pfv., but without *po*- it would be ipfv., since -*yva*- is the last one added before *po*- in the derivational history (cf. Tatevosov 2009: 94, from where this example is cited).

(5) Russian

pfv[po-[pfv[dо-pfv[zа- [pis]ipfv]] -yva]ipfv] -t' ≈ 'record a little bit ∣ ∣ ∣ ∣ ∣ ∣ additionally' (infinitive) pfx3 pfx2 pfx1 'write' sfx inf supralexical lexical [ 'record' ]

Such chains resemble increasing scope relations as in syntactic constituency with recursive insertions. In fact, this kind of analysis is useful in showing analogies between syntactic and morphological structure; above all, the order of appeareance of affixes in a derivational chain is important. This kind of analysis has been favored by linguists who emphasize the derivational character of word form structure and explain morphology in terms of syntax. Such "constructivist" thinking (see Section 4.2) largely disregards, or ignores, paradigmatic structure. This approach works reasonably well when morphological structure is concatenative. It is also sensible in view of the fact that the productive derivation of stems renders hard if not impossible any attempt at "catalogizing" stems into lexical units: given the productivity of affixes attaching to stems one can hardly determine the number of complex stems that are "part of the language". This number may by far exceed the inventory of verbal lexemes registered in even the most comprehensive dictionaries (Tatevosov 2015: 247).<sup>8</sup> Even more, it appears futile to try to establish which stems are related as representatives of identical lexical units (see Section 3).

The issue of lexical identity may be unimportant (or even unintelligible) for those who are just interested in the derivational possibilities of a language, but do not inquire how stems might be organized in a paradigmatic way. However, even

<sup>8</sup> Slavic languages differ in their liability for multiple prefixation and for secondary imperfectivization. Although these parameters have remained investigated insufficiently, Russian and even more so Bulgarian can be regarded as the "leaders" in both respects.

if the question of what is a lexeme, or what counts as its representative, cannot be answered exhaustively, this does not disprove the existence of paradigmatic relations. In the first place, there is an issue of how paradigmatic relations arise. This question is intimately connected to the issue of how the productive and diachronically stable stem-derivational properties of Slavic languages came to establish a grammatical aspect opposition. We now turn to this question.

## **2.2 Grammatical classification in the choice of stems**

However the association between morphological derivation and the meaning relation of the involved stems may be characterized, one crucial point needs to be answered: why are those stems members of different aspects and opposed to each other in grammatical terms?

The answer lies in the fact that morphologically and lexically related stems do not distribute randomly over grammatical forms and contexts, but underlie constraints (of different strength) which either block certain combinations of stems with verbal morphology or their combination with other word forms, or which restrict the interpretation of such combinations. The sums of what is called pfv. and ipfv. stems constitute classes whose grammatical distribution can be captured via their distribution over complementary sets of functions and restrictions on different levels of constituency. Thus, regardless of how morphologically complex (or simple) a verb stem is, and regardless of whether the involved stems are closely related lexically, every stem (with few exceptions) belongs to the *class* of either pfv. or of ipfv. stems *by virtue of restrictions of functions and combinatorial possibilities*.

The function sets and restrictions are not arbitrary, but constitute a complex network in which more peripheral (or specific) functions can to a large extent be motivated from basic functions associated with pfv. and ipfv. aspect. The choice is always binary (pfv. or ipfv.), and it cannot be avoided, since aspect is a property of the stem and thus determined for any form and every discourse token of a verb (see Section 2.1). These properties have led to more or less rigid restrictions in the interaction with other verbal categories, on clause level (complex predicates) and in clause combining. Thus, the grammatical behavior of stems depends on their membership in one of two opposed classes which are, in turn, defined via inventories of functions and restrictions of syn- or paradigmatic combinations; this includes the interpretation of combinations when pfv. and ipfv. stems "compete" with each other. After all, these distributional properties condition the classification of stems based on sum totals of properties which stretch from the core grammar (e.g., tense and mood) via clausal semantics (e.g., modality) to discourse functions (e.g., illocutions, presupposition management).

### 4 No paradigms without classification

Many (if not most) of these properties do not depend on the identity of the lexical concept which may unite morphologically related stems, but these properties can be best illustrated with stems which, being connected via morphological derivation, preserve extreme closeness, or identity, of lexical meaning. These are usually called "aspect pairs". Contrary to "ordinary" synonyms, the use of these derivationally related stems is constrained by complementary functions, and they can (or must) replace each other under predictable conditions, up to the point that even actionality properties (central for the definition of aspect) are "sacrificed". The most famous example of such replacements (at least in East Slavic, Polish and Bulgarian) is the requirement to employ ipfv. verbs in narrative uses of the present tense. See examples from Polish, which denote the same narrative sequence:

(6) a. Narrative past

W 1832 roku Mickiewicz *przyjechałpfv* do Paryża, a dwa lata później *ożenił*pfv *się* z Celiną Szymanowską. Żona *urodziła*pfv mu sześcioro dzieci.

'In 1832 Mickiewicz *arrived* in Paris, two years later he *married* Celina Szymanowska. His wife *gave birth* to six children.'

b. Narrative present

W 1832 roku Mickiewicz *przyjeżdża*ipfv do Paryża, a dwa lata później *żeni*ipfv *się* z Celiną Szymanowską. Żona *rodzi*ipfv mu sześcioro dzieci.

'In 1832 Mickiewicz *arrives* in Paris, two years later he *marries* Celina Szymanowska. His wife *gives birth* to six children.'

In (6b) ipfv. verbs are chosen, because in Polish (as in Russian) the default association of present tense forms of pfv. verbs with the future has become too strong; therefore pfv. verbs are excluded, although the context is about a sequence of singular events. Simultaneously, ipfv. verbs just "copy" the actionality property of their pfv. counterparts: they denote the same events as do the latter in the past tense (see 6a) and, thus serve as placeholders of their pfv. lexical counterparts. For this reason, this contrast, and corresponding tests, are considered trivial, and the respective relation between the pfv. and ipfv. stem is a trivial one. This is but an extreme manifestation of lexical identity (with complementary grammatical distribution), and this is why replacements between narrative pfv. past and ipfv. present tense are used as a test of aspectual pairings (at least for Polish and Russian). Other such "trivial" tests (with varying reliability, depending on the language) are

	- (7) Russian
		- a. Segodnja Vanja *vyšel*pfv uže v 5 časov. 'Today Vanja *went out* already at 5 o'clock.'
		- b. Obyčno Vanja *vyxodil*ipfv tol'ko v 8 časov. 'Usually Vanja *went out* only at 8 o'clock.'
	- (8) Russian
		- a. *Rasskaži*pfv emu, čto ty videl. '*Tell* him what you saw.'
		- b. *Ne rasskazyvaj*ipfv emu, čto ty videl. '*Don't tell* him what you saw.'

(8a) expresses a command (or request), (8b) is prohibitive. Again, both stems mark the same type of event, but (in combination with negation) their illocution differs. Remarkably, in North (= East+West) Slavic the pfv. stem can be used in the negated imperative (9a), but only in another illocution, namely a warning (preventive meaning). The ipfv. stem, in turn, can be used in unnegated imperatives as well (9b), but it then contrasts with the pfv. stem (see 8a) in that it indicates the speaker's assumption that the intended action is presupposed:

	- a. *Ne rasskaži*pfv emu (slučajno), čto ty videl. '*Don't tell* him (inadvertently), what you have seen.'
	- b. Nu, čto ty tam videl? (Ty obeščal mne rasskazat'pfv.) *Rasskazyvaj*! ipfv 'Well, what have you seen there? (You promised to tell me.) *Tell* me.'

That is, choice of aspect in the unnegated imperative is indicative of speaker's assumptions about absence (pfv.) vs presence (ipfv.) of knowledge and expectations shared with the interlocutor (cf. Padučeva 1996: 71–80 for Russian). This kind of presupposition management also works for contrasts of aspect choice in the future tense, e.g. in Polish (cf. Błaszczak et al. 2014: 193–199), and with modal auxiliaries (for instance, (10a) may be used if such a presupposition is implied).

Aspect choice can also differentiate modal functions, e.g. in minimal pair conditions under the scope of modal auxiliaries as in (10a–10b).

4 No paradigms without classification

(10) Russian, cf. Padučeva (2008) a. Zdes' možna *perexodit*' IPFV ulicu. 'You are *allowed* to cross the street here.' deontic, IPFV b. Zdes' možna *perejti*PFV ulicu. 'It is *possible* to cross the street here.' circumstantial, PFV

Relevance of aspect choice for modal functions requires two interrelated caveats. First, as pointed out above, (10a) can also be read with the implicature that the speaker considers the action (crossing the street) to be presupposed. This implicature need not contradict the modal interpretation, quite the contrary: permission given by the speaker is fairly compatible with the assumption that the interlocutor is waiting for allowance. Obviously, implicatures triggered by aspect choice under particular grammatical and communicative conditions may "overlap". Second, the differentiation of modal functions might be a side effect of other factors which prove stronger than the distinction between subdomains of modality. For instance, from her study of Russian, Polish and Croatian based on a parallel corpus Divjak (2011: 81) concluded: "Although type of modality remains a significant contributor to aspectual choice, the fact whether the option, permission, order etc. has been given in a generic or specific way outperforms the type of modality in predicting the choice of aspect for the infinitive."

Implicatures can be strengthened and eventually conventionalize. The latter happened to pfv. present tense in North Slavic, which by default has been reanalyzed as pfv. future. Such an implicature has not been entrenched in South Slavic, where pfv. present and pfv. future exist side by side; however, pfv. present tense forms are severely restricted to contexts of suspended assertiveness (otherwise subsumed under "irrealis" meanings). In addition, in Balkan Slavic present tense forms of pfv. stems in main clauses are dependent on the verbal proclitic *da* (an ubiquitous irrealis marker); cf. Ivanova (2014), Wiemer (2014), Todorović (2015). In turn, in North Slavic the default reanalysis [pfv. present > pfv. future] has been accompanied by the restriction of the inflected future auxiliary (*bǫd*-, with different phonological realizations) to ipfv. stems (infinitives). Another restriction on the syntagmatic axis applies to almost all contemporary Slavic varieties (except colloquial Upper Sorbian): phasal verbs ('begin', 'continue', 'finish, stop') can combine only with ipfv. stems.

Already these few examples demonstrate manifold functional cross-relations between forms that belong to core sections of the standard paradigm of the Slavic verb (e.g., the imperative, the future), and these cross-relations constitute a larger network of functional choices and constraints for which aspect choice is a sufficiently reliable indicator. The degree of reliability differs, and many of the con-

texts triggering these choices are not directly related to aspectuality or temporality, but they can usually be motivated by features like the limiting function of pfv. stems and the potential of ipfv. stems to defocus limits. In many environments relevant for aspect choice telicity does not play a role; also atelic stems (or readings of stems) require a distinction by aspect (and therefore, correlated patterns of stem derivation) under predictable conditions. For instance, the trivial conditions for the imperative illustrated in (8a–8b) hold true also for atelic stems (see 11a–11b), and the same effect of presupposed actions ("overlapping" with allowance) applies with atelic ipfv. stems (vis-à-vis their pfv. counterparts), as in (12a–12b):

### (11) Russian

	- a. V lesočke ostanavlivaet mašinu, žestom priglašaet vyjti. *Poguljaj*pfv nemnogo, jablok narvipfv. – A možno? – Konečno, možno. 'In the woods, he stops the car, with a gesture invites him to leave. *Walk* a little bit, pick some some apples. – May I? – Of course, you may.'
	- b. A čto, nel'zja? *Guljaj*, *guljaj*ipfv, tol'ko učti, sjuda podxodit' zapreščeno. 'So what, can't I? – *Walk*, *walk*, but remember, it is forbidden to come here.'

Moreover, atelic activities can be integrated into sequences narrated in the past tense by using corresponding pfv. stems. Compare atelic pfv. *pomieszkał* 'lived' in a sequence with telic pfv. *schował się* 'hid' and *zadomowił się* 'settled':

(13) Polish (PNC; Polityka, 15.04.2006) Czarny Kot najpierw *pomieszkał*PFV w Zielonym Baloniku, potem *się schował*PFV (...), a w końcu na dobre *zadomowił*PFV *się* w Piwnicy. 'The Black Cat first *lived* (a while) in the Green Balloon, then he *hid* (…), and finally *settled* in the Piwnica for good.'

### 4 No paradigms without classification

In a sense, this looks like the opposite of the first of the "trivial" tests discussed above with telic verbs (see example 6a–6b): the pfv. stem is used instead of the ipfv. one, since the narrative context forces a sequential reading. The consistency with which ipfv. stems denoting atelic activities undergo phase modification by prefixes varies across Slavic,<sup>9</sup> but the point here is that, regardless of (a)telicity, lexical concepts are adapted to functions to fit a particular type of context or construction. This adaptation considerably enhances the amount of stems, however one determines their relation to a lexicon, or a "constructicon" (in Goldberg's terms; see Section 4.1), of the given language. In this sense, we can say with V. Lehmann (1999: 229, 2004: 174) that "the grammaticality of aspect is based on the maximal extension of derivational affixation", i.e. on the degree of approximation toward this maximal extension. However, we cannot (and need not!) assume that each lexical meaning is represented by a pfv. and an ipfv. stem (i.e. by aspect pairs). There are plenty of stems which are to be considered *tantum*-stems.<sup>10</sup> The amount of stems furthermore increases by multitudes if we consider all possibilities of multiple prefixation (see (5) in Section 2.1). Most of these derivatives will never make it into dictionaries (for what sake?), as their derivation is arguably rule-based and many (most?) of them are ephemeral.

Now, all these intricacies of stem derivation, their lexical relatedness and lexicographic status are of secondary concern if the perspective is reversed, i.e. if this is to be looked at from the sets of functions and constraints (some of which have been discussed above) and their distribution over pfv. and ipfv. stems *in toto*. That is, instead of asking how lexical meanings (or concepts) are adapted to contexts by derivational means, let the question be which functions and constraints are assigned to (or characteristic of) which aspect. Under this angle, pfv. and ipfv. stems, respectively, are understood as classes whose members underlie specific sets of grammatical restrictions and which have a restricted amount of functions from which to choose. Sets of functions and restrictions are defined *for each class*, not for lexical concepts, and the grammatical classes as such arise from, and are strengthened with, these sets of distributional properties. Thus, the formation of opposite (pfv. vs ipfv.) classes, on the one hand, and sets of complementary functions and restrictions, on the other, are mutually dependent. One cannot think the one without the other.<sup>11</sup>

<sup>9</sup> In this respect, the languages in the east are more consistent than in the west; cf. Petruxina (2000: 141–230), among others.

<sup>10</sup>This applies even if we admit telic and atelic triplets (as in 3 and 4) as well as pairs of atelic activities (as in 11a–12b), not accepted by traditional Slavic aspectology.

<sup>11</sup>Notably, also the notion of grammatical recategorization, by which a lexical concept is transferred from one class into an opposite one (Lehmann 1999, Mende 1999), presupposes that such classes exist, in the first place.

The system properties of the pfv.:ipfv. opposition cannot be explained from derivational patterns alone, although these patterns provide the morphological basis for the recognition of stable form-function mappings and collocational restrictions. We may assume that form-function mappings of morphologically frequent and transparent patterns have been extended to stems with less frequent and/or productive morphological relations (e.g., Russ. *liš-i-t'*pfv *– liš-a-t'* ipfv 'deprive', Pol. *kup-i-ć* pfv *– kup-owa-ć*ipfv 'buy') and to stems with lexically specific affixes (as with a nasal suffix applied only to multiplicatives to yield semelfactives; see Section 2.1). These peculiarities, as either obsolete and unproductive or semantically too specific, have been "absorbed" in more abstract classes. An extreme manifestation of analogical expansion is the rise of suppletive aspect pairs, i.e. the inclusion of etymologically unrelated stems denoting identical lexical concepts, but with complementary distribution over the function sets discussed above; compare Russ. *lovit'*ipfv *– pojmat'*pfv 'catch', *brat'*ipfv *– vzjat'*pfv 'take', *klast'*ipfv *– položit'*pfv 'put', Pol. *widzieć*ipfv *– zobaczyć*pfv 'see', *mówić*ipfv *– powiedzieć*pfv 'say, tell'. The suppletive behavior of stems does not differ in principle from suppletion as described for inflection (cf. Veselinova 2006). After all, debates about inflection and derivation turn out to be unrevealing, if not misleading (cf. Wiemer 2020b for a discussion).

Figure 2 summarizes the insights discussed above (with the two most productive patterns of stem derivation). The function inventories (at the bottom) mean to include all sorts of restrictions applying to pfv. and ipfv. stems, respectively; part of such restrictions were illustrated above.

Members of aspect pairs (or partners, or stems united in actionality groups; see Section 3) belong to opposite sets, but are lexically most closely related. Technically, stems forming one aspect pair, including suppletive pairs, may be coindexed between their sets (= pfv. vs ipfv. class) for the respective lexical meaning. A certain amount of pfv.:ipfv.-pairs united by morphologically transparent and/or productive patterns is necessary, but these patterns only supply the backbone of the system to guarantee it some regularity in form. Probably, most stems do not allow for co-indexation, because their lexical relatedness is weaker or absent, but they nonetheless belong to either the pfv. or the ipfv. class; this includes *perfectiva* and *imperfectiva tantum* stems. Thus, crucially, the tendency toward complementary distribution over opposite sets of heterogeneous functions applies to pfv. vs ipfv. stems as two generalized and abstract classes, not to particular stems (or their pairs, triplets, groups). The distributional properties apply irrespective of aspect pairs or other considerations concerning lexical relatedness between stems.

function inventory PFV

a. simplex ⇒ prefix+simplex b. prefixed stem ⇒ [prefixed stem]+suffix c. simplexipfv — prefixed stempfv (e.g., Pol. *gotowa-ć* ⇒ *u-gotowa-ć* 'cook') d. prefixed stempfv — [[prefixed stem]+suffix]ipfv (e.g., Pol.*prze-kaza-ć* ⇒ *prze-kaz-ywa-ć* 'convey'; *s-praw-i-ć* ⇒ *s-prawi-a-ć* 'cause') + formation of two classes (= ipfv. vs pfv. stems) acquiring increasingly complementary distribution over function sets, regardless of lexical (non-)identity of concepts, of derivational patterns, and of (a)telicity: reinterpretation: (c) identical lexical concept (d) different gram. distribution analogical expansion IPFV: {F<sup>1</sup> , F<sup>2</sup> , … F } PFV: {F<sup>1</sup> , F<sup>2</sup> , … F } IPFV STEMS PFV STEMS

Figure 2: Productive patterns of stem derivation and complementary inventories of functions and constraints

function inventory IPFV

This, in turn, does not imply that every stem belonging to either the pfv. or the ipfv. class occurs with every single function associated to this class; it only entails that all those functions that are realized belong to the respective inventory, not to the opposite one. In other words: many stems realize only subsets of the function inventory (and constraints) that are characteristic of their class (pfv. vs ipfv.) in their respective entirety; a function may not be compatible with the lexical meaning of the stem or be inappropriate for communicative reasons.<sup>12</sup> Concomitantly, the strength of the restrictions (and the degree at which invited implicatures have conventionalized) varies, from obligatory choices to more probabilistic distributions.

All these considerations emerge from a thorough re-assessment of diverse observations made about the Slavic aspect system, which are in need of a synthesis in order to adequately describe its architecture. At its core, this synthetic re-assessment testifies to Slavic aspect as a grammatical system which can best be captured as a classificatory category, in line with Plungjan's definition cited in [1] at the beginning. However, strictly speaking, it should be applied to stems, not to words. Moreover, pointing out that the stem-derivational basis of this category leads to an assembly of word-formation patterns, is as unrevealing as are debates about the derivational, inflectional or mixed character of the pfv.:ipfv. opposition.

How do these considerations fit in with standard assumptions about the relation between paradigms and lexemes, and how would they modify them? We now turn to these issues.

## **3 Lexemes and paradigms for pairs, partnerships, groups and networks**

According to Plungjan's definition of grammatical classification adduced in [1], it is not forms of one lexeme that distinguish ipfv. and pfv. aspect in Slavic, but different lexemes, each of them having its own paradigm of forms. This corresponds to another well-known definition given by Zaliznjak (1967), who contrasts classificatory and inflectional categories as follows:

[2] "grammemes [= values of grammatical categories; BW] of inflectional categories characterize particular word forms of a paradigm by opposing

<sup>12</sup>For instance, many ipfv. stems are unable to express progressive meaning (Lehmann 1998), and many of the functional oppositions discussed above are "available" only for telic or punctual stems (see Section 2.1). Or stems with multiple prefixes (as in 5) are usually hardly imaginable in the imperative (for communicative reasons).

### 4 No paradigms without classification

them to other word forms of the same paradigm. By contrast, a grammeme of a classificatory category characterizes the entire paradigm by opposing it to other paradigms." (Zaliznjak 1967: 31–32; my translation)

Plungjan and Zaliznjak assume a "classical" notion of "paradigm", which implies (i) that, on the one hand, there is a practically open set of lexical units (Russ. *leksemy* 'lexemes'), (ii) these units are represented by stems (or roots, if this is a justifiable unit); on the other hand, (iii) there is a restricted and closed set of morphological changes on these stems and/or by auxiliaries (iv) which map onto a set of homogeneous (Russ. *odnorodnye*) functions, (v) but leave unaffected the lexical integrity of the manipulated units. What lexical integrity means is never really specified and rather taken for granted, often as intuitively clear. Regardless of this, the paradigm of a word in one particular meaning (= lexeme) describes a set of matches between forms of that word and their functions as well as their combinatorial restrictions. "Forms of a word" in practice refers to changes of word forms that apply to its stem.

This said, it may be generalized that for those who want to characterize the pfv.:ipfv. opposition as an inflectional category, stems which form aspect pairs share into *one* paradigm, though a complex one, *because* they represent the same lexeme. By contrast, those who prefer to describe this opposition in terms of grammatical classification, would say that each stem of an aspect pair corresponds to a lexeme. These stems are synonymous, but they behave differently with respect to grammatical conditions; and *since* they represent different lexemes, they also have *different* paradigms. Although this is not said explicitly, different paradigms may intersect (e.g., for their past tense or forms of mood), but the same lexical concept can also be expressed in different grammatically relevant contexts (namely, in those parts for which the two paradigms do not intersect). These two viewpoints are shown in Figure 3.

Crucially, both treatments are a necessary consequence of the tight connection between lexeme and paradigm. These notions are assumed to be in a 1:1-relation, in fact they are correlative in that they entail each other.<sup>13</sup> Consequently, if one wants to say that pfv. and ipfv. stems have different paradigms, even provided they represent the same lexical meaning (however captured) and are morphologically related, it must be said that they are different lexemes. An alternative position is argued for in Breu (1984a,b), who assumes that an aspect pair represents *one* lexeme whose paradigm is composed of the (finite and non-finite) forms of two verbs (stems) which differ only in aspect (1984b: 128–129). Here a complex

<sup>13</sup>This makes them similar to converses ('left' vs 'right', 'parent' vs 'child', 'give' vs 'take').

Figure 3: Contrasting views on the grammatical status of aspect pairs

set of forms for one lexeme is assumed, and it does not matter whether this set is conceived of as one complex paradigm or the intersection of two paradigms. At any rate, no strict 1:1-relation between lexeme and paradigm is required. After all, it can be seen that the connection between lexical relatedness, morphological relatedness and membership in one of two grammatically defined classes is completely opaque.

Furthermore, in Figure 3 the inventory of word forms is shown to comprise both finite and non-finite forms. For the present concern, the question whether to include non-finite forms into the paradigm or not would be off the point, because this decision does not hinge on the difference of aspect which is ascribed to stems (i.e. each of the stems, regardless of aspect, has finite and non-finite forms). Thus, the issue of whether transpositional categories (participles, converbs, action nouns and, notabene, the infinitive) are to be considered parts of the verb's paradigm, or whether this extension of the paradigm is to be considered inflectional or derivational (as discussed in Haspelmath 1996), is irrelevant for the relation between pfv. and ipfv. stems. This shows that, however one may want to capture the relation between inflection and derivation, it is an entirely different matter than the distinction between inflectional and classificatory categories. Nonetheless, it is justified to assume non-lexicalized non-finite forms as part of an extended paradigm (see Section 5), since they are formed regularly, do

### 4 No paradigms without classification

not affect lexical integrity and are often part and parcel of grammatical constructions considered to be parts of the grammar (e.g., participles suffixed with -*n/t*employed in the passive and other actor-demoting constructions, or the infinitive as a component of the ipfv. future in North Slavic).

Now, Plungjan (2000, 2011) emphasizes that the Slavic PFV:IPFV. opposition is a very suitable example of a classificatory category, based on the assumptions expressed in [1] and [2]. Recently, Tatevosov (2016) has adopted this view and very consistently showed that aspect systems regarded as inflectional (including "analytic inflection", as in English) are functionally equivalent to classificatory systems (as in Russian). While such equivalence had been pinpointed earlier (without however considering properties of classificatory systems), Tatevosov's equivalence relations are based on a rigid procedure of determining actionality classes for which verbs unite into groups with an identical lexical meaning (cf. already Tatevosov 2002). Tatevosov claims to be agnostic as for whether productive derivational patterns may be considered inflectional, and his procedure does not rely on lexical identity defined by metalinguistic periphrases (Tatevosov 2016). Instead, his analysis is based entirely on model-theoretic considerations and extensional (formal) semantics, combined with a generative, morpheme-centric approach which relies on assumptions typical of IA-models (Tatevosov 2015) or, more broadly, on "constructivist" reasoning (see Section 4.2). This is why no consequences follow from his recognition of the Russian (Slavic) aspect system as a classificatory category: grammatical classification simply remains a label, but no account is given of its properties and consequences for the analysis of such an aspect system.

In turn, representatives of Cognitive Grammar have advocated networks of Russian verb stems as adequate representations of units which are very closely related as lexical concepts (Janda 2007). Here lexical identity is relativized and consciously regarded as fuzzy. Similarly, there have been attempts in which the notion of aspect pairs is relativized as just one (although very salient) type of derivationally marked relation between stems with some shared lexical concept. Such an approach has been developed, first of all, by V. Lehmann (1988 and subsequent publications). These relations are captured as partnerships between pfv. and ipfv. stems, so that aspect pairs can be redefined as the tightest and grammatically most rigid kind of partnership. In general, aspect partners are divided into core, peripheral and only contextually determined ones (for the most comprehensive account based on Russian cf. Mende et al. 2013). Furthermore, an account is given of semantic motivation, which conditions actionality types and explains the direction of semantic derivation.

After all, regardless of whether one deals with aspect pairs, with aspect partnerships (à la Lehmann), with actionality groups (à la Tatevosov) or with networks (à la Janda), what is missing in all these approaches is an account of the behavior of pfv. and ipfv. stems beyond the domain of actionality and eventexternal pluractionality. Also missing is an explanation of how categorial syncretisms (like the present > future shift of pfv. stems in North Slavic) and partially very rigid replacement conditions arising from such syncretisms are to be treated in a network of functions and restrictions which themselves can be conceived of as paradigms (on a more abstract level) and which, nota bene, underlie a reasoning like V. Lehmann's (1999) "grammatical recategorization" (see fn. 11) or Breu's (1984b) "grammatical homonymy".

As the sketchy analysis in Section 2.2 should have shown, whatever the pattern of stem derivation looks like, its "output" (= ipfv. or pfv. stems) only provides the "input" for further operations with other units on a syntagmatic (morphological or, mostly, syntactic) level for which the choice of ipfv. vs pfv. stem is constrained, either by blocking one of the two or by yielding different interpretations. In other words: ipfv. and pfv. stems (and their paradigmatic opposition) enter into constructions, i.e. units of a larger format than stems, but aspect choice (or the PFV:IPFV. opposition) as such is *not* itself a construction; it is simply a central factor (or component) within possible templates for those larger constructions. The question to pursue now is whether templates that describe aspect choice can themselves be integrated into paradigms, albeit in an unorthodox and complex way. Does this make sense? And to what extent is work in Construction Grammar and Word-and-Paradigm models helpful?

## **4 Word-and-Paradigm and Construction Grammar**

In order to assess the possibilities, but also the limits, of Construction Grammar and Word-and-Paradigm morphology to capture the grammatical character of the Slavic aspect opposition, the basic tenets of both families of approaches to grammar and lexicon will be summarized in this section. This includes highlighting their intersection. I concentrate on those claims which allow me to take stance if there are "aspects of Slavic aspect" that cannot be integrated into these approaches, or that can only be integrated if certain issues are amended or modified.

## **4.1 Construction Grammar**

Construction Grammar (CG) implies a theory of linguistic knowledge: "a construction is a generalisation that speakers make across a number of encounters

### 4 No paradigms without classification

with linguistic forms" (Hilpert 2014: 9). This knowledge "includes both items and generalizations, at varying levels of specificity", which pertain to "conventional, learned form-function pairings at varying levels of complexity and abstraction". The latter means to "highlight the commonality between words and larger phrasal units" (Goldberg 2013: 16–17). Words differ from grammatical constructions only in terms of their internal complexity (Michaelis & Lambrecht 1996: 216). By the same token, there is no strict divide between grammar (morphosyntax) and the lexicon, or between rules and an inventory of ready-made units to which the rules apply to yield more complex units. Instead, grammar and lexicon condition each other in that linguistic knowledge is organized in hierarchies of schemas. These, in turn, result from abstractions that are generalized from experience on the basis of already acquired exemplars. For instance, "the coinage of new words depends on abstractions over sets of existing words and word forms in the lexicon of a language" (Booij 2010: 3). Analogical proportions are crucial, but it is not necessary that complex forms are "broken down" into smallest meaningful units (a.k.a. morphemes); rules are a subsidiary notion, while schemas are primary.

Constructions, in general, are defined as any linguistic pattern for which "some aspect of its form or of function is not strictly predictable from its component parts or from other constructions recognized to exist. In addition, patterns are stored as constructions if they are fully predictable as long as they occur with sufficient frequency" (Goldberg 2006: 5). The last sentence is important, since – contrary to an earlier definition in Goldberg (1995: 4) – even entirely predictable (and semantically transparent) patterns of form-function pairings can count as constructions. This holds true in particular for morphology, i.e. for units recognized as words: conventionalized complex words are listed in the lexicon even if they result from very productive patterns, but they are linked by inheritance relations (so that their idiomatic part can be calculated by subtracting the recurrent, inherited properties); cf. Booij (2013: 257–264). Inheritance relations reach down to morphemes, i.e. parts of word forms. However, although it is recognized that words can consist of smaller parts (= morphemes), morphemes are not given an independent theoretical status outside sets of word forms;<sup>14</sup> they are rather considered fillers of positions, or slots, within word forms: "A constructional idiom is a (syntactic or morphological) schema in which at least one position

<sup>14</sup>Goldberg (2013: 15) regards morphemes as "partially filled words". Booij (2010: 15) emphasizes that "the minimal linguistic sign is the word", and "bound morphemes form part of morphological schemas, and their meaning contribution is only accessible through the meaning of the morphological construction of which they form a part". Furthermore: "surface forms are regarded as basic morphotactic units of a grammatical system, with roots, stems, and exponents treated as abstractions over a lexicon of word forms" (Booij 2010: 256).

is lexically fixed, and at least one position is variable" (Booij 2013: 258). This again underpins parallels between morphological structure and syntactic constituency (cf. also Booij 2010), but the crucial point is that this structure is not composed bottom-up from smaller building blocks; instead, parts are defined, or inferred, from wholes. This holds true not only for the syntagmatic, but also for the paradigmatic level: any units contained in constructions (of word format or larger units) are interpreted with respect to larger units in which they are a part of (syntagmatic axis) and with respect to other units of the same format with which they stand in a replacement relation for a slot in a construction (paradigmatic axis). Both dimensions support the recognition of patterns which provide cues for analogical transfer between morphologically related word forms (whose segmentability may differ and whose semantic relation may also be opaque in form, e.g. because of diverse morphonological processes).

This top-down approach has several advantages, among others it allows for unified accounts of concatenative and non-concatenative morphology (e.g., reduplication, apophony) and of paradigmatic relations between word forms. Here a distinction between inflection and derivation becomes irrelevant. For either type of morphology, correspondences between schemas can spell out analogies even if form-function matches display irregularities, e.g. if the common parts of correlated word forms show a good deal of variation (otherwise called allomorphy), if there are "gaps" between particular instantiations of patterns, or if base forms are lacking. This can be illustrated with words formed by Engl. -*ism* vs -*ist*, as in (14):


It is easy to generalize the meaning relation between the words on the right and on the left by the formalization in (15):

(15) 〈[x-ism]Ni ↔ SEM<sup>i</sup> 〉 ≈ 〈[x-ist]Nj ↔ [person with property Y related to SEM<sup>i</sup> ]j 〉

The parts in 〈...〉 represent constructions, ↔ indicates a form-meaning correspondence, SEM stands for a semantic representation, or paraphrase. Importantly, ≈ points out the paradigmatic relation between the words to the left and the right, but since these words have a common part (*x*), their alternating parts (-*ism* vs -*ist*) also stand in a paradigmatic relation to each other; simultaneously,

### 4 No paradigms without classification

they can be interpreted as devices that mark some kind of semantic, or functional, change, regardless of whether *x* exists in isolation or not (see below). Moreover, this change can be quite concrete and palpable (e.g., agent of an action, or member of a group) or very abstract and unrelated to the lexical content of the entire unit. The examples in (14), and their generalization in (15), are closer to the concrete pole.

The list in (14) could be made considerably longer by the productive application of (15), but even the few instantiations given in (14) demonstrate that not only -*ism* and -*ist* do not exist in isolation, but also the "bases" of these word formation devices are often lacking. These two parts are thus mutually dependent, although each of them can be combined with other instantiations of the respective other part. It is this productive combinatorics, based on analogy, which makes these words segmentable (or: which make speakers infer about their parts). In particular cases, e.g. *social-ist* or *marx-ism*, a base exists, but the complex word does not derive from it semantically, since *social* and *Marx*, respectively, mean something different. Thus, the meaning of these complex nouns "is not simply a compositional function of their constituent parts but contains the meaning of a related word *with the same degree of complexity*" (Booij 2010: 33, emphasis added). In general, models based on bidirectional correspondences of word forms yield more satisfactory results on non-concatenative patterns (including suppletion); in addition, they are often even more convincing for concatenative patterns like, e.g., cross-formations and back formation (Haspelmath & Sims 2010: 47–51). In Section 5 I will come back to this kind of reasoning and the formalization in (15), but here parallels with patterns of stem derivation characteristic of Slavic aspect become obvious. The crucial difference compared to "usual" cross-formation and back formation is that, with Slavic aspect, stem derivation can lead both to synonyms (a) without a change of syntactic class, but different grammatical distribution, and to synonyms (b) with a change of syntactic class ("word class changing inflection" in Haspelmath 1996), so that the same lexical concept is made available for different syntactic contexts. Thus, the functional change between the paradigmatically related word forms is considerably more abstract than in the word form pairs in (14) and the schema in (15).

CG has also been employed in studies on grammaticalization. Representatives of CG have emphasized that grammaticalization often starts with the conventionalization of constructions (mostly in a narrower, discourse-to-syntax oriented understanding). Here it is, among other things, prefabs in the sense of Bybee (2010: 55), i.e. "[l]exically filled, instance-based constructional patterns occurring with sufficient frequency" (Nikiforidou 2009: 26–27), which lead to conventionalized

constructions. A related issue has been elaborated on by Diewald (2009): constructional frames supply the loci for critical contexts in which either particular elements (words, phrases) or abstract constructions as such (e.g., some clausal frame with a specific word order) gain significance and strengthen pragmatic implicatures (first of all, by resolving erstwhile meaning conflicts between different components of the constructional frame), so that eventually these implicatures become conventionalized meanings of the given construction. Diewald particularly focuses on paradigmatic contrasts between constructions. These consist either in the presence/absence of particular elements, which yields a binary opposition, or in the choice of one of the members from among a relatively closed set (e.g., modal auxiliaries, propositional modifiers like epistemic particles), which renders the opposition multiple. These two types of paradigmatic choices correspond to "transparadigmatic" and "intraparadigmatic" variability, respectively (cf. Chr. Lehmann 1995: 138–139). Diewald also emphasizes the role of obligatoriness (cf. also Diewald & Smirnova 2010); this, in fact, is but a concomitant of paradigmatic tightening and a loss of transparadigmatic variability.

While, for the present purpose, grammaticalization is of secondary concern, it is worth considering whether and how the relation which Diewald establishes between paradigm structure and obligatoriness, on the one hand, and between the conventionalization of constructions and pragmatic strengthening, on the other, might be applied to the paradigm structure of stem-derivational and classificatory aspect. Below it will become obvious that such a transfer is not feasible unless one considerably modifies the notion of paradigms (or of paradigmatic structure) and, above that, does justice to the particular conditions which make the choice of aspect obligatory and strengthen its function inventory in Slavic languages (see Section 5).

## **4.2 Word-and-Paradigm morphology**

The area of intersection between CG and Word-and-Paradigm (WP) approaches in morphology is large. This already becomes evident from the fact that WP "establishes correspondences between different sets of grammatical properties and the different forms of a word that realize each one of these sets of properties" (Fábregas & Scalise 2012: 31). That is, "words match a schema, and a schema subsumes words"; this schema is based on "morphological correspondence" between the phonological realizations and the functions of the word forms (Haspelmath & Sims 2010: 46–47). Thus, for English nouns the schema in (18) is a generalization over (17), and (19) represents a correspondence rule referring to (18) (cited from Haspelmath & Sims 2010: 46–47):

### 4 No paradigms without classification


$$\begin{bmatrix} \text{'begz/N} \\ \text{'bags'} \end{bmatrix} \qquad \qquad \begin{bmatrix} \text{'} \text{'bijz/N} \\ \text{'key'} \end{bmatrix} \qquad \qquad \begin{bmatrix} \text{'godz/N} \\ \text{'gods'} \end{bmatrix} \qquad \qquad \begin{bmatrix} \text{'rbrz/N} \\ \text{'robs'} \end{bmatrix}$$

(18) Word-Schema

$$\begin{bmatrix} / \text{X}z/\text{n} \\ \text{'purality of xs'} \end{bmatrix}$$

(19) generalized

[ / /n '*x*' ] [ / /n 'plurality of *<sup>x</sup>*s' ]

What is called "words" here might better be labelled "word forms". While WP-approaches do not deny that words (or word forms) are often composed of smaller units, and that these units usually have meanings of their own, their endeavor is not to decompose word forms into morphemes in order to construct from them the meanings of whole word forms (in a more or less compositional manner). WP approaches do not pursue a bottom-up procedure of this kind, but, conversely, they abstract away from particular phonological realizations (and concomitant alternations in the form of purported units on a subword level) of word forms and rather analyse in a top-down manner by comparing the shape of particular word forms with their variation according to some homogeneous functional parameter(s). This principled difference – constructing larger units from smaller ones vs abstracting away from particular forms and asking for functions due to which these forms enter into replacement relations – is the reason why Blevins (2016) distinguishes between "constructive" and "abstractive" models of morphology. WP approaches are clearly abstractive, while Item-and-Arrangement (IA) and Item-and-Process (IP) approaches are "constructive" because they rest on the basic assumption that larger units (words, phrases) are constructed from smaller ones according to certain rules.

From a "constructivist" viewpoint, paradigmatic relations are of no primary concern, often they are neglected or even denied altogether. In turn, "abstractivists" may not feel forced to assume any such units like stems (or, even more so, roots), although word schemas and correspondence rules suggest that word forms can be analyzed into smaller parts. Constructivist reasoning accepts morphemes as basic units of analysis, it has introduced allomorphs as a concept (and

morphonology as an intermediate structural level) by which different phonological realizations of meaningful distinctions in word forms correlate with their phonological environment on morphological conditions. IA-models may be sufficient for purely concatenative morphology, but morphonological alternations leading to a lack of perceptual transparency already require an IP-model (Plungjan 2000: 72–73). Up to here, no paradigmatic relationship between (the variation among) the involved units need be assumed, but no later than with suppletion a notion enters the scene which forces us to assume paradigmatic relations.

WP models have usually been called "realizational", since their primary interest lies in the discovery of patterns of replacement (given some sufficiently defined contextual conditions) from among a set of forms which share some lexical meaning, but also match variation with regard to certain function(s). In order to disclose such correspondences the match between word form variation and function variation must be predictable (to some minimal degree), and the more regular the pattern is in formal expression, the easier it is to discern. More traditional WP-varieties have discovered such matches on a descriptive level, while more recent WP-varieties move further by demonstrating that members of a paradigm have different weight, since some of them betray a higher degree of reliability, so that on their basis one may predict other members. That is, paradigms are often asymmetric, and parts of them are interdependent in that they provide cues for implications concerning the structure of the entire paradigm. WP-varieties which focus on these relations can be called "implicational". They show that paradigms supply structures which should be investigated from the point of view of information theory (and discriminative learning); cf. Blevins (2016).

Remarkably, neither the mainstream of "constructive" models nor the many varieties of "abstractive" models are very explicit about what they take to be a lexical unit, and how such units are to be identified (irrespective of form). Simultaneously, discussions about adequate models of morphology, or morphosyntax, have circled around inflection (or what is considered to be inflection), and the application of the proposed models is usually considered to be problematic for (whatever is considered) derivation.<sup>15</sup> The reason appears to be that derivation traditionally denies the lexical identity of the involved word forms (see Section 6). Regardless of this, apart from the attention paid (or not paid) to paradigmatic relations, a main difference between abstractivist and constructivist thinking is the format of the units that are assumed to be basic (words or, maybe, also stems vs

<sup>15</sup>Spencer (2013) seems to be an exception. Characteristically, Spencer (2020, forthcoming) argues for a tight mutual connection between the notions of "paradigm" and "lexeme" without "bothering" too much about an inflection-derivation divide.

### 4 No paradigms without classification

morphemes, including roots) and the direction of analysis ("from wholes to parts" or "from smaller to larger units"). As for these basic assumptions, WP sides with CG.

After all, IA-, IP- and WP-models (differentiated for realizational and implicational varieties) can be ordered on a gradient (Blevins 2016: 14–17, also Plungjan 2000: 71–78), and probably none of them is able to lay claim to an adequate picture of linguistic reality covering all types of form-function variation in the morphosyntax of all languages. Consequently, there is per se nothing bad combining theoretical premises and analyses, provided the following alternatives are equilibrated: are grammatical oppositions (or: categorial oppositions in morphosyntax and/or on discourse level) better inferred from the combination of distinct units and rules of their combination? Or are they captured more adequately by a hierarchy of schemas and patterns for which units of lower formats are inferred via analogical proportions and replacement conditions for slots (see Section 6)? This includes the question whether the "output" of combinations is transparent (= compositional) or not, but even more so two other things: first, one must define the format of the units that may be combined and, second, one needs to understand what triggers the grammatical (or, more broadly: categorial) opposition, i.e. which syntagmatic and/or paradigmatic cues are responsible as reliable indicators (or even predictors) of matches between oppositions in form and functional distinctions. Such considerations provide the background for the following subsection.

## **4.3 Units, choices, and conditions of replacement**

To resume, the derivational patterns relevant for Slavic aspect (Section 2.1) yield stems belonging either to pfv. or to ipfv. aspect by virtue of opposite sets of functions and constraints with complementary distribution (Section 2.2). On the "morphological surface" the patterns look very heterogeneous. This is not a problem for WP- or CG-models, since they focus on schemas and correspondences. However, as was concluded in Section 3, stem derivation only provides the "input" for templates of larger constructions in which aspect assignment is not "visible" as such, since it depends on the morphological and lexical relatedness of the given stem to other stems, and it is only this relation and the membership in one of two classes (defined via opposite sets of constraints and functions) by which a binary contrast arises. This contrast is highly abstract, both in form and function. Lets therefore have a closer look at schemas and correspondence rules used in CG and WP and ask whether they can implement this contrast. See first

the word-schema in (18) above and the related correspondence rule in (19), the latter replicated here for convenience:

$$\begin{aligned} \text{(20)} \quad & \left[ \begin{smallmatrix} \text{/}X/\text{\textasciic} \\ \text{\textasciic} \end{smallmatrix} \right] \longleftarrow & \longmapsto \left[ \begin{smallmatrix} \text{/}Xz/\text{\text{\textasciic}} \\ \text{\textasciic} \end{smallmatrix} \right] \end{aligned} $$

Here two segments are assumed: a unit bearing some lexical meaning (= *X*), and a unit indicating a value from some grammatical category (= *z*). The latter notion may be extended to any kind of categorial distinction (or opposition). In this example, the complex expression *Xz* is classified as a form of a noun; *X* may be considered a stem. Now, regardless of whether *z* indicates some value of another categorial opposition as well (i.e. whether it is a portmanteau morpheme) or not, it acquires some paradigmatic value if it stands in a replacement relation with another segment (say, *y*) marking, for instance, that a word form *Xy* denotes a duality of *X*s. Lack of either *z* or *y* in a word form based on *X* can then be interpreted as the denotation of a single entity, provided either *z* or *y* otherwise always appear if a plurality or duality of *X*s is referred to. This is how the standard analysis of a simple paradigm of number of nouns goes, in a simplified manner and abstracting away from possible differences in phonological realization (alias allomorphy), but implying obligatoriness.

Note that in the last example the "identifier" of the grammatical function (*x*, *y* or zero) belongs to a subword level. But nothing changes in principle if a correspondence rule as in (19) is applied to units of other formats, for instance to words as possible parts of clause frames, e.g. German modal particles in declarative or interrogative sentences, as in the following made-up example (21b):

	- b. Ich habe *ja* schon abgewaschen. 'I have {ja} already done the dishes.'

By using the particle *ja* the speaker reminds the interlocutor(s) that the information ('I have already done the dishes') should be known to them. The particle thus functions as a signal that the speaker assumes the content of their message to be presupposed in the communicative space shared with the interlocutor(s). Diewald (2006) and Diewald & Ferraresi (2008) show that this function is a general property of German modal particles16. In addition, these particles mark the

<sup>16</sup>Recently, Panov (2020) has demonstrated that such particles ("enimitives") are better characterized as means to frame a proposition as uncontroversial. However, this slight shift in functional definition has no impact on the argument pursued here.

### 4 No paradigms without classification

utterance as non-initial in a discourse, which seems to follow by implication from the presupposition assumption. On this basis, it is easy to create a correspondence rule in which a modal particle (mp) contributes this kind of meaning to a very general type of construction, namely to sentences (<sup>s</sup> ) which denote propositions (*prop*) or actions (*act*),<sup>17</sup> and marks them as non-initial. The form of such a correspondence rule would be analogous to (19); see (22), ⊃ means 'implies':


Analogous correspondence rules could be created for the meaning contribution of modal auxiliaries or of any other modifiers on the level of predication, clause or sentence. Especially if their set is small, they may enter into replacement relations and, thus, form a paradigm, at least in a loose sense.

There is, of course, a difference between the "morphological" example to which the schema in (19) applies, on the one hand, and modifiers on higher levels of constituency, on the other. On the morphological, i.e. word level, those parts which indicate some grammatical value (*x*, *y*, *z* in the example above) often become obligatory in a strict sense. For instance, in German, English or Russian nouns cannot remain unspecified for number (in contrast to other languages, e.g. Turkish), even if arithmetic count proves insensible (e.g., with mass nouns). As a consequence, lack or change of some phonological segment in the word form is indicative of some value in the relevant functional domain (here: '1' vs '>1' or, if there is a dual, '1', '2', '>2') or reinterpreted in accordance with whatever number marking means in non-trivial (i.e. non-arithmetic) usage contexts. In structuralist terms, this reasoning implies an equipollent contrast, while privative contrasts would not yield such effects, i.e. the functional value (trivial or non-trivial one) remains just unspecified. In "constructivist" approaches this reasoning leads to the postulation of zero morphemes. However, as a rule of thumb, the larger the format of constituents above word level, the worse the conditions get for postulating zero marking, and this mirrors the decrease in strictness with which one can speak of paradigmatic tightening and with which slots (within units of larger format) can be discerned. Consider modal auxiliaries (as Engl.*can*, *may*, *must*): intraparadigmatic replacement conditions between them may become tighter, but nonetheless it is difficult to impossible to pinpoint syntactic conditions which

<sup>17</sup>Some modal particles are also used in directives, e.g. *doch*: *Mach (doch)!* ≈ 'Just do it, will you!'.

make their use compulsory. One may possibly formulate communicative conditions under which modal auxiliaries are more likely to appear,<sup>18</sup> but this does not lead to the same level of predictability which can be observed with for the explicit distinction of many of the categories that are marked on words. So-called analytical inflections (e.g., periphrastic tense or mood forms) are no counterexample to this consideration, they rather confirm it, because their "auxiliary words" fill out positions in paradigms for functions that are expected beyond some communicative needs.<sup>19</sup> Expectability drawn simply from communicative needs would still leave some freedom of choosing the categorial distinction as such to the speaker.

Thus, even if transparadigmatic tightening for clause- or utterance-level modifiers may occur it normally still leaves the speaker some leeway,<sup>20</sup> including the possibility to mark the categorial distinction by some other means. For instance, Diewald argues that German modal particles are compulsory for the function described above and that, consequently, the contrast between (21a) and (21b) implies that (21a) does not convey this function (since it is, as it were, zero-marked). However, although modal particles are a very convenient, frequent and, thus, expectable way of expressing the speaker's presupposition in German, it seems too categorical to deny utterances without modal particles the possibility to induce such presuppositions, e.g. just by intonation, i.e. to deny that utterances without modal particles can be used if the speaker wants to express such a presupposition.<sup>21</sup>

After all, regardless of how this issue may be further settled, "morphological examples" like number marking on nouns and modifiers of higher levels of constituency have one thing in common despite all other differences: they can be spelt out by pointing to distinct elements (traditionally called morphemes, words, etc.) or at least to constructions (with different complexity of constituency). This sets them apart from obligatoriness conditioned by the choice of verb stems, which in Slavic languages is inevitably connected to aspect; this choice is binary and, apart from a limited amount of biaspectual (or anaspectual) stems, one cannot circumvent it. The latter property, namely: lack of transparadigmatic variabil-

<sup>18</sup>Compare the distinction between language-internal and communicative obligatoriness made by Diewald & Smirnova (2010: 5).

<sup>19</sup>Moreover, slots filled by periphrasis usually build on already established paradigmatic relations between "synthetic" word forms (Haspelmath 2000, Plungjan 2011: 61–66, Popova & Spencer 2015).

<sup>20</sup>To continue reasoning in structuralist terms, one may say that a privative opposition (marking vs not-marking) has not yet turned into an equipollent contrast.

<sup>21</sup>Modal particles are a bad example also for the reason that, at least in German, they can be combined in the same utterance; they are thus not even organized in stricter paradigmatic replacement conditions.

### 4 No paradigms without classification

ity, is similar to number on nouns (in Slavic, Germanic, and Romance languages), but different from what can be found with, for instance, modal auxiliaries and arguably with modal particles as well. Concomitantly, it is not bound morphology as such that triggers the distinction between aspect, but derivational patterns (i.e. the relation between two or more stems) correlated with sets of functions and restrictions.

Moreover, the affixes which relate verb stems to each other attach prior to any tense or person/number marking (or suffixes deriving non-finite forms). That is, although the patterns work on a subword level, it is not the simple concatenation of morphemes that is crucial but the relation between stems, which is intermediary between word and morpheme. In addition, one has to account, first, for the degree of lexical closeness between the involved stems and, second, often also for the step at which the respective affixes attach in a derivational chain (see Section 2.1). That is, what is implied by these patterns cannot be captured simply as constructions or by correspondence rules of the type illustrated above. Rather, templates are needed which, to a certain extent, include rule-based information about, first, the procedure of how to "get" one stem from another and, second, about alternations in phonological form (which are frequent and predictable, for instance, at the border between stem and imperfectivizing suffix). However, regardless of whether one allows for rules (operating on stems, not on words) or relies only on schemas, it has to be admitted that, in analogy to (19), what corresponds to *Xz* cannot be subdivided into a part which specifies the aspect value (pfv. vs ipfv.) and a part which bears the lexical meaning.<sup>22</sup> That is, neither an abstractive (WP, CG) nor a constructive (IA or IP) approach brings us to the ultimate goal; one seems better advised to combine elements of both (see Section 4.2). What the *Xz* corresponds to is most often only infinitive or present/non-past tense stems ("allostems"), e.g. Russ. *pisa*- or *piš*- 'write.ipfv' or Pol. *przepis-ywa*- or *przepis-uj*- 'write anew.ipfv'. The same applies if something precedes *X*, i.e. a prefix, as in *na-pisa*- vs *na-piš*-.pfv (see 1–2). Moreover, even if the notion of allomorphy is acknowledged in one's analysis, it is appropriate for morphonological changes that distinguish the present (or non-past) from the infinitive stem, but entirely inappropriate for variability in the choice from 15+ verbal prefixes, which in most cases change lexical meaning, but also may serve to simply indicate that the unit is a pfv. verb (see Section 2.1). That is, a correspondence rule like /X/ ('*x*') ↔ /pfxX/ ('pfv. of *x*') would apply only to a limited number of aspect pairs and triplets; it does not reflect the dual character of verbal prefixes as changes of (partial) word-forms that indicate pfv. aspect, but in most

<sup>22</sup>Or, in analogy to (22), into a part denoting propositional content and its modifier.

cases also modify the lexical meaning (which jeopardizes the relation between *X* and pfx*X* out of a paradigm of forms which realize the same lexical unit). Nor does it help to understand in which cases, nonetheless, pfx*X* just means 'pfv. of *x*' (as with Natural Perfectives), among a couple of other issues.

Nothing changes for all these considerations if, instead, a more complex correspondence rule is applied, e.g. a schema as in (15). That the applicability of correspondence rules (or schemas) may depend on the format of the units which, in some way or another, have to be assumed on a subword level. They cannot simply be transferred from a morpheme level to stems in particular if the change of the stem itself carries information about the value of the grammatical opposition (pfv. vs ipfv.), i.e. subtractive of tense and agreement or non-finiteness markers, which are only added to these stems.<sup>23</sup> In addition, the aspect value hinges on patterns of stem changes that are not unified: not only are there two predominant patterns with different directions of derivation (see Figure 1), but many idiosyncratic ones (mostly obsolete remnants of earlier layers of stem derivation); there is even one pattern which is based on a monofunctional suffix creating pfv. stems, but only in the confines of a specific semantic class of atelic stems (or lexemes, for that matter), namely semelfactive {nǫ} (see Section 2.1 and Section 3).

Even from a strict "constructivist" point of view it would be totally off the mark to try to subsume such a variation of patterns on the "morphological surface" under allomorphy (and presumably nobody has tried to do so). For "abstractivists" the problem differs, namely: can a common paradigm for some pfv. and ipfv. stem be imagined, provided there is reason to assume that they share an identical lexical concept or are even close synonyms? This problem cannot be tackled with correspondence rules. Therefore, it seems reasonable to consider whether paradigms may be based not so much on the form of particular stems (and of how they are composed from morphemes) and not too much tied to specific patterns of stem derivation, but, instead, primarily built on categorial restrictions on different levels of morphosyntax and discourse which yield a sufficiently reliable distribution, or patterns, of oppositions tied to the choice between pfv. and ipfv. stems. These patterns may be described with templates, but in a form which at present hardly anybody would want to call "constructions".

How can these insights be exploited for modeling the architecture of Slavic aspect? And, conversely, what can be gained for morphological theory, or a theory of grammatical categories? The last two sections explore these questions.

<sup>23</sup>Even from a constructivist perspective, one would not say that an unprefixed ipfv. stem has a "zero prefix" by which imperfectivity is marked, or conversely that prefixed pfv. stems have a "zero suffix" since their ipfv. counterparts are often marked by an extra suffix (at least, I am ignorant of any such attempt).

4 No paradigms without classification

## **5 Extended notion of paradigms: A proposal**

The proposal rests on two pillars. First, aspect choice is obligatory, and since this choice amounts to making a decision between verb stems it is these stems which are in paradigmatic opposition for an abstract feature, namely PFV:IPFV. Then, second, the question is what determines this choice in the first place. As explained above (mainly in Section 2.2), verb stems are ascribed pfv. or ipfv. aspect according to sets of functions and grammatical constraints over which they distribute in a (more or less) complementary manner. By virtue of these sets verb stems belong into one of two opposite classes, so that a binary opposition is established. These conditions justify considering Slavic aspect as a classificatory category. By the same token, issues like how many aspect pairs (or triplets) there are, and how regular the morphological relations between lexically close stems are, are important only to the extent that there must be some backbone of the system in order to (i) supply regular patterns of derivation to be applied productively, and in order to (ii) establish paradigmatic replacement conditions between stems denoting identical (or very close) lexical concepts, i.e. to create minimal pair conditions (as exemplified throughout this paper). This backbone in terms of formal patterns and of lexical relatedness provides the basis for analogical transfer, both for the productive application of rules (or schemas) and for relating stems of opposite aspect with obsolete or less frequent patterns, briefly: for the integration of new and old stems into a system which distributes them over two classes defined via sets of constraints and functions.

Consequently, there is a maximally simple paradigm to start with, which is conditioned by an inevitable binary choice: either a member of the ipfv. class or a member of the pfv. class, whatever other grammatical categories might be expressed by a verb, and in whatever discourse context. Whenever a verb is involved in a categorial (grammatical or pragmatic) contrast, this influences aspect choice – since speakers cannot avoid it. The associations between aspect choice and the value of the contrast are reliable (and predictable) to different degrees, so core and peripheral (or stronger and weaker) conditions (or factors) may be distinguished. Concomitantly, there is no general rule saying that a particular morpheme indicates pfv. or ipfv. aspect as such. Thus, the aspect value is a factor that should be accounted for not only as a distinct element of constructions, but as something that can be visualized with templates, which I take to be sets of properties ordered by levels, or components. This distinguishes templates from constructions, or schemas, which are primarily characterized by their syntagmatic "outfit" and which normally lack a complex structure of levels (see the

representation in (15) and the correspondence rules above). By contrast, the templates require up to five components:


Components (i) and (ii) are indispensable, since they are always specified morphologically (as any verb form in Slavic). Components (iv) and (v) are also indispensable, but only in the sense that these properties are inherent to any utterance, even without explicit specification. Elements of component (iii), in turn, are optional. Simultaneously, (ii) and (iii) represent distinct linguistic material (as units on a word and subword level) which interfere with aspect (as inherent to the stem) on a syntagmatic axis; (ii) and (iii) can specify parts of larger constructions (on predication or clause level) accessible for a description in CG terms or for correspondence rules in a WP fashion. The other components are non-distinct and, in this respect, abstract. After all, each of the components itself implies paradigmatic contrasts (between forms or functions), but the assignment of aspect, i.e. (i), provides the basic binary paradigmatic distinction.<sup>24</sup> In this sense, aspect choice is like a pivot, since in combination with the other components it participates in the formation of minimal pair contexts (part of which was illustrated in Section 2).

A template can be created for each categorial distinction for which aspect choice is a sufficiently reliable indicator, or by which it is restricted; the other factors which also contribute to this distinction (or condition the restriction) are listed as components (ii-v). The "nature" of the categorial distinction is used as a label of the template (maybe together with the language or subdivision of Slavic

<sup>24</sup>For this reason no additional level has to be assumed (contrary to what one of the reviewers suggested): by the choice of ipfv. or pfv. aspect the opposition to the other aspect (= grammatical class) is determined ipso facto (*tertium non datur*). See also the peg-metaphor below.

### 4 No paradigms without classification

to which it applies); it is, as it were, the minimal common denominator for the contrast conditioning aspect choice. The symbol "—" means that no specification is required, properties in parentheses are optional, additional information is given in brackets.

An illustration of a relatively simple template would be the formation of the future tense of ipfv. and pfv. stems in North Slavic; see Table 1.



Here, aspect choice bears on the choice of grammatical forms and their interpretation, regardless of which functions might be associated to pfv. and ipfv. future; that is why (iv) and (v) are left unspecified.

For South Slavic the situation changes insofar as non-past pfv. stems do not yield a default future reading. Instead, a distinct future marker (Bulg. *šte*, Mac. *ќe*, inflected Srb.-Cr. *ću* and Sln. *bom*) combines with either aspect and the nonpast of pfv. stems is tightly associated with irrealis functions (first of all habitual, modal, conditional); it must be combined with distinct irrealis markers, among which for most of these languages *da* is the predominant one.<sup>25</sup> Thus, for South Slavic non-past pfv. stems the template is as in Table 2.

Another minimal pair contrast, widespread all over Slavic, is aspect in the imperative +/− negation. See (8a–8b), adduced in Section 2.2 and replicated here as (23), and the corresponding template in Table 3.

### (23) Russian


<sup>25</sup>Alternatively, one might say that future belongs to the irrealis domain and that, correspondingly, the future morpheme itself marks irrealis. The consequence would be that South Slavic does not have future marking, or that future is but a standard (or generalized) implicature of non-past + the respective irrealis marker. This, however, would not change anything essential in the template (see Table 2).

Table 2: Non-Past in South Slavic


Table 3: Directive speech acts


For the ipfv. imperative (with negation) the referential condition is given in brackets because prohibitives do not imply the exertion of social force for single occasions; they can be (and often are) uttered as a general command (e.g., *Don't eat with your fingers*). However, provided the directive speech act refers to a single situation (with a concrete illocutionary concern), pfv. and ipfv. stems are in an ideal paradigmatic replacement condition, and this applies to virtually all Slavic languages: the illocutionary background (a deontically, i.e. socially motivated directive speech act) does not change, only negation makes the difference and "switches" the aspect.

A complication arises inasmuch as the negated ipfv. imperative is used for other purposes as well (see Section 2.2). A similar point holds for unnegated ipfv. imperatives which, among other grammatical forms of unnegated ipfv. stems, are employed to signal that the speaker assumes the relevant action to be presupposed (also by the interlocutor). Other grammatical contexts without negation in which ipfv. stems are associated with this discourse function are modals

### 4 No paradigms without classification

(compare ex. 10a–10b) or the future tense. Slavic languages obviously differ as for the prominence of this function, but it anyway often interferes with other functions associated with ipfv. aspect. An analogous point concerns event-external pluractionality, in particular the denotation of unrestricted repetition, which systematically conflicts with actional defaults and the limiting function of the pfv. aspect.<sup>26</sup>

Furthermore, as templates like Table 2 show, one of the two aspects may not require any further specification, i.e. its use is rather unrestricted, while the other aspect is subject to quite severe restrictions. Such an asymmetry also holds true for more complex cases, as with negated imperatives in which the employment of pfv. stems requires very specific conditions (which, in addition, may be more salient only for a particular subarea of Slavic) that are not visible just "on the surface" (see Wiemer forthcoming). In other clear cases of asymmetry one of the aspects is altogether blocked, not because of some specific (and shaky) context conditions, but for a more straightforward reason. This is the case with aspect in the scope of phasal verbs<sup>27</sup> which, as mentioned in Section 2.2, allow only for ipfv. stems in any Slavic language (except colloquial Upper Sorbian). The template looks, therefore, like Table 4; \* symbolizes blocking.


Of course, the blocking of pfv. aspect by phasal verbs is motivated, as the general meaning of pfv. aspect consists in setting limits, and this meaning conflicts with the semantics of phasal verbs. In fact, there is reason to argue that most (if not all) of the functional contrasts and constraints on aspect choice are motivated from the basic categorial distinction between setting (or foregrounding)

<sup>26</sup>This conditions an inner-Slavic differentiation of the factor hierarchy relevant for aspect choice (cf. Dickey 2000: Ch. 2, Wiemer 2008: 399–403, among others).

<sup>27</sup>Note that phasal verbs themselves distinguish aspect, i.e. most of them come in pairs, so that their own aspect is indicative of the same functions and constraints as for other verbs.

limits of situations (→ PFV) and backgrounding them (→ IPFV), and only part of the contrasts and constraints can additionally be motivated by telicity.<sup>28</sup>

There is no place (or need) to continue with illustrations of what templates might look like if one wants to capture not only the formal properties of constructions and involved verb forms, but also their functional interpretation for different types of utterances. Hopefully, the general idea has become clear. The crucial point is that every template is based on a choice between pfv. and ipfv. stem; this choice cannot be avoided once verbs are involved, and all properties from (ii) to (v) are "linked" to a pfv. or ipfv. stem (= (i)) like pieces of clothing hanging on pegs, either as unequivocal decisions or as salient tendencies. These pegs (= ipfv. vs pfv. stems) provide the basic paradigmatic contrast, regardless of whether the function (or constraint) concerns (plur)actionality features or bears relevance to temporal deixis (as with the default present > future shift for pfv. stems in North Slavic), or the functional contrast is at best remotely related to these core domains of aspect oppositions.

Conflicts between these factors are unavoidable since the pfv.:ipfv. opposition supplies only a binary choice, and since sense has to be made out of this choice even if actionality or pluractionality features are irrelevant or remain in the background. However, analogous conflicts arise with other binary oppositions and obligatory choices (i.e. if transparadigmatic variability is minimized or lost), such as singular-plural distinctions for nouns in most European languages or a definite article, e.g. in Balkan Slavic. The difference, again, is that these categories (and the corresponding paradigmatic contrasts) are not marked by stems (as is Slavic aspect).

The templates are able to integrate correspondence rules (or schemas), if aspect choice reliably hinges on some syntagmatic condition, for instance on some contextual element like *bud*- for the ipfv. future in Russian or Czech, or on an irrealis marker (like *da*) for pfv. present in Balkan Slavic. However, since simple constructional approaches are unable to capture the classificatory properties of Slavic aspect arising from sets of functions and constraints, and since this opposition is morphologically based on different derivational patterns, only templates can do the job of relating the paradigmatic opposition of pfv. vs ipfv. stems to the functional contrasts and grammatical constraints which have been discussed in the literature on Slavic aspect and in a flashlight manner throughout this article.

This said, a further step can be taken by reinterpreting *sets* of such templates as members of paradigms of aspect choice. That is, *each template*, regardless of how complex its internal structure, *equals a paradigm cell*, but elements of its internal structure are linked to other layers of the overall paradigm. The entire

<sup>28</sup>In other words, telicity provides a condition for subsets of the inventories of functions and constraints associated to ipfv. vs pfv. aspect.

### 4 No paradigms without classification

paradigm would then consist of two layers (see Figure 4).<sup>29</sup> Stems (each with its aspect) distinguish gamuts of finite and non-finite forms which make up paradigms in the traditional sense, and these constitute the first layer. Whether one wants to deal only with the set of finite forms, or whether one includes also nonfinite forms (and which ones) into the paradigm, is of secondary concern. The smaller set of finite forms may be called "narrow" or "classical paradigm", the larger set which includes non-finite forms "extended paradigm", and the latter is still needed. The issue of periphrastic forms which fill out cells of traditional paradigms ("analytical inflection"), such as, e.g., the future of ipfv. verbs in North Slavic (person-number inflected auxiliary *bud*- plus infinitive of ipfv. stem), is of secondary concern, too. The reasons were discussed in Section 4.3.

The templates of the type illustrated above provide the second, more abstract layer. Component (i) of each template is the paradigmatic opposition between ipfv. and pfv. stem itself, i.e. the pegs on which everything hinges. As inherent to stems, it ties together all templates from the same column (= vertical axis). Component (ii) cross-references forms from the extended paradigm (= first layer), it thereby specifies which of these forms are relevant for the given template and, together with component (i), supplies a connection between first and second layer.<sup>30</sup> The basic structure of such a complex, two-layer paradigm is shown in Figure 4. As shown above, the relation between the pfv. and ipfv. halves of the templates may be asymmetric, either in form-related conditions or in the functional conditions applying to one of the stems in the paradigmatic relation.

Each column in toto is opposed to the respective other column, just as, for instance, in a traditional paradigm of inflected nouns singular and plural are opposed for the category number "across" morphological cases and for stems of different gender. Admittedly, this analogy is not perfect since a traditional paradigm of inflected nouns (or verbs) has a closed set of cells, while the number of templates specifying the conditions of aspect choice can be less easily limited; it increases with every grammatically or pragmatically definable contrast for which aspect choice proves relevant. However, the amount of templates hardly constitutes an open class, either, since these contexts cannot increase *ad libitum*; otherwise aspect choice would not be salient and reliable enough. Apart from this, there are other paradigms in "hard core" grammar whose closedness is debatable; consider, for instance, voice-related distinctions, paradigmatic relations

<sup>29</sup>This proposal remotely resembles Leino's "metaconstructions" (Leino 2022 [this volume]), as far as analogy is at play. However, whereas metaconstructions are conceived of as generalizations of constructions, two-layer paradigms as developed here are much more characterized by internal relations between particular components which depend on the binary choice between pfv. and ipfv. stems.

<sup>30</sup>How this might be done technically should be considered separately.

Figure 4: Complex, two-layered paradigms of Slavic verb stems. Broken line/arrow = cross-references with element(s) in a component (ii) of templates

### 4 No paradigms without classification

between preverbs (e.g., in Germanic or Latvian), or German modal particles (see Section 4.3).

Another objection might be that the templates are not mutually exclusive, as many of them are either interrelated (by common motivation), or they can conflict with one another (see examples above). This makes the inventories of opposed templates dissimilar to cells in traditional paradigms. However, recent work in WP-morphology on traditional paradigms has shown that paradigms are often asymmetric in that their members betray an unequal status, in particular. Because some of them are better predictors for others but not vice versa. In general, paradigm members are better characterized as a network (see Section 6). Such network relations can as well be found in the sets of templates and the complex two-layer paradigms of Figure 4. Therefore, the architecture of traditional paradigms and the principles organizing complex two-layer paradigms do not seem that different. After all, it is an unavoidable binary choice between pfv. and ipfv. aspect which makes this network arise and stabilize.

Now, if such complex, two-layered paradigms can be acknowledged, what follows from this for (morphologically related) stems of opposite aspect which are so closely related in their lexical meaning that they can be considered synonyms? This question arises regardless whether one speaks about pairs, triplets or larger groups of stems. Why not assume that stems united in this way under one lexical meaning actually share into one (though more complex) paradigm? And if the answer is positive, does this entail that these stems are to be considered representatives of the same lexical unit (= lexeme)? The last question arises because synonyms are usually treated as distinct lexemes, however, the synonyms under consideration here are morphologically related and show complementary distribution over grammatically and/or pragmatically defined contexts.

As pointed out in Section 3, the assumed 1:1-relation between lexeme and paradigm has forced many to interpret the different stems as inflection, with diverse artificial and ad hoc "solutions". Alternatively, if treated as a classificatory system, each stem can be ascribed its own paradigm, but this alternative is based on the same 1:1-assumption between lexeme and paradigm. Notably, nothing changes with suppletion, since suppletion itself presupposes tight paradigmatic relations and extreme closeness of lexical relatedness. Actually, suppletive aspect pairs force us to acknowledge that the grammatical value (pfv. vs ipfv.) is a property of the stem<sup>31</sup> (cf. Wiemer 2020b: 149–150). Therefore, as concerns stems that

<sup>31</sup>As a reviewer remarked, since both WP and CG treat word forms as wholes, they can deal with suppletion, syncretic and analytic forms in the same way. While this is correct, it should be remembered that both syncretic and analytic forms, as well as suppletive aspect pairs, presuppose stems. Thus, as far as aspect in Slavic is concerned, stems are more basic than anything else.

are morphologically related and can be considered lexical synonyms, but show complementary grammatical behavior, I do not see any inherent reason why one should not admit for complex, two-layered paradigms composed of the extended traditional paradigms and sets of templates assigned to two related stems. These stems can each be considered separate lexemes which complement each other. Thus, instead of either of the representations given in Figure 3 (see Section 3), consider the one in Figure 5.

Figure 5: Alternative view on the grammatical status of aspect pairs

As for aspect triplets (of either kind discussed in Section 2.1), the only thing to be admitted further is to include a third stem. For the competing pfv. or ipfv. stems one may observe different biases for subsets within combinatorial restrictions and/or the function inventory, i.e. for properties specified in templates.<sup>32</sup> The same reasoning can be extended to actionality groups consisting of stems of either aspect, as argued for by Tatevosov (2016); see Section 3.

The assumptions underlying Figure 5 also do justice to the classificatory character of the PFV:IPFV opposition. Simply those verbs for which no morphologically related stem with a close enough lexical meaning exists (*perfectiva* and *imperfectiva tantum*), certain parts of the paradigm, defined via word forms (= traditional paradigms), constraints and functions are absent. This can be compared to analogous cases, for instance to *pluralia* and *singularia tantum* (or nouns with a strong bias to either singular or plural in non-arithmetic contexts) in relation

<sup>32</sup>Compare, for instance, the usage-based case study in Wiemer et al. (2020), which shows that Pol. *dzielić*ipfv1 and *rozdzielać*ipfv2 (pfv.*rozdzielić*) and their Czech cognates *dělit*ipfv1 and *rozdělovat*ipfv2 (pfv. *rozdělit*) 'divide, separate' have different preferences for stative (IPFV1) vs habitual (IPFV2) contexts.

### 4 No paradigms without classification

to the values of number in nouns, or with the paradigms of verbs that do not passivize in a language with an otherwise fully developed distinction between active and passive voice. Like inventories of word forms, function inventories and sets of combinatorial restrictions (i.e. the "ingredients" of templates) are defined for a maximum range of pfv. and ipfv. stems, but they do not (and cannot) apply to every single representative of these classes; rather they are like repositories supplying admissible functions for these representatives.

One final question, alluded to above, remains. The templates which constitute the second layer of the complex paradigms can be listed; but is there some internal order between them? In particular, can some templates be regarded as more important inasmuch as they serve as predictors for other templates? Similar key functions of members in paradigms have attracted attention particularly in most recent implicational varieties of WP (see Section 4.2). It would be instructive to learn whether there exist implicational relations between templates relevant for aspect choice that parallel (or are analogous to) such asymmetries in traditional paradigms, or between constructions that are organized more tightly in paradigmatic terms.<sup>33</sup>

In fact, some templates relevant for aspect choice are certainly more important, either because the functions and/or constraints which they capture are more frequently encountered and less restricted by lexical meaning (e.g., most pluractional meanings are rather insensitive to actionality properties, including telicity), or because they are more firmly associated with their context conditions (e.g., with communicative intentions) or more difficult to suppress by conflicting conditions than constraints and/or functions specified by other templates. These constraints and functions would play a more central role in the grammar. However, relations between them are often hard to pin down, let alone to quantify. Of course, we may start with certain "hard core" constraints like, for instance, the compulsory use of ipfv. stems instead of pfv. counterparts in the narrative present tense in East Slavic and Polish, in contrast to, e.g., Czech or Slovene (see Section 2.2). But then the problem is whether such factors of aspect choice correlate with others like, for instance, the restriction of the "inchoative" future (with bǫd-) to ipfv. stems (typical of all North Slavic languages), or functional contrasts of aspect choice in the scope of modals (which show an overlay of different contrasts that can easily conflict with each other). Finally, even if sufficiently robust correlations can be disclosed, the question arises in which direction the implication goes (or whether it is bidirectional).

<sup>33</sup>I am unaware of any attempt in CG to disclose implicational hierarchies, or asymmetries, among constructions (apart from standard assumptions about inheritance relations; see Section 4.1).

These questions are intriguing also for the reason that implicational relations between members of traditional paradigms seem to be arbitrary in the sense that no semantic motivation can be found; they usually just serve as an internal principle of the organization of paradigms,<sup>34</sup> possibly conditioned by the reduction of conditional entropy (Milin et al. 2009, Blevins 2016: 7). Conditional entropy is connected to expectability which ensues from the relation between frequency patterns of paradigm members. Even if some day it might be possible to describe conditional entropy for factors that influence aspect choice and can be captured by templates, one would certainly expect these factors to be related by semantic motivation (including communicative purposes). Therefore, contrary to asymmetries between cells of traditional paradigms, asymmetric relations between templates describing conditions of aspect choice are obviously of a different nature.

This brings us to the conclusions.

## **6 Conclusions and outlook**

The proposal made in Section 5 amounts to extending recent reasoning in implicational WP models to the properties of a stem-derivational aspect system. As shown in Section 2, productive stem derivation is only the morphological prerequisite for the formation of the Slavic PFV:IPFV opposition as a classificatory category. Basically, extending WP-reasoning to this case amounts to a transfer of paradigmatic order from the word level to the level of templates for constraints and functional oppositions connected to the choice of morphologically related stems. Blevins (2016: 75) states:

Treating paradigms as fundamental units of grammatical organization conveys the same kinds of advantages as treating words as the basic grammatical signs. Just as words may have properties that cannot be assigned to their parts, sets of words may express information that cannot be associated with individual words.

The analogy with Slavic aspect becomes clear if, in this quote, one replaces *words* by *stems* and adds that sets of templates may express information that cannot be associated with individual stems as well. This is obvious particularly for stems of opposite aspect that can be organized in pairs, triplets or actionality groups, since they are able to function as synonyms for different grammatical and

<sup>34</sup>Cf. Haspelmath & Sims (2010: 172–174) on Priscianic formations.

### 4 No paradigms without classification

communicative purposes, and the patterns of their morphological relatedness are regular to an extent that they can serve as a basis for analogical transfer between stems with less transparent and/or obsolete morphological ties.

The analogy with WP-models becomes even more straightforward with the next quote from Blevins (2016: 80) about the formation of conjugations and declinations conceived of as classes:

In the same way that stems are not basic units in a classical WP model, but are instead abstracted from a set of forms, classes [of conjugations or declinations; BW] are not "properties" of items but are abstractions over sets of paradigms that exhibit congruent patterns of form variation. The class of an item is exhibited via characteristic patterns of form alternation.

Here it suffices to admit that stems may be the basic units conveying grammatical information (i.e. pfv. vs ipfv. aspect) and to rephrase as follows: classes of pfv. and ipfv. stems are not properties of particular stems but are abstractions over sets of templates that exhibit congruent (i.e. consistent and predictable) patterns of mutual replacement for stems of the opposite aspect. The class to which a stem belongs is exhibited via characteristic patterns (or: sets) of templates which capture constraints and functions.

This amounts to saying that classes which constitute pfv. and ipfv. aspect in Slavic are more abstract, but also more important, than conjugational or declensional classes. Although the latter are of a rather formal nature, their interference with various levels of grammar and pragmatically motivated distinctions on utterance level is considerably lower (or even absent) in comparison to the far-reaching consequences that follow from the choice of a pfv. or ipfv. stem in Slavic languages. Inflectional paradigms represent an extreme case of predictive patterns, but they also represent a simple (probably the simplest) case of such patterns. Paradigms provide speakers with a "maximally reliable analogical base for deducing new forms based on previously encountered forms" (Blevins 2016: 12). While this applies to productive derivational morphology as well, this kind of analogical base would concern only the morphological prerequisite, so to say: the stem-derivational mechanism which is necessary, but not sufficient to explain the architecture of Slavic aspect. The analogy supplied in this case relates not simply to new forms, but to a fundamental paradigmatic contrast based on the class membership of verb stems to pfv. or ipfv. aspect, as argued for throughout this article.

Furthermore, Blevins doubts that WP approaches are suitable to deal with derivational morphology. He argues that at least more traditional realizational

models "are less applicable to the variable structure exhibited by 'families' of derivational forms" (Blevins 2016: 159). His argument primarily relates to wordclass changing derivation (e.g., verb ⇒ agent noun), which usually does not allow for the delimitation of "a finite set of forms within a uniform feature space". Moreover:

Given a list of derivational processes active in a language, it is of course possible to assign a uniform family of "potential" forms to all of the members of a word class. Yet the uniformity achieved is deceptive, because it collapses a *critical distinction between those forms that are established in a language and those that are merely possible in principle*. (Blevins 2016: 159, emphasis added)

First of all, the caveat expressed here would be justified equally well for many complex inflectional systems, especially if periphrastic forms ("analytic inflection") are accounted for. Consider, for instance, the complex verbal paradigms of Bulgarian, the Kartvelian languages, French, or even English. So there is no real difference between productive inflection and productive derivation (however one may wish to draw the line), and this can be maintained all the more for derivation which does not change the syntactic class, as is the case for Slavic aspect.

More importantly, implicit to Blevins' point, highlighted in the last quote, is a main bone of contention between word-based ("abstractive") and morphemebased ("constructive") morphology, namely the rule-versus-list fallacy: "the unwarranted assumption that linguistic constructs are either generated by rule or listed" (Booij 2010: 4, with reference to Langacker 1987). Why shouldn't human beings be capable of doing both: to store some ready-made units in their memory and to apply rules by which more complex units are composed "on the fly" from less complex ones? CG attempts to integrate regular and transparent constructs into a "constructicon" of a given language, on condition that they are sufficiently frequent (see Section 4.1). In general, researchers unanimously agree that, on the one hand, units (of different formats, i.e. on word, sub-word and multi-word level) are probably stored because they are more frequently encountered and easy to isolate from their immediate syntagmatic environment. On the other hand, without the productive, *ad hoc*-application of rules it would be difficult to understand how new complex forms (among them many hapax formations) can arise, apart from the fact that postulating myriads of complex linguistic constructs to be stored in memory (= lexicon) is not only uneconomic in linguistic description, but seems to be inadequate from a psychological perspective.

### 4 No paradigms without classification

Therefore, the problem pointed out by Blevins is justified, but it rather begs the question of how to achieve an adequate equilibrium between potential and ready-made forms in one's model of linguistic activity. This includes the question whether a lexicon may consist of both words and morphemes. Haspelmath & Sims (2010: 70–74) give convincing arguments in favor of a "moderate wordbased model", which combines words and morphemes, although the former are given primacy over the latter. In practice, a similar consequence follows from Goldberg's (2006) definition of constructions (see Section 4.1), according to which a "mixture" of idiomatic (non-compositional) forms and of some (frequent) complex forms is regarded to be stored in the lexicon. Experience with Slavic aspect may add crucial insights to this discussion. First, the basic unit on which Slavic aspect operates is verb stems,<sup>35</sup> i.e. units of a format intermediate between words and morphemes. However, by far not all stems occurring in real discourse are registered in dictionaries, instead many are certainly not stored as ready-made units in speakers' memories, but construed on the spot, often remaining ephemeral. Thus, a "moderate stem-based model" is called for, which basically follows the assumptions of WP approaches, but constructivist elements are to be included inasmuch as sufficiently transparent morphology (e.g., with productive suffixes employed to derive ipfv. stems) and sufficiently obvious rules of concatenation are an issue. However, since systematic morphonological alternations can be observed not only in the relation between non-past and infinitive stems, but also between related pfv. and ipfv. stems, the rule-based part of the model should be rather of an IP- than of an IA-type (see Section 2).

Second, the backbone in the architecture of Slavic aspect nevertheless resides in the relations between well-established (and frequent) stems (pairs, triplets, actionality groups) which can be regarded as units entrenched in an active lexicon, notably both as ready-made units and as products of the application of rules or schemas. The derivational patterns and functional distribution between stems united around lexical meanings can be best described as schemas, however of a very irregular type. We thus have to abstract away from particular patterns on the surface and recognize the paradigmatic relations which hold for stems by virtue of their membership in one of two opposed classes (pfv. vs ipfv.). This property cannot be captured by usual schemas, or correspondence rules, and this is where CG and WP come to their limits (see Section 3). Instead, templates are called for, and this brings the WP-based approach advocaSection 2).

Therefore, following Blevins (2016), paradigms can be conceived of as limiting cases of network relations in which certain members show some predictive

<sup>35</sup>Rather infrequently, stems may coincide with roots.

value for other members of the network. In Slavic aspect the morphological form of these members can be predicted only to some extent, and one always has to consider patterns of derivation against lexical closeness (this concerns particularly the role of prefixes added to ipfv. simplex stems). However, the sets of constraints and functions characteristic of each aspect lead to paradigmatically tight oppositions with regard to classes, hard constraints are predictable from the interaction with other grammatical categories (e.g., tense), and the selection of functions by specific representatives of a class can be predicted with a certain reliability on the basis of the actionality class of the stem and an account of (sometimes complex) conditions of the current discourse.

To conclude, it is one issue whether CG- or WP-approaches become interested in pursuing the path proposed here, and thereby try to integrate the lesson told by the architecture of Slavic grammatical aspect. This would demand an application of paradigmatic structure on a more abstract and complex level than even in recent implicational WP-models, which basically have remained restricted to inflectional paradigms. This understanding of abstract paradigm structure also reaches beyond CG-approaches to paradigmatic structure, mainly defined via inheritance relations between different levels of complexity that is measurable in terms of elements belonging to a schema. Moreover, obligatoriness – as the opposite of high transparadigmatic variability – for Slavic aspect arises on a different basis than it does in word-based or construction-based descriptions.

Regardless of whether such an extension of paradigm structure will be accepted in the mentioned theoretical frameworks, a complementary issue should be pursued. Namely, more should be learned about internal implications between (templates describing) constraints and functions relevant for aspect choice. Such an examination would greatly increase our understanding of the architecture of this category in Slavic, and probably of classificatory categories in general. For this purpose, it is worth considering whether and how conditional entropy between different factors relevant for choice aspect might be determined.

## **Acknowledgements**

Research has been funded by German Science Foundation (WI1286/19-1) and Polish Science Foundation (2016/23/G/HS2/00922), programme *Beethoven II*.

### 4 No paradigms without classification

## **Abbreviations and symbols**

## **In glosses and other examples**


## **For languages and corpora**


RNC Russian National Corpus (https://ruscorpora.ru/new/)

## **References**


Booij, Geert. 2010. *Construction Morphology*. Oxford: Oxford University Press.

Booij, Geert. 2013. Morphology in construction grammar. In Thomas Hoffmann & Graeme Trousdale (eds.), *The Oxford handbook of construction grammar*, 255– 273. Oxford: Oxford University Press.




Wood, Esther Jane. 2007. *The semantic typology of pluractionality*. Berkeley: University of California.

Zaliznjak, Andrej A. 1967. *Russkoe imennoe slovoizmenenie*. Moscow: Nauka.

## **Chapter 5**

## **Recursion and paradigms**

## Tabea Reiner

### LMU München

This paper sketches the current status of morphology and paradigms in linguistic theorising. In particular, it is shown that from a constructionist as well as from a compositional perspective, morphology including paradigms tends to dissolve. The former might be less obvious; however the paper argues that a constructional deconstruction of paradigms and morphology follows directly from Haspelmath's (2011) take on Booij (2010) and related approaches in the realm of Construction Morphology (CxM). The latter is more obvious; in particular, proponents of Distributed Morphology (DM) regularly emphasise that morphology is but an interface and paradigms are epiphenomenal. Throughout the paper I assume some familiarity with Construction Morphology and Construction Grammar more generally whereas I introduce DM specifically. However, the paper is not intended as a thorough discussion of the approaches presented (nor do I take sides); rather it is their shared detachment from paradigms that is at stake here. Consequently, also what is sometimes called Autonomous Morphology is addressed in the paper: a rather recent approach that advocates morphology as an irreducible level of description and upholds the paradigm as a format of description in its own right. The balance of the paper is rather pessimistic for morphology and paradigms but eventually I come up with a presumably new argument in favour of regarding paradigms as fundamental: restrictions on inflectional recursion fall out naturally from them.

## **1 Introduction**

In the last decades, morphology as an independent level of description has come under pressure from two sides. In construction-oriented approaches it runs the risk to dissolve somewhere in the middle of the lexicon-syntax continuum: "it's constructions all the way down" (Goldberg 2006: 18). In certain compositionoriented approaches, in turn, its traditional tasks have been transferred to syntax:

Tabea Reiner. 2022. Recursion and paradigms. In Gabriele Diewald & Katja Politt (eds.), *Paradigms regained: Theoretical and empirical arguments for the reassessment of the notion of paradigm*, 127–171. Berlin: Language Science Press. DOI: 10.5281/zenodo. 5675847

### Tabea Reiner

it's "syntactic hierarchical structure all the way down" (Harley & Noyer 1999: 3). The last quotation is from Distributed Morphology, which does have *Morphology* in its name but refers to a mere interface between syntax and phonology by this (Halle & Marantz 1993: 114, Bobaljik 2017: 1).

Together with morphology, a certain format of description appears to be discarded, which has been central to at least what is traditionally termed inflectional morphology:<sup>1</sup> the paradigm. Throughout this paper, *paradigm* is to be understood in the following sense, unless noted otherwise:

[…] a set of cells; each such cell is the pairing of a word form with the set of morphosyntactic properties which that word form realizes. (Stump 2002: 147)

I consider this definition quite general, since, in principle, it does not exclude an incremental instead of a realisational approach (for an overview of morphological theories cf. Stewart 2016): after completion, the string can be regarded as a whole. Please note, however, that the definition given here differs from Politt's (2019) contemporary concept of paradigm in two respects. First, it is narrower in that it refers to merely word forms (the notion of word is a point to be discussed below); second it is broader in that it does not take a stance on whether paradigms are part of native speaker knowledge (another point to be discussed below).

To give an impression of how the discussion in the present paper is going to proceed, I pick out two quotations from the literature, which coincide in rendering paradigms superfluous eventually. First, consider Haspelmath (2011) on constructions:

Clearly, the form–meaning relationship is often straightforward and compositional, but it is also often more complex. For the latter cases, morphologists have used paradigms and realisation-based rules, and syntacticians have used constructional idioms. The similarity between realisation-based morphology and construction-based syntax has recently been emphasized especially by Gurevich (2006) and Booij (2010). As far as I have been able

<sup>1</sup>This is the kind of morphology on which the present paper focuses; however, most points before Section 4 carry over, mutatis mutandis, to derivational morphology: word formation can be modelled using paradigms (Hathout & Namer 2019) but it can probably be modelled just as well using constructions or rules, including special rules. On a related note, let me add a word on terminology: in the present paper, the root *deriv*- is sometimes used meaning ʻrelated to the formation of new words' and sometimes used meaning ʻrelated to composition' (especially when presenting DM); I assume that in each case the context suffices for disambiguation.

5 Recursion and paradigms

to determine, the differences between them mostly derive from different traditions, not from any substantive differences. (Haspelmath 2011: 59)

I conclude from this quotation that paradigms can be rewritten as constructions. Incidentally, the reverse is also true, provided that the definition of paradigm is relaxed to include all sorts of paradigmatic relations in a structuralist sense: imagine a micro-paradigm like *be all ears* vs. *be all eyes*. However, from a constructionist perspective, constructions are needed anyway, so it is the paradigm that one can do without. In practice, the rewriting of paradigms as constructions can be thought of as conceptualising every single cell from a paradigm as the small semiotic entity that it is. For example, Table 1 from German can be rewritten in the manner of Figure 1.

Table 1: Present indicative of German *sein* ʻbe'


Figure 1: Constructions instead of a paradigm

The rewriting can also be accomplished in a less space-consuming way (cf. Section 2.1.); however I use a partly Saussurean way of presentation here in order to highlight the fact that every cell is a sign. Needless to say, these signs (= constructions) may form networks among themselves, which in turn might equal paradigms. Crucially however, paradigms are no more than emergent from this perspective.

This coincides surprisingly well with the view on paradigms following from an otherwise very different theoretical perspective, i.e. from the perspective of Distributed Morphology (henceforth referred to as *DM*). Consider the following footnote in a recent paper.

[…], we use paradigms only for representational issues. As is well-known, in DM, paradigms are epiphenomenal. […] (Pomino & Remberger 2019: 473)

### Tabea Reiner

Thus, paradigms are not rejected as such but they are rejected as parts of mental grammar (which is the only area of interest in DM). The most detailed argument to that effect might be found in Bobaljik (2002). More recently, David Embick (2015) even managed to write a book-long introduction to DM without even using the term *paradigm*(*atic*) beyond the bibliography and one footnote (Embick 2015: 232). These researchers' view on paradigms derives directly from the architecture of grammar assumed in DM, which will be laid out in Section 2.2 of the present paper.

Summarizing for now, the theoretical significance of paradigms may be seriously called into question from otherwise very different theoretical perspectives. The present contribution aims to portray these positions in some depth (Section 2), explore a common defence strategy applied by Autonomous Morphology (Section 3), and eventually come up with one task in linguistic theorising that only paradigms appear to fulfil directly, i.e. delimiting recursion (Section 4). Hence, the key term *recursion* will not reappear until the last section.

## **2 No morphology, no paradigms**

## **2.1 From a construction-oriented perspective**

If the introductory example had not been from German but from Turkish or another language with considerably less fusion and suppletion than German (Aikhenvald & Dixon 2017: 4, Dressler 1985: 334 but also cf. Bacanlı 2011), both the paradigmatic as well as the constructional notation would have displayed forms that are systematically segmentable to a large extent. So the question arises whether such forms should be described by rules rather than by paradigms or constructions. The present section aims at answering this question from a constructionist perspective. To anticipate the answer, compositional word forms are to be described by constructions like compositional syntax is to be described by constructions.

For a start, consider the Turkish example to which Haspelmath (2011: 59) alludes, stemming from Hankamer (1989: 403): one Turkish verb root yields 1,830,248 different forms when counting (what would traditionally be called) inflection and (what would traditionally be called) derivation, not even allowing for iterations. Obviously, this is too much to write down in paradigms. And things get worse when iterations are eventually taken into account. (1) provides an example.

5 Recursion and paradigms

(1) Turkish (Hankamer 1989: 396, emphasis added) daya prop\_up -n rfl -ış rcp *-tır caus -t caus* -ıl pass -a pot1 -mı neg -yabil pot2 -ecek asp -ti tns -k agr<sup>2</sup> 'we might not have been able to be made to make someone else practice mutual aid'

When, starting from (1), one tries to imagine all other possible combinations of affixes, this example gives an impression of the enormous size and systematicity any Turkish verbal paradigm would have. So using paradigms just does not seem to make sense for Turkish verbs.<sup>3</sup> Are things different when counting exclusively inflection? Yes and no. Yes, when only counting those forms that are usually considered inflectional in the language (i.e. those expressing aspect, tense, passive, mood, agreement, and negation) then the total number amounts to 576 (Kornfilt 1997, my count), which does exceed the number for, e.g., Latin easily (up to 120, Matthews 1972: 396, my count) but still seems to be manageable. Now, writing down the (reduced) paradigm still appears to be pointless as the realisation of the cells is largely predictable, quite different from the situation in Latin. As an example, consider the future and past forms of Turkish *yapmak* ʻdo' in Table 2.


Table 2: Future and past forms of Turkish *yapmak* ʻdo' (Kornfilt 1997, Ch. 2.1.3)

<sup>2</sup>Abbreviations (in order of occurrence): rfl – reflexive, rcp – reciprocal, caus – causative, pass – passive, pot1 – first potential, neg – negation, pot2 – second potential, asp – aspect, tns – tense, agr – agreement.

<sup>3</sup>The same holds for other categories in the language. For example, when Kornfilt (1997: 311–320) presents demonstrative and interrogative pronouns, she does spell out quite a few paradigms but only in order to illustrate the high degree of systematicity with which the pronominal items are followed by separate suffixes for number and case.

### Tabea Reiner

The root (*yap*) stays the same throughout, the tense suffixes (-*AcAK* and -*DI* respectively) merely undergo phonological alternations,<sup>4</sup> and the agreement suffixes are clearly separable. In fact, the paradigm above may be replaced by two simple instructions, cf. (2).

	- a. {root; tns:fut} ⇒ root-*AcAK*-agr1
	- b. {root; tns:pst} ⇒ root-*DI*-agr2

Read: for expressing (an instantiation of) the meaning on the left hand side in Turkish, use (an instantiation of) the form on the right hand side. The manner of notation is equivalent to the graphical way of representation in Figure 1. Note that the instructions can be read as either rules or constructions, which is, in fact, not a contradiction (Rostila 2011: 123–124). However, since I am portraying constructionist thinking here, the natural choice is for constructions.

Admittedly, the existence of more than one set of agreement forms in Turkish introduces some irregularity into the picture, especially since, in total, four such sets are posited (Kornfilt 1997: 382). Kornfilt refers to these sets as paradigms; in fact, this is one of the few places in the book where she uses the term at all. Although this usage is not in line with the definition adopted in the present paper – where paradigms are not about morphemes but about whole word forms – it already indicates that the purview of what we call paradigms is the moderately irregular. This becomes more tangible when we shift from Turkish to Latin. Consider Table 3, which is an (approximate) translation of Table 2.

Table 3: future and past ("perfect") forms of Latin *facere* ʻdo' (Panhuis 2009: Ch. 7)


Instead of two clearly distinguished tense suffixes we find merely one (-*e* in the future) plus a root vowel change between the two tenses (*fac*- > *fec*-). Moreover, the future marker -*e* becomes -*a* in the 1st person singular and in all future forms an -*i* slips in between the root and tense. Now, this appears to be exactly the

<sup>4</sup> In particular, note that the change from <k> to <ğ>, i.e. from [k] to ∅ (with a lengthening of the previous vowel), appears to be purely phonologically conditioned (Ketrez 2012: 13–14, 91).

### 5 Recursion and paradigms

kind of situation for which paradigms have been invented in the first place: they provide an economic way to write down the unpredictable (e.g., the -*i*) coupled directly with the predictable (e.g., 1pl -*mus* but also *f* V*c*-). To this extent, paradigms are a convenient analytical tool and at the same time a concise format for instructed L2-acquisition.

However, paradigms are not the only option. In order to capture the unpredictable as well as the predictable in a very general fashion, Construction Grammarians have long developed other means, i.e. constructions. Crucially, these are not only meant for syntax but for the whole syntax-lexicon continuum, including words and even morphemes (Goldberg 2006: 5). For example, the rules/constructions chosen in (2) above for the Turkish data from Table 2 can be transferred to the Latin data from Table 3, cf. (3).

(3) Latin


Admittedly, these are three lines for one verb instead of two lines for a whole range of verbs like in (2). We cannot even generalise from *facere* to all fifthconjugation verbs, since not all of them show the root-vowel change, e.g., *cupere* ʻdesire', *fugere* ʻflee', *rapere* ʻplunder, seize' (Bennett 1918, Greenough et al. 2021).<sup>5</sup> However, in principle, the notation is possible. It is like listing idioms, complete with their schematic parts. In contrast, the same kind of notation for Turkish was more like stating syntactic rules or writing down highly abstract constructions. The crucial point is that the same kind of notation is apt for both types of data.

In this sense, constructions can replace paradigms: paradigms cells *are* constructions. In order to elaborate on this idea, I am going to discuss additional examples in the following paragraphs. Most of the examples are adopted from the literature referenced in Haspelmath (2011: 58–59), i.e. from Spencer (2001), Gurevich (2006), and Booij (2010). Importantly, the first one of these authors, i.e. Andrew Spencer, does uphold the paradigm as a theoretically relevant notion. He belongs to a school of thought which not only holds that there are genuinely morphological phenomena (not reducible to something else, especially syntax) but also maintains that these phenomena can be described best by using paradigms. This school of thought seems to thrive especially within the Surrey Morphology

<sup>5</sup>Note that the fifth conjugation is also called third-*io* conjugation.

### Tabea Reiner

Group and is called *Autonomous Morphology* here.<sup>6</sup> In any case, what Spencer (2001) presents are, at the same time, realisations of paradigm cells *and* constructions. One of my main aims will be to demonstrate in detail how his examples can indeed be written down as constructions; a fact, whose further theoretical consequences will be explored in Section 3. By contrast, Olga Gurevich as well as Geert Booij explicitly opt for a purely constructionist approach to morphology with paradigms being merely emergent. As a whole, the following paragraphs may be read as a fleshing out of Haspelmath's rather brief remarks on the equivalence of realisational morphology and constructionist syntax, partially quoted above in the introduction (Section 1). Having said this, Haspelmath himself does not explicitly state that his observations render the traditional paradigm superfluous; rather this is the conclusion that I have drawn above (in particular with respect to the examples from Turkish and Latin) and that I will substantiate in the course of the following discussion.

First, consider some examples based on Spencer (2001), starting with an extended version of his example for cumulative exponence.

	- a. {ʻsing'; ps:1, num:sg, tns:prs, mood:indic} ⇒ *canto*
	- b. {ʻsing'; ps:1, num:sg, tns:pst, mood:indic} ⇒ *canté*
	- c. {ʻsing'; ps:1, num:sg, tns:prs, mood:sbjv} ⇒ *cante*
	- d. {ʻsing'; ps:1, num:sg, tns:impf, mood:indic} ⇒ *cantaba*
	- e. {ʻsing'; ps:2, num:sg, tns:impf, mood:indic} ⇒ *cantabas*

Here, the predictable part is the structure [*cant*- + X] and the unpredictable part is whether and how the respective feature values are expressed cumulatively, i.e. together in one morph. For example, the -*o* in context (4a) *canto* appears to realise 1st person, singular, present, indicative (and active) all at once, while the -*abas* in (4e) *cantabas* might be split into a thematic vowel (-*a*-), an imperfective past tense marker for the relevant inflectional class (-*ba*-), and an exponent of agreement (-*s*). So while the latter form seems to be compositional and apt for a non-constructionist morpheme-by-morpheme description, the former escapes such a description (provided that we try to avoid null elements). Here, the holistic pairing of form and meaning, i.e. the conception as a construction, presents

<sup>6</sup>Accordingly, I will call the practitioners *Autonomous Morphologists*, accepting the bracketing paradox. The classic reference is Aronoff (1994); later publications include Maiden et al. (2011). Also Stump's Paradigm Function Morphology belongs here (Stump 2016); however I will not treat this theory in any detail in the present paper since this would require another introduction (in addition to the one to DM).

### 5 Recursion and paradigms

itself as the only option as opposed to an incremental approach. Accordingly, the format of presentation chosen in (4) is simply the same one as in (2) and (3) above: {meaning} ⇒ form.

Since this format is equally apt for forms with a higher degree of compositionality (cf. the discussion above), it has been chosen in (4) throughout, even for (4e) *cantabas*. However, please note that on closer inspection not even this form meets the agglutinating (Turkish-style) ideal: in contrast to the other feature values, indicative mood is not signalled by a dedicated suffix but has to be inferred from the fact that *cant*- belongs to the -*ar* inflectional class, which would have -*e* as its thematic vowel in the present subjunctive. So this, like (4a) above, is a situation we would usually describe by putting the form as a whole into a paradigm cell – while it can be captured equally well by setting up a construction.

Next is Spencer's (2001) example for extended exponence, again written down as a construction here.

(5) Spanish (Spencer 2001: 286) {ʻeat'; ps:1, num:sg, tns:impf, mood:indic } ⇒ *comía*

According to Spencer, the feature value imperfective is signalled twice within this word form: by the imperfective marker for verbs of the -*er* inflectional class, i.e. -*í*, as well as by the -*a*, since the latter is a first person singular marker only in the imperfective (provided that the -*a* in the present subjunctive form *coma* is a thematic vowel, not a person/number suffix). This is extended exponence: the marking of one meaning extends over more than one morph. Again, this is a situation that is a) hard to capture by an incremental approach, b) traditionally captured by drawing a paradigm, and c) equally well captured by writing down a construction.

The same is true for Spencer's examples of zero exponence, e.g. (6).

	- a. {ʻtravel, drive'; ps:2, num:sg, tns:prs} ⇒ *brauc*
	- b. {ʻtravel, drive'; ps:2, num:sg, tns:pst} ⇒ *brauci*
	- c. {ʻtravel, drive'; ps:2, num:pl, tns:pst} ⇒ *braucat*
	- d. {ʻtravel, drive'; ps:3, num:sg, tns:prs} ⇒ *brauc*
	- e. {ʻtravel, drive'; ps:3, num:sg, tns:pst} ⇒ *brauca*
	- f. {ʻtravel, drive'; ps:3, num:pl, tns:prs} ⇒ *brauc*
	- g. {ʻtravel, drive'; imp} ⇒ *brauc*

### Tabea Reiner

What can be seen in *brauc* is essentially what you have to expect for any construction due to its nature as a linguistic sign: polysemy. Compare (7) to (8).


To be sure, ʻ[…] 2nd *or* 3 rd person […]' and ʻoperate a vehicle *or* motivate the process' represent very different kinds of meanings (in constructionist terms: a finite verb represents a partly schematic construction while a lexical entry represents a substantive construction, Hoffmann & Trousdale 2013: 2). However, the underspecification in both examples can be neatly captured by writing them down as constructions. So, ironically, a constructionist analysis can do something that also DM strives for (albeit in a different way, cf. Section 2.2): treating syncretisms<sup>7</sup> as cases of underspecification.

Please note that, strictly speaking, the constructions presented so far are only halves. For being full constructions, they would need a double arrow, signalling that not only the respective meaning triggers the respective form but also the other way round. However, against the background of syncretisms as treated above it is clear that the back arrow would require a more complete picture of the languages at hand than can be given here. For example, after ensuring that through the entire verbal paradigm of Spanish the form *canto* (without stress on the final vowel) is really only 1st person, singular, present, indicative (and active), (4a) could be rewritten as (9).

(9) Spanish (Butt et al. 2019: 170–712) {ʻsing'; ps:1, num:sg, tns:prs, mood:indic} ↔ canto

Let me add another word on modes of presentation. When I refrain from adding morpheme boundaries and associating the resulting units with individual meanings I take the following passage from Spencer (2001) seriously.

[…], in a sense it's a mistake to speak of meanings being concentrated in one morph or spread across several morphs or realized by zero morphs. (Spencer 2001: 287)

<sup>7</sup> Syncretism = in a given context two feature values are not overtly distinguished although they are overtly distinguished in another context in the same language (adopted from Kramer 2016: 96).

### 5 Recursion and paradigms

So in my running text wordings like "the -*o* in context (4) *canto* appears to realise […]" should be taken with a grain of salt: from the constructionist perspective, which I am portraying here, it is not necessarily the -*o*, not even the -*o* in a certain context, but *canto* as a whole that has some meaning to begin with. Similarly, if the instructions for Turkish in (2) above are read as constructions, the hyphens indicate what have turned out to be internal semantic entities (probably aligning with internal distributional entities); however neither Autonomous Morphology (based on paradigms) nor Construction Grammar (with emergent paradigms) depends on the presence of any semantic parts below the level of the word form.<sup>8</sup>

Before turning to his main example – auxiliary structures in Slavic – Spencer (2001: 287) mentions a last general group of examples in favour of the realisational approach to morphology: meaningless morphemes. More specifically, he judges Spanish -*ar* to be a case in point: in one class of verbs this element appears after the root in all three, the infinitive, the future, and the conditional. I am not sure whether we are rather dealing with a case of polysemy rather than meaning*less*ness here. However, a more obvious example may be found in the diachrony of German vs. Dutch:


That is, German retains a thematic vowel where Dutch does not – and this vowel does not appear to have any meaning (anymore), not even indicating inflectional class. Admittedly, the -*e*- does distinguish the 3rd person singular past (*rettete*, /retətə/) from the 1st person singular present (*rette*, /ʁɛtətə/ = /ʁɛttə/ = /ʁɛttə/ = /ʁɛtə/). However, distinguishing between meanings is not the same thing as having a meaning (recall the classical definition of phonemes vs. morphemes). The only "meaning" that could be assigned to the -*e*- in *rettete* would be ʻif you have a choice between a 3rd+past reading and a 1st+present reading, choose the former'. This piece of information does not count as a meaning since, as far as I can see, it is not directly evoked by the -*e*- in native speakers (let alone vice versa). As a result, we get a meaningless element in an otherwise more or less segmentable string, cf. (12).

<sup>8</sup>This does not only hold for inflection but also for word formation, cf. Booij (2016: 428).

### Tabea Reiner

(12) German rett-e-te rescue-?-1/3sg.pst ʻrescued'

Thus, again a realisational notation appears to be practical. My point is that such a notation does not intrinsically need paradigms but solely constructions, cf. the construction in (13).

### (13) German

{ʻrescue'; ps:1 *or* 3, num:sg, ((tns:pst, mood:indic) *or* mood:irr)} ↔ rettete

As an aside, from example (7) onwards in this paper, one-to-many relations between form and meaning have been treated as cases of polysemy; however, a constructionist account is even able to distinguish between polysemy and homonymy.<sup>9</sup> Consider Table 4, of which the light shaded cells are presented as a case of polysemy in (14) and the dark shaded cells are presented as a case of homonymy in (15). The motivation for drawing the distinction is that 1st and 3rd plural share a positive feature value (plural), while 3rd singular and 2nd plural do not; moreover, the former syncretism runs through all verbal forms in German (and extends to the present infinitive) while the latter dissolves in the past tense as well as in the present tense of umlaut verbs.

Table 4: Present indicative of German *kaufen* ʻbuy' (Helbig & Buscha 2001: 23)


### (14) German

{ʻbuy'; ps:1 *or* 3, num:pl, tns:prs, mood:indic} ↔ kaufen

	- a. {ʻbuy'; ps:3, num:sg, tns:prs, mood:indic} ↔ kauft *and*
	- b. {ʻbuy'; ps:2, num:pl, tns:prs, mood:indic} ↔ kauft

<sup>9</sup>Both of which can be special cases of syncretism as defined in the present paper, cf. fn. 7.

5 Recursion and paradigms

To be sure, technically speaking, also the dark shaded cells of Table 4 could be rendered as a case of polysemy, cf. (16).

### (16) German

{ʻbuy'; ((ps:3, num:sg) *or* (ps:2, num:pl)), tns:prs, mood:indic} ↔ kauft

However, as argued above, in the specific case at hand, other facts from the language cast doubt on this latter analysis. So here I would opt for the former analysis in terms of homonymy rather than for the latter in terms of polysemy. In sum, while both analyses make use of underspecification they are apt for different kinds of syncretisms.

Turning now to Spencer's main example, auxiliary structures in Slavic, it seems natural that his realisational treatment of not fully compositional word forms is transferred to not fully compositional strings of words that appear to realise similar meanings (e.g., tenses or aspects). It is especially this generalisation that is picked up by Haspelmath (2011: 59) and, among other issues, contributes to Haspelmath's overall thesis that any universal morphology/syntax boundary is elusive for the time being. Against this background, it will not come as a surprise that also auxiliary structures (or periphrases) may be written down as constructions, so I will not go through this here. However, please note that Section 4 of the present paper will provide a fundamentally different look on such structures.

Summarising my fleshing out of Haspelmath's (2011) reference to Spencer (2001) for the moment, there are reasons to use realisational rules rather than incremental procedures; however it does not seem to matter whether the feature specifications are conceived of as a grid, establishing paradigm cells to be realised, or simply as meanings of constructions (accordingly, the sounds/characters materialise a cell or provide the signifiant of a construction). The notation used in Spencer (2001) is, in fact, similar to the one used above. Going one step further, there is a choice between one representational format that is needed anyway from a constructionist perspective, i.e. the construction, and another representational format that is not needed anyway (though might emerge from applications of the former). Theoretical parsimony requires that we stick to the format needed anyway.

As a last reflection on Spencer (2001) let me note that the author is perfectly aware of the connection between realisational and constructionist approaches. For example, consider the following quotation on auxiliary structures.

We are dealing here with constructional idioms much like phrasal verbs. (Spencer 2001: 283)

### Tabea Reiner

In the same vein, consider the following paragraph, where he compares (inferential-realisational) Word-and-Paradigm models (which, needless to say, he favours) to Item-and-Arrangement models.<sup>10</sup>

Inferential models can be contrasted with lexical models in which the mapping between form and meaning is specified in the lexical entry of a morpheme. On such a theory, the -*z*-suffix contributes its own [Num:Pl] ʻmeaning' to the word form by being combined with the (numberless) form *dog*. In this respect, the plural suffix is a Saussurean sign, a pre-compiled pairing of form and meaning. It is precisely that conception that is denied in paradigmatic and inferential-realisational approaches to morphology. Inferentialrealisational models by their nature cannot involve lexically listed morphemes-as-signs. As a consequence, inflectional formatives turn out to be simply morphophonological ʻmarkers' on stems, signalling (realising) some subset of the feature set to be expressed. In the simplest cases there is a one:one mapping observable between form and function and the formative then has the appearance of a classical morpheme, but it's important to realise that this is just one extreme of the form-function mapping (and in inflection a rather rare occurrence at that). (Spencer 2001: 280–281)

Note the affinity of this approach to morphology with constructionist approaches to syntax: the primary signifiant is the expression as a whole; its subparts have their function within the expression but they do not necessarily have the same function in other environments as well (cf. the reflection on Spanish theme vowels above). A remaining difference seems to be that inferentialrealisational morphology still refers to superordinate rules for realisation, e.g.: if there is no dedicated expression for a given combination of feature values then use the default form (Spencer 2001: 280, 289). As an alternative to such rules, in Construction Grammar there are, however, relationships between constructions (Goldberg 1995: 109). In fact, also realisational rules can be captured by such relationships, e.g.: given a set of meaningful expressions (i.e. constructions) organised in an inheritance hierarchy, use the one that maximally specifies the meaning you want to express (and the maximally specific construction available might turn out to be rather unspecific). Such a conception, including further developments, is run-of-the-mill in Sign-Based Construction Grammar (Boas et al. 2012: 9–14).

In sum, the inspection of Spencer (2001) shows that constructionist thinking is influential in morphology and, from an outside perspective, that constructions

<sup>10</sup>Item-and-Arrangement (i.e. incremental), Item-and-Process as well as Word-and-Paradigm are families of morphological theories, for their characteristics cf. Stewart (2016).

### 5 Recursion and paradigms

might even fully replace paradigms. I can be briefer about the two other works mentioned in Haspelmath (2011: 59) and pursued here, i.e. Gurevich (2006) and Booij (2010). For what had to be worked out carefully with regard to Spencer (2001) is utterly explicit in in these works: Construction Grammar applies to morphology just as it applies to syntax.<sup>11</sup> So I confine myself here to a central figure from Gurevich (2006: 170) and an exemplary formula from Booij (2010: 241).

Gurevich (2006) is a book-long treatment of the so called version in Georgian. This is a pre-radical vowel with a whole range of functions, crucially depending on context (Gurevich 2006: 6–13). For example, -*i*- as a pre-radical vowel indicates that the indirect object is affected (in the presence of an overt 1st or 2nd person indirect object) or that the subject is affected as a beneficiary (in the absence of an overt indirect object) or that there is no affected participant (in the passive) or that the subject is 1st/2nd person (in the evidential perfect). In a way, this is a scaling-up of what was demonstrated above for thematic vowels in Spanish: a given element in a word form does make a specific contribution to the overall meaning of the word and clause but in order to identify that contribution the hearer has to have information far beyond the element as such.

Now, let's see how Gurevich (2006: 170) combines all these functions of version vowels (of which the example above was but a snippet) into a constructional network. Her presentation is reproduced here as Figure 2.

Syncretisms<sup>12</sup> like the one involving -*i*- can easily be read off the network.

Having addressed syncretisms (as well as other kinds of non-bijective formmeaning relations) several times in the present paper, some standard problems for classical Item-and-Arrangement approaches to morphology are covered. Another standard problem from the wealth of non-bijective form-meaning relations, however, has not been stated explicitly so far, i.e. allomorphy. This is tackled by Booij's (2010) last chapter. For example, he gives the following formula for capturing the ablaut in *sing* – *sang*:

$$\text{[X i Y]}\_{\text{V, [-past]}} \approx \text{[X a Y]}\_{\text{V, [+past]}} \text{ (Booij 2010: 241)}$$

Crucially, none of the forms is derived from the other or from some underlying form but they are equally stored in the mental lexicon according to Booij (2010: Ch. 10).<sup>13</sup>

<sup>11</sup>It does become explicit in Spencer (2004: 84): "Essentially, we need to think of inflected words as akin to constructional idioms".

<sup>12</sup>I assume that all of the distinctions are marked overtly elsewhere in the language.

<sup>13</sup>I concentrate on this last chapter since the other chapters are rather on word formation than on inflection whereas the present paper focusses on the latter, to the extent that the difference can be upheld, cf. Haspelmath (2011).

### Tabea Reiner

Figure 2: Constructional network for Georgian version vowels (Gurevich 2006: 170)

In principle, true suppletion (Mel'čuk 1994) may be captured the same way as far as I can see, e.g. for the partial paradigm given in Table 1 (viewed synchronically):

$$[\text{bi Y}]\_{\text{V, [1/2\,sg]}} \approx [\text{is Z}]\_{\text{V, [3\,sg]}} \approx [\text{sind}]\_{\text{V, [1/3\,PL]}} \approx [\text{see Z}]\_{\text{V, [2PL]}},$$

where Z is the regular agreement suffix (phonetically) and Y is either the regular agreement suffix or something else (Plank 2016: 6–7). Taxonomically speaking, the formula involves four sister constructions (Jackendoff & Audring 2020: 111).<sup>14</sup> The important point about these interconnected constructions is that they can, and do, serve to restate paradigms (Jackendoff & Audring 2020: 151–154). In fact, Jackendoff & Audring (2020) provide a very nice example of what is going on with paradigms in Construction Morphology (CxM): in a first step, they are acknowledged as central to inflection (Jackendoff & Audring 2020: 133–134) but in a second step, they are restated, one by one, as interconnected constructions (e.g., Jackendoff & Audring 2020: 151–154). This is reminiscent of my argumentation above that cells are in fact constructions and paradigms are (specific) relations between constructions.

<sup>14</sup>Also note the relevance of networks in another recent constructionist approach, i.e. Crysmann & Bonami (2017).

### 5 Recursion and paradigms

The extension of Booij (2010) concludes my fleshing-out of Haspelmath's (2011: 58–59) remark that, essentially, realisational morphology and constructionist syntax do the same thing. To summarise: against the background of strictly agglutinating languages, certain phenomena familiar from fusional languages (fusion as such, syncretism, allomorphy, …) suddenly appear fairly idiosyncratic – just like against the background of compositional syntax idioms appear quite idiosyncratic. For the latter case Construction Grammarians concluded that we simply had the cart before the horse, i.e. that we have to start conceiving of the whole as primary to its parts – and now their solution is transferred to morphology. For Haspelmath (2011), this expansion of the constructionist idea raises, together with a range of other considerations, the question where to draw the line between the two levels of description generally. In any case, traditional inflectional morphology's favourite format, i.e. the paradigm, seems to be replaceable by constructions, integrated in networks.

We will see in the next section that DM tries to capture exactly the same problematic groups of phenomena but opts for a radically different solution, i.e. the strict separation of meaning and form (at the outset).

## **2.2 From a composition-oriented perspective**

### **2.2.1 A short introduction to DM**

Distributed Morphology (DM) is a strictly compositional approach to morphology, presupposing GB or minimalist syntax. Its origins lie in Halle & Marantz (1993), fairly recent overviews are Embick (2015) and Bobaljik (2017), for a critical review of the former cf. Spencer (2019). The following summary is mainly based on these works; however, occasionally other publications will be referred to as well.

The key feature of DM is not so much the distribution of the morphology as the distribution of the lexicon, which is divided into three lists (Bobaljik 2017: 28). These lists are called A, B, and C below:


### Tabea Reiner

linked to the root √buy in list A. So the purview of list C is handling the idiosyncratic – even up to idioms. For example, if the aforementioned link does not fit the context, another link can be established like ʻaccept as true' – √buy (cf. *I don't buy your conclusion*).

These lists are the cornerstones of a derivational model for inflection and word formation. The derivation starts by selecting items from list A that attach to the terminal nodes of an appropriate syntactic representation, yielding an output like √buy-3sg. This output is then handed over to list B where matching phonological strings are inserted into the positions, e.g. /bai/-/s/. For less straightforward derivations, certain processes directly before or after insertion may be appealed to in addition. One of these options, i.e. impoverishment, will be presented in more detail below. First, let's visualise the basic architecture and say a few more words about the individual steps of the derivation, in particular about vocabulary insertion. For the latter objective it is suitable to pick a language with more agreement than English, so I chose German and the verb *kaufen* ʻbuy' whose past indicative forms are given in Table 5. Needless to say, the paradigmatic presentation is only for convenience, the forms being represented differently in the visualisation to follow.

Table 5: Past indicative of German *kaufen* ʻbuy' (Helbig & Buscha 2001: 23)


Now consider Figure 3, where *kauften* {ʻkaufen', past, 3pl} is derived.<sup>15</sup>

In list A it can be seen that there are several roots, including √kauf, as well as several feature values, including past, 3rd person, and plural. Crucially, the roots are neither concepts (list C) nor forms (list B). As to the feature values, I have taken the liberty to write them down as such, i.e. [feature:value], whereas it seems more common in DM to write the values only and refer to them as*features*. In any case, this is the sole kind of structure (apart from the distinction between roots and feature values) that enters list A. Otherwise this list is meant to be an unstructured cloud. In particular, the individual items are not pre-arranged into paradigms.

<sup>15</sup>Note that the corresponding present tense example would be more difficult to handle (Spencer 2019: 231).

### 5 Recursion and paradigms

Figure 3: example *kauften* {ʻkaufen', past, 3pl}, showing the basic DMarchitecture Bobaljik (2017: 6), Albright & Fuß (2012: 248–250)

The next step in Figure 3 is the selection of a root and several features, all of which attach to the terminal nodes of the syntactic tree. From my perspective as an outsider to the theory, a mystery is connected with this step: what kind of mechanism, apart from communicative intent, determines which feature values are selected? For example, somewhere in the model there must be information on which contexts require which tense-aspect categories to be realised in the language at hand (cf. English *He is eating* vs. German *Er isst*perfective-or-imperfective). In any case, the root and the feature values are arranged by the syntax, independently from any phonological information. So the intermediate result is an abstract string like √kauf-past.3pl.

Now, vocabulary insertion can take place. The phonological form /kaɔf/ is one of the forms over which √kauf is an abstraction, so they match (at least this is what I have been able to conjecture; insertion for roots in simple cases is not extensively discussed in the DM-literature). Also for the second part of the abstract string, i.e. past, there is just one matching vocabulary item, i.e. [past] ↔ /tə/, so this item is inserted. Things get more interesting and more difficult with

### Tabea Reiner

respect to [3pl]. Several vocabulary items are available for realising AGR and, crucially, these are ordered from most to least specific. Now the mechanism runs through this ordered list and first tries the highest (= most specific) vocabulary item, i.e. [2pl] ↔ /t/. However, 2nd person does not match 3pl, so this item is discarded. The second highest item is [pl] ↔ /n/, which does match 3rd person plural (i.e.: there is no contradiction). So this item is inserted. Crucially, the same item matches 1pl, hence it would likewise be inserted in the derivation of *kauften* {ʻkaufen', past, 1pl}.

What can be seen here are two uses of underspecification at once. First, it serves to order the list in such a way that, by browsing the list downward, insertion may operate on the Paninian Principle (= Elsewhere Condition = Subset Principle). That is: given two rules, where one is more specific than the other, application of the more specific one – here: insertion of the more specific item – blocks application of the less specific one (for an overview on the principle cf. Anderson 1992: 132). Put differently: the most general rule – here: insertion of the least specific item – is the default and whenever there are more specific rules/items available, the most specific one of them will overwrite the default. The "only" job of the data-tackling linguist is to find the most economic list for the data set at hand. With respect to the German past indicative forms, I adopted the list from Bobaljik (2017: 6), although from a German Linguistics point of view it might be debatable whether the schwa belongs to the tense or agreement suffix (pro Bobaljik's solution: the verb *tuen* ʻdo'). Moreover, as far as I can see, the order of [pl] ↔ /n/ and [2] ↔ /st/ is arbitrary since none of them is more specific than the other.

The second, ensuing, use of underspecification here is modelling syncretism. As mentioned above, the vocabulary item [pl] ↔ /n/, matching 3pl as well as 1pl, may be inserted into both potential strings, √kauf-past.*3*pl (cf. Figure 3) as well as √kauf-past.*1*pl. Likewise, when we derive the realisation of √kaufpast.3*sg* or √kauf-past.1*sg*, the mechanism runs through the list, discards [2pl] ↔ /t/, [pl] ↔ /n/, and [2] ↔ /st/ as not matching and ultimately reaches the least specific pairing [] ↔ ∅. So for both persons, AGR is realised as null. Please note that this way systematic syncretisms may be captured without referring to paradigms.<sup>16</sup>

However, underspecification is not the only way to model syncretism in DM (although the most desirable one according to Harley 2008: 253). In particular,

<sup>16</sup>The 1/3 conformity really is a syncretism in the sense of fn. 7 since the two person values are distinguished overtly elsewhere in the language, viz. in the singular of the present tense (*kaufe* vs. *kauft*).

### 5 Recursion and paradigms

syncretisms that appear to be even more systematic than the ones treated above may be modelled by a process called impoverishment. This is the deletion of features/feature values in the abstract string that has been issued from the syntax and awaits vocabulary insertion (Embick 2015: 139). For example, German determiners distinguish three gender values in the nominative singular (nominative: *der*, *die*, *das*) but neutralise this distinction in the nominative plural in favour of the feminine form (nominative: *die*),<sup>17</sup> in accordance with Greenberg's well-known universal no. 37. So all three output strings, √d-nom.*masc*.pl, √dnom.*fem*.pl, and √d-nom.*neutr*.pl have to yield *die* in the end; likewise for the other case values (*derer*/*n*, *den*, *die*). In this situation it seems reasonable to delete the gender value from the string altogether. A rule expressing this is shown in (17) for the example at hand.

(17) [+masc/fem/neutr] ⇒ fem/\_pl read: any gender value is set to the feminine under the condition of a plural environment

Having addressed syncretism, which is handled by underspecification or impoverishment in DM, I have to touch upon allomorphy, too, since these two types of phenomena – syncretism and allomorphy – appear to be the standard testing ground for any morphological theory. Allomorphy then can be handled by adding conditions on vocabulary items (Embick 2015: 169). For example, the German vocabulary item [past] ↔ /tə/ works well in a derivation like the one in Figure 3 above but would yield ungrammatical forms for strong verbs, e.g. the past 3rd person plural of *gehen* ʻwalk' would be wrongly predicted as \*/gehtən/. In order to derive the correct form /giŋən/ we may use the adopted vocabulary item in (18).

(18) /giŋ/ ↔ √geh/\_past

read: /giŋ/ is the form to be inserted for √geh under the condition of a past environment

Additionally, the insertion of [past] ↔ /tə/ has to be suspended in some way or other. A remaining problem of this analysis is how to supply the schwa of /giŋən/ when assuming Bobaljik's (2017) segmentation.

At this point I have presented a snapshot of DM in action, not even mentioning further processes like fusion under locality conditions, fission, or readjustment

<sup>17</sup>Cf. Helbig & Buscha (2001: 214). Note that the following analysis works only if *die* is viewed as feminine per se (Meinunger 2017, but cf. Leiss 1994: 291–292 on why this feature value has little to do with femininity as such). For relevant considerations cf. also Kramer (2019: 184).

### Tabea Reiner

(also note that linearisation is not standardly predictable from syntax).<sup>18</sup> To conclude for the moment, DM's slogan, it's "syntactic hierarchical structure all the way down" (Harley & Noyer 1999: 3) is to be taken with a grain of salt: it's syntactic hierarchical structure all the way down until vocabulary insertion and its satellite processes. That is, many things can happen on the way from syntax to final spell-out and it is this many-faceted interface that deserves the name *morphology* in DM. Crucially, however, DM strives to make even these processes as predictable as possible.

In any case, the architecture as a whole permits reading paradigms off individual derivations (e.g., Table 5 may be read off the derivation depicted in Figure 3) but there are no pre-designed paradigms in the model. That is: to the extent that DM provides a model of language representation and processing, paradigms do not have any psychological reality in the theory (cf. also Bobaljik 2002: 53). This negative attitude towards paradigms will be summarised and put into perspective in the following section (Section 2.2.2).

### **2.2.2 The status of paradigms in DM**

This section is not a real section but a convenient synopsis for those readers who skipped the introduction to DM. So the main point of the section is summarising what was laid out in Section 2.2.1. with respect to paradigms in DM. There are at least two places in the architecture at which paradigmatic structures would be expected by the average linguist but in fact do not play a role.

First, feature values (list A) are not organised into paradigms. Obviously, they could be (though, with a placeholder for roots/stems); however, according to the theory this is just not necessary for deriving correct forms. And unnecessary presyntactic structure is to be avoided in a strictly compositional approach (Embick 2015: 17).

Second, vocabulary items (list B) are indeed ordered but not according to intersecting feature values (as in paradigms) but according to specificity. It is striking that DM is all the same able to capture syncretism and allomorphy – phenomena that are otherwise thought to be inextricably linked with paradigms (cf. also Bobaljik 2002: 54).

All of this does not mean, however, that practitioners of DM do not use paradigms for presentation. To pick a random example, Harley (2008) is full of para-

<sup>18</sup>Fusion = combination of two sister nodes into one, which retains the features of both input nodes but has no internal structure (Bobaljik 2017: 15); mnemonic: fusional morphology in the typological sense. Fission = splitting of a single node in the syntax into two nodes in the morphological representation (Bobaljik 2017: 19); mnemonic: multiple exponence. Readjustment = phonological alternation after vocabulary insertion (Bobaljik 2017: 7); mnemonic: remedy for everything else that leaves derivation ill-formed.

### 5 Recursion and paradigms

digms. The important thing is that they do not have a primary status in the architecture. To use the food metaphor from Trommer's (2016) title: from a DM perspective, compiling paradigms is like pre-sorting the ingredients in your kitchen cupboard according to nutrients – useful for some purposes but not necessary (and hence not desirable) for cooking tasty meals.

I conclude this micro-section by a synoptic quotation, even more explicit than the corresponding one in Section 1:

Importantly, paradigms are epiphenomenal in DM. They have no theoretical status and they are never referred to by morphological operations. (Kramer 2016: 97)<sup>19</sup>

## **2.3 Comparison**

This section serves to compare the two approaches regarding their shared detachment from paradigms and their shared tendency to reduce inflectional morphology to something else (constructions or the syntax/phonology interface). To this end I will first address their respective standard data and then delve into a more theoretical discussion. For completeness, also Autonomous Morphology is occasionally integrated into the picture before it takes centre stage in Section 3.

While Haspelmath (2011: 58–59) refers to Turkish for showing that morphology can look like syntax, proponents of DM adduce Swahili for the same reason. However, Swahili inflection does not only display a high degree of compositionality but it also displays purely positional contrasts – something that we tend to expect from syntax exclusively. For example, consider the pair of examples in (19a) and (19b), taken from Trommer (2001: 18).<sup>20</sup>

(19) Swahili (Atlantic-Congo, Tanzania et al.)<sup>21</sup>


To be sure, this does not work equally well for all forms, cf. Table 6.

<sup>19</sup>One reviewer remarks: "Interestingly, they [= paradigms] seem to organise those morphological operations into meaningfully related sets, which would place them above those operations. In this sense, they are not epiphenomenal but even more abstract than the abstract operations". From a DM-perspective, this is not a contradiction: the meaningfully related sets might be constructed by the linguist or the L2-teacher; however they are not part of any L1-speaker's mental grammar. This position gains some plausibility from anecdotal evidence: it appears hard to write down L1-paradigms if one is asked to do so for the very first time.

<sup>20</sup>In a similar vein cf. Crippen (2019).

<sup>21</sup>https://glottolog.org/resource/languoid/id/swah1253

### Tabea Reiner

Table 6: Swahili person forms (Almasi et al. 2014: 15, 102; only M-/WAclass)



Table 8: English nominal case forms


Still, writing down the first line of the paradigm appears to make as much sense as writing down nominal case forms for English, compare Table 7 to Table 8.

Obviously, here it is not the paradigm that tells the language user which form is the subject or object – it is the combinatorics (here: surface linear order). Switching to a constructionist perspective taken to the extreme, even the second and third line of the Swahili paradigm may be treated very much like the first line: it is still the relative position (slot) of a given form that determines its syntactic function – and makes it vary in idiosyncratic ways (e.g. *u* ⇒ *ku*).

So to some extent both kinds of data – Turkish-style and Swahili-style – support the idea that paradigms might be quite parochial a format: apt for Latin type languages but hardly beyond.

To repeat, with regard to the constructionist perspective, the above conclusion is my own one, neither Haspelmath (2011: 58–59) nor related works plainly oppose against paradigms. To the contrary, Haspelmath (2000) even allows periphrases as paradigm cells and Booij (2016) reconceptualises paradigms as "second order schemas". However, let's have a closer look at these two conceptions.

Haspelmath (2000) argues against a gap-filling account of periphrasis and in favour of a grammaticalisation-based account.<sup>22</sup> Crucially, he does not only allow periphrases as paradigm cells but, by extension, also entire clauses:

<sup>22</sup>Some of Haspelmath's (2000) arguments against a gap-filling account may be countered by the criterion of feature intersection (Brown & Hippisley 2012: 250–252, Reiner 2020); however this is beyond the scope of the present paper.

### 5 Recursion and paradigms

However, it is not difficult to find syntactic phenomena that provide a striking analog of inflectional paradigms, gaps, and periphrasis in morphology. Again, a good example comes from English, where only a small subclass of verbs can occur without complications in interrogative and negative clauses. In (16), this well-known pattern is represented in such a way that the similarities with morphological suppletive periphrasis become apparent. […] Clearly, "periphrastic *do*" is periphrastic in much the same way as the cases of morphological periphrasis, but the filled gaps in (16) are not morphological monolectic forms. *Did you see* is a syntactic phrase which replaces the impossible syntactic phrase \**saw you*. (Haspelmath 2000: 662)

Here is Haspelmath's example number (16), reproduced as Table 9.


Table 9: English (Haspelmath 2000: 662)

This extension seems simply logical; however, it raises the question whether there is anything at all that can*not* be described by paradigms in this sense, i.e. by oppositions. In fact, a strictly constructionist perspective mandates that any set of clauses (transparent or not) is viewed as a set of constructions between which the language user may chose, hence as a paradigm in the above sense. This notion of paradigm, then, is so abstract that it becomes vacuous: if everything is paradigmatic it is pointless to state that such and such linguistic phenomenon (e.g., inflection) is, indeed, organised paradigmatically.

Turning now to Booij's CxM, as summarised in Booij (2016), the first thing to note is that inflection as well as word formation and phrasal idioms are captured by constructional schemas. (20) is an example for inflection.

(20) English (Booij 2016: 440, number (37) there) 〈[(x<sup>i</sup> )ω-j ↔ [N<sup>i</sup> , +sg]<sup>j</sup> ↔ [SG [SEM<sup>i</sup> ]]j 〉

Read: x<sup>i</sup> constitutes a phonological word (ω), which is associated with a certain morphosyntactic structure, which is associated with a certain semantic structure; the indices show identity relations.<sup>23</sup>

<sup>23</sup>Inflectional class information can be integrated into the morphosyntactic structure or modelled by second order schemas (to be introduced below).

### Tabea Reiner

Schemas that relate to each other (i.e.: share at least one element) constitute second order schemas, e.g. (21).

(21) English (Booij 2016: 440, number (39) there) 〈(x<sup>i</sup> )ω-j ↔ [N<sup>i</sup> , +sg]<sup>j</sup> ↔ [SG [SEM<sup>i</sup> ]]j 〉 ≈ 〈(x<sup>i</sup> -z)ω-j ↔ [N<sup>i</sup> , +pl]<sup>j</sup> ↔ [PL [SEM<sup>i</sup> ]]j 〉

Such schemas correspond to traditional paradigms in an obvious way and, crucially, they are said to organise language in their own special manner (for a lucid example from Saami cf. Booij 2016: 442). However, paradigms in this sense, central as they are, do not constitute the fundamental building blocks of morphological theory as Stump or Spencer would have it. Rather, they emerge from a more general organisational principle, which is the constructional schema: as soon as two or more schemas share one or more elements, they constitute a second order schema and in this sense a paradigm. To put it in a nutshell: paradigms are relevant but not basic (also cf. Marzi et al. 2020: 239–240, 257).

To conclude on Haspelmath (2000) and Booij (2016), while they do use paradigms in some sense, I argued that, implicitly, they deprive paradigms of any fundamental theoretical status: the former extends the notion to such a degree that it becomes void and the latter's approach allows reducing paradigms to a mere consequence of the fact that there are shared elements between constructional schemas.

Note that both takes on paradigms just presented differ from the position of Autonomous Morphology. There, periphrases are allowed into paradigms, too; however only under a very restricted notion of periphrases, which ensures that they are mere surrogates for true word forms (Brown & Hippisley 2012). When, moreover, Stump (2002: 147–148) speaks of "syntactic paradigms", he means something different: word forms seen as instantiations of a lexeme, among which a given syntactic context may chose the appropriate one.

Coming back to CxM, with regard to the non-fundamental status of paradigms it is surprisingly close to DM, except that the demotion is explicit in DM. However, in another regard, it is DM that implements considerations otherwise basic to CxM as well as to Autonomous Morphology. Prima facie non-compositional phenomena like syncretism, allomorphy (including suppletion) or polyfunctionality seem to call for a constructionist account and/or might constitute the irreducibly morphological in language (more on this relation below). Yet, also DM has developed means to deal with such phenomena: some are handled directly by underspecification, others may require processes like impoverishment, fusion, fission, or readjustment (cf. Section 2.2.1). Although even these processes are designed to be as predictable as possible, one cannot deny that this is much more

### 5 Recursion and paradigms

than just syntactic derivation plus vocabulary insertion. Put differently, if morphology is only an interface between syntax and phonology (as DM says) then it is a quite rich one. To be fair, not all processes (or rules) mentioned above are embraced equally by all proponents of DM. For example, Trommer takes a very critical stance on the accumulation of rule types in DM, summarised in Trommer (2016: 61). In the same volume, Haugen (2016) argues explicitly against the use of readjustment rules, in particular when modelling stem allomorphy. Also other practitioners of DM try to restrict the purview of readjustment rules (e.g. Pomino & Remberger 2019: 475). In fact, my example of /giŋ/ above has been designed in this spirit.

As an interim summary, I compared CxM and DM with respect to their take on paradigms (implicit vs. explicit demotion) as well as with respect to their conception of morphology vis-à-vis other levels of description (explicit vs. implicit acknowledgment, but see below). In the rest of this section, I will extend the comparison to three further dimensions, ordered in growing distance from the actual topic of paradigms and independent morphology: the notion of word, the question of psychological reality, and the kind of restrictiveness found in the respective approach.

*The notion of word* is discussed at length in Haspelmath (2011), from which I have cited but a snippet up to now. In brief, his overall discussion concludes that criteria for a universal definition of word remain elusive<sup>24</sup> – and so does a definite border between morphology and syntax. This fits well with the general constructionist idea of a lexicon-syntax continuum. However, from my perspective as an outsider to the theory, any commitment to this idea is in conflict with acknowledging morphology as a level of description qualitatively different from both, syntax and the lexicon: how can we draw solid lines in the middle of a continuum? Thus, reflecting the notion of word casts some doubt on the extent to which constructionist accounts are able to acknowledge "morphology by itself" (to allude to Aronoff's 1994 title). This may be a dividing line between CxM (Cx"M"?) and Autonomous Morphology.

Concerning their own definition of word, Autonomous Morphologists are hard to pin down, though. Working in the tradition of Word-and-Paradigm approaches, they seem to accept that for any given string we can tell whether it constitutes a word or not – no matter if we are looking at data from Latin, Turkish, Swahili or West Greenlandic. The only explicit pertinent discussion I am aware of is in a footnote in Luís & Spencer (2013: 127–128), where they elegantly delegate the task to Canonical Typology (more on this relationship in Sections 3 and 5).

<sup>24</sup>But cf. Gil (2020) for a recent proposal.

### Tabea Reiner

Turning now to DM, the theory is not particularly obvious with respect to its attitude towards a universal notion of word. Certainly, it is universalist in spirit and its hallmark is the (far-reaching) structural isomorphism between word structure and phrase/clause structure, which seems to leave no room for words as different from phrases/clauses. However, even in DM there is an endpoint of the derivation, e.g. /baiz/; and other terminals in the same clause structure tree host their own derivations, e.g. /hi:/. In this sense, there are words as opposed to larger phrases. So, latently, DM does recognise words.

*The question of psychological reality* can be phrased as: do formats of description (possibly including paradigms) have a role to play in our mental representation of language? This question is the least obviously answered in Autonomous Morphology. Again, the topic does not appear to receive much attention in the literature. My impression is that Autonomous Morphologists silently follow a weak version of a Language-as-an-abstract-object approach (Katz 1981), i.e. it is simply not their intention to describe any mental representation of language as a system (Chomsky's competence or Saussure's langue), rather they focus on abstractions over the parole. For example, this position is suggested by Aronoff (2016: 197–198), where the high usefulness of paradigms in language description is enough to justify their role as a central tool for the linguist. Note that being useful for the linguist and being represented in the language user's mind are not necessarily the same thing (Haspelmath 2018: 92, fn. 7).<sup>25</sup>

This is in sharp contrast to DM, which inherits the demand for providing a psychologically real model from its background in GB/Minimalist syntax. The demand concerns every aspect of the model, including the irrelevance of paradigms. However, pertinent psycholinguistic evidence seems to be scarce. Even Barner & Bale (2002), who do collect many pieces of psycholinguistic evidence from the literature, only cover one aspect of the model (viz. roots lack syntactic category information). There is a need for more comprehensive, custom-tailored experiments, in particular for ones checking whether or not speakers draw on ready-made paradigms in language production and/or comprehension.

In comparison to DM, the psycholinguistic evidence adduced in favour of CxM seems to be much more encompassing. In particular, it has been shown that word forms can be both, computable as well as holistically stored, without contradiction (Zwitserlood 2018, Masini & Audring 2019: 7–8, Jackendoff & Audring 2020: Ch. 7).<sup>26</sup> The crucial point from the perspective of CxM, though, is not that

<sup>25</sup>This is one of the few aspects of Haspelmath's comparative concepts that I embrace, cf. Reiner (2021).

<sup>26</sup>Typological coverage might be better; however this is a problem of current psycholinguistic research more generally, which may be overcome in the future (cf., e.g., http://www.llf.cnrs.fr/ labex-efl (as of 24.03.2020)).

### 5 Recursion and paradigms

word forms *can* be stored but that they *are* in fact stored (different from DM, cf. McGinnis-Archibald 2016: 392, fn. 3). If stored forms relate to each other, they may constitute paradigms in Booij's (2016) sense.

*The attitude towards restrictiveness* seems to differ vastly between the theories discussed here. The difference is often said to lie in striving for maximal empirical coverage (CxM + Autonomous Morphology) vs. striving for testable predictions (DM). For example, Kramer (2016) draws this line. However, I will argue that the difference is rather about *where* to formulate restrictions.

Surveying work in Autonomous Morphology, it would be utterly wrong to say that these researchers do not formulate restrictions: every generalisation over data is a falsifiable prediction to the effect that new data are hypothesised to comply with the generalisation. For example, Baerman et al.'s (2005: 220–221) types of syncretism restrict the range of syncretisms we expect to find in the languages of the world. In short, there *are* restrictions and they reside in generalisations over (descriptions of) data, with paradigms being used as a central tool for description. Another – more obvious and more recent – example of restrictiveness is Herce (2019).

Similarly, practitioners of CxM or Construction Grammar more generally use the construction as a maximally flexible tool of description *before* they start looking for cross-linguistic tendencies in the data thus described (e.g., Goldberg 2006: Ch. 7–9). The tendencies are then to be explained by general cognitive principles (Goldberg 2006). Crucially, cognitive principles restrict the range of what we expect to find in human (linguistic) behaviour but there is no need to build the restrictions into one's descriptive tools.

This is different from DM, which, as a generative theory, intends to model language as a specific competence, identified by specific restrictions, which, consequently, have to be part of the model. Thus, anything that cannot be derived by the model is predicted not to be accepted by native speakers and vice versa (derivable ↔ accepted).

So both kinds of approach acknowledge restrictions (and I am agnostic as to which way of doing so is the better one). Incidentally, both also embrace the liberties of language, at least to a certain extent. For DM, this might be not so obvious; however recall that they have list C at their disposal, although this part of the architecture seems to be the one that is worked out least of all.

Taking stock of this section, it appears that, surprisingly, Autonomous Morphology never patterns with CxM, cf. Table 10.

### Tabea Reiner


Table 10: Summary of theory comparison

## **3 Defending morphology and paradigms**

Recalling Table 10, one might expect that Autonomous Morphologists defend their approach against CxM sense much more forcefully than they defend it against the various incarnations of DM. However, the opposite is true. The present section will elaborate on this remarkable situation and evaluate tentatively whether the Autonomists' defence of morphology by itself – including paradigms – succeeds.

Virtually everyone who advocates the idea of Autonomous Morphology includes in their overview publications some words on why DM fails. For example, Brown & Hippisley (2012: 19–29) present DM as an alternative but inferior approach. Other examples include Spencer (2004: 73–89), Aronoff (2016: 194–195) and, avant la lettre, Aronoff (1994: 82–85). Additionally, Spencer's (2019) review of Embick (2015) represents a recent argument against DM from an Autonomist's perspective.

One of the central arguments is that there are phenomena that cannot be described in purely syntactic or lexical terms, i.e. so called morphomes (for a recent overview of the notion cf. Enger 2019: 160–166). Crucially, the most prominent examples of morphomes directly refer to paradigms: inflectional classes and patterns of stem allomorphy (Maiden 2009). For substantiation, consider an instance

### 5 Recursion and paradigms


of the latter case, viz. Maiden's (2009) L-pattern in Romance verb morphology. Table 11 represents the pattern abstractly, Table 12 gives an example.

Crucially, this pattern, among two additional ones, pervades irregular verbal morphology in Romance languages and comes without any obvious lexical or syntactic core: why should certain verbs show the same stem for exactly these feature value combinations and not for others? In particular, it is hard to imagine a semantic or syntactic property that is shared by, e.g. *digo* and *digan* but not *digo* and *dicen*. What the pattern does refer to, however, are the pairings of word forms and feature values they realise, i.e. the cells of a paradigm: it is always the same set of cells that shares a stem. So while the stem as such is not predictable, its distribution is and, crucially, this distribution refers to paradigm set-up. In short, morphomes seem to provide a very clear indication that, after all, there is something genuinely morphological about language, more precisely something that must be captured by paradigms in Stump's sense (cf. Section 1).<sup>27</sup>

In a way, this line of research is continued by Ackerman & Malouf (2013), who adduce evidence that it is the complexity of paradigm structure rather than the complexity of individual distinctions and realisations that predicts learnability (so here also the question of psychological reality is touched upon). However, their results have recently been called into question (Johnson et al. 2021).

Moreover, recall that also in DM there is more between syntax and spell-out than just vocabulary insertion (Section 2.2, Section 2.3). Against this background, it does not come as a surprise that by now also proponents of DM have started to embrace morphomes. Trommer (2016) provides a "postsyntactic morphome cookbook", not even using the full DM-machinery. To be prudent, Trommer's

<sup>27</sup>For a typological survey of morphomic structures cf. Herce (2020a).

### Tabea Reiner

paper is the only DM-treatment of morphomes that I am currently aware of. Future research will show whether Autonomous Morphologists will be beaten at their own game.

A more urgent, but largely unrecognised threat comes from the "friendly takeover" by constructionist accounts (Section 2.1 and Section 2.3), especially when considering that even morphomic patterns might be conceived of as second order schemas involving homonymy. While, e.g., Spencer (2001, 2004: 84–86) welcomes constructionist thinking, he does not seem to recognise that this family of theories tears down the very boundaries he tries to defend: as indicated in Section 2.3 above, Construction Grammar does not recognise any qualitative difference between syntax and the lexicon, let alone a discrete level in between to be justly called morphology (cf. also Goldberg 2006: 5, 2013: 17, Hoffmann & Trousdale 2013: 1).

As a consequence, it does not appear to be a coincidence that precisely those morphologists who adopt a constructionist perspective most consistently and most explicitly (e.g., Booij 2010) are those who are *not* at the same time Autonomists. Moreover, while (Autonomist) Network Morphology (Brown & Hippisley 2012) stands side-by-side with Canonical Typology in the Surrey Morphology Group,<sup>28</sup> the latter theory is free to develop an integrative perspective on autonomy. That is, canonical (ideal) morphology might be regarded as *non*autonomous, reducible to either syntax or the lexicon or both, whereas deviations from this ideal represent autonomous morphology – with the deviations often presenting themselves as patterns in paradigms. To the extent that it is the deviations rather than the canonical ideal that we expect to find in real languages, Canonical Typology and Network Morphology fit well together indeed.<sup>29</sup> This relationship, however, does not prevent Autonomous Morphology, including its commitment to paradigms, from being largely absorbed by constructionist approaches, as described above.

In sum, so far the attempts at defending morphology (as a distinct level of description) and defending paradigms (as defined in the introduction of this paper) have not ultimately succeeded – or the attempts are missing altogether. Does this mean that paradigms have nothing to offer for any linguistic theory? To my mind, one use of paradigms remains in any case. This is the topic of the next section.

<sup>28</sup>Even literally so: https://www.smg.surrey.ac.uk/approaches/ (as of 24.03.2020).

<sup>29</sup>These deliberations are what I understand from Gaglia & Hinzelin (2016), Hippisley (2017) and Herce (2020b).

### 5 Recursion and paradigms

## **4 A remaining use for paradigms: Restricting recursion**

The marginalisation of paradigms as fundamental organisational units has been noted and criticised recently by Diewald (2020), focussing on constructionist approaches. Arguing from a diachronic perspective, she reminds us that [+ paradigmaticity] and [− paradigmatic variability] are two of Lehmann's (2015: 132) six famous criteria for grammaticalisation (and that speaking of grammaticalisation is pointless without an analytical distinction between grammar and non-grammar). Thus, in Diachronic Construction Grammar paradigms are both: marginalised as well as fundamentally needed. To escape from this dilemma, she suggests the following concept of paradigms: as constructions-of-constructions they are complex signs, which may constitute nodes in a constructional network. She notes that this is similar to Booij (2016); however, as far as I can see, Diewald (2020: 297– 301) goes one step further in that she explicitly excludes open class paradigmatic relations from the notion of paradigm. This makes, I would argue, a huge difference, in particular from a diachronic perspective. For example, consider German *R*/*richtung* ʻdirection/to' in (22) vs. German *R*/*riesen* ʻgiant' in (23).

	- ʻI am walking towards the station.'
	- a. Dort ist eine Riesen-Statue. there is a giant-statue ʻThere is a giant statue.'
	- b. Dort ist eine riesen Statue. there is a giant statue ʻThere is a giant statue.'

The former would qualify as grammaticalised on Diewald's (2020) account since it enters the closed class of prepositions. Note that I am assuming here that this class may be described by pairs of feature values. The latter, by contrast, would not qualify as grammaticalised on Diewald's (2020) account since there is no closed class of either "grading initial parts of compounds" or adjectives.

Thus, Diewald's (2020) conception of paradigms might reintroduce them into Construction Grammar as fundamental organisational units. Crucially, however,

### Tabea Reiner

as nodes in constructional networks they have to be psychologically real by themselves (not merely being a consequence of some other psychologically real entities and processes). Here, Diewald (2020: 306–310) provides a rough design for a psycholinguistic study as well as a core linguistic argumentation to the effect that certain grammaticalisation processes are hardly conceivable without mental representations of paradigms in her sense.

So again, the question whether paradigms (in whichever sense) are psychologically real turns out to be pivotal. To the extent that this remains a tricky question, I suggest that anyone arguing in favour of a fundamental status of paradigms should look for arguments independent of this question, i.e. for a use of paradigms that is just that: fundamental to the linguist but not necessarily reflected in the language user's mind (very much in the spirit of Aronoff 2016: 197–198, quoted above). In the rest of this section, I will present such a use.

There is one kind of data that neither CxM nor DM is good at handling, i.e. limits on recursion. The following paragraphs provide a snapshot of such data, explains why they are problematic for the two approaches and, finally, presents the paradigm in Stump's sense as a solution.

Basically, CxM and Construction Grammar more generally share with Autonomous Morphology the strategy to seek restrictions in generalisations over data rather than building the restrictions directly into their descriptive tools (cf. Section 2.3). This strategy adheres to Haspelmath's (2004) postulate of a sharp distinction between description and explanation. At the same time, this strategy requires the descriptive tools to be as flexible as possible: they must be able to capture whatever may be found in a language. So the often heard accusation that "everything is a construction" (e.g., Van Valin 2007: 236) misses the point since here lack of restrictiveness in descriptive tools is a virtue, not a weakness.

Against this background, it is clear that CxM can handle every kind of data well: all we need is an association between a form and a meaning, i.e. a construction. This means that individual morphs like, e.g., thematic vowels are not forced to have any meaning in isolation. It is the verb form as a whole that has a meaning to begin with and this association of meaning and form constitutes a construction. Likewise it is the clause as a whole that has a meaning and again, this association constitutes a construction. Incidentally, a construction may have constituent constructions and in this sense there is internal structure; however the internal structure is not generated bottom-up via valency or subcategorisation.

So verb forms are constructions, clauses are constructions, and the rare case of a monofunctional affix (that is, rare in SAE) is a construction, too. Ideally, we obtain a comprehensive network of constructions for each language. These

### 5 Recursion and paradigms

networks, I hold, are again organised in a network to the extent that they share certain properties.<sup>30</sup> Crucially, these properties are abstractions over individual minds rather than having any psychological reality in their pure form. In principle, the most encompassing of networks might be read as set of generalisations about language. As argued above (Section 2.3.), generalisations *are* predictions, including predictions about what there is *not*. To take an example from Autonomous Morphology, if there is no language to be found with syncretism of 1sg, 2du, and 3pl subject agreement,<sup>31</sup> we may predict that this kind of syncretism is impossible. The next step is gathering more data (from the same as well as other languages) in order to see whether the prediction of absence holds. However, most practitioners of CxM and Construction Grammar more generally do not seem to be concerned too much with that "dark matter" (I am borrowing the metaphor from Werner 2018 on word formation here). To pick a recent example, the ongoing project "FrameNet & Konstruktion des Deutschen" focuses, according to its self-description, on the wealth of what there is, not on finding systematic gaps.<sup>32</sup> This might seem to be a mere matter of emphasis – but if we do not even care for the potential absence of certain phenomena we will not be able to explain (in any sense of the word) that absence.

Beside a certain type of syncretism, another example for a potentially nonoccurring kind of phenomenon is the one I am concerned with in the present section: meaningful iteration of affixes (presuming here that we know in each case what counts as an affix as opposed to a clitic or function word and that we can always decide whether the affix is inflectional or derivational). Consider (24) based on (25).

(24) Turkish (p.c., Seda Yilmaz Wörfel) yap-tı-m do-pst-1sg ʻI did'

(25) Turkish (p.c., Seda Yilmaz Wörfel) \* yap-tı-tı-m do-pst-pst-1sg intended: ʻI had done'

Judging from the "syntax-like" systematicity of Turkish verbal forms and from the possibility to iterate derivational (causative) affixes (cf. Section 2.1), the unavailability of a systematic iteration in (25) is surprising. Why not have a past

<sup>31</sup>Surrey Person Syncretism Database, https://www.smg.surrey.ac.uk/personsyncretism/.

<sup>30</sup>For a pertinent but rather non-constructionist proposal, cf. Reiner 2021.

<sup>32</sup>https://gsw.phil.hhu.de/; the project relates to parallel projects for other languages.

### Tabea Reiner

form (*yaptım*) as the input to another past formation (-*tı*), yielding a past-of-past meaning? In other words: if we define recursion as the application of a given structural-semantic operation to an output of some former application of the same operation, example (25) represents the limits of recursion. In fact, I am not aware of any language that allows for recursion of inflection (or what is usually described under the rubric of inflection).<sup>33</sup> So the limitation seen in (25) might be quite general. For present purposes, however, the important point is that there is a limitation at all.

Thus, this is a kind of language fact that CxM will capture implicitly but not care for. What about DM? Striving for a restrictive model, this approach is expected to predict the ungrammaticality of (25) explicitly. At first sight, this is true: as long as there is only one T-node in the syntax, the pattern in (25) is indeed excluded, since there is simply no terminal for the second -*tı* to attach to.

However, in DM the problem resurrects at a later stage of the derivation: after vocabulary insertion has selected the most specific candidate available, how can we stop the mechanism from starting over again, yielding for example \**play*-*eded* (Embick 2015: 97)? Note that this is not just an empty iteration of forms, since every vocabulary item, including [past] ↔ /əd/, comes with a function. More precisely, a vocabulary item is the pairing of a phonological form and information on where this form may be inserted (cf. Section 2.2.1.), with this information consisting of morphosyntactic feature values. So when [past] ↔ /əd/ attaches to the single T-node multiple times (in principle, ad infinitum), it adds [past] every time. Embick's solution to this problem is a stipulation called *uniqueness*: "In a derivation, only one Vocabulary Item may apply to a morpheme" (Embick 2015: 98). So DM can handle data like (25) but only by means of a stipulation.

According to Spencer's (2019) review of Embick (2015), the need for the uniqueness stipulation arises from the "attempt to derive word structure (directly) from syntactic structure" (Spencer 2019: 218). I tried to show above that syntax is not the problem here. However, I agree with Spencer (2019) that paradigm-based theories are not affected by the problem of recursion. The reason is that the paradigm as such restricts recursion: since rows and columns represent different pieces of information, iteration is excluded automatically. For example, since past will not appear in both, a row and a column, no cell can contain a realisation of past-past. Consider the partial paradigm for Turkish in Table 13.

<sup>33</sup>But cf. Van der Voort (2016) for potential examples. Importantly, it has to be the whole structural-semantic operation that is applied twice. Hence, for instance reduplicative plurals do not count, since here we are dealing with one structural operation having one semantic effect.

### 5 Recursion and paradigms


Table 13: Partial paradigm for Turkish, gathered from Kornfilt (1997: Ch. 2.1.3).

Thus, the paradigm can do what Construction Grammar fails to do and what DM needs a stipulation for: modelling the limits of recursion. Another advantage of paradigms is that no one needs to postulate them. They do have a merely secondary status in CxM as well as in DM (cf. Section 2); however, they are there.

Before concluding the paper, let me draw your attention to a limitation of my reflections: I focussed solely on phenomena that are usually considered to be inflectional in a narrow sense. Though presenting theories that can as well cope with periphrasis and word formation, I did not have much to say about these phenomena. Indeed, these appear to be inherently different from inflection when it comes to recursion: they do permit it, to a certain extent. For recursion in periphrasis cf. Rothstein (2012, 2013a,b) on double futures and the references therein to the wealth of works on double perfects. For recursion in word formation cf., e.g., Brattico et al. (2007).

## **5 Conclusion**

The "Morphome Debate" (Luís & Bermúdez-Otero 2016) is far from settled and it remains to be seen whether morphomes provide an ultimate argument for morphology as a distinct level of description, organised by paradigms in Stump's sense. In the present paper I only presented an outline of this debate, not even elaborating on inflectional classes as morphomes. The main point of the paper was showing that even if paradigms are becoming secondary in current theorising, the traditional paradigm in Stump's sense still serves an apparently unique function: it provides an economic way of modelling restrictions on recursion in inflection. More precisely, the habit of having the rows and columns host different features (and a fortiori different feature values) prevents any given feature value from operating on some former application of itself like in \*[*play*-*ed*]-*ed*.

### Tabea Reiner

However, recall that paradigms in Stump's sense are pairings of word forms and the morphosyntactic properties they realise. Thus, an old problem raises its ugly head again, i.e. the very problem with which Haspelmath (2011) is concerned: how can we tell what constitutes a word (form) in the first place? I doubt that the delegation of this task to Canonical Typology (cf. Section 2.3) is sufficient. The only thing that, in this respect, Canonical Typology can tell us about a given string in a given language is this: it is a canonical word (form) to such and such a degree. However, in order to decide whether the string can realise a paradigm cell, we need to know whether it *is* a word or not, categorically.

## **Abbreviations**

Please note that abbreviations in examples adopted from other authors are spelled out in footnotes throughout.


## **References**

Ackerman, Farrell & Robert Malouf. 2013. Morphological organization: The low conditional entropy conjecture. *Language* 89(3). 429–474.

Aikhenvald, Alexandra Y. & R. M. W. Dixon. 2017. Introduction: Linguistic typology – setting the scene. In Alexandra Y. Aikhenvald & R. M. W. Dixon (eds.), *The Cambridge handbook of linguistic typology* (Cambridge Handbooks in Language and Linguistics), 1–35. Cambridge: Cambridge University Press.

### 5 Recursion and paradigms


### Tabea Reiner


5 Recursion and paradigms


### Tabea Reiner


5 Recursion and paradigms


5 Recursion and paradigms


## **Chapter 6**

## **Redundant indexicality and paradigmatic reorganisations in the Middle Danish case system**

## Bjarne Simmelkjær Sandgaard Hansen

University of Copenhagen

The Danish case system changed profoundly throughout the Middle Danish era. Based on examples from mainly three texts written in East Danish (Scanian dialect), I describe the steps of and stages in these changes and claim that they were caused neither by an unstressed-vowel-neutralising sound law nor by language contact as often assumed, but by various interrelated processes of grammaticalisation. I focus on one of these processes, viz., that the fixed topology of the Middle Danish noun phrase simply made noun-phrase internal agreement by means of case marking redundant and caused the loss of the indexical relations signalling this agreement, which, in turn, contributed to the gradual phase-out of case marking. Moreover, I relate this phase-out to two general linguistic principles, viz. those of markedness agreement (Andersen 2001: 27–37) and single encoding (Norde 2001: 258– 261). Finally, based on Nørgård-Sørensen et al. (2011: 5–6) and Nørgård-Sørensen & Heltoft's (2015: 262–263) five criteria for what constitutes a grammatical paradigm, I also demonstrate that, irrespective of the existence of some level of free variation, the Middle Danish case system may be described paradigmatically and, correspondingly, that the changes it undergoes constitutes an instance of paradigmatic and thus grammatical change.

## **1 Introduction**

Grammaticalisation as defined by Nørgård-Sørensen et al. (2011: xi, 71–72) and Nørgård-Sørensen & Heltoft (2015: 261–262) equals paradigmatisation. This implies that, in order to count as an instance of grammar, any linguistic phenome-

Bjarne Simmelkjær Sandgaard Hansen. 2022. Redundant indexicality and paradigmatic reorganisations in the Middle Danish case system. In Gabriele Diewald & Katja Politt (eds.), *Paradigms regained: Theoretical and empirical arguments for the reassessment of the notion of paradigm*, 173–200. Berlin: Language Science Press. DOI: 10.5281/ zenodo.5675849

### Bjarne Simmelkjær Sandgaard Hansen

non must me paradigmatic, and any grammatical change must be describable as a paradigmatic change.

In the present paper, I will account for one type of grammatical and thus paradigmatic change, viz., the change of the case system in Middle Danish of the Scanian dialect from having four fully functional cases to, virtually, none in the nominal paradigm. I shall focus mainly on the changes that happened to the original genitive and dative and the change in relations between these and the accusative, since Jensen (2011) has already effectively accounted for the changes happening to the relations between the original nominative and accusative.

More specifically, I will shed light on one among several factors causing this change, viz., redundancy in the indexical relations of noun-phrase internal agreement, and present the paradigmatic consequences of the changes caused by this and other factors, including in particular the paradigmatic consequence of seemingly free variation in the case system during the period of change. Finally, in tandem with that, I will discuss the consequence of free variation to the aforementioned scholars' understanding of paradigmatic essence.

In order to fulfil these tasks, I will first provide synchronic descriptions of the Middle Danish use of case with examples from Middle Danish texts in Section 2, after which I will seek to explain the developments of the case system by means of processes of grammaticalisation in Section 3. These two sections, which build on – and constitute a concise version of – my previous outline of these matters in Hansen (2021), serve as the starting point for my discussion of the paradigmatic consequences, which will occupy Section 4, while Section 5 will constitute the conclusion of the article.

## **2 Data from Middle Danish**

## **2.1 Relevant texts**

As I have described in detail in Hansen (2021: 282–289) and will now recapitulate here, a comparison of three texts written in the East Danish (Scanian) dialect in the first half of the 15th century, viz., *Skånske Lov* (SkL) after Cod. E don. var. 136, 4<sup>o</sup> , *Sjælens Trøst* (SjT), and *Søndagsevangelier* (SdE), reveals the existence of multiple simultaneous systems of case application in this period. What follows is therefore a brief outline of the systems found in these texts. For a full description of the details in the systems, see Hansen (2021: 282–289).

### 6 Redundant indexicality and paradigmatic reorganisations in Middle Danish

## **2.2 Case marking on both nouns and on typical modifiers**

In one system, genitive and dative case is marked on all members of a noun phrase, i.e., both on nouns and on typical modifiers (adjectives, articles, pronouns, numerals and other determiners), in situations with a potential for genitive or dative government.

Marking for dative thus appears both on the possessive pronoun *sinum* 'their' and the noun *thiænarum* 'servants' in (1), where the function of the noun phrase as the indirect object of *budho* 'they commanded' triggers the use of dative, and similarly on both the indefinite pronoun *ene* 'a', the adjective *longe* 'long' and the noun *iærnlenkio* 'iron chain' in (2), where the preposition *mæth* 'with' governs the dative. In the 15th century, this system prevails with the feminine singular and with the (genitive/dative) plural for all genders, but especially in SkL, this system is also found with other case forms, as evidenced by (3) which represents a case of a preposition governing the dative.

(1) SjT: 70<sup>25</sup>

buth-o command-pst.3pl sin-um their-m.dat.pl thiænar-um servant(m)-dat.pl<sup>1</sup> 'they commanded their servants [that …]'


## **2.3 Case marking on typical modifiers only**

A second system is the one found in examples like (4–6) where all members of a noun phrase but the typical noun-phrase head, i.e., the noun itself, are marked for case.<sup>2</sup>

<sup>1</sup>All category labels used in interlinear glossing follow the Leipzig Glossing Rules, the only additions being obl and c, which signify "oblique" and "commune" (common gender), respectively.

<sup>2</sup>Although nouns do not inflect for case in this system, they still inflect for number; hence, I still need to mark them for inflectional endings as per, e.g., the ∅-ending in (4–6).

Bjarne Simmelkjær Sandgaard Hansen


gifv-in give-imp.2pl th-øm it-acc.pl fatig-o poor-n.dat.sg folk-∅ people(n)-sg 'give them to poor people'

(6) SjT: 122<sup>9</sup> mæth with en-um a-m.dat.sg stor-um great-m.dat.sg hær-∅ army(m)-dat.sg 'with a great army'

As revealed by a comparison of (3) and the structurally almost identical example (4), at least SkL after E don. var. 136, 4<sup>o</sup> had some level of apparently free variation between the former system of case marking on all members of the noun phrase and this system of case marking on all members but nouns. Because of this paradigmatic choice, it may also be, however, that the seemingly case-uninflected noun *thing* 'moot' actually does inflect for case, but merely expresses the traditional dative content with the endingless accusative instead of the historically expected dative. Only in a linguistic system with no option left for case marking on nouns may one establish with certainty that an endingless noun is indeed also uninflected.

Examples (5–6) stem from SjT, and both display the same system of no case marking on nouns. Here, it seems more certain that the nouns *folk* 'people' and *hær* 'army' are, indeed, uninflected for case, since in this text, singular forms of masculine and neuter nouns ending in a consonant never enter into a paradigmatic opposition with variants that are inflected for case.

On the surface, example (7), which stems from SdE, resembles (4–6) by marking for case on typical modifiers only, not on nouns. Here, however, the traditionally dative-governing preposition *met* 'with' suddenly governs the accusative instead. For that reason, the endingless form *renlik* 'cleanliness' in (7) might actually, at least theoretically, have represented the accusative of the noun rather than the noun stripped for case marking, just as with the noun *thing* 'moot' in (4). The second member of this prepositional phrase, i.e., *fasta* 'fasting', clearly represents the form uninflected for case, however, since the accusative would have been expressed by the oblique form \**fasto* or \**fastæ*/*faste* (with unstressed-vowel neutralisation).

### 6 Redundant indexicality and paradigmatic reorganisations in Middle Danish

(7) SdE: 42<sup>27</sup>

met with lang-en long-m.acc.sg renlik-∅ cleanliness(m)-(acc.?).sg ok and fasta fasting(f).sg 'with long cleanliness and fasting'

The most remarkable feature of (7) is, however, not the status of*renlik* and *fasta* as either accusatives or bare nouns uninflected for case, but the unambiguous use of the accusative in the adjective *langen* 'long'. As I have already mentioned, the preposition *met* 'with' traditionally governed the dative, not the accusative. If this were a standalone example of failure of choosing the historically expected case, I might have dismissed it as a simple error, but seeing that one may come across several structurally similar examples – e.g., (8–9) with the preposition *til* 'to', which traditionally governed the genitive – this is hardly the case.

(8) SjT: 3227–28

af of hørdh-a-na herdsman(m)-pl-the.m.acc.pl 'from the herdsmen'

(9) SjT: 96<sup>18</sup>

til to en a-m.nom/acc.sg hælgh-an holy-m.acc.sg abod-a abbot(m)-obl.sg 'to a holy abbot'

Rather, the explanation for the use of the accusative in (7–9) lies either with prepositions simply allowing accusative government in addition to their traditional government, i.e., *met* 'with' may govern either the accusative or the traditional dative, or with the accusative simply taking optionally over for the dative and genitive in all regards, i.e., so-called participation (Hjelmslev 1935, 1970: 87, Bjerrum 1966: 8–10, 38–40, Andersen 2001: 46, Heltoft 2010: 16–18, Jensen 2012, among others). The concept of participation constitutes a typical example of markedness relations. In an opposition between two members of a paradigm, one member is restricted in its functions (in this case: the genitive and the dative), whereas the other member both covers its own restricted functions and may participate in the functions of the first member (in this case: the accusative). Needless to say, accusative participation may also constitute an important trigger for the loss of case marking on Middle Danish nouns seen in (4–6), since in most nominal classes, the accusative ended in *-∅* and was thus formally identical to the bare stem; see also Norde (2001: 250–251) on a similar situation in Old Swedish.

Bjarne Simmelkjær Sandgaard Hansen

## **2.4 No case marking**

Finally, I have recorded a third – or is it a fourth? – system in Middle Danish in which case marking is neither present on nouns, nor on the typical noun-phrase modifiers, as evidenced by (10–11). This is largely reminiscent of the nominal system recorded in modern Danish, which (12–13) serve to illustrate.

(10) SjT: 122<sup>9</sup>

mæth with en-∅ a-m.sg stor-∅ great-m.sg hær-∅ army(m)-sg 'with a great army'

(11) SdE: 178–9

fran from all-∅ all-f.sg køtlik-∅ corporeal-f.sg lust-∅ lust(f)-sg 'from all pleasures of the flesh'

(12) Modern Danish

jeg I giv-er give-prs e-t a-n.sg klog-t wise-n.sg barn child(n)\sg bog-∅-en book-sg-the.c.sg 'I give the book to a wise child'

(13) Modern Danish til to e-n a-c.sg ung-∅ young-c.sg dreng-∅ boy(c)-sg 'to a young boy'

I cannot rule out completely, though, that (10) represents not the lack of inflection, but instead a traditional nominative outside of its original domain and with the novel function of marking foreground information (Jensen 2011: 264), but (11) is an unequivocal example of true caseless forms in both the modifier and the noun, as are (12–13) from modern Danish.

## **2.5 Older and younger stages of Danish**

When compared to the situation in an older manuscript of SkL (such as Cod. Holm. B 74, 4<sup>o</sup> from the first half of the 13th century) and to later linguistic stages such as modern Danish, it is evident that what the nominal system undergoes is a development from case marking with the historically expected case on all members of a noun phrase, as in (1–3), to no case marking at all, as in (10–11).

For instance, as witnessed by (14–16), SkL after Cod. Holm. B 74, 4<sup>o</sup> displays many instances of traditional case marking, even in contexts where traditional

### 6 Redundant indexicality and paradigmatic reorganisations in Middle Danish

case marking would normally not occur on nouns in the 15th century, viz., in the masculine and neuter singular. I have also found examples of accusative participation on the noun as in (17), where the accusative form *brofial* 'plank' replaces the historically expected dative form \**brofialu* or \**brofialo* (Bjerrum 1966: 39). This form, to which the participating accusative form *brofial* would stand in paradigmatic opposition, is not attested in Cod. Holm. B 74, 4<sup>o</sup> , however, but cf. the dative form *brofiæle* in (18) from an addendum to *Skånske Lov* (Add.SkL), printed in a manuscript (Cod. Holm. B 73) that exhibits neutralisation of some unstressed vowels.


Conversely, the frequency of case marking decreases significantly in a 16thcentury post-reformation text such as the first full Danish bible translation, *Christian 3.s danske Bibel* (Chr.3.B) from 1550. Examples (19–20) from Chr.3.B illustrate the absence of case marking even in such contexts where case distinctions were kept for the longest time, viz., in the feminine singular and in the plural, respectively.

Bjarne Simmelkjær Sandgaard Hansen


## **2.6 Stages in and developments of the Middle Danish case systems**

When taking into account both my outline of the situation prior and posterior to the 15th century and the possible steps of and pivots for reanalysis, I may sum up the potential stages in and developments of the Middle Danish case systems as follows.


As I have already shown, these systems and subsystems existed side by side within one and the same text. A simple comparison of, e.g., (1), (6), (8) and (10), which all stem from SjT, serves to illustrate this.

## **3 Explaining the developments of the Middle Danish case system**

## **3.1 Traditional views**

Traditionally, two views have prevailed on how the changes in the Middle Danish case system came about, the former generally more accepted – or at least more frequently mentioned – than the latter.

### 6 Redundant indexicality and paradigmatic reorganisations in Middle Danish

First, viewing the reductions in the Danish case system as a result of sound laws, above all the Danish unstressed-vowel-neutralising sound law *e*/*a*/*o* > [ə], has long been the prevalent position among historical linguists (Falk & Torp 1900: XIV–XV, 16–17, Meillet 1922: 71, 95–100, 113, Skautrup 1944: 266, etc.). This reductionist or "phonology-first" view holds that the coalescence of unstressed vowels (i.e., typically vowels in non-first syllables) resulted in homophony and syncretism of many inflectional endings. Such an explanation may theoretically work not only for Danish, but also for other languages with phonological reductions in non-first syllables; see, e.g., Barber et al. (2009: 167–168) on English.

Second, Wessén (1954: 27) regards Middle Low German influence as (one) reason for the Scandinavian case-system reductions:

Vi har stor anledning att tro, att det främmande inflytandet har sträckt sig jämväl till ordens böjning och till uttalet. Då fornspråkets rika formsystem mot medeltidens slut upplöses och förenklas, har man med skäl sökt en av orsakerna därtill i att de inflyttade tyskarna aldrig kunde lära sig att rätt bruka de gamla kasusformerna och ändelserna; deras förenklade ordböjning smittade efterhand av på landets egna barn.<sup>3</sup>

Although admitting that it is difficult to establish the exact extent of the Middle Low German influence, Haugen (1976: 65) agrees with Wessén by noting that both English and Scandinavian underwent case-system reductions while dominated by other languages and that Low German has a structure similar to that which the mainland Scandinavian languages adopted. Several scholars still include this specific influence or similar types of language contact in their list of causes for the mainland Scandinavian case-system reductions; see, e.g., Norde (2001: 243) on Swedish.

## **3.2 Challenging the traditional views**

Appealing as these two traditional explanations may seem, they suffer from some major deficiencies.

Taking the reductionist view first, the neutralisation of unstressed vowels and the reductions in the case system simply do not seem to be connected. One would

<sup>3</sup>My translation: "We have great reason to believe that the foreign influence has encompassed even the inflection of words and the pronunciation. Since the rich morphological system of the ancient language is dissolved and simplified around the end of the mediaeval period, scholars have reasonably regarded the following circumstance as one of the reasons for that, viz., that the immigrating Germans could never learn to use the old case forms and endings correctly; their simplified inflection gradually rubbed off on the country's own children."

### Bjarne Simmelkjær Sandgaard Hansen

expect this sound law to be operational in texts where the case system is in the process of change, but as all the examples from SjT reveal, this is certainly not the case. For instance, (9) shows an instance of accusative where the traditional system would have dictated a genitive, even though both *hælghan* 'holy' and *aboda* 'abbot' preserve the unstressed *-a*, and in (5), *fatigo* 'poor' keeps its unstressed *-o* in spite of the absence of the historical dative singular ending \**-i*/\**-e* on *folk* 'people'. The reductionist view meets the exact same challenge when attempting to explain the case-system reductions in Swedish. As Jensen (2011: 18) points out, the Swedish case system has been reduced and changed to the same extent and more or less with the same result as the Danish one, but Swedish does not display any substantial weakening of unstressed vowels.

SdE also demonstrates the mismatch between the unstressed-vowel-neutralising sound law and the case-system reductions, but in the opposite manner. In SdE, unstressed vowels have been neutralised in many positions, as revealed by, e.g., *langen* (< *langan*) 'long' in (7), but this neutralisation has not prevented a by-and-large retention of the Danish case system, illustrated again by the accusative form *langen* in (7). Following Baechler & Pröll (2018: 4–5), I will stress this point even further by drawing attention to a language like standard German, in which the process of unstressed-vowel neutralisation is just as advanced as in modern Danish despite the preservation of a functional distinction between all four historical cases: nominative, accusative, genitive and dative.

Finally, leaving the possibility of a causal correlation between the unstressedvowel-neutralising sound law and the case-system reductions aside, this sound law is not capable of explaining the loss of consonantal endings. Neither the functional changes and subsequent loss of the historical nominative singular ending *-Vr* in the mainland Scandinavian area (Jensen 2011: 18) nor the Faroese and dialectal Norwegian and Swedish loss of the genitive in *-s* (Enger 2013: 6–7) finds any catalyst in reductionist sound laws. To sum up my objections against the reductionist view, I will cite Loporcaro (2018: 42) who concludes on some instances of gender agreement in Italian dialects that "there is no deterministic impact of sound change on morphosyntax".

Turning now to the second traditional explanation of the mainland Scandinavian case-system reductions, i.e., language contact, I will call attention to Ringgaard's (1986: 177–182) highly valid objection of a mere chronological mismatch. Followed by Askedal (2005: 2–3) and Enger (2013: 13–14), Ringgaard claims that Middle Low German would exert its allegedly system-changing influence too late on Danish and the remaining mainland Scandinavian languages for it to constitute a factor. The Middle Low German influence was most pervasive in the 14th century, but one may register case-system reductions already in the earliest

### 6 Redundant indexicality and paradigmatic reorganisations in Middle Danish

manuscripts of the Danish regional laws, the language of which may have been settled as early as the end of the 12th century. In all fairness, while Ringgaard's criticism is indeed relevant for Danish, it may be slightly less so for Swedish, for which Wessén (1954: 27) postulated his claim originally.

## **3.3 Processes of grammaticalisation**

Scholars like Andersen (e.g. 2010: 143–144), Heltoft (e.g. 2010, 2022 [this volume]), and Petersen (2018) offer an alternative to the traditional explanations of the Danish case-system changes. They all attribute certain instances of language change, including those of the Danish case system, to processes of grammaticalisation, which they define as processes of change in the function and contents of the grammatical signs and in the paradigmatic oppositions between them (Andersen 2006, 2010: 123, Nørgård-Sørensen et al. 2011: xi, 7–8, 11–17, Nørgård-Sørensen & Heltoft 2015: 261–262, etc.).<sup>4</sup>

For instance, Heltoft (2010: 13–22) describes the changes in the Middle Danish case system as part of a larger process that turns the Old Scandinavian noun phrase into a determiner phrase in the modern mainland Scandinavian languages. Jensen (2011: 201–232, 283–311), in turn, focuses specifically on a reanalysis of the relationship between the nominative and the accusative, resulting in nominatives marking only such subjects and subjective complements that also provide foreground information. Finally, Petersen (2018: e.g. 63–89) connects all the changes

<sup>4</sup>By entailing both grammation, regrammation and degrammation (i.e., the rise, change and dissolution of grammar, respectively) and also insisting on syntax (topology and constructions) and syntactic changes forming part of grammaticalisation (Nørgård-Sørensen et al. 2011: 43– 45), this definition of grammaticalisation goes beyond the mainstream unidirectionality hypothesis advanced by, e.g., Hopper & Traugott (2003: 7) and Lehmann (1995: 12, 121–123). According to the mainstream definition, grammaticalisation equals the rise of grammar by means of the movement of a linguistic element down the cline of grammaticality, i.e., a unidirectional development of grammar from syntax to morphology going through the stages from content item via grammatical word and clitic to inflectional affix, revealed by accompanying features such as phonetic reduction, increased syntactic bonding, desemanticisation, use in new contexts and increasing frequency. As a consequence of the definition by Nørgård-Sørensen et al. (2011) etc. of grammaticalisation as processes of change in the function and contents of the grammatical signs and in the paradigmatic oppositions between them, their type of grammaticalisation cannot limit itself to such a change from syntax to morphology, but must comprise also, e.g., the rise of morphologically and/or syntactically expressed grammatical sign oppositions from a reanalysis of formerly lexical items (grammation) and restructurings of existing morphologically and/or syntactically expressed grammatical oppositions (regrammation). Such regrammations may also comprise changes from morphological to syntactic expression of grammatical content oppositions (as in the grammaticalisation processes suggested in the present article) and not only changes from syntactic to morphological expression as per the limitations of the mainstream unidirectionality hypothesis.

### Bjarne Simmelkjær Sandgaard Hansen

mentioned above to the rise of definite and indefinite articles and the subsequent shift in markedness between inflected nouns with and bare nouns without an article as well as to the rise of unity stress. According to Petersen, all these changes form part of the process that gives rise to the concept of incorporation in Danish.

## **3.4 Redundancy in the indexical relations of noun-phrase-internal agreement**

The three grammaticalisation-based approaches to explaining the Danish casesystem changes mentioned in Section 3.3 by no means contradict each other. On the contrary, they complement each other, each contributing one important factor to the complex overall explanation of the systemic changes. Neither do they provide the full explanation, however. As I have advocated for in Hansen (2021), one additional factor must be added, viz., an apparent desire among the language users for eliminating redundancy in noun-phrase internal agreement as evinced by, e.g., the mere existence of the Middle Danish case-system stages 3–4 in Section 2.6. 5

Most grammaticalisation-based explanations of the case-system changes focus on noun-phrase-external relations and functions of case marking. For instance, Jensen (2011) regards the functional change of the historical nominative from marking subjects and subjective complements to signalling foreground information as a decisive factor. However, although the primary function of case marking is, indeed, noun-phrase external in terms of 1) indexical reference to the valency of a predicate (i.e., revealing the argument status of the noun phrase in question) or to the government of an adposition and 2) symbolic reference to location, direction, means, etc. (Blake 2004: 1080–1086, Andersen n.d.: 2, Heltoft 2019: 154– 155), case may also point indexically within the noun phrase. This is what creates noun-phrase-internal or endophoric agreement, i.e., that multiple members of a noun phrase inflect identically (on the functional level; the formal expression of the endings may differ) and point indexically to each other (Andersen n.d.: 2, Haspelmath 1996: 52, Nielsen 2010: 82, see also Nielsen (2010: 86–89) on the general principle of what he labels *conditioned agreement*).<sup>6</sup>

Example (21) serves to illustrate this double function of case marking. Please note both the noun-phrase-external relations, because the noun phrase *ondom quinnom* 'evil women' in the dative points indexically to the dative-governing preposition *for* 'against', and the noun-phrase-internal or endophoric agreement,

<sup>5</sup> See also Section 3.5 on the application of Norde's (2001: 258–261) principle of single encoding.

<sup>6</sup> For an outline of the general distinctions between symbolic and indexical sign relations, see Andersen (1980: 4–5, 27–30, 2010, n.d.) with further reference to Peircean sign theory.

### 6 Redundant indexicality and paradigmatic reorganisations in Middle Danish

because the dative ending *-om* of *ondom* 'evil' and *quinnom* 'women' point indexically to each other, signalling that they belong together as members of the same noun phrase.

(21) SjT: 52<sup>28</sup>

for for ond-om evil-f.dat.pl quinn-om woman(f)-dat.pl 'against evil women'

This type of noun-phrase internal agreement may be instrumental for language users when they attempt to group individual words of a sentence together in phrases, but this would hold true mostly for languages where a fixed topology, including juxtaposition of constituents belonging to the same phrase, is not the general rule. For instance, in (22), the noun phrases *contiguās domōs* 'neighbouring houses' and *altam urbem* 'upper city' are separated by other constituents (Andersen n.d.: 2, Nielsen 2010: 89–93), and readers of that sentence must therefore rely heavily on case marking in order to group the words together correctly.

### (22) Latin

contigu-ās neigbouring-f.acc.pl tenuere they.lived dom-ōs, house(f)-acc.pl ubi where dīcitur is.said alt-am high-f.acc.sg coctil-ibus bricked-m.abl.pl mūr-is wall(m)-abl.pl cinxisse to.have.surrounded Semīram-is Semiramis(f?)-nom.sg urb-em city(f)-acc.sg 'they lived in neighbouring houses where Semiramis is said to have surrounded the upper city with brick walls'

In languages with a fixed topology and with juxtaposition of constituents that belong together, this type of noun-phrase internal agreement becomes redundant. As Diderichsen (1941: 93–107) has demonstrated, Middle Danish is such a language, at least when it comes to noun-phrase-internal topology. Juxtaposition is standard in the Middle Danish noun phrase, and the position of the modifiers (pre-head or post-head) follows from a set of fixed rules that I may sum up as follows.<sup>7</sup>

1. Modifier in pre-head position: Determiners (quantitative adjectives, indefinite pronouns etc.) and numerals as well as characterising or emphatically used adjectives and possessive pronouns

<sup>7</sup> For an elaborate presentation of these rules, see Diderichsen (1941: 93–107) and Hansen (2021).

### Bjarne Simmelkjær Sandgaard Hansen

2. Modifier in post-head position: Simple descriptive adjectives, adjectival appositions, participles equivalent of subordinate clauses, "superfluous"<sup>8</sup> possessive pronouns, and partitive genitives

Hence follows that the Middle Danish case-marking system is redundant in one of its two functions, viz., noun-phrase-internal agreement, leaving nounphrase-external reference as its sole non-redundant function. In actual fact, nounphrase-internal agreement is marked in three ways: (1) by juxtaposition and fixed rules of noun-phrase-internal topology, (2) by morphological case marking, and (3) by morphological marking of gender and number. One could reasonably argue, therefore, that Middle Danish displays double redundancy in noun-phraseinternal agreement. In light of this triple marking or double redundancy, it is hardly surprising that one of the ways of marking this agreement, viz., case marking, would become prone to loss, as evidenced by stages 2–5 in Section 2.6.

Figures 1–6 express the loss that happened in terms of grammatical sign relations and changes in these relations, referring concretely to the prepositional phrase *for ondom quinnom* 'against evil women' in (21). Figure 1 9 illustrates the traditional representation of the original situation with morphologically marked case on every member of the noun phrase.

Figure 1: Symbolic sign relations morphologically expressed (case focus only)

<sup>8</sup>This term and the examples given in this footnote stem from Diderichsen (1941: 100, 206). Superfluous possessive pronouns comprise cases such as *faþær sin* 'his/her father', *kuna sin* 'his wife', *barn sit* 'his/her wife' where the possessor is self-evident from the context and can easily be left out as in *at barn uar føt æftir faþur*, lit. 'that child was born after father' with omission of *sin* 'his'.

<sup>9</sup>All figures and tables: CC-BY 4.0 Bjarne Simmelkjær Sandgaard Hansen. In Figures 1–6, the label "not coinc." stands for "not coincidental", referring to the noun-phrase-internal position of the head and the modifier not being coincidental, but governed by a set of fixed rules (Diderichsen 1941: 93–107).

### 6 Redundant indexicality and paradigmatic reorganisations in Middle Danish

As I have already shown, however, topology also plays a decisive role in the marking of noun-phrase-internal agreement, to which point I will add that the order of constituents within a prepositional phrase like *for ondom quinnom* 'against evil women' is also fixed: preposition first, noun phrase second (Diderichsen 1941: 109). In order to illustrate the grammatical relations in further detail, I will therefore need to add an additional topological layer to the presentation of Figure 1, for which see Figure 2, where thicker arrows represent topologically marked relations, and thinner arrows those that are morphologically marked.

Figure 2: Symbolic sign relations morphologically and topologically expressed (case focus only)

So far, the model does not reveal much about the claimed redundancy in nounphrase internal agreement. In order for that to be illustrated as well, I will need to extent the model even further. Figure 3 therefore adds morphologically marked indexical relations, including such that represent noun-phrase-internal agreement, and Figure 4 is even further augmented with those indexical relations that are topologically marked. Full lines in blue colour represent symbolic relations (as in Figures 1–2), whereas dotted lines in red represent indexical relations. As illustrated especially in Figure 4, both the noun-phrase-internal agreement and the noun-phrase external relations to the preposition are doubly marked, viz., both morphologically and topologically. One of these layers, the morphological one, is therefore dispensable and subject to gradual phase-out by the language users.

Figure 5 illustrates the grammatical relations in my model with the morphological level (i.e., marking by means of case) phased out. It has now become evident that the topological level alone is fully capable of marking both the noun-phraseinternal agreement and the noun-phrase-external relations to the preposition. In other types of situations, e.g., when a noun phrase originally marked for dative did not form part of a prepositional phrase with a dative-governing preposition but functioned as an indirect object, the topological level would be capable of

### Bjarne Simmelkjær Sandgaard Hansen

marking noun-phrase-internal agreement only, seeing that the Middle Danish topology is not fixed on the level of sentential constituents.

Figure 3: Symbolic and indexical sign relations morphologically expressed (case focus only)

Figure 4: Symbolic and indexical sign relations morphologically and topologically expressed (case focus only)

Figure 5: Symbolic and indexical sign relations morphologically and topologically expressed after the removal of case inflection (case focus only)

Despite what one may have deduced and thus been led to believe from the model in Figures 1–5 so far, one must not forget that, even after the loss of case marking, noun-phrase-internal agreement is still marked in two ways, viz.,

### 6 Redundant indexicality and paradigmatic reorganisations in Middle Danish

by juxtaposition and fixed rules of noun-phrase-internal topology and by morphological marking of gender and number. Redundancy thus remains, but only singly, not doubly as in the original system prior to the loss of case marking. Figure 6 serves to illustrate that, besides expressing the number and gender of the elements of a noun phrase symbolically, number and gender marking also express indexical relations within the noun phrase, i.e., noun-phrase-internal agreement.

Figure 6: Symbolic and indexical sign relations morphologically and topologically expressed (gender and number focus)

## **3.5 Gradual abandonment of case marking**

In Section 3.4, I presented one of the factors that motivated the Middle Danish case-system changes, viz., redundancy, but it remains unexplained so far why the language users simply did not remove such redundancy at once, but did it gradually instead. In what follows, I will therefore outline why case marking remained longer in the feminine singular and in the plural than in the masculine and neuter singular, and longer, too, on typical modifiers than on nouns.

Andersen's (2001: 27–37) principle of markedness agreement may account for some of this asymmetry. According to this principle, elements that are marked similarly behave identically. Marked forms behave in the same way as other marked forms, and unmarked forms in the same way as other unmarked forms. When it comes to linguistic change, Andersen (2001: 36) claims that one would expect

[…] the innovation to occur earliest in environments with equivalent markedness value and to subsequently gain ascendancy first in such contexts and then, as it loses its novelty, in the complementary contexts with opposite markedness value.

This corresponds well to the Middle Danish situation, where the case-system innovations occur first in the unmarked environment, i.e., in nouns (Andersen

### Bjarne Simmelkjær Sandgaard Hansen

1980: 44) and in the masculine and neuter singular. The marked environments, i.e., the non-substantival nominal parts of speech, the feminine singular and the plural, keep case marking longer.

One additional factor that may account for the marking of case on only one (type of) noun-phrase member, viz., the typical modifiers, at the intermediate steps of the Middle Danish development (represented by stages 3–4 in Section 2.6) may be the application of a principle of single encoding (Norde 2001: 258–261). This principle, which entails that noun phrases, for instance, only inflect for case once and not on every single (type of) noun-phrase member, also operates in the well-known way of marking case in German noun phrases. Here, only one nounphrase member distinguishes fully for case, while the remaining case markers remain underspecified. In (23), for instance, case is expressed explicitly only on the adjective, leaving the indefinite article underspecified, whereas in (24), case is expressed explicitly only on the definite article, leaving the adjective underspecified.

(23) German

ein-∅ a-m.nom.sg gut-*er* good-m.nom.sg Mann man(m).sg 'a good man'

(24) German

de-*r* the-m.nom.sg gut-e good-m.nom.sg Mann man(m).sg 'the good man'

Returning to the application of this principle in the Middle Danish noun phrase, one may attribute the preference for inflection on typical modifiers over inflection on nouns to the circumstance that the adjectival and pronominal paradigms historically contain more and clearer distinctions than the nominal paradigm.

## **4 Paradigmatisation**

## **4.1 Theoretical viewpoint**

Now that I have described and explained some of the possible reasons for the Middle Danish case-system changes, it is not only interesting, but also necessary to witness their paradigmatic consequences. As mentioned briefly in Section 1, I follow the theoretical viewpoint of Nørgård-Sørensen et al. (2011: xi, 71–72) and Nørgård-Sørensen & Heltoft (2015: 261–262) that grammaticalisation equals

### 6 Redundant indexicality and paradigmatic reorganisations in Middle Danish

paradigmatisation, meaning that one cannot have grammar and changes in grammar without also having paradigms and changes in paradigms; see also Diewald & Smirnova (2010: 2–4). Consequently, in accordance with this theoretical viewpoint, the changes of the Middle Danish case system described in Section 2 and explained in Sections 3.3 and 3.4 cannot be grammatical changes unless they can be formalised paradigmatically.

In Nørgård-Sørensen et al. (2011: 5–6) and Nørgård-Sørensen & Heltoft (2015: 262–263) understanding of what constitutes a grammatical paradigm, any grammatical paradigm must meet five criteria. First, they state, it must be closed, in principle, thus containing only a fixed number of members. Second, it must be possible to specify the domain of the paradigm, i.e., the syntagmatic context to which the paradigm applies. Third, in close correspondence with the domain introduced by the second criterion, any paradigm must have a semantic frame within which the content of the specific members of the paradigm is defined. In other words, the semantic frame reveals what type of oppositions the members of a paradigm serve to express. Fourth, the choice between the members of the paradigm is obligatory, meaning that, when producing an utterance that activates the domain of a grammatical paradigm, the language users cannot avoid choosing one of its members. Fifth and finally, grammatical paradigms tend to be asymmetric and thus to distinguish automatically between marked and unmarked members, the latter being the one without a specific semantic load. For that reason, the unmarked paradigm member may sometimes participate in the functions of the other, i.e., the marked, members; see also Section 2.3 for a further discussion of and references to the concept of participation.

## **4.2 Paradigmatic consequences of the Middle Danish case system changes**

The question to be answered now is how the notion of a paradigm outlined in Section 4.1 fits the data presented in Section 2 and analysed in Section 3.3 and Section 3.4. In order to answer this question, I will first attempt to set up a paradigm of the original Middle Danish case system with historically expected case on both nouns and typical modifiers (represented by stage 1 in Section 2.6) in Table 1, the focus of which lies on the paradigmatic opposition between the accusative and the dative. Similar and more extensive tables may be set up for the inclusion of the nominative and the genitive, but for the sake of clarity, the accusative-dative opposition will suffice.

This paradigm contains a syntagmatic domain (Middle Danish noun phrases consisting of a modifier and a noun), a semantic frame ("case", i.e., indexes of noun-phrase external government as well as noun-phrase-internal agreement)

### Bjarne Simmelkjær Sandgaard Hansen

and a closed set of case endings as its members.<sup>10</sup> In addition, the language users cannot avoid choosing between an accusative and a dative ending in utterances relevant to this opposition.

Table 1reveals that the accusative consistently marks historical accusative contexts, and dative the historical dative contexts. Consequently, no overlapping occurs between the accusative and the dative in this system.

> Table 1: Paradigmatic visualisation of the system of historically expected case marking both on nouns and on typical modifiers


As I mentioned in Section 3.4, the Middle Danish noun-phrase-internal word order was fixed even at the earliest attested stage of Middle Danish when the application of this historical case system was most widespread. In order to illustrate the grammatical potential of this topological system, I have entered it into a paradigm, as well, as represented by Table 2.

So far, both tables have represented clear-cut grammatical paradigms with no vacillation between the members. However, the paradigmatic representation attempted in Table 1 of each stage in a separate paradigm does not depict the actual situation, since all the case-system stages of Section 2.6 actually do occur within one and the same text; cf. again, e.g., (1), (6), (8) and (10), which all stem from

<sup>10</sup>Please note that the expressional label acc covers a wide array of endings from different inflectional classes such as *-an* (adjectival m.acc.sg), *-a* (masculine *n*-stem noun acc.sg) and *-∅* (vowel-stem noun acc.sg), while dat covers endings such as *-u* (feminine *n*-stem noun dat.sg or adjectival n.dat.sg), *-i* (adjectival f.dat.sg or masculine *a*-stem noun dat.sg) and *-um*/*-om* (dat.pl).

### 6 Redundant indexicality and paradigmatic reorganisations in Middle Danish


Table 2: Paradigmatic visualisation of the Middle Danish predictability of noun-phrase-internal topology

SjT and may reveal up to four competing systems. Focusing on the many different ways to express indexes of historical dative contexts, one may therefore be tempted to produce a paradigm like that of Table 3 to check if it would constitute a more precise rendition of the actual situation.

This representation creates an entirely novel issue, viz., the introduction of free choice within the paradigm. Admittedly, the existence of a free choice need not necessarily violate Nørgård-Sørensen et al. (2011: 5–6) and Nørgård-Sørensen & Heltoft's (2015: 262–263) fourth criterion that the choice between the members of a paradigm is obligatory, for, as they further state (Nørgård-Sørensen et al. (2011: 5), "[t]his choice may be free or bound, but will ultimately be determined by the content of the forms constituting the paradigm."

In this statement lies the real problem with the paradigm of Table 3. Even though it seems to contain a domain, a frame, a closed set of members and an obligatory, yet free choice, it does not entail an opposition of content, unless one may assume that the content opposition is one of indexing variation within the language users' personal register, i.e., within the language users' range of varieties between which they may choose at different times (Halliday 1994: 77).<sup>11</sup>

<sup>11</sup>I am greatly indebted to Henning Andersen for pointing this possibility out to me during the discussion round at the SLE workshop "Paradigms regained" where I first gave the presentation upon which this article is based.

### Bjarne Simmelkjær Sandgaard Hansen

Table 3: Attempt at a paradigmatic visualisation of four competing case-system stages in noun phrases based on the content of indexing historical dative contexts


If such a register-focused content opposition is not present, this paradigm is invalid, for a grammatical paradigm must oppose grammatical signs, i.e., linguistic units consisting of both a content side and an expression side. In other words, the absence of a content opposition equals the absence of a grammatical opposition, which, in turn, equals the absence of a paradigmatic opposition. In that sense, this may rather be a case of allomorphy in its broadest sense than a case of paradigmatic opposition.<sup>12</sup>

Notwithstanding any considerations on allomorphic variation, a more effective way of presenting the paradigm in question would be that of Table 4, which reintroduces the content opposition between expressing indexes of historical

<sup>12</sup>Whether one may really label an apparently free choice between ways of expressing indexes of historical dative contexts a case of allomorphy depends on one's definition of this term. As Bauer (2003: 17, 113–114) points out, the prototypical allomorph is a phonologically, grammatically or lexically conditioned variant of the same morpheme. Consequently, one could regard the different realisations of the historical Middle Danish dative singular ending in nouns as grammatically (gender) or lexically (inflectional class) conditioned allomorphs of a morpheme that expresses indexes of historical dative contexts; see fn. 10 for examples of these different realisations. Since the choice presented in Table 3 is free rather than phonologically, grammatically or lexically conditioned, I should be able to rule out allomorphy here, at least in its prototypical sense. Bauer (2003: 113–114) adds, however, that in a broader context, allomorphs may be conditioned by the choice of register as in the choice between the English plural forms *tempos* and *tempi*. If one accepts this expansion of the definition of allomorphy, I would be able to regard it as a case of allomorphy.

### 6 Redundant indexicality and paradigmatic reorganisations in Middle Danish

accusative and historical dative contexts. Here, it becomes evident that what seems to be (and is indeed) free variation between different ways of expressing indexes of historical dative contexts also represents an instance of difference in markedness relations, i.e., the fifth criterion of Nørgård-Sørensen et al. (2011: 6) and Nørgård-Sørensen & Heltoft (2015: 263). The option for using historical accusative forms or uninflected forms in historical dative contexts simply follows from the general unmarkedness of these forms; see again Section 2.3 on accusative participation. In contrast to that, the historical dative forms remain marked and are applicable only in their original contexts. What Table 4 does not reveal, however, is the relative markedness of the accusative forms and the uninflected forms, since the rendition of such a distinction would require the inclusion of a larger set of data and forms (the nominative and the genitive), which lies outside the scope of the present article.



### Bjarne Simmelkjær Sandgaard Hansen

## **5 Conclusion**

In this article, I have analysed linguistic data from mainly three Middle Danish texts with respect to case use and presented a development of the Middle Danish case system divided into possibly five stages:


I have demonstrated that the Middle Danish case-system changes result neither from reductionist sound laws, nor from linguistic simplification due to language contact. Rather, various processes of grammaticalisation, i.e., processes of change in the function and contents of the grammatical signs and in the paradigmatic oppositions between them, are responsible for the changes. One of these processes is the change from double to single redundancy in noun-phraseinternal agreement. After the fixation of noun-phrase-internal topology had rendered the use of case for expressing noun-phrase-internal agreement superfluous, this type of indexical reference was phased out gradually in general accordance with both Andersen's (2001: 27–37) principle of markedness agreement and Norde's (2001: 258–261) principle of single encoding.

Based on Nørgård-Sørensen et al. (2011: 5–6) and Nørgård-Sørensen & Heltoft (2015: 262–263) five criteria for what constitutes a grammatical paradigm, I have also shown that the Middle Danish case system may be described paradigmatically and, correspondingly, that the changes it undergoes constitute an instance of paradigmatic change. The existence of an intermediate transitional period with competition and seemingly free variation between different Middle Danish case-system stages does not challenge this claim, since these stages do not only represent free variation, but also an instance of difference in markedness

### 6 Redundant indexicality and paradigmatic reorganisations in Middle Danish

relations, i.e., the fifth criterion of Nørgård-Sørensen et al. (2011: 6) and Nørgård-Sørensen & Heltoft (2015: 263).

After subjecting the Middle Danish case-system changes to this paradigmatic test, I dare now claim that the changes in the Middle Danish case system are indeed grammatical changes – and thus represent a process of both grammaticalisation and paradigmatisation – in Nørgård-Sørensen et al. (2011: xi, 71–72) and Nørgård-Sørensen & Heltoft's (2015: 261–262) sense.

## **References**


### Bjarne Simmelkjær Sandgaard Hansen


Wessén, Elias. 1954. *Om det tyska inflytandet på svenskt språk under medeltiden* (Skrifter utgivna av Nämnden för svensk språkvård 12). Stockholm: Nämnden för svensk språkvård.

## **Chapter 7**

## **The semantic reorganisation of case paradigms and word order paradigms in the history of Danish**

## Lars Heltoft

University of Copenhagen

This article is a study of the relation between the paradigmatic organisation of case and the paradigmatic organisation of word order in late Middle Danish (1300–1500) and in Modern Danish. A content analysis of these paradigms shows a typological difference, even if the older pronominal case system looks exactly like the modern system. Middle Danish preserves inactive (impersonal, traditionally) constructions with an inactive argument 1, and at this stage, the dimensions of case, position, argument hierarchy and subjecthood (still) combine freely. The case system is still indexical. The alignment of subjecthood, status as argument 1, position and nominative case with symbolic meaning is a development of post-Reformation Danish.

## **1 Introduction**

It is no surprise that a morphological category like case should be organised in paradigms, since this usage of "paradigm" has been current since antiquity. It may come as more of a surprise that a syntagmatic aspect of language like word order can have paradigmatic organisation; yet, this refers to contrasts of meaning between word order patterns and is therefore not a new idea either, but rather a neglected aspect of language. What I will try to add, is the possibility of co-organisation between morphological content and alternating constructional

Lars Heltoft. 2022. The semantic reorganisation of case paradigms and word order paradigms in the history of Danish. In Gabriele Diewald & Katja Politt (eds.), *Paradigms regained: Theoretical and empirical arguments for the reassessment of the notion of paradigm*, 201–244. Berlin: Language Science Press. DOI: 10.5281/zenodo.5675851

### Lars Heltoft

organisation. Case meaning can interact with constructional hierarchic organisation of arguments, forming what could be called second-order paradigms (or hyperparadigms), see Christensen (2007), Nørgård-Sørensen et al. (2011), and Juul Nielsen (2016).

After Sections 1.1 and 1.2, I present in Section 2 the system of Middle Danish inactive constructions in some detail, including important differences from the parallel West Norse (Old Icelandic) system. Section 3 deals with the paradigmatic organisation of case in its interplay with transitive and inactive. The function of case is to point to the governing verb and thus to determine the semantic role value of the arguments. Section 4 documents that topology (word order) is not included in these constructional paradigms, since there are no positions reserved for subjects and objects. This is very different from the system of present-day Danish, with specific positions for subjects and objects (see Section 5) and a case system that has lost its coding of semantic roles.

Some readers may find it easier to skip initially Sections 2.2–2.5 and go directly to the overview in Section 2.6 and then on to Section 3, to return later to the details of the data in Section 2.

## **1.1 Inactive constructions**

A central problem in the analysis of case morphology in constructional contexts lies in its function in so-called impersonal constructions, in my terminology inactive<sup>1</sup> constructions, as in (1–2).<sup>2</sup>


Example (1) has a one-place predicate *thik* 'thee', in the oblique case, manifesting the semantic role Inactive. Since this role is basic to this constructional paradigm, I name the Inactive argument A1, the primary argument.

<sup>1</sup> *Inactive construction* bears resemblance to the contrast between active features and inactive features in so-called active languages, cf. Lehmann 1989, 1993. When the *active voice* is meant, I will be explicit about this and simply write "active voice".

<sup>2</sup>Line two is rendered in the normalised orthography of the collections of the Old Danish Dictionary, see Section 6.

### 7 Semantic reorganisation of case paradigms

Argument hierarchy reflects basic semantic choices. The Inactive role applied to the A1 of this construction excludes agentive meaning, as also emphasised by Faarlund (2001, 2004) for Old Icelandic. Agentive meaning is characteristic of transitivity, and neither (1) nor (2) can convey agentive meaning.

Examples (1–2) show a difference in valence within inactive constructions. Example (2) is a two-place predicate; the A1 is again *thik*, likewise in the oblique case, manifesting a variant of the inactive role, what other case grammarians have called the Experiencer role. Here, the A2 is an embedded clause: *at thu haver ille gjorth*, denoting the content of the mental impact on the referent of the A1. Halliday (1994) speaks of this semantic role as the Phenomenon role. The term A2 reflects an extension of the construction, possible with certain verbs.

The paradigmatic organisation of inactive constructions will be laid out in Section 3, esp. Section 3.3. I will not go into details here, since a clear exposition will call for a comparison with especially transitivity and other organisation principles. But it must be pointed out now that argument hierarchy reflects the semantic valence of predicates, and this differs from the level of sentence members and thus from the grammatical functions subject and direct object. In principle, the subject function can apply to either A1s or A2s, and similarly, direct objects to either A1s or A2s. Again, this means that one should not just assume that one of these levels can reduce to one of the other levels, for instance of subjects to A1s. On the contrary, a claim that these levels are or have been aligned must be the outcome of the application of empirical criteria and cannot be taken for granted a priori.

The A1s of (1) and (2) have the oblique form, and this is of course a case of what other traditions call oblique subjects (among many others Allen 1995, Barðdal & Eyþórsson 2003, 2018, Eyþórsson & Barðdal 2005, Kiparsky 1997). The analysis of oblique case subjecthood has been advocated for many older Indo-European languages by esp. Jóhanna Barðdal and Þórhallur Eyþórsson. I shall not in this context discuss their views in detail, nor will I refer to the sometimes-polemic discussions between different positions. Barðdal and Eyþórsson have a specific definition of subject as a starting point, namely the identification of subject and A1. As they see it, the A1 is the subject, or rather, the universal definition of a subject is taken to be the status as an argument 1.

One potential subject definition that we have used as a working definition since Eyþórsson & Barðdal (2005), is to view the first argument of the argument structure as being the syntactic subject.<sup>3</sup> (Barðdal & Eyþórsson 2018: 263)

<sup>3</sup>The following quotation will illustrate their view: "The reason that we have proposed such a subject definition is that when generalizing across the subject tests, we have found that it is

### Lars Heltoft

A discussion of their subject criteria and of similar approaches (e.g. Sigurðsson 1989) and the way they are operationalised<sup>4</sup> must be the topic of another article (Heltoft 2021b).

In the hierarchical-linear configurations of generative grammar, such reductional notions of a subject will always come out as a Specifier of something, normally generated as a Vp-specifier, next upgraded to I-Spec, or for V2-languages, all the way to C-Spec. This presupposition, that the subject holds the upmost position, inherently linear and hierarchical at the same time, is shared also by linguists (esp. Kiparsky 1997) who try to combine and reconcile syntax and morphological case by ascribing syntactic features to the arguments and case features to morphology, and thereafter, working out unification procedures for the respective feature clusters. Kiparsky refers to Cynthia Allen for the insight that Old English had IP available as a category since "it had dative subjects, in the sense that oblique experiencers were structurally parallel with nominative subjects", interpreting this as "at least a *prima facie* indication of Spec-IP positioning" (Kiparsky 1997: 12). Behind this, we also find the identification of subject and A1, meaning that A1 is the hierarchically upmost argument. Instead of this assumption, as mentioned in Section 1.1, I hold that argument hierarchy should be seen as organised by valence; see further Section 2.

Of course, the sign-oriented approach adopted here determines part of what is possible. Sign-oriented grammars such as Croft (2001), Traugott & Trousdale (2013) or Danish Functional Grammar (Engberg-Pedersen et al. 1996) must respect the sign limits delineated by the expression side and must therefore seek for models that do not presuppose reduction attempts of Inactive A1s to underlying subjects across the sign boundaries.

I will return to the importance of linearity in Section 4, but the road to there will go via an analysis of the structure and function of simple traditional case paradigms like the ones behind examples (1) and (2), consisting of oppositions of number, deixis and a case distinction of just two: nominative and oblique.<sup>5</sup>

Apart from orthography and sound change, and a few later shortenings (of *hanum* to *ham*, *ither* to *jer*), the expression system of Table 1 is exactly the same as that of the modern language. The 3p.sg/pl forms are attested in the Jutish Law (of 1241, oldest manuscript from 1284), and this has led to the traditional

always the first argument of the argument structure that is targeted by the subject tests. In that sense, our approach is bottom-up; we have arrived at a subject definition on the basis of the subject tests, a definition which can then be applied independently of the individual tests." (Barðdal & Eyþórsson 2018: 263–264)

<sup>4</sup>The only separate treatment of Danish known to me is a brief article by Hrafnbjargarson (2003), using the criteria of Sigurðsson (1989).

<sup>5</sup>The genitive is only used in possessive constructions and is therefore not part of this paradigm.

### 7 Semantic reorganisation of case paradigms


Table 1: Pronominal case in 14th century western Middle Danish

assumption that Middle Western Danish<sup>6</sup> had already introduced roughly the modern pronominal case system (e.g. Karker 1991: 129, Karker 1993: 198). As we shall see, however, when properly analysed at the level of content, the Western Middle Danish two-case pronominal system turns out to be typologically different from the modern system. Some generative grammarians (Sigurðsson 2006, 2012b,a, Parrott 2012) use the term 'case impoverished' for such modern Germanic languages that have reduced their case inventory to pronouns and there to a minimum of two cases, and insofar as they speak of case as an expression system, this term might apply to Middle Danish as well. However, what matters is not quantity, but the quality of the content organised in such minimal case paradigms. I will claim that a content analysis of the Middle Danish case paradigm will show that it is clearly typologically different from the modern Danish case paradigm, and secondly, that this analysis demands a thorough analysis of the way Middle Danish case paradigms are integrated in more complex paradigms interlocking morphology and constructional alternations.

To conduct this analysis, we must take the semiotic function and content of even a reduced case system seriously. We cannot simply assume that case has no meaning potential and relegate it to a status as part of the expression system, or, in the generative terminology, to phonological form.<sup>7</sup> Nor can we assume that its content is simply the positions defined by an abstract, a priori given syntactic configuration. One part of the exercise will consist in determining the content system of the Middle Danish case paradigm, and contrary to most other presentday approaches, I will not accept any a priori distinction between syntactic and lexical case. Given convincing arguments, the discussion is open to the possible

<sup>6</sup>The written tradition of Middle Danish falls in two main dialects, Western Middle Danish in Jutland and the central islands and the more archaic Eastern Middle Danish (Scanian) in the provinces east of the Sound, in present-day Sweden.

<sup>7</sup>Not all generative grammarians buy the reduction of morphology to PF, of course. Among them especially Kiparsky (1997), but also Sigurðsson (2006, 2012b,a) realises this is a weak point.

### Lars Heltoft

conclusion that even a reduced system like the one under scrutiny can manifest a semantic role system, and just that. This is the topic to be investigated in Section 2.

## **1.2 The word order systems: Why include them?**

The word order systems of late Middle Danish and Modern Danish will be investigated and compared, too, as a way to determine whether the oblique A1s share properties with nominative subjects. The model to be used is the so-called *sentence frame model*, a descriptive model with Scandinavian and German roots (Diderichsen 1946, Faarlund 1989, 1990, Faarlund et al. 1997; Heltoft 1992). This positional model does not intertwine syntactic hierarchy and linearity. It presupposes a nonlinear dependency model for syntax but consists of concatenated positions in itself. Some are characteristic or even definitional (Mel'čuk 2014) of their syntactic category, others are open positions for a set of syntactic categories. Such open positions can express a separate content system independently of the categories that may fill them. One relevant example for the present agenda is illocutionary force, or better: *illocutionary frame*, the speech act potential coded in word order; another example is background-focus structure.

In Modern Danish, subjects – in the nominative form, if possible – are confined to a limited number of positions, namely two: the open initial position, the socalled fundamental field of Danish topological tradition (the P1 of Simon C. Dik, see Dik 1997: 408–416), and the third position immediately after the V2 position.

	- b. Hende.(1pos.) her-obl beundrer.(2pos.) admires *han*.(3pos.) he-nom 'Her he admires'
	- c. (1pos.empty) ∅ Beundrer.(2pos.) admires *han*.(3pos.) he-nom hende?.(post-subject-pos.) her-obl 'Does he admire her?'

The Modern Danish system is clearly an XVSO-system, and in traditional terms, the contrast (3a–3b) vs. (3c) codes declarative function vs. interrogative function. The basic structure of the paradigm can be laid out in terms from Peircean semiotics, namely symbolic and indexical meaning. The main expression contrast is between a filled-in position 1 (see 3a–3b) and its zero opponent

7 Semantic reorganisation of case paradigms

(3c). The symbolic contrast is between constative (pos. 1 filled-in) and interrogative meaning (pos.1 zero), and there is an indexical function to notice as well, namely the position 2 filled by the finite verb.

So in Table 2, position 1 is the locus of the contrast zero vs. X, position 2 the indexical identification of this locus. Position 1 holds the symbolic, illocutionary frame contrast of the paradigm and is thus the locus of the frame of the paradigm; position 2 indicates the locus for this frame and defines the domain of the paradigm.<sup>8</sup>

Table 2: The indexical function of position 2 in Modern Danish word order


Notice that the subject's unique position is position 3, and that this position must be filled in to form the yes-no question (3c). And since the subject in active clauses can in the modern language readily be identified with the argument 1 (A1), the case system and the positional system are clearly related.

There is every reason to ask whether the medieval language had a characteristic, let alone definitional subject position in the way the modern language has it, that is, whether position plays a role for the identification of subjects and objects, and furthermore, of the arguments A1 and A2. Thus, after an analysis of the role of case in the inactive construction, I will suggest in section 4 an analysis of the word order paradigm for late Middle Danish.

## **2 The inactive construction in late Middle Danish**

The inactive construction of late Middle Danish falls in a number of subtypes, of which I shall deal with three. It is a continuation of a common Norse (and Germanic, further back Indo-European) set of constructions that deviate in important ways from transitive constructions. Late Middle Danish differs from Ice-

<sup>8</sup> For the terminology of this paragraph, see Nørgård-Sørensen et al. (2011), Nørgård-Sørensen & Heltoft (2015), Heltoft (2019).

### Lars Heltoft

landic as well, but a detailed comparison is not available, so I will restrict myself to dealing with one basic difference, see Section 2.6.

The constructional set comprises 1) verbs that are inherently semantically inactive, that is: their stems will construe with an oblique A1; 2) verbs with transitive stems, needing an inflectional modification, namely the middle voice form, to form an inactive construction, 3) verbs with neutral stems, construing either with a nominative A1 or an oblique A1, that is, verbs with semantically different case constructional potential, but no morphological change of the stem to mark this difference, and 4) a type with no obvious difference between active voice and middle voice. The subtypes have been selected from a list of lemmas (Bom 1954) for the Old Danish Dictionary (not yet completed), and from a next to complete collection of quotations (card copies in electronic form, GldO).

Two basic issues: 1) The inactive system's interplay with the voice system must be clarified. Some stems allow the inactive construction with the middle voice only, see Section 2.2; others allow it with the active voice, see Section 2.3; again, some apparent mergers of voice allow inactive construction both with the active and the middle voice, but at least in some cases, this distinction expresses a semantic contrast between two subtypes of inactive constructions. 2) Like many other older Indo-European languages, Old Scandinavian, including Old and Middle Danish, allows zero arguments, meaning that NPs at all levels can be let out, or better, replaced by zero. This leads to a methodological problem of how to determine whether an argument is a valence-governed actant of a verb stem that has been optionally replaced by zero, or whether it could instead be considered a free syntagmatic extension of the semantic nucleus of the clause (cf. Nielsen & Heltoft to appear); see Section 2.2 and Sections 2.4–2.5 for details.

## **2.1 Verbs that are inherently semantically inactive**

The verbs belonging to this subcategory take an argument 1 (A1) denoting an animate referent that is causally affected, be it by bodily demands, by mental or social impression or by incidents of fate. I call this semantic role *Inactive*, and constructions comprising it *inactive constructions*.

Some are one-place verbs, excluding the possibility of an argument 2 (A2), for instance: *hungre* 'starve', *thyrste* 'thirst', *værke* 'feel pain' (see 4); *lithe* 'do, fare'; *fare ille/væl* 'have a misfortune/have good fortune'.

(4) then thæn the timæ time time mek mik me.obl hungrudæ hungrethe starved tha tha then gauæ gave gave i i 2pl megh mik me at at to ædæ æte eat 'When I was hungry, then you gave me something to eat' (Luc 69v 8–10)

### 7 Semantic reorganisation of case paradigms


Two-place: *æve* (forms with breaking: *jave*, *jæve*) 'doubt, be in doubt'; *tvivle*<sup>9</sup> 'doubt', *skilje* 'disagree'. The A2 of these three verbs must have predicational value, either through clausal form as in (8) or through a predicational noun (8b).

(8) a. iafuær jav-er doubt-prs.act them thæm they.obl um um about oc ok and skil skil disagree-prs.3sg them thæm they.obl um um about hwat hvat what hældær hælder either hun hun she ær ær is mæth mæth with ællær æller or ey æj not 'If they (appointed good women) doubt and disagree whether she is with (a child) or not' (DgL V. 5,3)

<sup>9</sup>*Tvivle* is a 15th century Low German replacement loan for *æve*. Sources show both inactive and transitive construction and thus, the continuous productivity of the inactive pattern. A handful of later manuscripts have *tvivle* for *æve* in example (7); of these, 4 retain an inactive construction, 3 are transitives, according to the edition's critical apparatus.

### Lars Heltoft

b. hwaræ hvare where sum sum rel mæn mæn man.pl æuær æv-er doubt-sg um um about sannænd. sannende truth thær thær there skal skal must logh logh law lethæ lethe guide hwilt hwilt which ræt ræt right ær ær is 'Where people are in doubt about truth, there the law must guide which is right' (CCD X, 3v)

Example (8b) is included because it shows a secondary morphological effect of the construction's semantics. There are hosts of medieval manuscripts of this text, the prologue of the Jutish Law, but not a single variant of this reading showing a plural form *æv-e* to agree with *mæn* 'men'. The inactive construction does not allow concord between A1 and verbal number, only transitive constructions with a nominative A1 allow this, and even though nouns no longer inflect for the nominative vs. oblique distinction, the concord rules are still maintained<sup>10</sup> , banning concord with inactive constructions.

## **2.2 Middle voice inactive verbs**

Some verbs need a middle voice form in order to construe inactively. The verb *te* is from te-a, Icelandic *tjá*, and its active voice forms are transitive only (9), the *s*-form has a clearly passive variant (10).

(9) Ok ok and ther te-r show-prs.act.sg thu thu 2sg hanum han-um he-obl thitt thit poss.2sg wredhe vrethe angry anledhe, andlete face Tha tha then ær ær is thet thæt it ey æj not taknemælight, taknemlikt evident hwat hvat what got got good thu thu you gør gør do hanum han-um he-obl

'And if you show him your angry face, then it is not evident what good you are doing to him' (Sydr 161, 18–19)

<sup>10</sup>Bjerrum (1949: 166) writes: In "impersonal constructions" into which it is impossible to interpolate any subject (…) the verb is invariably in the singular, e.g. *skil børn with mothær* (51<sup>5</sup> ) si mater et pueri discordant … ", that is: 'if the children disagree with their mother'

7 Semantic reorganisation of case paradigms

(10) oc ok and ænglæ engle angles føræ føre carry foræ fore in front of hanum hanum him korss kors cross tegn, tekn, sign, ath at that thet thæt it skal skal will theræ thære there thees te-s show-pass foræ fore for al al all mankøn. mankyn mankind 'And angels will carry the sign of the cross in front of him, so that it be there shown to all mankind' (Luc 69r 7–10)

The middle form in East Norse *-s* (West Norse *-sk/-st*) has four semantic variants (Dyvik 1980; Heltoft 2006), of which the passive is but one. The middle voice functional varieties are the reflexive function, the reciprocal function, and the detransitive function. The reflexive and the reciprocal functions are transitive variants, so the relevant function for the discussion of the inactive construction is the latter, detransitive one11. Examples are (11–13):

(11) Tees tea -s show-middle thic thikA1a 2sg.obl thet, [thæt, that.nom thic [thikA1b 2sg.obl wel væl well liger, likerb\_\_A2b]]A2a, likes, tha tha then ladh lat let sighe sighje say messe misse the masses de de of trinitate trinitate Trinity (…) 'If you behold that which pleases you, then let say the masses of Trinity' (Bønneb III, 122, 17)

In (11), both verbs are inactive. The verb form *tes* governs the arguments subscribed with an *a*, the verb *liker* those with a subscribed *b*. In both cases, the A2 is an embedded clause. In (12) and (13), the A2's cannot be read as agents and hence they are not transitive, but inactive.

(12) Meg MikA1 1sg.obl thee-s te-s appear-prs.middle twæne [tvænne two honde hande kinds of folck folk]A2 people 'I see two kinds of people before me' (JBB kap.7, b5v)

<sup>11</sup>The reflexive function is demonstrated in (i) *Gudh alsommæctigste teedes henne* 'the almighty God showed himself for her' (Bønneb II, 133, 15); the reciprocal function in (ii) *the tordæ æy tees førræ æn the brudæ kostæ oc skyuldæ tøm met* (Luc 76v 7–10) 'they (Adam and Eve) dared not show themselves to each other until they had broken off twigs to hide themselves with'.

### Lars Heltoft

(13) ogh ok and ther thær there thedhes te-th-es appear-prt-middle them thæmA1 3pl.obl stiærnen stiarne-nA2 star-def i geen, igen, again, efter æfter after ad at that hun hun she borthe borte gone war var was 'And there the star appeared to them again, after it had been gone' (Vejl Pilgr 220, 12)

In (14) and (15), I address the problem of zero arguments. In Old and Middle Scandinavian, NPs at all levels can be replaced by zero, and as premises for assuming a zero, I posit either conceptual necessity or linguistically well-defined ellipsis, and (14) will show conceptual necessity. In (14), the A1 is represented by zero, since it is referentially unspecified. The A2 is specified: 'then some sign (A2) would appear (to whoever might be the perceiver, A1)', a conceptually necessary A1 referent, in the present case generic and therefore also textually omissible.

(14) vare var-e be-subj han han he saan san guilty at at as saken, saken, charged tha tha then tedess te-th-es appear-prt-middle e e always noget noket some teken teknA2 sign i i in hans hans his andlade, andlete, face (...) (...) (...) æn æn but vare vare be-subj han han he vsan, usan, not guilty tha tha then tediss te-th-es appear-prt-middle icke. ækkeA2 nothing 'If he should be guilty as charged, then some sign would appear in his face

(…) but should he be not guilty, then nothing would show' (HellKv 8, 1)

Apart from the omissibility of A1 (a zero argument, again of the verb *tethes*), example (15) is included to document the existence of actantless predicates (here: *ræghne* 'rain') in Middle Danish, in the sense that they have *zero valence*, that is: no actant at all. This proves that Middle Danish, like so many other old Indo-European languages, does not have categorical NP-VP structure as a necessary structural principle. The context is: … *that from Adam's time and until the day of Noah …*

(15) Tha Tha then regnedhe ræghnethe rained aldrigh, aldrigh, never, Ok ok and teddes te-th-es appear-prt-middle ekke ække not regn bwæ ræghnbugheA2 rainbow pa pa in

7 Semantic reorganisation of case paradigms

hemmelen hemelen the sky 'then it never rained, and no rainbow appeared in the sky' (Sydr 51, 11–12)

I have interpreted (14) and (15) according to the classical rules of zero arguments in Old Scandinavian; see Heltoft (2012) and Faarlund (2004). Theoretically, they could be seen as bridging examples allowing also the modern intransitive reading with a subject A1. In both cases, they would show subjects in a position later than the third structural position, cf. Section 4.2.

## **2.3 Neutral stems**

Some stems are neutral with respect to the transitive-inactive contrast, examples being: *thrængje* 'put a strain on, bother' • 'need, be in jeopardy'; *varthe* 'be responsible for, guard' • 'concern, be somebody's task or obligation'. Such verbs allow inactive construction with the active voice, and the opposition between transitive and inactive is manifested by the syntagmatic argument hierarchy only. Notice that (16a–16b) are transitive constructions, so the A1s are subjects, the oblique case arguments are A2s and direct objects.

(16) a. Mæn Men but vndher underA1 miracle haffde havthe had swa sva such trængth thrængth overwhelmed hannum, han-umA2 he-obl at at that han han he wisthe viste knew ey, æj not hwat hvat what han han he skulle skulle should sighæ. sighje say 'But the miracle had overwhelmed him so that he knew not what to say' (ML 152, 19–153, 2) b. Nar nar when ikten iktenA1 the gout trængher trængher bothers tegh thikA2 2sg.obl tha tha then strygh strygh smear tegh thik yourself wel væl well om um around medh mæth with salffuen salven the balm

> 'When the gout bothers you, then smear yourself well with the balm' (Lægeb Thott 47, 30)

### Lars Heltoft

c. oc ok and skal skal must han han he bevi[se] bevise show them thæm them ydermer ythermer more vinskap vinskap friendship, om um if them thæmA1 3pl.obl threnger thrænger are in distress eller æller or vetherthorvæ vitherthurv-e need-prs.pl 'And he must show more friendship to them, if they are in distress or they need this' (3/8 1442 Varberg)

In (16c), however, the oblique case argument *thæm* is the A1 (the A2 is probably zero = *ythermer vinskap* 'more friendship').<sup>12</sup>

The verb *varthe* is transitive in (17). It has number agreement between the nominative subject and the finite verb, and the A2 in in the accusative, as in unmarked transitive patterns. Example (17c), however, is an inactive construction on the basis of the same verb stem in the active voice.<sup>13</sup> The use of the cataphoric nominative pronoun *thæt* is not obligatory, it is not a formal subject marker, and this construction therefore consists of an A1 in the oblique case, and a predicational A2 (*at the hava æj vin*). The A1 has inactive semantic role meaning (in this case as the Obliged in a relation of duty or relevance coming from the outside).

(17) <the owner of a pond may bar his fellow-villagers' access to the pond>


<sup>12</sup>The verb *vitherthurve* 'be in need of something' and its simplex *thurve* 'need' are not inactive verbs, and the GldO has no examples. The conjunction between *thrænger* and *vitherthurve* does not prove anything about subject status for the A1, since oblique A1s cannot agree with verbal number. *Vitherthurve* can easily be read as a zero-argument transitive: (they) are in need (of this) (i.e. friendship). There is nothing in Old Scandinavian like Modern English or Modern Danish gapping rules.

<sup>13</sup>Similarly in Old Icelandic, with an accusative A1: (*at segja þér þat*) *er þik* (acc) *varðar* 'to tell you what concerns you'.

<sup>14</sup>In the Scanian Law, the transitive interpretation of the verb *vartha* governs a dative object: *Elde* (d) *sin-um* (d) *scal man vartha* (CCD III 93r) 'a man must safeguard (or 'be responsible for') his fire'. The West Danish example could either match the Old Icelandic situation where *varða* in the sense of 'guard, watch' governs the accusative, or it could be an instance of the general loss of verbal government of the dative case. I retain acc here, since the form indicates that this source preserves the accusative (*han*) vs. dative (*hanum*). There are no examples known to me of inactive constructions in Western Middle Danish that preserve a distinction between the accusative and the dative.

7 Semantic reorganisation of case paradigms

b. Hwat hvat what waardher varthar concerns thet thætA2 it.nom miik [mik me.obl eller æller or tik, thik]A1 2sg.obl at [at that the the they.nom.pl hawa hav-a have-pl ey æj not wiin, vin]A2 wine, (...) (...) 'How does it concern me or you that they have no wine' c. Thet thætA2 it-nom wordhar varthar concerns them thæmA1 3pl.obl som sum rel os os us.obl hawa hava have budhit, buthit asked oc ok and ey æj not os, osA1 us.obl thet [thæt this at at that the the they hawa hava have ey æj not wiin. vin.]A2 wine 'It concerns those who have invited us, and not us, that they have no

A fourth example of a neutral stem would be the verb *skilje*, meaning (transitive) 'divide', (reflexive) 'part, divorce' and (inactive) 'disagree'. The inactive function is exemplified in (8).

## **2.4 An apparent voice merger**

wine' (Post 46, 9–13)

Some inactive verbs construe inactively as such irrespective of voice, that is, both the active voice and the middle voice can be used. I will discuss the verb *thækje* 'learn, find reasonable' • 'like, please', which allows an A2 of either type: nonpredicational or predicational. In the active voice, the inactive construction of *thækje* means that 'somebody knows or learns something', or that 'somebody finds something reasonable', as in (18–19).

(18) vthæn uten but standæ stande stand moth mot against høymot høghmot haughtiness oc ok and bældæ bælde arrogance met mæth with mywgdom, mjukdom meekness tho uær men tho-at-hvarem even if hanum han-umA1 he-obl tekker thækk-er learn-3prs.sg thet thætA2 that at [at that han han he vorthær varther becomes forsmoth forsmath despised ther thær there aff af from fore fore for værdæn. værden.]A2 the world. 'But he must resist haughtiness and arrogance with meekness, even if he learns he is despised for this by the world' (Luc 65r 14–17)

### Lars Heltoft

(19) æn æn even ther thær there forudhen foruthen in addition ma ma may man man one delæ dele charge hannom hanum him fore fore for hærwirke hærwirke armed sagh, sak robbery oc ok and æn æn even ydermere ythermere more vm um if hannom han-umA1 he-obl thecker thækker seems reasonable 'And in addition to this, one may charge him with armed robbery, and even more if he finds this reasonable' (Thord Degn text 2, 122, 20)

The middle form of this verb is *thækkjes* 'to please, to satisfy', in religious texts a most frequently discussed relation to God and Jesus, and therefore one of the best documentations of the distribution of case forms, including word order.

(20) Oc ok and æy thes mynne æj thæs minne nevertheless gøre gøre do the the they ther æffter thær æfter thereafter alt alt all thet thæt that them thæmA1 they.obl thækkes thækk-es \_\_A2 please-middle 'And nevertheless they do thereafter [after the Holy Communion] anything they please.' (Fragm 107, 15–16)

Examples (21–22) have 2sg nominative A2s.


'She (Anna) lived completely in righteousness, and therefore she pleased God.' (Bønneb III, 61, 8–10)

Notice that (21–22) cannot have the transitive reading 'do something to please'. They mean 'A1 finds pleasure in A2'.

In the case of *thækkje* there was a clear semantic difference between the lexical meanings realised, in the active and the middle voices, respectively. In all probability, some instances of genuine mergers are also found. In addition to example (7), there is also the following version of a poetic formula:

7 Semantic reorganisation of case paradigms

(23) jen stwndh en stund at a time fors for-s fare.prt-middle.3sg hanum han-um he-obl fwld ful full wæl væl well adh at along 'For some time, he (a rich king) fared very well (i.e. he had great luck).' (RD I, 147, 4620 – 4621)

Other candidates would be *thykje(s)* 'think, seem' and *hope(s)* 'hope, wish'. However, space will forbid a more thorough investigation, and the full overview is – as I see it – no precondition for the present line of thought: to lay out the paradigmatic organisation of case and constructions in Middle Western Danish, in order to relate this to some basic differences between the paradigmatic organisation of word order in Middle and Modern Danish.

## **2.5 The loss of the accusative-dative distinction**

Even a brief comparison of Middle Danish with Old Icelandic will show a major difference, namely the loss of a clear accusative vs. dative distinction in Danish. The archaic Scanian dialect preserves clear datives in cases like (24) and (25).


Notice the number concord in (24) between the subject *the logh* and the finite verb *brista*. Whether this construction belongs to the inactives, will be discussed below.<sup>15</sup>

(25) Sama sama same nattena nattena night tha tha when hon hon she var var was dødh døth dead tha tha then tedhis te-th-is show-prt-middle abodanum abod-a-numA1 abbot-\*-def.dat.sg en en [a.nom andelik andelik spiritual-nom syn. syn vision]A2 'On the same night when she had died, then the abbot had a spiritual vision.' (SjT 34, 5–7)

<sup>15</sup>In (24–25), \* = a syncretism of acc/dat/gen, characteristic of the an-stems and ōn-stems.

### Lars Heltoft

Clearly, (25) documents the distribution of case with this type of inactives in the archaic Scanian dialect, but what is hard to document in Danish is not the use of explicit datives for the A1 of inactive constructions, it is the accusative. The earlier, presumably Common Norse system was preserved in Old Icelandic (and to a large extent even in Modern Icelandic), and here the A1s can appear in the accusative. I shall compare the situation with two verbs, in Section 2.5.1. the verb OIcel. *reka*, Middle Danish *vreke*, *vrake*; in Section 2.5.2. the verb OIcel. *bresta*, Middle Danish *briste*.

### **2.5.1 A difference from Old Icelandic**

Old Icelandic has the inactive construction type (26) (cf. Sigurðsson 1989, 2006):

(26) bát-a-na boat-acc.pl-acc.pl.m rak drift.prt.sg til lands ashore 'The boats drifted ashore.'

The archive of the Dictionary of Old Danish lists as comparable verb forms transitive *vreke* 'drive out, expel' • 'open a lawsuit', from \*wrekan (Ablaut type 5), and a parallel (mainly East Danish) form *vrake*, corresponding to Germanic \*wrakan, but possibly a relatively recent remodeling to Ablaut type 6.<sup>16</sup> The intransitive meaning 'drift' and the inactive construction is not found in the data in the active voice but has apparently been replaced by a mediopassive intransitive. Such intransitives as (27a–27b) can have nominative subjects.


<sup>16</sup>East Norse preserves Germanic \*w- in front of r-, compare Old Danish *vrēth* 'angry' to Old Icelandic *reiðr* 'angry'. There is even a -jan-formation *vrekje* 'expel', from \*wrak-jan, to be disregarded here.

### 7 Semantic reorganisation of case paradigms

True, the pronominal form *thæt* does not distinguish the nominative from the accusative but judging from Old Icelandic this distinction is clear-cut. Example (28a) is the inactive construction, while (28b) is a reflexive construction with a nominative subject.


The correct strategy here is to postulate only inactive constructions where inevitable. The data are scarce, but it seems likely that this type of inactive construction has been replaced in Danish, in this case by an intransitive middle form.

### **2.5.2 The verb** *briste/bresta*

The polysemous verb *briste* 'burst, split' • 'fail' • 'miss, lack, be short of' (Old Icelandic *bresta*) is yet another illustration of the way the Danish construction has been reshaped. In the sense of 'lack, miss', *bresta* is documented with an accusative A1 *mik* (the dative is *mér*):

(29) eigi not brest-r lack-prs.3sg mik 1sg.acc áræði courage 'I do not lack fighting spirit.' (ONP 750 Vatnsdæla saga)

Even in the most archaic Danish data, I have found nothing similar with any type of NP, so the accusative type has been merged with the dative type, as typical of almost all other occurrences, as in (30):

### Lars Heltoft

(30) førstæ fyrste first them thæm them.obl brøster brist-er lack-prs.sg wobn vapn weapon i i in strid strith combat tha tha then holdæ halde hold the the they met mæth with æn en one hand hand hand oc ok and slaa sla punch met mæth with then thæn the annen annen other 'As soon as they lack a weapon in combat, they grip (the enemy) with one hand and punch with the other one.' (Luc 60r 21–23)

This means we can ask whether constructions with the other senses of *briste* should also be analysed as inactive constructions. Consider (31) (sense 'burst') and (32) (sense 'fail'):

	- af. tha …
	- from, then …

'But if it happens that he who drives that the rein bursts for him, or he who rides that the bridle should split for him, and the cart or horse run with him and (this) man meets his death from this, then ...' (DgL V 202, 9)

Where the sense of 'burst, split' is concerned, there is no conceptual necessity that an oblique actant should be part of the valence schema. We can have *tyghlæther brister* 'the bridle splits' and *bughe brast* 'the bow burst', Old Norwegian *Jorðin oll brestr oc rifnar* (ONP 2: 750) 'the whole earth is bursting and quaking', without implying an extra Afficiary<sup>17</sup> actant. The Norwegian example has a nominative subject and documents that the verb is intransitive in this sense. An Afficiary actant may of course be added, but then freely, as a free oblique argument with the Afficiary Role as the A2, in the present case the Maleficiary

<sup>17</sup>The terms Afficiary and Maleficiary are from Zúñiga (2011).

### 7 Semantic reorganisation of case paradigms

variant. In (32), the meaning of *briste* is 'fail, not succeed', clearly implying an argument ascribing the notion of a Maleficiary to its referent.

(32) en æn but brister brist-er fail-prs.3sg hannum han-um he-obl takk. tak guarantee eth æth or skotæ. skote proof tha tha then gøme gøme guard bondæ bonde landowner sialf sialf self sin sin refl thiuf thjuv thief

'But if guarantee or proof fail him (a suspected thief), then the landowner may himself alone take his thief into custody.' (JL CCD X, C 37, 45r)

On the basis of this line of argument, I group the types (24) and (30) together with (32) as synchronically belonging to the inactive constructions. The A1 is in the oblique case, and neither subordinate sense is compatible with any notion of agenthood where the semantic roles are concerned. What we obtain, is a new variant of the A1 Inactive role, namely the Afficiary role, in addition to the Experient role. Notice that in transitive constructions, Afficiary meaning can only be ascribed to A3s, since the dative with verbs like *thakke* 'thank', *skathe* 'do harm to', *møte* 'meet', *varthe* 'be responsible for' has been lost in Middle Western Danish.

## **2.6 Case roles of the inactive constructional system**

In this survey of inactive Middle Danish predicates, the categorisation below seems to cover most of the occurrences. No Agentive meanings are coded, and the Inactive semantic roles apply to animate referents that could in a different constructional context very well carry Agentive meaning. The inactive role differs from the patient role in that the latter applies freely to animate referents and inanimate referents alike, the former only to potential agents.

	- a) A1 (Experient), A2 (External factor), such as: *angre* 'repent', *drøme* 'dream', *hope(s)* 'hope', *minnes* 'remember', *sjunes* 'seem', *tes* 'appear', *thryte* 'regret', *thækkje(s)* 'know, learn'; 'please'.

### Lars Heltoft


## **3 The paradigmatic organisation of pronomial case in Western Middle Danish**

So far the analysis has shown that we cannot know the actual content of the case forms without checking their valence bearer, i.e. the verb stem governing them. Actants in the oblique case are polysemous as far as the content of the case form is concerned. Case forms with one-place verbs are simple, since a nominative actant will be checked against an intransitive verb and the abstract, open semantic role (the classical, general function of the nominative) will be selected, for instance, *the gape* 'they gape, open their mouths wide' has the nominative *the*, and since both agentive and non-agentive readings are possible, the stem *gap-* confirms that this nominative must be read in the open, unmarked sense. In *mik thyrster* 'I am thirsty' the oblique form *mik* will point to the stem *thyrst*to acquire the inactive role reading, excluding the patient reading. In the case of two-place stems, let alone the polysemous ones like *thrængje*, *varthe* and *skilje*, the argument hierarchy helps to determine which variety (bound variant) of the case meaning is the relevant one, and it is therefore part of the paradigmatic organisation. Case meaning and constructional meaning must both be included in the paradigmatic analysis. Say that the semantic roles relevant for transitive constructions are Unmarked role (very often Agent), calling for the nominative case, and Patient, calling for the oblique case. This pair of roles will not apply as case meanings for the arguments of the inactive constructions such as *hanum thækker thæt* 'he learns this', cf. (18). The oblique form *hanum* must manifest an A1 and hence this case form must denote the inactive role, an animate referent, with two-place predicates, influenced by some external factor, for instance: a phenomenon perceived, a norm to be complied with, or some state-of-affairs related to what is in one's interest or need. Notice again: It is excluded from any meaning of agenthood or intentional action.

The form *thæt* 'that/it' denotes the external factor leading to the state of satisfaction on behalf of A1's referent, that is, it is a nominative A2 with a very specific meaning. Syntagmatic hierarchy and case oppositions go together, and such combinations of morphological contrasts and syntagmatic systems were called connecting paradigms by Nørgård-Sørensen et al. (2011), since they consist of both morphologically determined meaning potential and constructional

### 7 Semantic reorganisation of case paradigms

determination of the choice between options given by the polysemous case system. Thus, the structurally determined meaning of the members of the case paradigm is the result of an intersection between morphology and construction, and case meaning has both a morphological expression system and a syntagmatic, constructional one.

To see this in uncomplicated practice, take the German dative case form. This will receive different semantic interpretations from different predicates. In (33– 35), a well-known type of example, case meaning differs along with argument hierarchy.


Schematically, cf. Table 3. <sup>18</sup> The present analysis of Middle Danish can be represented as Table 4.


Table 3: Hierarchy of German dative

The status of the nominative A1 as unmarked is of course fully compatible with the expectation that the majority of lexically, not grammatically, determined roles will be Agents, but the nominative in transitive constructions does not insist on this.

<sup>18</sup>It is not important to discuss here whether the dative A2 means Afficiary or Patient, or whether dative verbs like *begegnen*, *begleiten and folgen* take a Comitative A2.

### Lars Heltoft

Table 4: Case paradigm and argument hierarchy in Middle Western Danish


## **3.1 The indexicality of case**

The semiotic function of pointing between signs is well-known from C.S. Peirce's semiotics as a subtype of the indexical function. This notion has been applied especially to morphology by Andersen (1980) and Anttila (1975), with a clear indication that it will apply to syntactic and topological issues as well.

When indexical, case forms point to their governing predicate as the locus where their exact semantic function is determined. The predicate determines the relevant argument status and the relevant variety of semantic role. With the oblique form, the choice is between Inactive and Patient; with the nominative, it is between Unmarked role (very often lexically filled in as Agent) and External factor (Experiencer, Afficiary or Obliged). Thus, the nominative of the two-place inactive construction points to an inactive verb and receives A2 status, with a very specialised semantic role meaning potential.

In conclusion, indexical case means the case form depends for its actualised meaning on its predicate. Importantly, indexical case structure is but one typological organisation of case. To some extent, Old Indo-European languages have symbolic case structure19, and as we shall see, Modern Danish has in fact abolished indexical case to replace it by a simple symbolic opposition, see Section 5.

## **3.2 Subjects and objects**

Up to this point, I have by and large avoided the issue of grammatical relations in the sense of subjects and objects. The argument hierarchy is laid out as pro-

<sup>19</sup>Where symbolic case is concerned, the case form alone bears case meaning. A well-known remnant of simple symbolic case in Latin is found in *cave canem* 'beware of the dog' vs. *cave cani* 'take care of the dog'. The case opposition specifies the meaning potential of the verb stem cave-, in itself neutral to this opposition. Case is normally indexical in Latin. In *signa … detracta lucis* 'emblems carried out from the groves' (Tacitus Germ. 140, 3), the case ending -is indicates the stem detract- from where the ablative sense of the case ending is determined.

### 7 Semantic reorganisation of case paradigms

jections of valence structure, and a priori assumptions of a connection between A1 and subject, A2 and direct object has been deliberately shunned.

Mel'čuk (2014) suggests a set of universal syntactic criteria for (*not* features of) the universality of subjects, applied by me in Heltoft (2021a,b). To the criteria of Keenan (1976), he adds a distinction between definitional criteria (necessary for a given language) and characterising criteria (frequent, but not necessary). Very briefly, his definitional criteria are laid out in 1–7. The *subject candidate* (SC) must be checked against the following parameters:


On the basis of Section 2, we can now determine the subject criteria relevant for Middle Danish and compare them to the criteria relevant for Modern Danish.

Criterion 1 applies to all instances of A1 and A2, both in transitive and in inactive constructions20, and where criterion 2 is concerned, all arguments are omissible. Thus, neither of these parameters are relevant for Middle Danish.

It must be an open empirical question whether the subject candidates hold a particular linear position, and I will deal with this in Section 4. To reveal the conclusion already here, Middle Danish does not have a subject position, whereas the modern language certainly has developed one, cf. Section 1.2. This means that we are referred to morphological criteria, namely to numeral concord and to case rection (government). In transitive constructions, the A1 must be in the nominative case; inactive constructions, by contrast, take the A2 in the nominative. As

<sup>20</sup>I omit here a discussion about the status of Predicative complements as Main Predicates; see Heltoft (2017), in general Hansen & Heltoft (2011).

### Lars Heltoft

a general principle, nominative DPs agree with the finite verb in number,<sup>21</sup> cf. Section 2.1. These criteria point to the nominative DPs as the subjects of Western Middle Danish. Parameter 7 is relevant as far as it determines the application range of nominative government. Voice cannot count as a defining feature, since inactive constructions do not have an active vs. passive voice contrast.

## **3.3 A constructional typology: Case, grammatical relations and argument structure**

The outcome of the analysis is that the overall distribution of case defines the subject in Middle Danish, whereas the argument status is responsible for the ascription of semantic role variety within the case system. There is no traditional term for a grammatical relation corresponding to the A1 inactive, since the idea of a direct object is intimately connected with the transitive pattern. We can illustrate the two types in Table 5.


Table 5: Transitive and inactive constructional typology

To add to the relevance of the distinction between arguments and grammatical relations, I include two further possible interaction types between morphology, grammatical relations and argument structure, namely the constructional option found in both English and Modern Mainland Scandinavian, somewhat confusingly named "ergative" by Halliday (1968, 1994). Hansen & Heltoft (2011) call this pattern the incausative pattern, and Danish verbs construing in this way are: *brænde* 'burn', *dreje* 'turn', *standse* 'stop', *vælte* 'turn over'*, øge* 'increase', etc., the translations immediately offering English parallels.

	- She.nom burned them.obl 'She burned them.'

<sup>21</sup>Some details omitted, especially about the singular substituting for the plural, never vice versa.

### 7 Semantic reorganisation of case paradigms

The incausative structure is shown in Table 6. It is a combination of ergative argument structure with transitive grammatical relations and transitive case morphology. The modern case morphology involved is different from that of Middle Danish, see below Section 5.2, in that it no longer marks semantic role. It is an example of ergative argument articulation in combination with what looks like transitive morphology. To make this point stand out, I add classical ergative structure, as represented by Greenlandic in Table 7, examples (37–38).


Greenlandic has always number and person concord between subject and finite verb, and in transitive clauses even between direct object and finite verb. In transitive clauses, the intransitive concord is maintained and yet another layer of concord is added. In elementary Greenlandic:

	- b. piniartu-t sealer-abs.3pl piniar-put hunt-indic [3pl(subj)] 'The sealers are/were hunting.'
	- b. piniartu-p sealer-rel.3sg puisi seal-abs.3sg pisar-aa catch-indic [3sg(subj).3sg(obj)] 'The sealer catches/caught the seal.'
	- c. piniartu-p sealer-rel.3sg puisi-t seal-abs.3pl pisar-ai catch-indic [3sg(subj).3pl(obj)] 'The sealer catches/caught the seals.'
	- d. piniartu-t sealer-rel.3pl puisi-t seal-abs.3pl pisar-aat catch-indic [3pl(subj).3pl(obj)] 'The sealers catch/caught the seals.'

### Lars Heltoft

## **3.4 Summary**

Summarising Section 3, the main point is that inactive constructions cannot be reduced to transitive constructions, and the semantic role ascription to their A1s cannot be reduced to that of transitive subjects. The polysemy of the members of the case category is resolved by indexical pointing to the predicate as the valence bearer. In Section 5, we shall see that this system was replaced by a symbolic, nonvalence governed case system, mirroring at first syntactic relations alone, later also phoric distinctions.

## **4 Inactive constructions and the topology of Middle Danish**

In this section, we return to Melčuk's criterion 3 (Mel'čuk 2014) and the question whether Middle Danish subjects can be positionally identified. One point here is Melčuk's distinction between definitional criteria, which are necessary for a given empirical language, and characterising criteria, for instance prototypically relevant features, and thus also standard identifications of subject positions as the position held in unmarked clauses (the more marked positions then being transformationally derived). What we are asking, then, along with Melčuk, is whether some positional criterion is unique for the Middle Danish subject. For instance, Modern French subject topology is unique, in that this language has a position reserved for subjects, and furthermore, an obligatory one.

I shall add the question whether Middle Danish had a particular linear position for subjects, and next, whether subjects contribute to the content side of the word order paradigm for Middle Danish. We have already seen in Section 1.2 that Modern Danish certainly has a semantically coded subject position.

I have claimed elsewhere (Heltoft 2003, 2011,Nørgård-Sørensen & Heltoft 2015) that word order can be paradigmatically organised. Just like with morphological paradigms, we must distinguish between the *frame* of word order paradigms: the semantic content zone coded in the paradigm, and the *domain* of a paradigm: the syntagmatic context where the paradigmatic contrast applies. For the old Scandinavian languages, the semantic frame of word order was not argument status, nor syntactic relations, but information structure.

## **4.1 The iconic focus pattern of Middle Danish**

Initially, all Old Scandinavian languages are verb second, but in relation to the non-finite verbs, they retain the possibility of OV-order, or more generally, XV-

### 7 Semantic reorganisation of case paradigms

order, X being all types of NPs, predicatives and adverbials. The finite and the non-finite verb define three topological zones, a prefield preceding the finite verb, a middle field between the verbs and a postfield, following the non-finite verb. I illustrate this through examples of transitive constructions, namely (39) showing pronominal object + V, (40) showing full NP object + V and (41), pronominal object + full NP subject + V. In (42), I add an example of V + negation + subject + V, in which the object holds the initial position.


'Not even fire can harm this gem.' (Harpestreng 191,13–14)

These examples document two points: 1) Focus operators such as *ække* 'not', *æj* 'not', *ok* 'also' and *hværken* 'neither' define information structural subzones, a background zone preceding the operator and a focus zone following it. 2) There is no specific subject position, and like objects, a subject can be in focus position. If there is no operator, the system predicts that an object or adverbial will precede a focused subject. The relevant portions of text can be laid out topologically as in Table 8.

### Lars Heltoft

Table 8: Information structure and word order in Middle Danish


In Peircean terms, the finite and the nonfinite verb indicate the middle field, the zone for word order to manifest symbolic information structural meaning, the opposition of *background* versus *focus*. In analogy with morphological paradigms, a given member cannot manifest both meanings; however, this symbolic paradigmatic contrast must be mapped onto a linear sequence, and this iconic sequence (Heltoft 2019, 2003) is then indicated again by the position of focus operators. The indexical function of verbs and focus operators define the domain of the paradigm.

$$\begin{array}{ll} \mathcal{V} \Rightarrow \begin{bmatrix} & & \text{Middle field} & & \end{bmatrix} \Leftarrow \mathcal{V} \\ \mathcal{V} \Rightarrow \begin{bmatrix} \text{Background positions} \Leftarrow \text{Operator position} \end{bmatrix} \Leftarrow \mathcal{V} \end{array}$$

Figure 1: Topological analysis of inactive clauses

Notice that there is no coded subject position. Subjects can occur anywhere in a clause, depending on the textual organisation. Again, what is structurally possible – not what is frequent – defines what is grammaticalised. No doubt, subjects in the 3rd position, immediately after the finite verb, have an overwhelmingly high frequency, but this fact can in all probability be derived from the fact that the A1s of the transitive system are very often lexically coded as Agents. At any rate, there is obviously no interlock between A1, 3rd position and subject, so the Middle Danish subject is clearly not topologically coded.

## **4.2 Inactive clauses follow the general pattern**

In this subsection, I consider a number of examples illustrating the positional range of A1s and A2s. Since the domain of the paradigm is the middle field, special attention will be given to examples where both A1 and A2 are in this field. Example (43) documents that A1 can hold the third position, A2 holding the open initial position; and vice versa, (18), partly repeated here as (44), documents initial A1 and 3rd positional A2.

7 Semantic reorganisation of case paradigms


(45) documents A2 preceding A1 in the middle field, (21–22) likewise, see above. For A1 preceding A2 in this context, see (13).

	- b. Nw Nu now sømmer søm-er befit-prs.sg thet thætA2 it.nom mik mikA1 me.obl icke ække not oc ok and jek jæk I kan ey kan æj cannot fanghe fange get noghet noket any andhet annet other klædhe klæthe. garment. 'Now this (anaphor = 'wearing no tunic') does not befit me and I cannot get any other garment.' (ML 407, 7–10)

### Lars Heltoft

æj not vin]A2 wine thætA2 it.nom varthar concerns thæmA1 3pl.obl sum rel os us.obl hava have buthit asked ok and æj not os us.obl [at [that the they hava have æj not vin.]A2 wine] 'How does it concern me or you that they have no wine? It concerns

those who have invited us, and not us, that they have no wine.'

(48) Teckes thækk-es please-prs.sg.middle ether [ither poss.2pl naadhe, nathe]A1 grace att [at that theres theres their egett eghet own budth buth\* messenger skall skal should føræ føre bring breffwet brevet the letter fræm frem forward tiill til to thee the the Lubskæ, lybske]A2 people of Lübeck, thaa tha then staar star is thet thæt that i i in ether ither Your naades nathes Grace's hendher; hænder; hands; tæckes thækkes pleases ether [ither your naade nathe]A1 grace ickæ ække not thet, thætA2 this, tha tha then haffwe have have wii vi we … … … 'If it pleases Your Grace that their own messenger should bring the letter

forward to the people of Lübeck, then this is in Your Grace's hands; if Your Grace is not satisfied with this, then we have (…)' (29/2 1512 (Halmstad; AarsberGeh VI, Till. 13)) [\*The Swedes' own royal courier, whether he should be granted transfer through Denmark.]

A template including these examples is Table 9. Examples (21) and (45) have both arguments in background position, the focus being on the adverb *mæst* 'most' in (21) and on the verb *sømer* 'is decent' in (45); examples (22), (46), and (47) have their A1s in focus position, but (48), by contrast, has the A2 in focus position.

The logic behind this does not include argument hierarchy or grammatical relations, but the middle field contains a purely topological grammaticalised system, consisting of focus and non-focus (background) positions, indexically identifiable through the position of the focus operators, esp. negation. Examples (45) and (48) both contain the pronoun *thæt* 'that', in (45) in background position, in (48) in focus position. The paradigm's coded contrast is between background and focus, since a linguistic element cannot have both of these information structural values at a time. In this type of paradigm, the contrast is mapped onto the syntagmatic axis, that is: onto word order, see further Heltoft (2019). The system

### 7 Semantic reorganisation of case paradigms

works without any assumptions of grammaticalised connections between topology (word order) on one hand and case morphology, argument hierarchy and subject-object articulation on the other. One could say this type of topological system is neutral with respect to transitivity and inactivity.

For a final argument, notice that in example (2), included in Table 9, the constituent in focus position is the deictic adverbial *nu* 'now'. The examples have mainly been of objects and subjects, but this position is also open to adverbials, should they be intended as the focused constituent.


Table 9: The topological frame for Middle Danish

I have added examples (14–15), in order to add to the number of subject A2s definitely not in the 3rd position.

## **5 Categorical clause structure and the loss of indexical case**

During the period app. 1400–1750 the inactive construction was reinterpreted as transitive constructions, including a shift in case marking aligning the relationship between arguments, grammatical relations and case selection. This actualisation process must be the topic of another study, and I will just give two examples by the same author, the lutheran bishop Palladius:

### Lars Heltoft

(49) derfor derfor therefore bør bør ought.prs.sg i i 2pl.nom at at to haffue have hold denne denne this sted sted place och og and kirke kirke church kierist kjærest dearest offuer over beyond alle alle all andre andre other steder steder places i i in verden verden the world 'Therefore, you ought to hold this place and church dearer than anywhere else in this world.' (Palladius 38, 18–19)

In the very same text, we find the older construction in a sentence otherwise identical:

(50) derfor derfor therefore bør bør ought.prs.sg eder eder 2pl.obl att at to haffue have love denne denne this sted sted place kierist kjærest the most 'Therefore, you ought to love this place the most.' (Palladius 39, 14)

The use of the nominative as the marker of the subject-predicate was abolished during the 16th century. For the details of the distribution of case in this post-medieval period, see especially Jensen (2017, 2018), with supplementary overviews and details in Heltoft (2019) and Jørgensen (2000).

## **5.1 Modern subject topology**

Returning to Melčuk's subject criteria, the difference between the medieval system and the modern one is striking. The modern language has a subject definable along parameters 1–3, and no longer by morphological binding by the finite verb. The subject is the only obligatory DP-constituent, in the sense that its positions must be filled in, if not by a referential DP, then by a formal marker (*det* 'it' or a deictic marker *der* 'there' or *her* 'here'), to facilitate the illocutionary system. A feature not mentioned by Melčuk is the interdependence (*catataxis*, in Hjelmslevian terms, *exocentrism* in Bloomfield's) between finite verb and subject, a relational type and criterion normally disregarded in modern grammatical theories and schools. In contrast to the predicate valence system of the medieval language, these relations are solely between grammatical categories, thus defining clausal structure as subject vs. predicate (in the wider sense), so-called categorical sense structure, the presumedly universal DP-VP dichotomy. This structure is again mirrored in the modern sentence frame, in which the middle field has lost its positions for objects and valence bound PPs. These go into the postfield, mirroring

### 7 Semantic reorganisation of case paradigms

the VP, subject positions illustrated by (51–53), a next to translation of (42). The focus operator *selv* is inserted to match better the meaning of (42), but it may well be let out.



The topological frame of Modern Danish (Diderichsen 1946, Heltoft 1992) mirrors categorical clause structure, in that the middle field contains the definitional subject position and the postfield the non-finite verb and the rest of the valence bound constituents.


Table 10: The topological frame for Modern Danish

The modern word order paradigm and the role of the subject in this paradigm was mentioned in Section 1.2. Given the present preconditions of the analysis, the Middle Danish has nothing similar, and there is no cogent reason to assume any underlying categorical structure.

## **5.2 From indexical case to symbolic case**

In symbolic case systems, case forms are self-dependent in the sense that their meaning can be identified on the basis of the case sign itself. When the indexical case system was lost with the inactive constructional alternation to transitivity, the nominative form (still only in the same handful of pronouns as before, see Table 2), lost its polysemy and could no longer carry semantic role meaning. It was left with the sole content of manifesting the subject function, in the sense of the

### Lars Heltoft

argument that the VP is predicated about, the categorical subject. The Modern Danish nominative case has the content 'subject' in all contexts22. Indexical case systems call for reference to the governing verb and the constructional level of argument hierarchy in order to resolve the polysemy of the case forms and identify the referents. The paradigmatic Table 11 shows modern case meaning, see Heltoft (2021b) for more detail.

Table 11: Danish symbolic case paradigm in categorical sentence structure


Notice that where the medieval transitive pattern had an unmarked nominative in relation to the relevant zones of semantic roles, the shift of content function leads to the reverse relation of a marked subject meaning in contrast to the non-subject function of the oblique case. In modern times, from app. 1900 forwards, the meaning of the nominative specialises even more, so that except for some formal registers, the nominative now means 'anaphoric subject' (Hansen 1967). The oblique form is used in subjects with all kinds of restrictive modifiers contributing to the identifiability of the subject referent, such as (54–55).


Thus, within the frame of the symbolic case paradigm the nominative has again specialised, the oblique form 'bleached', see Table 12.

<sup>22</sup>'All contexts' refer to all uses as 1st rank constituents as heads. I take examples such as the following to be 2nd rank constituents : Ham (obl) *og Peter kommer forbi i dag* 'him and Peter will pop by today'; *det er svært for mor og* jeg (nom) (lit. 'this is difficult for mummy and I').

### 7 Semantic reorganisation of case paradigms

Table 12: Danish symbolic case paradigm adding phoricity to its content


## **5.3 Positions indicate roles and arguments**

In the modern language, arguments and grammatical relations have been aligned, so that all A1s are subjects and all A2s are direct objects. The indexicality of the case category in relation to the predicate is gone, the general rule being that all A1s are subjects and prototypical subjects – whatever the predicate's semantics – are in the nominative. The topological system has changed from a more open and free information structural system to a case-like system with specific positions for the subject, the direct object (and in fact, for the indirect object as well). In this system, the subject position indicates the predicate as the category and stem determining the A1 and its meaning, the direct object position indicates the A2, and the indirect object position the A3,<sup>23</sup> with its more specific semantic role (Recipient).

The dimensions of linear position, case meaning and syntactic hierarchy have been aligned as definitional criteria for the identification of subject and A1 in the modern language.

## **6 Conclusion**

Middle Danish with its very reduced case system still retains the indexical character of Germanic case. In spite of the case system's simple morphological expression side, its content side is very complex. Both cases, nominative and oblique, differ in meaning, depending on their constructional context: inactive and transitive constructions, and these constructions and their case differences are distinguished indexically. The predicate's stem must be checked in order to identify the relevant contextual variety (bound variant) of the case forms.

Grammatical relations (subjects and objects) were not aligned in Middle Danish (or in the Norse languages in general). The core actant A1 is the subject of

<sup>23</sup>For a detailed analysis of the shift from symbolic to indexical function in the topology of the indirect object, see Nielsen & Heltoft (to appear).

transitive and intransitive constructions, but the object of inactive constructions, one-place or two-place. Both types are found with an additional A2, the oblique direct object of transitives, but a nominative subject in the inactive construction. Case assigns semantic roles according to the semantics of the predicate and the constructional pattern.

When the inactive construction was finally lost during the 18th century, the case paradigm also lost semantic roles as its content frame. In present-day Danish, the case system has turned symbolic, in that they code directly the relevant grammatical roles and argument status. Now, the nominative case in itself marks its status as subject and A1, the oblique form – now the unmarked form – has roughly the content non-subject and non-A1.

Topology (word order) has taken over the indexical function the case system had, but in a simpler version with no systematic polysemy. Positions, not case forms, point to the predicate stem.

## **Acknowledgements**

This work is also a part of the project *The Middle Danish language in the light of a modern theory of grammaticalisation*, grant DFF-7013-00045 from The Danish Council for Independent Research.

## **Sources**


7 Semantic reorganisation of case paradigms

C37 and C39, ed. by Peter Skautrup. Copenhagen: Munksgaard. 1961 and 1973.


### Lars Heltoft


## **References**


Faarlund, Jan Terje. 1990. *Syntactic change: Toward a theory of historical syntax* (Trends in Linguistics: Studies and Monographs 50). Berlin & New York: Mouton de Gruyter.

### Lars Heltoft


7 Semantic reorganisation of case paradigms


## **Chapter 8**

## **The Dutch modals, a paradigm?**

Jan Nuyts University of Antwerp

Wim Caers University of Antwerp

## Henri-Joseph Goelen

University of Antwerp

This paper addresses the question how to define the notion of a "paradigm", as a cognitively real phenomenon. The discussion is based on the case of a set of forms from a linguistic class that is not part of the traditional domain of "paradigmhood" (i.e. inflectional morphology): the modal auxiliaries in Dutch. The paper presents the results of a few studies into the diachronic evolution of these forms, grammatically and semantically, showing how a subset of them has gradually accumulated shared features and developed an internal division of labor, thus displaying active group behavior.

## **1 Introduction**

In this paper we address the question how to define the notion of a "paradigm", as a cognitively real phenomenon. We do so by means of the concrete case of a set of forms that may arguably be classified as a paradigm, from a linguistic form class that does not belong in the traditional domain of "paradigmhood" (i.e. inflectional morphology): the modal auxiliaries in Dutch. We present the results of a few studies into the diachronic evolution of these forms, grammatically and semantically, showing how a subset of them has gradually accumulated shared features and developed an internal division of labor, thus displaying active group behavior.

Jan Nuyts, Wim Caers & Henri-Joseph Goelen. 2022. The Dutch modals, a paradigm? In Gabriele Diewald & Katja Politt (eds.), *Paradigms regained: Theoretical and empirical arguments for the reassessment of the notion of paradigm*, 245–265. Berlin: Language Science Press. DOI: 10.5281/zenodo.5675853

### Jan Nuyts, Wim Caers & Henri-Joseph Goelen

The paper is organized as follows. In Section 2 we address the issue of how to define a paradigm, as a cognitively real phenomenon. In Section 3 we briefly lay out the methodology used in the studies into the diachronic evolution of the Dutch modals. Section 4 presents a bird's-eye overview of the grammatical evolutions in this set of verbs. Section 5 surveys the semantic developments. In Section 6 we formulate the conclusions.

## **2 Paradigms**

One of the central issues addressed in the present volume is: how cognitively real is the concept of a linguistic paradigm? The only way to answer this question is to look at the linguistic behavior of the members of a candidate for the label.

This, however, invokes another crucial, and maybe more controversial, issue: what are the necessary and sufficient conditions for categorizing a set of forms as a paradigm? We assume that a diagnosis should be based on three very elementary criteria: (i) Do the forms share properties or characteristics, and do they show a tendency towards increasing convergence over time? (ii) Is there a kind of internal organization within the group of forms, and do the members show developments sensitive to it, for instance in view of optimizing the division of labor between them? And (iii) does the set of forms occupy a distinct position, structurally and/or functionally, in the overall linguistic system of the language?

If the answer to at least some of these questions is "sufficiently positive" (see below), one is entitled to call the set of forms a paradigm. Moreover, one then has to consider it cognitively real: it is demonstrably a significant element in the organization of the verbal behavior produced by the cognitive systems for language use implemented in the brains of the speakers of the language, hence it is somehow "represented" in the latter (in a non-literal sense of "represented": it somehow has a specific, recognizable status in the language users' "cognitive grammar").

This answer raises at least three new questions, however, to which the answer is in part less straightforward.

First of all, what is a "sufficiently positive" answer to the above questions? How many and what kinds of features and tendencies do the forms have to share in order for the set to be called a paradigm? How stringent does the internal organization of the set have to be? How distinct should it be in the overall linguistic system? This may be the wrong way to formulate the issue, though: paradigmhood is not a black-and-white matter, but a graded one (hence the concept of "necessary and sufficient conditions" for paradigmhood is quite relative). Paradigms come in degrees of integration, internal organization, and distinctness in

### 8 The Dutch modals, a paradigm?

the linguistic system, and there is no hard cut-off point for membership of the category. This corresponds to how paradigms come into existence diachronically. They do not appear all of a sudden in one fell swoop. They emerge and develop gradually, potentially over a long period of time (and they will also gradually and slowly disintegrate and disappear again). It is pointless to try to determine one specific point on this developmental "cline" at which paradigmhood starts/stops.

Secondly, what types of sets of linguistic forms may be considered candidates for paradigmhood? The notion of a paradigm is traditionally more or less confined to obligatory inflectional systems in a language (cf. e.g. Diewald & Smirnova 2010, Blevins 2015). These nearly automatically satisfy the criteria of sharing sufficient features, showing a stringent internal organization, and taking a distinct position in the overall linguistic system of the language. Yet the definition of a paradigm in terms of the above diagnostic criteria does not imply in any way that only an inflectional system, or only an obligatory system, can count as such. There is no reason why other types of linguistic forms could not fulfil the criteria for paradigmhood as well, including non-inflectional grammatical forms such as auxiliaries, or forms of which the status as grammatical vs lexical is controversial such as adverbs and adjectives (cf. e.g. the modal adverbs/adjectives in languages), or even clearly lexical categories such as main verbs (cf. e.g. the perception verbs, communication verbs, or mental state verbs in languages), even if these are hardly ever obligatory elements in the grammar of a language. The fact that, diachronically, inflection often emerges from auxiliaries, which in turn typically develop out of main verbs (i.e. one of the classical examples of a grammaticalization path), further underscores this point (cf. the preceding issue regarding the gradual emergence of paradigms).

This raises a third question, pertaining to another central issue addressed in this volume: is there a necessary link between paradigms and grammaticalization? Probably not. There is a tendency in the literature to strongly associate paradigms with grammaticalization processes (cf. the fact that Lehmann 1982 considers paradigmatization a key feature of grammaticalization; see e.g. Diewald 2009, Nørgård-Sørensen et al. 2011). This is to a large extent justified: grammaticalization typically does result in, and may even be the most important trigger for, the formation of paradigms (all inflectional paradigms are due to grammaticalization). (This, however, raises a chicken-and-egg question: does the strive in a linguistic system to form paradigms trigger grammaticalization, or is the formation of paradigms a side-effect of grammaticalization?) Nevertheless, there is nothing in the three diagnostic criteria implying that the formation of paradigms must necessarily always involve grammaticalization.<sup>1</sup>

<sup>1</sup> Some authors would seem to consider grammaticalization a necessary condition for a set of

### Jan Nuyts, Wim Caers & Henri-Joseph Goelen

In this paper we offer an example of a set of forms that may arguably be called a paradigm and that demonstrates the import of the three questions just raised. It concerns a grammatical system in a domain other than inflection, which illustrates the gradualness of the notion of a paradigm as defined in terms of the three diagnostic criteria mentioned earlier, and which shows that grammaticalization is not a necessary correlate of paradigmhood. Our casus belli is the set of modal auxiliaries in Dutch. In the traditional view there are six such verbs in the language: *kunnen* 'can', *mogen* 'may', *moeten* 'must', *hoeven* 'need', *zullen* 'shall/will', and *willen* 'will/want' (see e.g. Duinhoven 1997: 383–437). Corpus investigations into the diachronic evolution of these verbs reveal that four of them, *kunnen*, *mogen*, *moeten* and *hoeven*, show systematic behavior, with a tendency to increase and strengthen shared properties over time, grammatically as well as semantically, while at the same time avoiding functional "conflict" within the set. The other two verbs, *zullen* and *willen*, do not participate in these evolutions. In this paper we will present an overview of the facts to this effect.

## **3 Data**

The considerations in this paper are predominantly based on corpus studies into the grammatical and (in part) the semantic developments of the six modal verbs. Although we will only offer a summary overview of the observations from these studies relevant for our present purpose, we should briefly sketch the method applied in them (more details can be found in the references given below).

Our primary source of information are investigations into the global evolution of each of the six modal verbs from the earliest known documents till the present (see Nuyts 2013, Byloo & Nuyts 2014, Nuyts & Byloo 2015, Nuyts et al. 2018, in preparation). These studies all used the same method and analytical categories.

They compared the grammatical and, in part (see below), the semantic properties of the verbs in samples of (in principle) 200 instances per modal from four main "time slices" (based on the generally accepted division in main stages in the history of the language): Old Dutch (OD, > 1150), Early Middle Dutch (EMD,

forms to be called a paradigm. We do not adopt this perspective. Paradigmhood is a basic analytical notion, required to describe a type of pattern observable in collections of forms in a language. Tying it up with another analytical notion, in casu grammaticalization, means reducing its value as a descriptive notion (and in a way even means making it superfluous, since what it describes is also covered by the notion of grammaticalization). It also means that if one encounters sets of non-grammaticalized forms with essentially the same basic properties (in terms of the criteria specified earlier), one needs to introduce another term for these, and one thereby looses the generalization that one is dealing with the same basic phenomenon.

### 8 The Dutch modals, a paradigm?

1250–1300, the start of the Middle Dutch period), Early New Dutch (END, 1550– 1650, the start of the New Dutch period), and Present Day Dutch (PDD, < 1950).<sup>2</sup> For PDD the studies worked with separate samples of 200 instances for the written language (PDDW) and the spoken language (PDDS).

The samples were drawn from a self-compiled balanced corpus of materials covering these periods. The selection of texts for inclusion in the corpus, and of instances for the samples, was based on criteria such as representativity (in terms of text genres and regional spreading, among others) and comparability across the periods. Within these confines, the selection was random.

The data were analyzed paying due attention to inter-rater-reliability and making use of statistical tools (Fisher's exact and the Spearman rank coefficient of correlation).

As a secondary source, we will occasionally refer to a follow-up study which focused on the grammatical developments from END onwards in *kunnen*, *mogen*, *moeten* and *hoeven* (see Nuyts & Caers 2021). In comparison with the earlier studies, it considered an additional language stage, half way between the stages of END and PDD, labeled New Dutch (ND, 1750–1850), and it worked with much larger samples of (in principle) 1000 instances. Otherwise the methodology was the same as in the earlier studies.

In this paper we will only present the relevant headlines emerging from these studies. For the full story (including tables with detailed frequency information) the reader is referred to the publications mentioned above.

The semantic evolution of *zullen* and *willen* was not investigated in the studies mentioned. Our discussion of this issue in Section 5 is based on the information about these verbs provided in the major general and historical dictionaries for Dutch: the *Oudnederlands woordenboek* (2012) for OD, the *Vroegmiddelnederlands woordenboek* (Pijnenburg et al. 2000) for EMD, the *Middelnederlandsch woordenboek* (Verwijs & Verdam 1885–1929) for Middle Dutch, and the *Woordenboek der Nederlandsche taal* (De Vries & Te Winkel 1864–1998) for the developments from Middle Dutch till the 20th century.

All examples provided in this paper are from the corpus data used in the studies mentioned above (they were not necessarily cited in the publications, though). The sample from which they were taken (OD, EMD, END, etc.) is mentioned between brackets after the example. The relevant form is boldfaced. In examples from the spoken data we maintain the transcription conventions used in the source corpus (no capitals, no punctuation, etc.; the spoken samples were drawn

<sup>2</sup>There is some dispute over the precise temporal demarcation of the stages, the studies followed van den Toorn et al. (1997). Occasionally samples are smaller than 200, because the materials for the period did not contain more instances of the modal.

Jan Nuyts, Wim Caers & Henri-Joseph Goelen

from the *Corpus Gesproken Nederlands* (2004), but the transcript has occasionally been simplified somewhat by omitting, among others, irrelevant pause fillers, repetitions, or back channel cues (without this being marked in the example).

## **4 The grammatical developments**

The grammatical developments of the modals, from OD till PDD, were outlined and discussed in detail in Nuyts (2013) for *kunnen*, *mogen* and *moeten*, in Nuyts et al. (2018) for *hoeven*, and in Nuyts et al. (in preparation) for *zullen* and *willen*.

Like most auxiliaries in the languages of the world, the Dutch modals have all emerged from main verbs through a regular process of grammaticalization (Hopper & Traugott 2003). (See further below in this section on the definition of the main verbal vs auxiliary status of the modals.) *Kunnen*, *mogen*, *moeten* and *zullen* are preterit-presents, verbs of which the current present tense stem was originally the past tense stem. *Hoeven* and *willen*, however, are not (although *willen* has adopted some of the characteristics of the preterit-presents, see De Vries & Te Winkel 1864–1998). The timing of the evolution towards an auxiliary is not the same in all the verbs, however.

In *mogen*, *moeten* and *zullen* the auxiliarization process is more or less completed already in the OD sources. There is no trace of the original main verbal use of *zullen* (which meant 'to owe something (to someone)') in our data for any of the periods.<sup>3</sup> In the data for *mogen* and *moeten* (which originally meant, respectively, 'to have power' and, probably, 'to measure') there are possible relics of the original main verb (with a substantially different meaning, though), even until today. But these are marginal, in all time slots, and these modals, too, were nearly exclusively auxiliary already in OD and EMD.

In *kunnen*, however, the auxiliarization process is in full course in OD and EMD. The auxiliary use is already dominant then, but the old main verbal use, with the meaning 'to know', is still prominently present. It even exists until today, even if it is fairly rare now – (1) is a PDD example.

(1) hij he *kon* could al already een a beetje bit Spaans Spanish (PDDS) [lit.] 'He could some Spanish already.' [i.e.] 'He already knew some Spanish.'

<sup>3</sup>The *Oudnederlands woordenboek* (2012) mentions one single instance, occurring in the *Mittelfränkische Reimbibel*. The linguistic status of this text is under dispute, however (Old Dutch or Old High German?). For that reason it was not included in the data in Nuyts et al. (in preparation).

8 The Dutch modals, a paradigm?

In *hoeven* the evolution is even more recent. This modal has only emerged around the start of the END period, out of the main verb *behoeven*, which meant, and still means, 'to need'. *Hoeven* shows clear traces of its main verbal origins until today, illustrated in (2) (*hoeven* can be replaced freely by *behoeven* in the example). Nevertheless, from its conception in END onwards it predominantly behaves as an auxiliary.

(2) Anderen others vonden found oplossingen: solutions ze they lezen read grootgedrukte large.print boeken books of or kijken watch enkel only naar to Nederlands Dutch gesproken spoken programma's programs waarvoor for.which ze they geen no ondertitels subtitles *hoeven*. need (PDDW)

'Others [elderly people with poor eye sight] found solutions: they read books in large print or only watch programs spoken in Dutch for which they do not need subtitles.'

*Willen*, finally, is predominantly an auxiliary from the oldest sources onwards, but the original main verbal use, meaning 'to wish', 'to desire', was still very prominent in EMD, and, even if declining through time, remains present until today. (3) is an example.


In sum, in terms of its origins the set of modal verbs in Dutch (as traditionally conceived) emerged only very gradually and unsystematically.

In more recent grammatical developments, however, a subpart of the traditional set (if it has ever been a real set at all)<sup>4</sup> – specifically, *kunnen*, *mogen*, *moeten* and *hoeven* – starts to behave in a very similar way, both in terms of the direction and of the timing of the changes. *Zullen* and *willen*, however, do not participate in the evolutions, or at least not clearly so. The fact that it concerns quite remarkable developments makes the observation even more significant.

Thus, in the course of the New Dutch period, *kunnen*, *mogen*, *moeten* and *hoeven* start showing a distinct tendency to regain independence, and to get used

<sup>4</sup>There are many more auxiliary verbs in Dutch, hence the fact alone that the six modal verbs have auxiliarized is not sufficient to call them a paradigm. As we will argue in Section 5, the semantic criterion often adduced for considering them a system is not convincing either.

### Jan Nuyts, Wim Caers & Henri-Joseph Goelen

again as an autonomous verb in the clause.<sup>5</sup> Moreover, the properties of the new autonomous uses are exactly the same in all four verbs. Strongly simplifying matters, these uses come in two main types.

On the one hand there are instances that are presumably still auxiliary, but with an elided main verb. Very occasionally, the elision is due to the fact that the implied main verb has been mentioned in the preceding clause, as in (4) (the elided main verb is [*in de tuin*] *werken* 'work [in the garden]'). In by far most cases, the implied main verb has not been mentioned in the preceding discourse, but is more or less clearly imaginable, as in (5) and (6) (in both examples the main verb *gaan* 'go' is understood).

	- [lit.] 'You must [implied: go] along with fashion and with the colors.'
	- [i.e.] 'You must follow fashion and the popular colors.'

The fully implicit type in (5) and (6) is special, though, and differs from the contextual type in (4), in that making the implied main verb explicit usually sounds unnatural and forced to native speakers. The sentence simply feels better without it. Also note that elision of the kind in (5) and (6) is impossible in English, unlike that in (4).<sup>6</sup> Possibly, the position of the main verb is getting unstable in these uses, and the modal verb is regaining independence.

<sup>5</sup>These changes make these four modal verbs in Dutch very different from their equivalents even in closely related languages such as English and German, which do not show comparable developments. They make them special even among the world's languages, since we appear to be dealing with a process of collective degrammaticalization, considered highly unusual in the literature (cf. Norde 2009; see Nuyts 2013, Nuyts & Caers 2021 for discussion of this issue).

<sup>6</sup>The use of a modal without a main verb in the clause (without the latter having been mentioned in the preceding discourse) is to some extent possible in other Germanic languages, including

8 The Dutch modals, a paradigm?

On the other hand, there are autonomous instances in which one cannot imagine a main verb next to the modal verb anymore, as in (7) and (8). In such cases, the modal itself must be considered the main verb of the clause.<sup>7</sup>

(7) Wat what u you doet do *kan* can helemaal at.all niet, not een a klooster monastery bouwen build op on het the grootste biggest Joodse Jewish kerkhof cemetery ter on wereld! earth (PDDW)

[lit.] 'What you are doing cannot at all, building a monastery on the largest Jewish cemetery on earth!' [i.e.] 'What you are doing is totally unacceptable, ....'

(8) En and rijdt rides er there tussen between het the feestgedruis festivities door through toch nevertheless nog still soms sometimes eens once een a trein, train dan then is is dat that mooi nicely meegenomen. taken.along Maar but het it *hoeft* need niet not meer anymore per per se. se (PDDW)

[lit.] 'And if there is occasionally still a train during the festivities, that is an asset. But it need not absolutely anymore.' [i.e.] '... But it is not absolutely necessary/indispensable anymore.'

The overwhelming majority of these new main verbal uses – including the examples in (7) and (8) – has a valency pattern that is very different from that of the original main verbal uses illustrated in (1) and (2) above (hence these new main verbs cannot be considered a continuation of the original main verbs, i.e. the development is not a case of "retraction" in Haspelmath's (2004) sense). The

German. Usually this concerns the elision of a motion verb in the presence of a directional phrase, as in German *ich muss jetzt nach Hause* [lit.] 'I must [implied: go] home now'. The possibilities to omit the main verb in Dutch extend far beyond those in German or other Germanic languages, however (instances such as (5) or (6), e.g., would seem impossible in German). The presence of a directional is not required. Main verbal uses of the type in (7–8) below would even seem entirely absent in other Germanic languages (and these never feature directionals). (See Nuyts & Caers 2021 on the role of directionals in the re-autonomization process in the Dutch modals.)

<sup>7</sup>There is not a sharp borderline between the autonomous uses of the type with an elided main verb and the main verbal type. There are quite a few cases of doubt between the two in the data. This may not be accidental, but may be a sign of a diachronic relationship: presumably, if a verb changes from an auxiliary to a main verbal status, or vice versa, it passes through the stage with an implied main verb.

### Jan Nuyts, Wim Caers & Henri-Joseph Goelen

original main verbs are transitive, in all four modals, with a first argument referring to an entity, very often a living being. But the new main verbal uses of the four modals are nearly always intransitive, and their only argument refers to a state of affairs, deictically, or in a complement clause or an equivalent. The difference between the old and the new main verbal use is illustrated again for *kunnen* in (respectively) (9) and (10).<sup>8</sup>

(9) Dat that voel many liede people sijn are die who en not geen no latijn Latin en not *conen* can noch nor en not verstaen. understand (EMD)

[lit.] 'That there are many people who can nor understand Latin.' [i.e.] 'That there are many people who do not know or understand Latin.'

(10) iemand somebody van of twintig twenty die who beroemd famous wil wants worden become dat that *kan* can op on allerlei various manieren ways (PDDS)

[lit.] 'Someone aged twenty who wants to become famous, that can in different ways.' [i.e.] '..., that is possible in different ways.'

There are minor differences between the four modals participating in the reautonomization process, in terms of the precise timing and/or in terms of how intensively they participate in the developments, as the follow-up study in Nuyts & Caers (2021) has shown (cf. Section 3). The process has started after 1850 in *kunnen*, *mogen* and *moeten*, but already between 1750 and 1850 in *hoeven* (parallel with the ongoing process of auxiliarization in this modal, which only started in END). Moreover, the increase in frequency of the autonomous uses is very substantial in all four verbs (it is statistically highly significant in all of them), but it is most prominent in *hoeven*, followed by *kunnen* in the written data, but by *mogen* in the spoken data, and it is overall weakest in *moeten*. 9

<sup>8</sup>The old and new main verbs are also semantically different. The original meaning of the old main verbal uses (mentioned earlier in this section) is entirely absent in the new main verbs. The latter exclusively feature modal and related meanings: sometimes they express dynamic modality, as in example (10), or in (8) above, but they are much more often deontic modal, as in (7) above, or directive (expressing a permission or obligation). (See Section 5 on these meanings; and see Nuyts (2013) for a more elaborate discussion of the meaning difference between the old and new main verbal uses of the modals.)

<sup>9</sup> See the references given earlier for detailed frequency data. To give an impression of the figures: in the data in Nuyts & Caers (2021), in *hoeven* the new autonomous uses jump from app. 1% of

### 8 The Dutch modals, a paradigm?

In spite of these small differences, there is a clear common line in the – from a regular grammaticalization perspective quite unexpected – developments in these four modals. They appear to behave as a system, displaying a collective dynamics, in which analogy between the individual members may play an important role.<sup>10</sup>

*Zullen* and *willen* do not show a comparable evolution, however. *Zullen* does have a new autonomous use, as illustrated in (11).

(11) ik I weet know niet not of whether ie he 't it heel very druk busy heeft has maar but 't it *zal* shall wel rather (PDDS)

[lit.] 'I do not know whether he is very busy but it probably will [implied: be] so.' [i.e.] '... but he probably is.'

It only occurs in the PDDS data, however, and it is very marginal even there (2% of the instances). It is moreover exclusively of the type with an implied main verb, so these instances are arguably still auxiliary. Also, it exclusively occurs in combination with the modal particle *wel*, as in (11). Maybe the few cases in the data indicate that this verb will ultimately join the others in the process of re-autonomization, but if so, it is at least highly reluctant to do so.

*Willen* also shows autonomous uses, but it featured them from the earliest documents onwards (their frequency fluctuates through time), and these uses have not changed in nature over time. They are all remains of the original main verbal use of the verb. New main verbal uses of the kind in (7) and (8) (or of any other type) do not occur, in any of the time slices, not even in the PDDS data. Hence this verb shows no signs of a participation in the group dynamics characterizing *kunnen*, *mogen*, *moeten* and *hoeven*.

## **5 The semantic developments**

The semantic developments, from OD till the present, are discussed in detail in Byloo & Nuyts (2014) and Nuyts & Byloo (2015) for *kunnen*, *mogen* and *moeten*,

all occurrences of the modal in END to slightly over 8% in ND and PDDW and to more than 28% in PDDS. (See Nuyts (2013) and Nuyts & Caers (2021) for arguments why the more drastic increase in the spoken PDD data, which occurs in all four modals, is not due to sloppiness but signals the direction into which the language is evolving.)

<sup>10</sup>The timing and intensity of the process in the four forms would seem to suggest that *hoeven* has the leading role in it. Yet the question is how this is compatible with the fact that this is by far the youngest among the modals, still in the process of auxiliarizing in the relevant period, as well as with the fact that it is much less frequent than the other modals, and occupies a special position among them as a negative polarity item. These remain open questions.

### Jan Nuyts, Wim Caers & Henri-Joseph Goelen

and in Nuyts et al. (2018) for *hoeven*. We have not made an equivalent diachronic meaning analysis for *zullen* and *willen*. Yet the rough outline of the semantic profile of these verbs in different stages of Dutch emerging from the historical dictionaries (see Section 3) will suffice for the present purpose.

The semantic developments in the different verbs perfectly mirror the grammatical ones described in Section 4: *kunnen*, *mogen*, *moeten* and *hoeven* show similar evolutions, but these are not shared, or at least not to the same extent, by *zullen* and *willen*.

Although the original main verbs from which *kunnen*, *mogen*, *moeten*, and *hoeven* emerged had quite different meanings (cf. Section 4: resp. 'know', 'have power', 'measure', and 'need'), as auxiliaries these verbs have developed more or less the same full range of modal and related meanings typically associated with modal verbs in the languages of the world (see Nuyts 2006, 2016 for elaborate definitions and discussion). This includes, in all four verbs, different types of dynamic modal meanings (an ascription of a capacity or possibility, or of a need or necessity, to a participant in the state of affairs, or of a potential or inevitability to the state of affairs as a whole), as in (12) (see also (4), (8) and (10) above).

(12) kijk look als if gij you een a huishouden household hebt have en and ge you *moet* must nog still vanalles different.things d'rbij on.top.of.it doen do dan then is is 't it heel entirely wat something anders else he right (PDDS) 'Look, if you are managing a household and you have to do several things on top of it, then you have a different story, right?'

It includes a deontic modal meaning in all four verbs (an assessment of the degree of moral acceptability of the state of affairs), as in (13) (see also (6) and (7) above).

(13) We we onderzoeken investigate nu now de the authenticiteit authenticity van of de the lak paint en and of if deze this geretoucheerd retouched is. is Bij with zo'n such.an dure expensive aankoop acquisition *mag* may je you geen no risico risk nemen. take (PDDW)

'We are now investigating whether the paint is authentic and has not been retouched. When buying something so expensive one shouldn't [lit. may not] take a risk.'

8 The Dutch modals, a paradigm?

Three of the four verbs have also developed an epistemic modal or inferential (evidential) meaning (an assessment of, respectively, the degree of likelihood of the state of affairs, or the degree of reliability of the information resulting in the postulation of the state of affairs), as in (14) (the inferential meaning occurs in *moeten*, the epistemic one in the other verbs). This meaning type is missing in *hoeven* – but it is also a very minor one in the three other verbs (in *mogen* it has even disappeared again in PDD).

(14) Iets somewhat minder less zon sun morgen, tomorrow en and er there *kan* can een a buitje small.shower vallen. fall (PDDW)

'Somewhat less sunny tomorrow, and there may be an occasional shower.'

All four verbs have moreover acquired a directive meaning (marking a permission, obligation, advice, etc.), as in (15) (see also (5) above).<sup>11</sup>

(15) ik I heb have begrepen understood dat that 't it in in het the paspoort passport niet not *hoeft* need te to worden become ingeschreven registered dus so als if je you dat that geheim secret wilt want houden keep kan can dat that (PDDS) 'I understand that it [one's marital status] need not be mentioned in the passport [i.e. it is not compulsory], hence if you want to keep it secret that is possible.'

Finally, three of the verbs have developed a volitional meaning (expressing a wish), as in (16). This use is missing in *kunnen*, but it is also minor in the other three verbs.<sup>12</sup>

(16) dat that zijn are gedemodeerde old-fashioned spullen things dat that je you zegt say ja yes dat that *hoef* need ik I niet not meer anymore hé right dat that *moet* must ik I niet not meer anymore hebben have in in feite fact (PDDS) 'Those are old-fashioned things, so you think "I don't want [lit. need] them anymore", right, "I don't want to [lit. must] own them anymore", essentially.'

<sup>11</sup>Directivity is often considered part of deontic modality, but there are good arguments to keep the two categories separated. See Nuyts et al. (2010). This is of no further importance here.

<sup>12</sup>Next to the meanings and uses mentioned above, some of the modals have developed yet other minor ones.

### Jan Nuyts, Wim Caers & Henri-Joseph Goelen

There are considerable differences between the four verbs in terms of when, and how fast, the developments towards these different meanings have happened. In *mogen* and *moeten* the full range of modal and related meanings sketched above is already present in OD and EMD, and the original main verbal meaning is obsolete even then. In *kunnen* a considerable part of the developments happened after OD and EMD. Only the dynamic modal meanings are present then, along with the original meaning 'to know'. The deontic, epistemic and directive meanings emerge in the course of the evolutions towards END and PDD. In *hoeven*, finally, the evolution happens helter-skelter: all modal and related meanings emerge more or less simultaneously, at the moment when the form arises as an auxiliary, around the start of the END period. These uses immediately assume a dominant position in the semantic profile of the verb, and they gain further ground in the course of the New Dutch period.

In sum, there appears to have been a semantic "unification" process among these four verbs, aiming to form a linguistic system that allows the expression of complementary meanings within the same range of modal and related semantic dimensions. The process seems to have stepped up after *kunnen* joined *mogen* and *moeten*: it stands to reason that the blitz evolution in *hoeven* is due to a strive in this form to semantically adapt as quickly as possible to the profile of the other three forms. It is significant that the re-autonomization process in the system, as described in Section 4, sets in while the semantic evolution in *hoeven* is still in full progress. This confirms that the four verbs at least from then onwards behave as a full-fledged paradigm.

There are indications that interactions between the members of the set already started much earlier, however. There are, for instance, semantic changes in the four verbs that may be the result of a tendency to avoid synonymy between them, within the range of meanings they share (cf. Nuyts & Byloo 2015). Thus, *moeten* has evolved from a weak modal (expressing ability, possibility, potential, etc.) to a strong modal (expressing need, necessity, inevitability, etc.) in OD (with last traces of the process in early EMD), possibly in order to avoid semantic overlap with weak *mogen*. Weak *kunnen* has gradually taken over meanings from weak *mogen* from OD and EMD onwards, which may explain why since then *mogen* is increasingly focusing on its directive meaning of permission. And strong *hoeven* may have developed into a negative polarity item (a process which started immediately when this modal emerged around the beginning of the END period) in order to avoid conflict with strong *moeten* (which subsequently has come to dislike negative contexts in PDD).

*Zullen* and *willen*, however, do not seem to participate in the systematic semantic evolutions observed in *kunnen*, *mogen*, *moeten* and *hoeven*.

### 8 The Dutch modals, a paradigm?

The earliest meaning evolutions in *zullen* happened before the time of the oldest documents, hence they are unknown. But from OD onwards, and continuously until today, this verb predominantly expresses a temporal meaning, as the marker of the future tense in Dutch. (17) is an EMD example (from Pijnenburg et al. (2000), entry *sullen* in the online edition).

(17) Wi we debroeders the.brothers ende and desustre the.sisters van of sente Saint ians John's hus house jn in ghent Ghent doen do cont announcement alledenghenen all.those die who dese these letteren letters *sullen* shall zien see dat that ... 'We, brothers and sisters of Saint John's house in Ghent, announce to all those who will see this letter, that ...'

This verb probably never developed a dynamic modal meaning (the *Oudnederlands woordenboek* (2012) offers one or two ambiguous examples in which one of the meanings could possibly be situational necessity, as a special type of dynamic modality). In OD and EMD it also featured a deontic and a directive meaning.<sup>13</sup> How important these were is hard to assess: the dictionaries offer numerous examples, but many or most of them can just as well be interpreted temporally. Moreover, these meanings have more or less disappeared since then (the directive meaning is still mentioned in dictionaries for the present day language, but the examples sound fairly archaic). The verb did develop an epistemic meaning (cf. 11), but the source is no doubt the temporal meaning (since the future is inherently uncertain, markers of the future generally show a strong tendency to acquire an epistemic meaning). This is unlike the epistemic (or evidential) meaning in the other modals, which emerged from other modal meanings (most probably from the dynamic modal ones, see Byloo & Nuyts 2014).

<sup>13</sup>In the literature it is often suggested that the temporal meaning (future) emerged from the deontic and/or directive meaning in this verb (see e.g. Duinhoven 1997: 428; a similar claim has been made regarding English *shall*, cf. Bybee & Pagliuca 1987, Bybee et al. 1991). Yet authors do not provide proof for this assumption: they only cite individual instances that existed side by side in the oldest documented stages of the language, but they do not offer evidence that demonstrates a diachronic order between the meanings. It is perfectly imaginable that both meaning types developed in a parallel evolution out of the original main verbal meaning 'to owe something (to someone)'. The development from the original meaning directly to the future meaning involves a straightforward metonymic path. If one owes someone something, this implies that one has to do something in the future to settle one's debt. The change to the future meaning is a small step (it follows the same logic as that offered in the literature for the presumed change from a directive to a future meaning). Offering formal proof for one or the other scenario may be impossible, however, at least for Dutch, since these meanings emerged before the oldest documents.

### Jan Nuyts, Wim Caers & Henri-Joseph Goelen

*Willen* does not seem to have had any distinct meanings other than the present: from OD onwards it marks a wish or desire, i.e. volition. (18) is an EMD illustration (from Pijnenburg et al. 2000, entry *willen* in the online edition).

(18) Aldaer when si they tesamen together waren, were so so sprac said Symon Simon Petrus: Petrus Jc I *wille* want gaen go veschen. fish

'When they were together, Simon Petrus said: I want to go fishing.'

The dictionaries do mention some other meanings, but these can more or less all be accounted for as contextual implicatures from the volitional meaning.<sup>14</sup> In any case, the verb never developed any of the prototypical modal and related meanings central in *kunnen*, *mogen*, *moeten* and *hoeven*, such as a dynamic, deontic or epistemic/inferential one. As indicated above, volition also occurs as a meaning in these other modals, but only as a minor one. Although it is arguably related to directivity (see Nuyts 2008), this meaning is not central to the system of the modal and related meanings.

In sum, *zullen* and *willen* have a semantic profile and development very different from that of *kunnen*, *mogen*, *moeten* and *hoeven*. Hence, also semantically, *zullen* and *willen* are not part of the system constituted by the other four modal verbs. Maybe the reason why they did not join the latter is that their dominant or exclusive meaning (future tense marking and volition), present from OD/END onwards, is too remote from the classical range of modal and related meanings,

	- do with good peace.

'So what they want to/will say and order about it, that I will obey and do in good spirit.'

<sup>14</sup>For example, Pijnenburg et al. (2000) mention 'to demand' as a meaning in EMD (i.e. directivity), but from the examples it is obvious that this always concerns a use of the verb with a volitional meaning, yet in a context in which fulfilling the wish is inevitable for the addressee (all examples are from legal texts such as wills, ordinances, or decrees). A somewhat more doubtful case is the future meaning in EMD listed by Pijnenburg et al. (2000) (which they call "infrequent"). Here, too, most dictionary examples can be read volitionally, but in just a few instances a temporal reading is more obvious than a volitional one. (i) is one of the very few illustrations (adapted from Pijnenburg et al. 2000, entry *willen* in the online edition):

So maybe there has been a minor tendency in this verb in Middle Dutch to develop a temporal meaning.

### 8 The Dutch modals, a paradigm?

and/or too "unnatural" as a source for developing these meanings. (Volition may emerge out of, but is an unlikely source for a dynamic, deontic or epistemic meaning. Time is an unlikely source for a dynamic or deontic meaning. See Byloo & Nuyts 2014.)

## **6 Conclusion**

Dutch *kunnen*, *mogen*, *moeten* and *hoeven* show clear signs of "group behavior", both in the grammatical domain (cf. Section 4) and in the semantic domain (cf. Section 5), hence there is every reason to consider them a paradigm. This conclusion is further underscored by the fact that *zullen* and *willen*, even though they are grammatically and semantically somewhat related (and therefore traditionally considered part of the system of the modals in Dutch), do not participate in the dynamics shared by *kunnen*, *mogen*, *moeten* and *hoeven*. The latter four verbs constitute a significant cluster in the linguistic behavior of speakers of Dutch, hence we cannot but conclude that they play a distinctive role in the cognitive grammar coded in the brains of those speakers.

As such, this set of four modal verbs also offers an illustration of the different issues raised in Section 2 regarding what may count as a paradigm.

First of all, the set illustrates the gradualness of the notion as defined by the three diagnostic criteria mentioned in Section 2. The four verbs clearly satisfy the criterion of increasing convergence. Over the past 1000 years, they have gradually grown closer together, at an increasing speed, both semantically and grammatically, showing more and more signs of system-bound dynamics, such as analogy effects, or avoidance of synonymy or functional overlap within the system. This process has gained momentum in the last 200 years, causing the helter-skelter grammatical and semantic evolutions observed in *hoeven*, as well as the rapidly evolving, and linguistically exceptional, re-autonomization process in all four verbs.

The fact that there are mechanisms of avoidance of synonymy and functional overlap at work also signals that there is at least to some extent a strive for a division of labor among the members of the set (cf. the second diagnostic criterion). As indicated in Section 5, the result in the present day language is that there are two strong verbs (*moeten* and *hoeven*) and two weak verbs (*mogen* and *kunnen*). The two strong verbs differ in terms of their relation with negation,<sup>15</sup> the two

<sup>15</sup>This is true at least in the standard language and in the Northern Dutch dialects. Interestingly, in the Southern Dutch dialects *moeten* does not show a preference for positive contexts, but *hoeven* barely exists in these language varieties hence there is not really a competitor for *moeten* in them (see Diepeveen et al. 2006).

### Jan Nuyts, Wim Caers & Henri-Joseph Goelen

weak verbs in terms of which modal or related meanings are most prominent in their semantic profile (cf. Section 5). The internal organization in this paradigm is not fully systematic (the division of labor is along different criteria between the weak and between the strong modals) nor very strict (*hoeven* is a pure negative polarity item but *moeten* does not entirely exclude negation; and *kunnen* and *mogen* share some meanings). As such the organization is probably weaker than, for instance, in an inflectional paradigm such as a case marking system in which (in the "perfect" situation) all participating forms are more or less complementary and together cover the entire functional "territory" (i.e. coding all semantically and/or syntactically relevant roles of participants in a state of affairs). Nevertheless, the four Dutch modals are relatively complementary in expressing aspects of the shared range of modal and related meanings, witness the fact that there are relatively very few usage contexts in which two of the forms would be mutually exchangeable without altering the meaning or communicative effect of an utterance.

Also the third diagnostic criterion, distinctiveness in the overall linguistic system, is satisfied to some extent. Again, the set of four forms is less unique in the system of Dutch than, for instance, an inflectional paradigm such as the case marking system in Latin, for which there are no alternatives at all in that language. Structurally, there are many other auxiliary forms in Dutch beyond the modals. And semantically or functionally, there are many other (sets of) forms for expressing modal and related meanings in the language, including adverbs and adjectives, and full verbs (e.g. some mental state predicates). Yet none of the latter types of expressive devices covers the same range of meanings as the four modal auxiliaries (modal adverbs and adjectives, for instance, typically focus on one modal category, most often epistemic modality or inferentiality; the same applies for the mental state verbs). Moreover, the modals are the only set of forms covering this semantic domain within the range of grammatical devices in the language (assuming that adverbs are not grammatical forms, a view that is not generally shared).

The gradualness of the notion of a paradigm, and of the three diagnostic criteria for it, is also manifest if one considers the diachronic evolution of the set of four modals in Dutch: it has taken a long time for this system to emerge, and determining a cut-off point on the historical cline for calling it a paradigm is arbitrary.

Secondly, the set of four modals in Dutch also illustrates that not only inflectional systems deserve to be called paradigms. As suggested above, it is beyond doubt that inflectional systems (obligatory or non-obligatory ones) more readily qualify for paradigmhood than systems of forms from other parts of speech. (Not

### 8 The Dutch modals, a paradigm?

all inflectional systems are necessarily "perfect" paradigms, though. For instance, in many languages, systems of verbal affixes for marking categories such as tense, aspect, mood, modality, or evidentiality, even obligatory ones, are not the unique markers for these meanings, and/or are not better organized internally than noninflectional systems.) Still, in our present case we are dealing with a system of non-inflectional grammatical markers, and not even very strongly grammaticalized ones (the Dutch modals are much less grammaticalized than their English counterparts, for instance).<sup>16</sup> Moreover, in view of the re-autonomization trend and the strong increase in main verbal uses of these forms in Dutch, the status of the system as a grammatical one would seem to be weakened.

Thirdly and finally, the latter point also illustrates the last issue regarding paradigmhood raised in Section 2: the link with grammaticalization. Although, as indicated, more grammaticalized systems no doubt stand a better chance to qualify for paradigmhood than less grammaticalized ones, the set of four Dutch modals shows that grammaticalization is not a necessary correlate of paradigmhood. The process of re-autonomization in the four forms, which substantially increases their mutual tie, hence strengthens their status as a paradigm, is even, arguably, an instance of degrammaticalization (see Nuyts 2013, Nuyts & Caers 2021), i.e. an evolution in the opposite direction.

## **Acknowledgments**

Research for this paper was supported by a project financed jointly by the Belgian and Dutch science agencies FWO-Vlaanderen and NWO ("lead agency" project G.0A18.15N).

## **References**


<sup>16</sup>For instance, unlike the English modals, the Dutch modals still have productive infinitive and past tense forms, and they are still inflected for person and number.

Jan Nuyts, Wim Caers & Henri-Joseph Goelen


*Corpus gesproken Nederlands. Version 2.0*. 2004. Leiden.


8 The Dutch modals, a paradigm?

Nuyts, Jan & Pieter Byloo. 2015. Competing modals. *Diachronica* 32. 34–68.


*Oudnederlands woordenboek*. 2012. Online edition. https://gtb.inl.nl.


## **Chapter 9**

## **Grammaticalisation, schematisation and paradigmaticisation: How they intersect in the development of German degree modifiers**

## Jakob Neels

Leipzig University

## Stefan Hartmann

University of Düsseldorf

Grammaticalisation research has identified the tight integration of structures into paradigms as the final state of many grammaticalisation processes. Construction grammar approaches are particularly suitable for modelling such cases of paradigmaticisation since they invite researchers to study constructions not in isolation but in a wider network context. Drawing on this theoretical perspective, this paper investigates the grammaticalisation of a whole family of constructions. Based on synchronic and diachronic corpus studies, it presents quantitative analyses on the interrelated German constructions [*ein wenig* X] ('a little'), [*ein bissche*n X] ('a bit'), [*ein Quäntchen* X] (lit. 'a quantum'), [*ein Tick* X] (lit. 'a tick') and [*eine Idee* X] (lit. 'an idea'). The paper discusses to what extent their individual and shared developments can be understood in terms of paradigm formation, concluding that early paradigmaticisation involves emergent paradigms of multiple orders in the sense of constructions at different levels of schematicity. Change appears to be guided mainly by associations to micro-constructions and lower-level meso-constructions (e.g. Traugott 2007). Only in advanced stages of grammaticalisation, when microconstructions become sufficiently homogeneous, do high-level abstractions – i.e. paradigms in the traditional understanding of the term – act as decisive organisational forces. The key forces at the domain-general cognitive level are argued to be efficiency-driven automation and analogy.

Jakob Neels & Stefan Hartmann. 2022. Grammaticalisation, schematisation and paradigmaticisation: How they intersect in the development of German degree modifiers. In Gabriele Diewald & Katja Politt (eds.), *Paradigms regained: Theoretical and empirical arguments for the reassessment of the notion of paradigm*, 267–296. Berlin: Language Science Press. DOI: 10.5281/zenodo.5675855

Jakob Neels & Stefan Hartmann

## **1 Introduction**

In constructionist approaches to language, the question of how the internal structure of the constructicon can be modelled has recently come to the centre of attention (see e.g. Diessel 2019, Sommerer & Smirnova 2020). The inventory of inheritance relations proposed by Goldberg (1995) has been complemented by various suggestions that put a particular emphasis on horizontal links, called "lateral links" by Norde & Morris (2018) or "sister" links by Audring (2019). Links between constructions at the same level of abstraction are particularly interesting from a usage-based point of view in that they can account for phenomena of language variation and change such as constructional alternations Perek (2015) and constructional contamination (Pijpops & Van de Velde 2016, Pijpops et al. 2018).

The conceptualisation of linguistic knowledge as a fine-grained taxonomic network of constructions leads us to the classic concept of paradigmatic, or, as de Saussure (1916) calls them, associative relations between linguistic units. As Boden (2005: 174) puts it, "[p]aradigmatic relations oppose a unit to others that could replace it in a given sequence". In diachronic research, grammaticalisation theory has identified the tight integration of structures into paradigms as a target of grammaticalisation processes. For example, Lehmann (2015) posits paradigmaticity as one of his well-known grammaticalisation parameters and paradigmatic integration, or paradigmaticisation, as one of the key processes involved in grammaticalisation. In this process, "grammaticalized elements join preexistent paradigms and assimilate to their other members" (Lehmann 2015: 144).

Overall, it is received wisdom that paradigmatic relations are essential to language structure in general, and that paradigms in the more specific sense of indexical, closed units of organisation are central to grammatical structure, most notably in categories such as tense, aspect or case. In usage-based construction grammar, paradigmatic associations are central as well, since they are tightly integrated into the taxonomic architecture of the constructicon. However, some usage-based findings also call into question the importance of paradigms as highly abstract generalisations, suggesting that language users rely on more local generalisations and lower-level schemas (e.g. Boas 2003, Dąbrowska 2008, Perek 2015: Ch. 5, Schmid 2020). Thus, the status of paradigms in usage-based construction grammar is ambivalent. The present paper therefore explores the question of how relevant paradigms are to grammatical structure from a cognitive-functional usage-based perspective. Theoretically, a range of answers to this question is conceivable, with opposing views such as the following: paradigms are merely an epiphenomenon of other motivations and mechanisms shaping grammar; or

they are essential organisational forces, and thus cognitive entities, themselves. The present study approaches this issue from the perspective of language change. It investigates to what extent paradigmatic forces are reflected in the diachronic development of a family of grammaticalising constructions. In particular, the selected test case is quantifier/degree-modifier constructions in German. This constructional family forms a layered domain of grammar (cf. Hopper 1991), with older members such as [*ein bisschen* X] ('a bit (of) X') and [*ein wenig* X] ('a little X') and less grammaticalised members such as [*ein Quäntchen* X] (lit. 'a quantum-dim (of) X') and [*eine Idee* X] (lit. 'an idea (of) X'). Through quantitative and qualitative corpus analyses, this study examines to what extent the grammaticalisation trajectories of each individual construction constitute independent changes or an interconnected process of paradigmaticisation influenced by network links as well as by overarching mid-level or high-level schemas. Despite some constructional individuality, the case study finds empirical evidence suggesting a scenario in which older constructions serve as attractor sets promoting the analogical recruitment of new members to an increasingly strengthened schema they collectively instantiate (cf., e.g., Amit 1989, Traugott 2008b, Verveckken 2015, Aaron 2016, De Smet & Fischer 2017). The organisational units contributing to this trend of convergence between multiple constructions appear to be partially schematic constructions at the mid-levels of abstraction. We reason that such a scenario is characteristic of early paradigmaticisation in the grammaticalisation of periphrastic constructions, and that high-level schemas – which come closest to the traditional idea of paradigms – exert an influence mainly in advanced stages of grammaticalisation and paradigmaticisation as typical of inflectional morphology.

The remainder of this paper is structured as follows. Section 2 discusses the theoretical background, focusing on the status of paradigms in (diachronic) usage-based construction grammar. Section 3 presents the corpus-based case study and interprets its results against this theoretical background. Section 4 provides a more general discussion of the process of paradigmaticisation from the perspective of the constructionist concept of schematisation. Finally, Section 5 offers a brief conclusion.

## **2 Background: Paradigms and usage-based construction grammar**

The structuralist notion of *paradigmatic* comes in broader and in more narrow senses, and not all of them have a close equivalent in the theoretical apparatus

### Jakob Neels & Stefan Hartmann

of usage-based construction grammar. In a broad sense, *paradigmatic* denotes a relation of choice between linguistic items that are associated with each other based on functional (e.g. synonymy) or formal (e.g. homonymy) features. As mentioned in Section 1, paradigmatic relations in this sense translate into links in constructional networks, such as subpart links and metaphorical extension links (e.g. Goldberg 1995) that connect constructions at the same level of schematicity (e.g. *arm* and *leg*; ditransitive construction and *for*-benefactive construction). In a narrow sense, *paradigmatic* and *paradigm* can be understood as sets with a fixed number of mutually exclusive forms whose meanings are indexical in that they are largely determined by their relations within those closed-class sets. Prototypical instances of this type of paradigm are found in grammatical domains such as tense, person, voice and case. Such grammatical paradigms clearly differ from loose sets of open-class items such as lexemes being paradigmatically related via various sense relations. Construction grammar understates this difference in tightness and closedness since it represents lexical items and grammatical items in a uniform format within a lexicon–grammar continuum (cf. Diewald 2020). In response to this problem, several construction grammarians have recently proposed theoretical add-ons: consider Booij's (e.g. Booij & Masini 2015) concept of second-order schemas, Audring's (e.g. 2019) sister schemas, Diewald's (2020) hyper-constructions and Leino's (Leino 2022 [this volume]) metaconstructions. Advances in constructionist theorising are needed to model paradigms more comprehensively.

Figure 1: Constructional levels, exemplified with a network of English

A classification that is very much in line with such network models is found in Traugott's (e.g. 2007) notion of constructional levels in terms of micro-, mesoand macro-constructions. Using the example of English future constructions (cf. Trousdale & Norde 2013: 36), Figure 1 visualises the basic idea behind these con-

structional levels. Micro-constructions, i.e. individual (largely) substantive construction types, can be subsumed under a single highly abstract schema, the macro-construction, if they share the same basic function. In Figure 1, the five formally diverse micro-constructions at the bottom can be taken to be weakly linked at the macro-level based on their shared basic function of marking futurity. In-between, meso-constructions, i.e. mid-level schemas, unite subsets of similarly-behaving micro-constructions with a shared structure. Of the five micro-constructions used for illustration, [*will* v] and [*shall* v] are structurally alike (e.g. v as bare infinitive) and so are [be *going to* v] and non-standard [be *fixing to* v] (e.g. progressive). Unlike [be *about to* v], each of these two pairs of micro-constructions can therefore be subsumed under one meso-construction, respectively. The higher the number of micro-constructions that collectively constitute one meso-construction, the more entrenched and productive this meso-construction tends to become. Meso-constructions are hypothesised to influence neighbouring constructions in the network through the domain-general cognitive process of analogy. Diachronically, constructional clusters headed by well-entrenched meso-constructions may attract other/novel micro-constructions, leading to the growth of a family of constructions. As detailed in the corpus study further below, Traugott's constructional levels thus qualify as a valuable descriptive tool for studying the reorganisation of the constructicon.

However, it has been a point of debate whether all of these constructional levels are psychologically plausible. Usage-based linguists have questioned whether patterns at the highest levels of abstraction/schematicity (e.g. fully schematic argument structure constructions) are represented in the minds of most speakers. Possibly, high-level generalisations are no more than linguists' constructs emerging from the analysis of aggregated usage data. As hinted at by some psycholinguistic and corpus-linguistic evidence (e.g. Boas 2003, Dąbrowska 2008, 2015, Perek 2015: Ch. 5, Schmid & Mantlik 2015: Ch. 5), the actual generalisations of individual speakers may stop at the mid-level of partially schematic constructions. In other words, mid- and low-level constructions appear to be the key material of linguistic knowledge thanks to their greater cognitive accessibility (i.e. entrenchment). From this perspective, high-level schemas, including paradigms, seem to be of relatively little psycholinguistic importance.

Yet, other findings from psycholinguistically minded grammaticalisation research indirectly stress the importance of paradigms. As argued most convincingly in Lehmann (2017), grammaticalisation is essentially a linguistic instance of cognitive automation, i.e. the domain-general process that turns more controlled, intentional activity into efficient, unconscious, rigid behaviour (cf. Schneider &

### Jakob Neels & Stefan Hartmann

Shiffrin 1977, Logan 1988, Givón 1989: Ch. 7; Moors & Houwer 2006). Automation streamlines the execution of recurrent tasks, and grammaticalisation does exactly that in language: it creates efficient solutions for frequently recurring communicative tasks, such as signalling futurity, plurality, possession or negation. Paradigms are an expected outcome of automation in language since they are rigid closed-class structures that can be executed with minimal attention. They allow speakers to maximise the processing capacities available for more complex, discursively primary communicative tasks (cf. Harder & Boye 2011). Humans' efficiency-driven capacity for automation thus seems to be a key factor underlying the trend that grammaticalisation processes are directed towards paradigms as a target state. From this perspective, paradigm formation is essential to language processing.

In short, *paradigm* and *paradigmaticisation* can be translated into usage-based constructionist concepts as follows. The synchronic notion of paradigms roughly corresponds to higher-level schemas and the "horizontal" associations between the constructions sanctioned by these schemas. It must be noted, however, that there is currently no single conventional solution in construction grammar regarding the exact types of schemas and links that fully capture grammatical paradigms in the narrow sense (see Smirnova & Sommerer 2020 for a problemoriented discussion of different conceptions of nodes and vertical as well as horizontal links in constructional networks). The diachronic process of paradigmatisation is then a process of schema formation, i.e. schematisation, and of the gradual convergence of subordinate constructions into an increasingly tight set (cf. Diewald & Smirnova 2012). Two basic cognitive mechanisms of paradigmaticisation appear to be analogy and automation; a basic motivation is processing efficiency.

Based on the theoretical considerations presented above, we derive the following working hypotheses for the present case study on paradigmaticisation:


In the case study presented below, our goal is to link these theoretical assumptions with observations based on the diachronic development and synchronic behaviour of one particular family of constructions. On a terminological note, we employ the term *constructional family* as an exploratory notion: unlike *paradigm* and constructional levels like *meso-construction*, which are meant to capture mental associations and representations, *constructional family* more loosely refers to a group of linguistic expressions with functional and/or formal commonalities at a pre-theoretical level.

## **3 Case study: The development of German quantifier/degree-modifier constructions**

Parts of the present investigation draw on findings from a previous case study (Neels & Hartmann 2018) focusing on the diachrony of the frequent German degree modifiers [*ein wenig* X] (' a little X') and [*ein bisschen* X] ('a bit (of) X'). In that study, we extended our account of the diachronic development to other structurally similar quantifier/degree-modifier constructions with low usage frequencies, but we did not provide systematic corpus data on these less established modifiers. The present follow-up study adds these data, providing quantitative corpus analyses on three more members of the constructional family: [*ein Quäntchen* X] (lit. 'a quantum (of) X'), [*ein Tick* X] (lit. 'a tick (of) X') and [*eine Idee* X] (lit. 'an idea (of) X'). Section 3.1 first introduces the data and methods used in this investigation. Section 3.2 sketches out three possible grammaticalisation scenarios leading to degree-modifier functions, before moving on to the results. Section 3.3 reviews the main results on *ein wenig* and *ein bisschen*. Section 3.4 details the results on the less grammaticalised expressions *ein Quäntchen*, *ein Tick* and *eine Idee*, and provides a synthesis of our earlier findings and the present analyses.

## **3.1 Data and methods**

Five members of the expanding family of German quantifier/degree-modifier constructions are under scrutiny in the present investigation.<sup>1</sup> These five microconstructions are exemplified in (1) to (5) below.

<sup>1</sup>Cross-linguistically, structurally similar constructions include English *a bit* (*of* ), *a shred of* and *a bunch of* (e.g. Brems 2007, Traugott 2008b, Shao et al. 2019), Spanish *un montón de* 'a heap of' and *un hatajo de* 'a herd of' (e.g. Verveckken 2015), and Dutch *massa*(*'s*) 'mass(es)' (e.g. De Clerck & Colleman 2013) among others.

### Jakob Neels & Stefan Hartmann

	- b. Ich tränke gern ein Glas, die Freiheit hoch zu ehren, Wenn eure Weine nur *ein Bißchen* besser wären. 'I'd love to drink a glass, in freedom's honour, if only the wine were *a bit* better.' (1790, DTA)
	- c. Rück doch mal *'n bischen* den Tisch! 'Move the table *a bit*!' (1890, DTA)

In present-day German, [*ein bisschen* X] ('a bit (of) X') can be considered the "ideal" representative of the constructional family, being highly frequent and productive and exhibiting all prototypical features. Typical family members are made up of (i) the indefinite article *ein*, (ii) a noun denoting a small unit, and (iii) an open slot that can be filled by items from various word classes. Different word classes hosted in the constructions are associated with different constructional functions. Partitive and quantifier uses are associated with noun modification, typically with concrete nouns and mass nouns (cf. Examples 1a, 3), respectively. Constructs modifying adjectives (1b, 2, 4, 5), verbs (1c) or other parts of speech generally fulfil degree-modifier functions. As detailed further below, individual members of this family of quantifier/degree-modifier constructions differ in their constraints and preferences regarding the syntactic categories sanctioned in their productive slot.

Usage data reflecting the diachronic trajectories and potential mutual influences of the five micro-constructions were extracted from several historical and contemporary German corpora. In particular, historical data on all five constructions was collected from the *Deutsches Textarchiv* (DTA; German Text Archive; Geyken & Gloning 2015); additional diachronic data on [*ein bisschen* X] and [*ein wenig* X] ('a little X') were extracted from the historical component of the *Deutsches Referenzkorpus* (DeReKo-HIST; German Reference Corpus; Kupietz & Keibel 2009); and we searched a tagged synchronic component of that corpus, DeReKo-Tagged-C, for present-day uses of the low-frequency constructions [*ein Quäntchen* X] (lit. 'a quantum (of) X'), [*ein Tick* X] (lit. 'a tick (of) X') and [*eine Idee* X] (lit. 'an idea (of) X'). For extracting these three younger constructions, the following queries were used:

```
REG(^Idee$|^Tick$|^Quäntchen$) /+w1 MORPH(A)
REG(^Tick$|^Quäntchen$) /+w1 MORPH(N)
REG(^Idee$|^Tick$|^Quäntchen$) /+w1 MORPH(ADV)
```
False hits as well as duplicates were removed manually, and the remaining data were coded for the part of speech of the modified element. In total, this procedure yielded 44 instances of *eine Idee*, 1,777 of *ein Tick*, and 1,694 of *ein Quäntchen* from the contemporary DeReKo-Tagged-C data. Historical data of these three constructions are fairly rare, as described in Section 3.4 further below. For (*ein*) *bisschen* and *ein wenig*, the dataset extracted from DTA amd DeReKo-HIST comprised 3,226 and 15,783 historical tokens, respectively,(see Neels & Hartmann (2018) for details).

## **3.2 Diachronic paths leading to degree-modifier functions**

Despite the structural similarity of the German degree modifiers, their individual origins and developmental paths seem to be fairly diverse. We consider three possible grammaticalisation scenarios giving rise to such micro-constructions. The plausibility of each scenario for the German constructions at hand will be discussed in the subsequent sections in the light of our corpus data.

In the first scenario, a lexical source construction grammaticalises "under its own steam" along a cross-linguistically attested grammaticalisation path. This scenario theoretically entails a highly gradual development with discernable chronological stages and with little to no analogical influence by extant grammatical constructions. It roughly corresponds to what Lehmann (2004, 2015) calls "pure" grammaticalisation or "innovation", contrasting with his notions

### Jakob Neels & Stefan Hartmann

of "analogically-oriented" grammaticalisation and "renovation". According to Lehmann (e.g. 2004: 161), only the former type, i.e. pure grammaticalisation, may give rise to a genuinely new grammatical category (but cf. De Smet 2014: Section 4.1). In such cases of category emergence, meso-constructions can be expected to play no significant role. In the case and time frame investiaged here, they potentially do since categories like quantifier and degree modifier are in place and instantiated by multiple expressions of German. Under these conditions, applying the theoretical concept of pure grammaticalisation is not expedient. We therefore define the first scenario by a feature that is empirically more accessible, namely the incremental progression of grammaticalisation along a cline. In line with previous research on the diachrony of similar degree modifiers in English (esp. Traugott 2008a,b), the grammaticalisation path applying to the German constructions is likely to be as follows: pre-partitive > partitive > quantifier > degree modifier (e.g. *the bite of an apple* > *bits of bread* > *a bit of work* > *a bit tired*).

In the second scenario, the micro-construction starts as a fixed idiomatic expression but expands and develops a productive slot. For example, it is conceivable that the degree modifier [*ein Tick* X] started out as a fairly fixed expression combining with a particular temporal adjective, as in *einen Tick schneller* 'a tick/bit faster' or *einen Tick zu spät* 'a tick/bit too late', which are conventional collocations and in which the noun *Tick* 'tick (of a clock)' retains a concrete timerelated meaning. Possibly, the expression has been gradually increasing its productivity through item-based expansion from one temporal adjective to others, to non-temporal adjectives and eventually to other word classes. This scenario differs from the first one in that it does not presuppose earlier stages with partitive and quantifier uses.

In the third hypothesised scenario, analogically driven grammaticalisation, a micro-construction directly joins an existing degree-modifier meso-construction, with firmly entrenched schema members serving as role models. As introduced in Section 2, it is this scenario that demonstrates the organisational force of paradigms-as-schemas most clearly. Whereas in the first scenario the process of paradigmatic integration is likely to become evident only relatively late in the grammaticalisation process (cf. stage IV in the model by Diewald & Smirnova 2012), paradigmatic integration is at the heart of the beginning of grammaticalisation scenarios that are chiefly determined by analogy.

Although the three scenarios are theoretically distinct, they are not entirely mutually exclusive. Mixed trajectories and influences are possible, and distinguishing the three scenarios on the basis of usage data is not straightforward. Still, there are some fairly reliable variables in usage that help disentangle the hypothesised scenarios. The following analyses focus on four parameters:


More than in the analogically driven grammaticalisation scenario or the scenario from fixed to productive expression, the isolated grammaticalisation of a micro-construction along a grammaticalisation path correlates with, and even depends on, increasingly high frequencies of use. The rationale behind this correlation is that high frequency – token frequency especially, but also cooccurrence and type frequency – appears to be a crucial causal force in independent/"pure" grammaticalisation, fuelling several underlying cognitive processes such as chunking, habituation, neuromotor automation and schema entrenchment (Bybee 2003, Krug 2003, Diessel & Hilpert 2016, Neels 2020). Thus, the scenario of independent grammaticalisation along a cline can be expected to differ from the other two scenarios with respect to diachronic token-frequency levels and, above all, with respect to the chronology of functional expansion. What makes the scenario of grammaticalisation by analogy empirically distinguishable from the scenario of a fixed expression turning productive is patterns of behavioural convergence between constructions and patterns of collocational range.

## **3.3** *Ein wenig* **and** *ein bisschen*

The quantifier/degree-modifier constructions *ein wenig* and *ein bisschen* can arguably be used largely interchangeably in present-day German, although the former variant – which is also the older one – may be seen as slightly more formal. *Ein wenig* has largely ousted the yet older variant *ein lützel*, which used to be fairly frequent in the Middle High German period<sup>2</sup> but had fallen out of use in most German dialects by the 17th century (see Neels & Hartmann 2018: 143).

*Ein bisschen* undergoes a development that bears striking similarities to the evolution of English *a bit* as outlined by Traugott (Traugott 2007, 2008a,b). Like its English equivalent, *bisschen* derives from the (diminutivised version of the)

<sup>2</sup>This study adopts the traditional periodisation of German language history: c. 750–1050 Old High German; 1050–1350 Middle High German; 1350–1650 Early New High German; 1650– today New High German.

### Jakob Neels & Stefan Hartmann

noun *Bissen* 'bite', and just like English *a bit*, it tends to combine with concrete nouns as in (6) at first before it comes to modify abstract nouns to an ever larger extent (7), as Neels & Hartmann (2018) show, taking data from the mid-17th to the mid-20th century into account.


While the construction tends to combine with nouns in the early stages of its development, the proportion of verbs modified by *ein bisschen* increases steadily (Neels & Hartmann 2018: 150), as in (8). A slight upward trend can also be seen for adjectives, as in (9), although the pattern is less clear here, especially due to data sparsity in the early time slices analysed.


One striking result of our previous corpus study is that the distributional characteristics of *ein bisschen* seem to align over time with those of the older *ein wenig* construction, especially with regard to relative frequencies of nouns, verbs and adjectives modified by the two constructions. Furthermore, we observed a diachronic shift in the variability of determiners in [det *bisschen* x] towards the increasingly fixed string *ein bisschen*. These changes rendered the *ein bisschen* construction structurally very similar to *ein wenig*, suggesting some analogical influence by the latter construction.

Other aspects of the diachrony of the *ein bisschen* construction, in contrast, are more indicative of a fairly independent grammaticalisation process. For one thing, the attested shift from concrete to abstract nouns and from noun modification to adjective and verb modification closely matches the steps on the cline

from partitive to quantifier to degree modifier. That is, the historical usage data examined in Neels & Hartmann (2018) support the conclusion that *ein bisschen* passed through the stages of the grammaticalisation path in successive order. Another relevant piece of evidence is the long-term token-frequency profile of *ein bisschen*. As plotted in Figure 2, the grammaticalisation of the *ein bisschen* construction is accompanied by a pronounced increase in absolute token frequency.<sup>3</sup> The written records are, however, likely to depict this increase with a considerable temporal delay owing to the fact that, until the mid-19th century at least, *ein bisschen* was evaluated as colloquial (see Tiefenbach 1987: 6).

Thus, for *ein bisschen*, the usage-based parameters introduced in Section 3.2 tentatively suggest a fairly typical case of grammaticalisation whereby one particular micro-construction incrementally emancipated from its lexical source through frequency effects – however, possibly with additional support from extant constructions acting as analogical models. Interestingly, the picture emerging for the younger German quantifier/degree-modifier constructions examined next is notably different.

## **3.4** *Ein Quäntchen, ein Tick, eine Idee*

The modifiers [*ein Quäntchen* X] (lit. 'a quantum (of) X'), [*ein Tick* X] (lit. 'a tick (of) X') and [*eine Idee* X] (lit. 'an idea (of) X') are clearly more constrained and much less frequent in contemporary usage than the *ein bisschen* and *ein wenig* constructions. Table 1 (page 280) provides a first overview of the usage patterns of *ein Quäntchen*, *ein Tick* and *eine Idee*. In the 1.5-billion-word corpus DeReKo-Tagged-C, none of the three constructions is attested much more frequently than about 1,500 times, which yields normalised token frequencies of merely about one occurrence per million words (cf. *ein bisschen* in Figure 2). The modifier [*eine Idee* X] is particularly infrequent, with 0.03 tokens per million words. Degree-modifier uses of the three expressions do exist, but their developments constitute cases of low-frequency grammaticalisation, a phenomenon that is somewhat problematic for frequency-effect explanations in the Bybeean tradition (Hoffmann 2004, Brems 2007, Neels 2020).

Accordingly, only few historical tokens of the three younger quantifier/degreemodifier constructions can be found in the smaller DTA database. The corpus

<sup>3</sup>The DWDS corpora are used here to visualise the increase in frequency as it is much larger than the database we used for Neels & Hartmann (2018); however, the findings based on our own data, covering the time span from c. 1650 to 1900, are very much in line with what Figure 2 shows (see Neels & Hartmann 2018: 148).

Figure 2: The normalised absolute token frequency of (*ein*) *bisschen* for the period from 1600 to the present; reproduced from the DWDS; based on aggregated data from the reference corpora and the newspaper corpora used by the DWDS (retrieved on 3 September 2020)


Table 1: Overview of the corpus data for the three younger constructions

searches returned no more than 7 tokens of [*eine Idee* X] and no quantifier or degree-modifier uses of *ein Tick* at all. As for *ein Quäntchen*, the historical picture is slightly different. The expression originated as a commercial weight, *Quentchen*, i.e. the fourth or fifth part of one lot.<sup>4</sup> About 90% of the 158 DTA tokens attested between 1756 and 1910 are clear instances of *Quäntchen/Quentchen* being used as a technical measure, as exemplified in (10); a few tokens are ambiguous, leaning towards more colloquial uses, as in (11).

(10) Die Kuhmilch enthält endlich an Ram zwanzig *Quentchen*, an fester Butter sechs *Quentchen*, an dichtem Käse drei Unzen, an eingedickter Wadikke zehn *Quentchens*

'Cow's milk ultimately contains 20 *Quäntchen* of cream, 6 *Quäntchen* of solid butter, 3 ounces of thick cheese, 10 *Quäntchen* of concentrated whey' (1756, DTA: Haller)

(11) Ein *Quentchen* Mutterwitz ist besser, als ein Zentner Schulwitz. 'One *Quäntchen*/A *bit* of mother wit is better than one centner of school wit.'

(1762, DTA: Rabener)

Through folk-etymological reanalysis, as the commercial weight of *Quentchen* became uncommon in the 20th century, the expression was interpreted as a diminutive form related to 'quantum'.<sup>5</sup> It survives as a lower-frequency quantifier and, to a certain extent, as degree modifier, as quantified below.

In addition to the total frequencies of *ein Quäntchen*, *ein Tick* and *eine Idee* in DeReKo-Tagged-C, Table 1 shows the distribution of parts of speech across the three constructions. This provides a first glimpse into their functional commonalities and relative differences in contemporary usage. Part of the information of Table 1 is condensed into Figure 3, which visualises the distribution of nouns, verbs and adjectives in the three constructions. Together, Table 1 and Figure 3 reveal that all three constructions exhibit clear preferences with regard to the parts of speech with which they combine. *Eine Idee* and *ein Tick* combine mainly with adjectives, more precisely with adjectives in the comparative or with the degree particle *zu* 'too'. *Ein Quäntchen*, on the other hand, shows a strong preference for nouns, while also combining with graded adjectives and other parts of speech from time to time. Thus, *ein Quäntchen* serves primarily as a quantifier whereas *ein Tick* and *eine Idee* are used mostly for degree-modifying purposes.

<sup>4</sup>Compare Latin *quintus* 'the fifth'.

<sup>5</sup>Note that the spelling was changed from *Quentchen* to *Quäntchen* in the German spelling reform in 1996. This is why most of the cited examples are spelled with an 〈e〉 instead of an 〈ä〉.

Figure 3: Distribution of nouns, adjectives, and verbs in the three younger constructions

At a more fine-grained level of analysis, Table 2 (page 284) lists the most frequent lexical items occurring in each of the constructions. Two things become immediately clear in these lists of collocates. Firstly, there is a considerable amount of overlap between the lexical items that preferentially occur in the three constructions. For example, *mehr* is a top-three collexeme in each construction, including the *ein Quäntchen* construction despite the fact that this construction is dispreferred as a degree modifier. Secondly, their use seems to be largely constrained to a relatively small set of lexemes, which sets them apart from the *ein bisschen* and *ein wenig* constructions discussed above.

Some collocations are particularly idiomatic and dominant in current usage. Most notably, combinations with the noun *Glück* 'luck' make up about 80% of all *ein Quäntchen* tokens found in DeReKo-Tagged-C, and *Glück* is more than 25 times more frequent than the second-ranked collexeme. Also note that, in line with the prototypical phrase *ein Quäntchen Glück*, the other top collexemes of *ein Quäntchen* listed in Table 2 are nouns that have positive connotations as well. This specific collocational pattern is not to be expected solely on the basis of the construction's past as a commercial weight. Based on this origin, one might expect an early grammaticalisation trajectory that involves high relative frequencies of (mass) nouns in general, but not necessarily the dominance of one particular noun. However, it seems that, to a large extent, the construction owes its present-day existence as a low- to mid-productive quantifier to the strongly entrenched collocation *ein Quäntchen Glück*. Conceivably, the low-frequency [*ein Quäntchen* X] construction also receives some analogical support through the firmly established *ein bisschen* construction, with which *ein Quäntchen* shares associative links thanks to similarities such as the diminutive suffix *-chen*. Overall, however, the most informative parameter for the diachrony of *ein Quäntchen* appears to be the parameter of collocational range. It strongly points to the second grammaticalisation scenario introduced in Section 3.2, i.e. a process whereby a fixed idiomatic expression develops an open slot with a certain degree of productivity.

Contemporary corpus data demonstrates that the *ein Quäntchen* construction is barely entering more advanced stages of grammaticalisation in which it expands from quantifier (95%) to degree-modifier uses (5%), whereas *ein Tick* and *eine Idee* are first and foremost degree modifiers and only rarely used as quantifiers. Their usage profiles thus yield the somewhat strange picture that *ein Tick* and *eine Idee*, as it were, skipped the presumably less grammaticalised quantifier stage on the grammaticalisation path presented in Section 3.2 (pre-partitive > partitive > quantifier > degree modifier). Interestingly, De Clerck & Brems (2016), studying similar English constructions with size nouns (e.g. *mass*(*es*), *heap*(*s*),


Table 2: Top 20 modified items for each construction (without hapax legomena).

*bunch*), also observe trajectories with unattested (supposedly) intermediate grammaticalisation stages. The parameter we labelled "chronology of functional expansion" suggests that the degree modifiers *ein Tick* and *eine Idee* have not evolved through independent grammaticalisation processes. Given the remarkably low absolute token frequencies of the two constructions, also the frequency parameter speaks against the scenario of independent grammaticalisation. Moreover, unlike in the case of *ein Quäntchen*, the analysis on collocational range has not revealed any highly dominant collexemes for [*ein Tick* X] or [*eine Idee* X]. This makes the grammaticalisation scenario from fixed to productive expression seem less likely for these two constructions. Both constructions, we argue below, owe their existence or emergence largely to pre-existing templates in the modern German network of quantifier/degree-modifier constructions.

Diachronic corpus data suggest that *ein Tick* and *eine Idee* gained some ground only around 1900. [*eine Idee* X] presumably emerged in the 19th century. The 7 tokens extracted from the DTA stem from that time; one of them is reproduced in (12).

(12) Ceara: Eine der Maranham sehr ähnliche Baumwolle, vielleicht sogar *eine Idee* besser.

'Ceará: a type of cotton very similar to Maranhão cotton, possibly even *a bit* (lit.: *idea*) better.'

(1889, DTA: Justi)

With not a single token in the DTA data, the modifier [*ein Tick* X] seems to be even younger than *eine Idee*.

The development of *eine Idee*, *ein Tick* or any other German quantifier/degreemodifier construction grammaticalising during the 20th century must be understood in the context of a variety of co-existing near-synonymous constructions. Not only are there the three other constructions analysed in the present study, i.e. *ein wenig*, *ein bisschen* and *ein Quäntchen*, but the constructional family comprises several more lower-frequency members, two of which are exemplified in (13) and (14).


### Jakob Neels & Stefan Hartmann

By the end of the 19th century at the very latest (but probably much earlier; recall the observations on the frequency profile and colloquial status of *ein bisschen* in Section 3.3), constructional network constellations had developed that assist the grammaticalisation of novel quantifier/degree-modifier constructions. Multiple micro-constructions, above all the highly grammaticalised *ein bisschen* and *ein wenig*, had entered into a paradigmatic relation. As discussed in Section 2, the coexistence of multiple formally and functionally similar expressions, some of which being highly frequent, assist the recognition and entrenchment of an overarching mid-level schema. In the present case, this schema or mesoconstruction features an emergent slot with the feature 'small/minor unit': [*ein* small/minor-unit x]. Such a meso-construction, together with some frequent members, can be conceived of as an attractor set for the development of novel micro-constructions. According to this line of reasoning, *ein Tick* and *eine Idee* were attracted to the existing, increasingly strengthened quantifier/degree-modifier meso-construction in a scenario of analogically driven grammaticalisation. This account is empirically supported above all by the observation that the diachronic context expansion of *ein Tick* and *eine Idee* did not proceed along a multi-stage grammaticalisation path but rather resembles an instant recruitment as degree modifiers.

The notion of a German quantifier/degree-modifier meso-construction, however, needs some refinement to do justice to the usage profiles of its hypothesised members. This meso-construction should not be conceived of as a single homogeneous mid-level schema but rather as an assembly with multiple subschemas. For one thing, not all associated micro-constructions are equally productive as both quantifiers and degree modifiers. As shown in the corpus analysis above, degree-modifier uses of *ein Quäntchen* are uncommon, while quantifier uses have low relative frequencies in the usage profiles of *ein Tick* and *eine Idee*. What is more, even within one of the two functional domains, individual micro-constructions may pattern more locally. For instance, the degree-modifier constructs of *ein Tick* and *eine Idee* in the corpus have been found to combine overwhelmingly with graded adjectives. Such construction-specific differences make a unified account seem somewhat problematic. Clearly, each micro-construction has its own unique properties and history. However, it is important to stress that there are certain dominant micro-constructions that freely participate in most or all quantifier and degree-modifier subschemas. The degree-modifier subschema with graded adjectives, which is central to *ein Tick* and *eine Idee*, is also served by *ein bisschen* (15) and *ein wenig* (16) with a substantial (absolute and relative) token frequencies.


Highly productive members such as *ein bisschen* and *ein wenig* can be expected to serve as prototypes of the constructional family. Other, less productive members with more specific constraints are associated with the prototype, and they are associated with each other through family resemblance (see Traugott 2008b: 33 for a similar account of English [n *of* np] patterns).

There is not only a great formal similarity between the family members but also a considerable functional overlap regarding the quantifier and degree-modifier subschemas served by them. This hypothesised network constellation with interrelated quantifier and degree-modifier subschemas at the meso-level can be argued to promote the analogical recruitment of novel micro-constructions as well as the analogical realignment of existing ones, turning some quantifiers into degree modifiers and vice versa. Without the support of a quantifier/degreemodifier meso-construction and subschemas linking extant micro-constructions, the observed emergence of a variety of lower-frequency degree modifiers and quantifiers would be hard to account for.

## **4 Schematisation and paradigmaticisation in morphosyntax**

The grammaticalisation of individual micro-constructions and constructional families involves schematisation in multiple ways. In micro-constructions, shifts towards grammatical functions translate into increasing schematicity not only at the level of constructional semantics but also at the level of the constructional slots as reflected by the types of collexemes entering these slots (cf. Perek 2020). In other words, the formerly more contentful, referential meaning of a grammaticalising micro-construction becomes more abstract and more schematic in this regard, and the slots can be understood as schematic categories constraining the host-class (Himmelmann 2004) of the construction. In the present case study, it has been illustrated that the slots of less grammaticalised quantifier/degreemodifier constructions are less schematic and more constrained. The [*ein Quäntchen* X] construction, for example, seems to be structured around one prototypical exemplar, *ein Quäntchen Glück*. The slot of the highly grammaticalised [*ein*

### Jakob Neels & Stefan Hartmann

*bisschen* X], in contrast, has hardly any semantic constraints and hosts virtually all parts of speech. Another dimension of schematisation is the formation and increasing entrenchment of higher-level schemas, which we refer to as meso-constructions (and macro-constructions) and which emerge as generalisations over multiple structurally similar micro-constructions. At high levels of abstraction, such as a potential macro-construction for German quantifiers and/or degreemodifiers, the micro-constructions studied here might be linked to formally dissimilar, but functionally related constructions such as the single-word modifiers *etwas* 'somewhat', *leicht* 'slightly', *sehr* 'very' and *viel* 'much'. However, there appears to be little empirical ground for the postulation of such a highly abstract link as a potential source of mutual influence in the development of such formally dissimilar micro-constructions. There is more reason to argue that change in the early grammaticalisation and paradigmaticisation of syntactic constructions is guided mainly by associations to micro-constructions and lower-level meso-constructions.

There are signs of analogical and thus paradigmatic forces, but this influence is limited in that the micro-constructions in our case study each have their unique diachronic trajectories and retain idiosyncratic properties. For instance, although the degree modifiers *ein Tick* and *eine Idee*, skipping the quantifier stage expected in a scenario of independent grammaticalisation, appear to have evolved through analogy, their use has been shown to be largely restricted to one specific subtype of degree modification, namely combinations with graded adjectives. Possibly, such construction-specific properties are effects of persistence (Hopper 1991) based on features of the original lexical source construction. They can, however, also be thought of as subschemas at the meso-level if there is a subset of two or more micro-constructions within the family that share a particular distribution. In our case study, this applies to *ein Tick* and *eine Idee*, and this also applies to *ein bisschen* and *ein wenig* to the extent that a considerable proportion of their uses match the hypothesised degree-modifier subschema for graded adjectives. From this perspective, the observation that some grammaticalising micro-constructions pattern more locally in usage than might be expected need not always result from micro-level persistence (or competition) but may be linked to meso-constructional subschemas (cf. Langacker 2000: 29). The lower the degree of schematicity, the more accessible a template tends to be – that is what a lot of usage-based research suggests (cf. Section 2). Accordingly, novel micro-constructions in analogically driven grammaticalisation are likely to be most strongly attracted to the lowest-level schemas entrenched at the meso-level.

Families of grammatical(ising) constructions form a layered (Hopper 1991) domain of grammar shaped by diverse, partly opposing forces. Older and younger

micro-constructions both support and compete with each other (cf. Delorge et al. 2014, De Smet & Fischer 2017, De Smet et al. 2018). The present case study has focused on the role of support, but the aspect of competition in the German quantifier/degree-modifier constructions is also evident, for example in the aforementioned disappearance of the old *ein lützel* construction accompanying the rise of *ein wenig*. Another possible outcome of competition is division of labour and thereby the specialisation of micro-constructions, which might be the fate of some of the lower-frequency expressions studied here (e.g. *ein Quäntchen*, *ein Tick*). At the same time, analogical forces cause micro-constructions to converge with respect to certain formal and/or functional features, creating more homogenous constructional families. It is remarkable that this convergence and growth has been occurring in the German constructional family under scrutiny, when considering that some of the micro-constructions originated from fairly diverse lexical sources, including concrete nouns like *Biss*(*chen*) 'bite', abstract nouns like *Idee* 'idea', a commercial weight in the case of *Quentchen/Quäntchen*, and even an adjective in the case of *wenig* 'minor, small'. The observed convergence and growth seem to be caused, above all, by the pervasive process of analogical thinking and pattern matching, i.e. the establishment of form–function associations across constructions (e.g. Fischer 2011, Traugott & Trousdale 2013: esp. Section 1.6.4.2). Collectively, the diachronic trajectories of the micro-constructions examined here amount to a macro-process that can be thought of as (early) paradigmaticisation.

The macro-process of paradigmaticisation is a long-term effect of numerous (low-level) constructional changes whose directionality is constrained above all by the domain-general cognitive processes of analogy and automation. Early and advanced paradigmaticisation seem to involve different kinds of paradigmsas-schemas and different degrees of automation typically linked to the transition from syntax to morphology. In early paradigmaticisation, as typical of sets of fairly heterogeneous periphrastic constructions in syntax, the parts of the network seized by this process are micro-constructions as well as lower-level schemas at the meso-level. As long as the members of the constructional family exhibit rather heterogeneous properties, high-level generalisations such as macro-constructions can be expected to be only weakly present (if at all) in speakers' minds. This stage of paradigmaticisation involves emergent paradigms that are still relatively open; micro-constructions that are part of such emergent paradigms may be quasi-synonymous, and their use is still optional (cf. Diewald & Smirnova 2012). This phase is witnessed in the case of the quantifier/degreemodifier constructions in present-day German. In more advanced stages of paradigmaticisation, the relations between the family members become more rigid.

### Jakob Neels & Stefan Hartmann

The members tend to be formally more homogeneous, but quasi-synonymous uses have given way to contrastive meanings; and the paradigms are essentially closed. This stage of paradigmaticisation corresponds to degrees of grammaticalisation that are prototypically associated with (inflectional) morphology rather than periphrastic syntactic constructions. It seems that only as a constructional family approaches such late stages in its life cycle do the relations of its members become defined by one organisational unit at a very high level of abstraction.

Grammatical(ising) micro-constructions are not teleologically determined to become more homogeneous and meet in paradigms; still, this direction of change is highly likely. Closing this section, we summarise what we argue to be the reasons for this long-term trend. One causal factor is how the domain-general cognitive process of analogy shapes language structure (e.g. Anttila 2003). The present paper has discussed the constructional network constellations – "attractor sets" involving meso-constructions – that promote the analogical emergence and/or change of micro-constructions. These analogical forces are certainly influential, but even beyond analogy as a cognitive basis of paradigmatic relations there are forces that lead grammaticalising micro-construction in similar directions, making them more alike as well. These directions are the well-known tendencies of semantic generalisation and morphosyntactic fixing and reduction. Cognitive mechanisms and motivations underlying these strongly directional changes in grammaticalisation include chunking, increasing ease of retrievability, habituation, neuromotor practice and frequency-induced predictability among others. Many of the underlying processes can be understood as concomitants of the even more general cognitive process of automation (Bybee 2010, Neels 2020). The asymmetry inherent in automated versus more controlled activity should be considered a cognitive key factor accounting for the unidirectionality of grammaticalisation: since highly automated grammatical operators are withdrawn from conscious control, they are very unlikely to be manipulated for more contentful, referential – i.e. more "lexical" – purposes (cf. Lehmann 2017, Haspelmath 1999). Automation serves efficiency; and, as pointed out in Section 2, grammatical paradigms represent efficient solutions in processing. In short, at the general level of language-related cognition and performance, paradigm formation is governed by efficiency, automation and analogy; at the level of linguistic representations, the locus of change can be modelled as clusters of interrelated schemas, above all micro- and meso-constructions.

## **5 Conclusion**

This paper has investigated a family of German quantifier/degree-modifier constructions against the theoretical background of construction grammar and in the light of research on grammaticalisation. Regarding the theoretical background, construction grammar has been shown to have a somewhat ambivalent relationship to paradigms: some proponents of constructionist approaches to language have rejected the concept, others have suggested that it might be theoretically useful for capturing horizontal relations in the constructicon, especially in the domain of mid-level constructions. Following up on a previous study showing that the development of *ein bisschen* and *ein wenig* bears many similarities to the development of English *a bit* in the sense of a gradual extension from more concrete to more abstract readings and from quantifier to degree-modifier uses, we have investigated the low-frequency constructions [*eine Idee* X], [*ein Tick* X], and [*ein Quäntchen* X] in contemporary German. The overall picture emerging from the corpus analyses is that the older quantifier/degree-modifier constructions served as attractor sets for an increasingly strengthened mid-level schema, i.e. meso-construction, thereby promoting the analogically driven grammaticalisation of younger micro-constructions. This scenario is supported by the finding that younger micro-constructions (i) approximate the usage patterns of more established ones, and (ii) "skip" stages on the cross-linguistically attested grammaticalisation path pre-partitive > partitive > quantifier > degree modifier. For modelling such linguistic developments, the constructionist idea of grammaticalising expressions being linked via meso-constructions has proved fruitful. Recognising types of grammaticalisation that are enabled by meso-constructions and analogy – as opposed to independent/"pure" grammaticalisation (Lehmann 2004) – moreover helps reconcile frequency-effect approaches with the phenomenon of low-frequency grammaticalisation. The infrequent modifiers *ein Tick* and *eine Idee* seem to owe their existence to an established template, but they also exhibit constraints that cannot be derived solely from a single overarching mid-level schema. Such properties specific to individual micro-constructions or subsets of constructions may be interpreted not just as persistence but as entrenched subschemas or local emergent paradigms. Early paradigmaticisation has been argued to involve emergent paradigms of multiple orders in the sense of constructions at multiple levels of schematicity. Change is guided mainly by associations to micro-constructions and lower-level meso-constructions. Only in advanced stages of grammaticalisation and paradigmaticisation, when microconstructions become sufficiently homogeneous, do higher-level meso-constructions and macro-constructions – which come closest to paradigms in the traditional understanding of the term – act as decisive organisational forces.

Jakob Neels & Stefan Hartmann

## **Data availability**

The annotated raw data are available at https://osf.io/t9j7p/.

## **Acknowledgements**

The authors would like to thank the editors of the present volume as well as the anonymous reviewers for helpful comments and suggestions. The usual disclaimers apply.

## **References**



### Jakob Neels & Stefan Hartmann



### Jakob Neels & Stefan Hartmann


## **Chapter 10**

## **Generics as a paradigm: A corpus-based study of Norwegian**

## Anna Kurek-Przybilski

Adam Mickiewicz University Poznań

This paper examines generics as a part of human cognition, rooted in speakers' language knowledge in form of a grammatical paradigm. The paradigm proposed in this paper is based on a broader understanding of the notion, as presented in Diewald (2009) and Diewald & Smirnova (2010) among others. The matrix of the noun forms used in Norwegian in generic contexts is based on the models proposed by Radden & Dirven (2007), Radden (2009) and Pettersson (1976). The models of generics were adjusted to the data from the Norwegian language, collected in this corpus-based study.

## **1 Introduction**

For many years, a *paradigm* was only associated with inflectional paradigms and morphology. However, a discussion on the matter in recent years has resulted in a new view on paradigms, a view that allows for a much broader understanding of the notion and the one that exceeds the branch of morphology.

The core of this paper is based on this new approach to grammatical paradigms which can be applied in examining different language phenomena. In this introductory section, the following questions will be addressed:


### Anna Kurek-Przybilski

In scholarly literature, one finds numerous mentions of paradigms, most of them understood as morphological systems. An example of that can be found for instance in Ackerman et al. (2009: 55), where paradigms are presented in a classic understanding of the term, namely as inflectional paradigms.

However, in this paper a different approach will be taken into account. Diewald (2009) and Nørgård-Sørensen et al. (2011) present a new view on paradigms, which will be discussed and utilised in this study.

In her discussion, Diewald (2009: 445) interprets paradigms as a representation of particular construction types. She points out that certain paradigms are obligatory, such as inflectional paradigms, whereas others are not. Genericity falls certainly in the second category – it can be expressed in many different ways and even though certain NP types occur in certain generic contexts, they are not limited to only such contexts.

Such an understanding of the notion allows for a claim that certain language phenomena, such as genericity, can in fact be rooted in speakers' language knowledge and consist of a well-organised system of available forms. One can therefore differentiate between the different forms and the interpretations they imply. Nevertheless, the obligatoriness of such a paradigm remains questionable.

Paradigms understood in a broader sense were also discussed by Nørgård-Sørensen et al. (2011). An example of Danish verbs that can be construed in different ways was provided by Nørgård-Sørensen et al. (2011: 72–73). For instance, the verb *skyde* 'to shoot' can occur with a direct object or with a preposition *på* 'at'. The same mechanism can be observed in the case of generics, especially when it comes to bare nouns (BNs) occurring in generic contexts. In Norwegian (and Swedish, which will be discussed later), where BNs are common, certain verbs can require BNs:


The verb 'to be' in (1) requires a BN because professions, nationalities and religious beliefs are expressed without any article in Norwegian (cf. Swedish as discussed by Pettersson 1976). This can be considered an obligatory paradigm that follows certain grammar rules. In contrast to this, example (2) shows a paradigmatic use of a BN with the verb 'to have'. Here the BN functions as a concept

### 10 Generics as a paradigm: A corpus-based study of Norwegian

of a dog, not a certain dog or a dog as the whole species. However, it is possible to say *å ha en hund* 'to have a dog', which does not change the meaning of the phrase. The examples illustrates therefore that certain paradigms, including the paradigm of genericity, are optional but nevertheless they structure the language.

Another notion connected to paradigms understood in a broader sense is paradigmaticity. According to Diewald (2009: 447), paradigmaticity is an essential property that distinguishes grammatical items from lexical ones. This view can also be found in Diewald & Smirnova (2010), where paradigmaticity and obligatoriness are perceived as vital elements of the grammaticalisation process, based on the model of Lehmann.

In his seminal work on grammaticalisation, Lehmann presents parameters of grammaticalisation. One of those parameters is paradigmaticity.

What is meant here by paradigmatic cohesion or paradigmaticity is the formal and semantic integration both of a paradigm as a whole and of a single subcategory into the paradigm of its generic category. This requires that the members of the paradigm be linked to each other by clear-cut paradigmatic relations, especially opposition and complementarity. (Lehmann 2015: 141)

The main requirement when it comes to Lehmann's model is the clear-cut relation between the elements of a paradigm. Complementarity or opposition might not seem as clear-cut in case of generics and different noun forms used with generic NPs. However, analysing generics in texts where a broader context was provided, proved that certain noun forms complement each other in some contexts or exclude each other in other contexts.

The theoretical framework used in this paper is based on Diewald (2009), Diewald & Smirnova (2010) and Lehmann (2015). The approaches to grammatical paradigms presented in those studies will be combined with empirical data of Norwegian in order to evaluate the existing paradigms of generics proposed by Radden & Dirven (2007), Radden (2009) and Pettersson (1976). The models will be discussed in greater detail in the following section.

Genericity is a language phenomenon that is present in every language studied to date Behrens (2000, 2005). However, there are no language devices used to express that phenomenon, despite the numerous theories on the matter (Liebesman 2011, Collins 2018). Certain researchers claim that a silent gen-operator exists (Carlson 1977, 1982 and Chierchia 1998 among others), whereas others opt for the so-called "simple view" on generics (Liebesman 2011) that does not take into

### Anna Kurek-Przybilski

account any operators and treats generics in the same way as quantified statements. Nevertheless, no empirical studies have shown the presence of a device that would be used solely to express genericity, even though numerous studies on the matter show that the phenomenon may be one of the language universals, present already in children's speech (Gelman & Tardif 1998, Leslie & Gelman 2012 among others).

In different studies concerning generics, the following claims on the paradigmatic nature of the phenomenon are made:


When it comes to the first claim, the fact that generics depend on one's cognition is closely connected to the paradigmatic view on the matter. Being able to interpret and produce generic NPs, sentences and texts means that certain aspects of the phenomenon are automatised – be it on the cognitive level or on the grammatical level.

Another claim is that genericity is a linguistic universal, as is proposed by Leslie (2007). Indeed, numerous studies confirm that children at a very early age are able to correctly interpret and produce generic statements (Gelman & Tardif 1998). This can also imply that generic knowledge is deeply rooted in one's cognition in the form of a paradigm (Leslie et al. 2011).

In the cognitive literature on the matter, one finds numerous studies with different models of genericity. For instance, Leslie et al. (2011) present a model of generic predications and types of references that may be construed with them in English. The model is based on the truth-value of the predications and can be applied when analysing generic sentences.

As has been mentioned, the models of generics that will be utilised in this study and modified according to the data, are those proposed by Radden & Dirven (2007), Radden (2009) and Pettersson (1976). The first two models are based on English where four noun forms can be used to express genericity, namely indefinite singular, indefinite plural, definite singular and definite plural. The noun forms and their generic meanings according to Radden's model are depicted in Table 1.

All types of genericity are illustrated by Radden (2009: 224) as follows:

### 10 Generics as a paradigm: A corpus-based study of Norwegian


Table 1: Types of generic reference (Table 2 in Radden 2009: 224.)


In the example (3), a member of the kind represents the whole species, indicating that having a bushy tail is a characteristic feature of most lions. Proportional generic is depicted with example (4) and it concerns prototypical members of the kind. Namely, most hedgehogs are shy and therefore the feature can be connected to the whole kind. In example (5) a reference to a kind is expressed with the use of indefinite singular NP. Kind-reference rarely allows for exceptions, therefore one may assume that a great majority of tigers hunts by night. The last category of generic expressions in English, namely delimited generic, is limited to human groups only, as can be seen in example (6). The definite plural generic in English cannot be used with other nouns, whereas in Norwegian this NP type occurs in many more generic contexts as will be discussed later in this paper.

A graphic representation of English generics is presented in Figure 1.

The generic noun forms, both definite and indefinite, singular and plural, have certain functions and interpretations assigned to them. For instance, indefinite singular is used when referring to a single member of a given group in order to state a generalisation about the whole species. A similar use is assigned to definite singular which is used mostly as a kind-reference. The two plural forms – indefinite and definite – also have their typical uses according to the proposed paradigm. Indefinite plurals are used when a certain proportion of members in

### Anna Kurek-Przybilski

Figure 1: Types of generic reference (Table 5.4 in Radden & Dirven 2007: 111).

a group (the majority) share a given feature, whereas the definite plural form is used only when talking about groups of people.

Both models proposed by Radden & Dirven (2007) and Radden (2009) are constructed according to the restrictions mentioned by Lehmann (2015). Namely, the functions of generic forms are complementary and/or opposite to each other. This results also in relatively low interchangeability of the forms. For instance, delimited generics used solely when talking about people in English cannot be utilised as kind generic, and so on.

In both models, bare nouns are not taken into consideration as they are not very common in generic contexts in English. The only situation when BNs are used generically is with mass nouns. However, generic use of this noun type in Norwegian (and Swedish as will be presented) is a lot more frequent.

Let us now consider the paradigm of Swedish generics proposed by Pettersson (1976) in Figure 2, <sup>1</sup> where BNs occur as one of the available generic forms.

The author proposes a number of possible readings of noun forms that can occur with generics. Such references are divided according to the features "limited +/−" and "specific +/−", according to which non-generic nouns are always specific (Pettersson 1976: 121).

Pettersson provides also a number of examples that illustrate generics presented in his model:

<sup>1</sup>My translation into English. For the original graph in Swedish the reader is referred to the mentioned paper (Pettersson 1976: 121).

10 Generics as a paradigm: A corpus-based study of Norwegian

Figure 2: Reference types in Swedish according to Pettersson (1976: 121)


All of the sentences above are generic, even though they include two NP types – singular indefinite and singular definite. The examples in (8) and (10) illustrate representative generic, where a single instance represents the kind, namely 'a dog' and 'a boy' have some features that can be assigned to the whole group (most/all dogs chase cats and none/almost none boys cry).

### Anna Kurek-Przybilski

The examples (9), (10) and (11), where definite singular NPs are used, can be seen as *classic* kind references, where a prototypical instance represents the whole kind. One can assume that the prototypical wren shares a certain behaviour of all wrens, just as the prototypical dog does and the prototypical politician supporting the Social Democratic Party.

What is more, Pettersson's model allows for a bare form in non-specific contexts. Generic references are indeed considered non-specific, as they are generalisations, not references to particular instances of a given kind (cf. Lyons 1977).

The examples of a non-specific use of bare nouns can be seen in the sentences below (Pettersson 1976: 127):


Professions, nationalities and religious functions are used without articles in Swedish (as well as in Norwegian). The reason for this, as Pettersson explains, is that they describe general concepts. A similar statement can be said about bare generics in Norwegian, as in example (14).

(14) Ola Ola vil will kjøpe buy hus. house 'Ola will buy a house.'

The NP 'house' is rather a concept than a particular house that Ola wants to purchase. The conceptual BNs in Norwegian are neither singular nor plural – they are neutral when it comes to number, as well as definiteness (cf. Halmøy 2016). They can also be used in generic contexts as will be shown later in the paper.

For the purpose of this study, all of the models presented above will be used to identify different types of generics in order to create a paradigm of the phenomenon in Norwegian. The following section focuses on the material used for the study, in Sections 3 and 4 the results of the analysis will be presented, whereas Section 5 concerns implications for further research in the field.

10 Generics as a paradigm: A corpus-based study of Norwegian

## **2 Material**

The study presented in this paper is based on a tailor-made corpus of generic texts. 170 texts (27 761 tokens in total) were retrieved from an online encyclopaedia *Store norske leksikon*. The encyclopaedia contains texts written both in *bokmål* and *nynorsk*, the two written standards of Norwegian. In this study, only texts written in *bokmål* are analysed.<sup>2</sup>

There were many reasons for the choice of encyclopaedic texts. First of all, generic NPs and sentences are not annotated in any of the available corpora of Norwegian so the phenomenon would have to be tagged manually anyway. Second of all, tagging genericity manually means that the texts chosen for the analysis must contain relatively many references of this sort in order for the material to be sufficient. This might not be the case should newspaper articles or literary texts be analysed, where the number of generic references is relatively low.

Finally, genericity is a context dependent phenomenon and it can occur both at the NP level, sentence level and text level (Behrens 2005). Therefore, choosing the texts that are primarily generic makes the analysis possible, also when it comes to the available resources.

The texts chosen for the study are at least one paragraph long and contain at least one generic reference. The texts are divided into five thematic categories:


Below follow the example generic sentences from each of the categories:

(15) Vitnet witness-def skal shall som as utgangspunkt rule gi give forklaringen explanation-def umiddelbart immediately for for den the dømmende sentencing rett. court

'A witness shall, as a rule, provide an explanation immediately to the sentencing court.'

<sup>2</sup>The reason for such a choice is that the *bokmål* language variant is used by the majority of the population (87.7% according to a study published by Gunnerud 2017).

### Anna Kurek-Przybilski


The category 'people' contains 20 texts, the categories 'animals', 'plants' and 'tools' contain 25 texts each, whereas the category 'other' consists of 75 texts. Differences in the number of texts in some of the categories is due to a few reasons. When it comes to the category 'people', finding a description of an ethnic group, a nationality or an official function that is written in *bokmål* and has the length of at least one paragraph, is a challenge. However, the difference of 5 texts in relation to the categories 'animals', 'plants' and 'tools' is not significant in terms of the results.

The category 'other' is the biggest of the whole corpus and it contains most diverse NP types. There are countable nouns and mass nouns, descriptions of objects, abstract notions, professions<sup>3</sup> , elements of the landscape and many others. Since the category contains so many different nouns that do not fit to any of the other categories, it also consists of many more texts. This way numerous generic contexts are taken into account without the need to divide the corpus into many separate subcategories.

The choice of source texts for the analysis is subjective and its main purpose is to investigate the use of noun forms with different types of nouns – animate, inanimate, countable and mass, as well as those recognised as well-established kinds (WEKs) and familiar nouns (c.f. Carlson & Pelletier 1995 and Borthen 2007 among others).

The example sentences with each of those NP types can be found below ((20) – animate, (21) – inanimate and countable, (22) – mass noun and (23) – WEK):

<sup>3</sup>Descriptions of professions are recognised as different from descriptions of groups of people. Professions are perceived more as concepts or a type of job, not people who perform a given task. Therefore the NPs of this sort are classified in the category 'other' and not 'people'. Official functions however, such as 'king', 'priest' etc. are classified in the category 'people'.

10 Generics as a paradigm: A corpus-based study of Norwegian


'Ink consisted originally of soot or dyes from minerals, plants and animals.'

(23) Stavkirke stave.church-∅ er is en a høyt highly utviklet developed kirketype,. church.type 'A stave church is a highly developed church type.'

All texts in the corpus were tagged manually for genericity and noun forms of the NPs in question. The material was annotated and analysed with the use of R software (version 3.6.1). A random sample of annotated texts was proofread by a native speaker of Norwegian in order to eliminate potential mistakes or misinterpretations of generic references. However, due to the fact that all texts in the corpus are generic, annotating genericity was a fairly simple task and no mistakes were found in the randomly generated sample.

## **3 General results**

In the Norwegian nominal system, one finds five NP types, namely: BN (*hund* 'dog'), indefinite singular (*en hund* 'a dog') and plural (*hunder* 'dogs'), as well as definite singular (*hunden* 'the dog') and plural (*hundene* 'the dogs'). Among the five NP types, BNs seem to have the least stable status. In recent years, it has been discussed whether BNs are marked for number and definiteness. According to one view, they can be considered indefinite and singular (Borthen 2003), whereas others claim that BNs are neutral in terms of definiteness and number (cf. Halmøy 2016). As will be presented in this section, the second view seems to be more plausible. BNs are not considered singular or plural, neither are they considered (in)definite.

All nouns tagged as generic were counted and summed up in Table 2.

### Anna Kurek-Przybilski


Table 2: Generic noun forms in the corpus – general results

As can be seen, the three most frequently occurring noun forms include bare nouns, indefinite plural and definite singular. The two other forms, namely indefinite singular and definite plural, were used less frequently.

The main tendencies for the use of different noun forms already show that certain of them are used in many more contexts than other ones. For instance, BNs were used in numerous examples throughout all corpus texts, both as first mentions and as subsequent mentions. What is worth pointing out is that first mentions in many of the corpus texts were both encyclopaedic entries and subjects of the first sentences. This may be one of the reasons for such a frequent use of BNs. However, in many of the texts BNs appeared also as subsequent mentions, following often other noun forms used earlier in the text.

The second most often used noun form was indefinite plural, seen often as *default* when it comes to expressing genericity in Germanic languages (cf. Carlson 1977, Carlson & Pelletier 1995, Chierchia 1998, Pettersson 1976 among others). The collected data proves that this claim holds also for Norwegian (see Section 3.4).

Another aspect that was also observed in the corpus, was the position of generic NPs in the corpus texts. The results are shown in Table 3.

Table 3: Generic noun forms – position in a sentence


The position of generic nouns in the corpus did not influence the choice of different noun forms, as there were almost as many initial NPs as there were non-initial ones.

### 10 Generics as a paradigm: A corpus-based study of Norwegian

What is more, the functions of the generic NPs were annotated. This way it was possible to see the grammatical roles the nouns had and whether there were some main tendencies. Indeed, as can be seen in Table 4, in the majority of generic sentences, the NPs in question occurred as subjects. Many generic sentences given as examples of genericity in the scholarly literature have a similar structure – generic nouns are much more often subjects than objects. This tendency was observed also in the corpus. Other grammatical functions, such as genitive modifiers, were much less frequent.

Table 4: Generic noun forms – function in a sentence


In this paper, questions concerning NP's position and function in generic sentences will not be considered. They are not central to the issue of paradigms and, as has been argued above, they do not seem to influence the use of different noun forms in generic texts. In the following subsections, all five noun forms will be discussed with reference to genericity in Norwegian.

## **3.1 Bare nouns**

Bare nouns used generically with countable nouns were frequent in the corpus and their use was often conceptual, as can be seen in example (24).

(24) I In middelalderen Middle.Ages var was det there badstue sauna-∅ i in alle all byer. cities 'In the Middle Ages there were saunas in all cities'.

Some researchers consider BNs to be marked for number and closely connected to singular nouns (see e.g. Borthen 2003), whereas others claim that the noun form is neutral in terms of number and definiteness (Halmøy 2016).

As has been mentioned before, the second interpretation is utilised in this study. BNs, such as *badstue* in the example above, cannot be interpreted as a specific sauna, rather a concept of it. Similarly as in example (14) mentioned

### Anna Kurek-Przybilski

before, most of the BNs from the corpus represent a concept and are therefore neutral when it comes to number and definiteness.

The noun 'sauna' in (24) does not refer to any particular sauna but rather to the concept of it. The BN used in the example is not marked for number or definiteness and it is therefore difficult to match that use to any of the generic categories from the models of Radden & Dirven (2007) and Radden (2009). However, the graph proposed by Pettersson (1976) accounts for such a context, depicting it as a non-specific, non-limited generic.

Another conceptual use of BN can be observed in example (25), where a profession is described with the use of this noun form.

(25) Psykolog psychologist er is en a person person med with utdanning education i in psykologi. psychology 'A psychologist is a person with education in psychology.'

In English, a similar example can be rendered with indefinite singular to suggest that any given psychologist needs to have education in the field in order to perform the profession. In Norwegian however, the reference is expressed with a bare noun, making the profession more of a concept than an actual person working in the field.

The bare noun form, most widely used throughout all corpus texts, appeared also in sentences with WEKs and familiar nouns, namely nouns that are recognised as well-known in a given language and/or culture, as can be observed in example (26).

(26) Stavkirke stave.church er is en a høyt highly utviklet develop.pst kirketype church.type (...). 'A stave church is a highly developed type of church (...).'

Again, the reference to a stave church is conceptual as the BN used in the sentence above does not suggest number or definiteness. What is more, familiar nouns are often used as BNs and/or definite nouns, which suggests that they function in a way as proper names (cf. Carlson & Pelletier 1995).

## **3.2 Indefinite singular**

Indefinite singulars were very rarely used in the corpus and in the two of the categories, namely 'animals' and 'plants' there were no instances of indefinite singulars. Indefinite singular, according to Radden (2009), is a representative

### 10 Generics as a paradigm: A corpus-based study of Norwegian

generic type, used when an instance represents its type. Such use is prototypical in the sense that one element of the group or one member of the whole kind is referred to in order to denote the whole group/kind. An example of such a use of indefinite singular can be found in (27).

(27) En A plateskrue flat-head.screw er is spiss sharp og and laget make-part av of herdet tempered stål. steel 'Flat-head screws are sharp and made of tempered steel.'

In the sentence about a plate screw, a characteristic feature of that object is described, making a generic reference possible. The essential attribute (Radden & Dirven 2007: 111) of an arbitrary instance (Radden 2009: 224) makes the reference prototypical.

Kind reference can also be expressed with indefinite singular, similarly as in the models presented in the previous section. However, such kind reference is also prototypical as it utilises a single referent in order to denote a whole kind. The features of a single member of a kind are then projected over the whole group.

## **3.3 Definite singular**

Definite singular is regarded by Radden (2009) as typical kind generic. Pettersson (1976) however treats definite singular generic as prototypical, similarly to indefinite singulars mentioned above.

In example (28), 'the ballpoint pen' serves as a prototype for all pens of this sort. It was not a single pen that was invented at that time but rather the whole kind.

(28) Kulepennen Ballpoint.pen-def ble was oppfunnet invented på in 1800-tallet. 19th.century 'The ballpoint pen was invented in the 19th century.'

Definite singulars were among the three most used noun forms in the corpus. They occurred as kind references as in the example above, but also with mass nouns, familiar nouns and WEKs. As has been already mentioned, the definite forms, both singular and plural, are often used with well-established kinds, as can be seen in example (29).

(29) Hunden dog-def var was det the første first husdyret domestic.animal vårt. ours 'The dog was our first domestic animal.'

### Anna Kurek-Przybilski

What one interprets as a familiar noun or a well-established kind, depends greatly on one's language intuition and the context in which the noun appears. However, certain nouns, such as 'the dog' in the example above, are widely used in everyday speech and are therefore recognised as familiar.

The interpretation of certain NPs as WEKs or familiar nouns does not influence the results presented in this study. The generic contexts are clear and analysing certain NPs as familiar or well-established enriches the analysis with additional contexts.

## **3.4 Indefinite plural**

Indefinite plurals are considered *default* generic noun forms in many Germanic languages. In the corpus, this noun form occurred very frequently, indicating clear-cut kind reference, as in examples (30) and (31) below.


According to Radden (2009: 224), indefinite plural indicates *a salient proportion of the type's reference mass*. In other words, when making a kind reference with this noun form, the speaker refers to the majority of the kind and generalises over the whole kind this way. Radden & Dirven (2007: 111) mention also a *characteristic attribute* that allows to make such generalisations. The attribute however, does not need to be applicable to 100% of the population to be valid. Such generic usage is also mentioned by Leslie et al. (2011) and is called for 'majority characteristic'. This type of generic expressions consists of a salient feature that allows for exceptions.

The exceptions for the examples above can be dogs without any fur or skates that may not be considered good edible fish by certain people. Nevertheless, a hairless dog or inedible skates would still be classified as members of their respective kinds, despite the lack of a certain feature.

It is nevertheless questionable whether such nouns, while referring to a certain proportion of the class, can occur as BNs in contexts that allow for exceptions. BNs, as has been discussed before, often express concepts, not particular entities that one can generalise about.

10 Generics as a paradigm: A corpus-based study of Norwegian

## **3.5 Definite plural**

The last of the noun forms and generic types mentioned in the models, are definite plurals. Definite plural generic in English is considered as delimited generic, since its use is rather limited. In Norwegian, many more generic contexts are possible. One of the contexts is the same as in English, namely it concerns groups of people.

Surprisingly enough, kind reference can also be rendered with definite plural, as can be seen in examples (32) and (33).


The sentences about bats and bumblebees are *classic* generic sentences, where a truth about a given kind is stated. The interpretation is very similar as in case of indefinite plural generic, and is translated this way into English.

Definite plural generic is nevertheless not an equivalent of indefinite plural and the kind reference expressed this way differs in interpretation. In the corpus, not all animal species were described with definite plurals. In the majority of the cases, the form was used in descriptions of small animals and animal species considered gregarious, as in the case of bumblebees.

Another uses of delimited generic included WEKs, as well as quasi-mass nouns. Quasi-mass nouns are nouns that are grammatically countable but can be perceived as mass. The example in (34) shows this phenomenon.

(34) Torvmosene peat.moss-pl.def har have en a eiendommelig homogeneous bygning. structure 'Peat moss has a homogeneous structure.'

In Norwegian, the NP 'peat moss' is grammatically countable but, as can be seen in the example above, is often regarded as a mass. Another explanation for the use of definite plural generic in this case is the fact that moss consists of many smaller elements and therefore gives an impression that there are many different plants seen as a whole.

### Anna Kurek-Przybilski

## **4 Discussion**

All of the noun forms and their generic uses discussed in the previous sections, are based solely on the corpus material. The list of available generic contexts for each of the forms is not exhaustive but it allows to see main tendencies when it comes to genericity in Norwegian. In this section, a model of Norwegian generics will be presented in the form of a paradigm, based on those tendencies.

Table 5 comprises generic contexts that were frequent in the corpus texts. The cells marked with "+" indicate that the feature was present, whereas those marked with "−" lack a given feature. Asterisks in certain cells indicate modifications of the contexts in question. The cells marked with asterisk in case of kind reference indicate a conceptual or prototypical reference discussed in the previous sections. Mass nouns expressed with definite plural and equally marked with asterisk refer to quasi-mass readings, such as 'the peat moss' in example (34).


Table 5: Paradigm of Norwegian generics

The presented paradigm of Norwegian generics differs from a feature matrix in that it comprises different NP types, as well as contexts in which they occur. For instance, if we take a look at BNs we see that this NP type is not only used with numerous types of nouns (countable, mass, WEK etc.) but it also expresses kind reference, can be used to describe groups of people or to express concepts. The semantic basis for this paradigm is therefore a generic interpretation of a given NP type together with the context in which such an interpretation occurs.

As can be observed, bare nouns were used in most diverse contexts throughout the corpus texts. Almost all of the aspects mentioned, except for prototypes and gregarious animal species, were rendered with this noun form in at least one of

### 10 Generics as a paradigm: A corpus-based study of Norwegian

the texts. Concepts were expressed only with BNs, accounting for the opposition between concepts and prototypes.

The three noun forms that express kind reference through prototypes, namely BN, indefinite singular and definite singular, are complementary with each other but stand in contrast to a *classic* kind reference rendered with indefinite plural.

Another interesting aspect of the proposed model are countable nouns. In the Norwegian data, in generic contexts, countable nouns are used with all noun forms. Such an abundance of possible noun forms does not mean that the forms are interchangeable in all possible contexts. For instance, when a kind reference is usually rendered with indefinite plural, but in certain contexts, other noun forms can also have such a reading.

One of the biggest changes with reference to the original models of Radden & Dirven (2007) and Radden (2009), are the contexts available for definite plural generics. Unlike English, in Norwegian a few other uses were found in the corpus, such as quasi-mass plants or references to gregarious animal species. Definite plural generic is still limited in Norwegian but to a lesser degree than it is the case in English.

The contexts in Table 5 are the most frequent ones. For instance, in certain examples sub-kinds and hyperonyms were accessed through definite plural generics but those reference types were not in the majority and are therefore lacking in the model.

Understanding genericity as a paradigm rooted in speakers' language knowledge and intuition implies that certain noun forms are grammaticalised in generic contexts and to a certain degree. For instance, a *default* kind reference is always rendered with indefinite plural, although a few other noun forms could take on such reading. The diversity of the available generic forms in Norwegian is still greater than it is in English generics but the observed tendencies prove that a paradigm of generic noun forms exists.

## **5 Implications for further research**

The study presented in this paper concerns one written standard of Norwegian and generic use of different noun forms in encyclopaedic texts. A possible extension of the analysis could include the other written standard of the language, namely *nynorsk*. What is more, different text genres could be considered in examination of generics, should the resources allow for such a wide analysis.

Another aspect that was out of the scope of this study was the possible influence of dialects on the written standard. The Norwegian dialects are rather

### Anna Kurek-Przybilski

difficult to study, especially when it comes to such subtle phenomena as genericity. One of the reasons for this is the diversity of all dialects, as well as the diversity within the main dialect groups. Finally, dialects are mostly restricted to the spoken medium and therefore a study on genericity in Norwegian dialects would have to be based on discourse analysis or interviews.

As has been shown in this paper, genericity is a complex phenomenon that is considered by some researchers as one of the language universals. A paradigmatic view on the matter may help to further develop the matrix of generics that would be applicable for all aspects of the language, both written and spoken.

Last but not least, the presented paradigm could be considered when analysing the process of grammaticalisation in Norwegian. Currently, the different noun forms that occur in generic contexts show a relatively low degree of grammaticalisation, compared for instance to Romance languages, where genericity is almost always definite (cf. Kabatek 2013). It is possible that as the grammaticalisation of articles proceeds, also genericity will also be grammaticalised to a certain degree.

## **Acknowledgements**

This work was supported by the National Science Center under grant number 2017/25/N/HS2/00003.

## **References**


10 Generics as a paradigm: A corpus-based study of Norwegian


## **Chapter 11**

## **The importance of paradigmatic analyses: From one lexical input into multiple grammatical paradigms**

## Kirsten Jeppesen Kragh

University of Copenhagen

I take as my starting point that when lexical entities grammaticalize, they enter preexisting paradigms. Therefore, grammatical paradigms are important for the understanding of the reanalyses leading to grammaticalization. In the line of Henning Andersen's thinking I propose to conceive of grammar as composed of sets of paradigms (Nørgård-Sørensen et al. 2011). The term *paradigm* is used not in the narrow sense of inflectional paradigm, nor entirely in the line of the "classical" grammaticalization approach of Lehmann (1985), but in the more general sense of a selectional set, composed of marked or unmarked members (Andersen 2008: 19). The lexical input that I use to illustrate my point is the French verb *voir* 'to see', in order to show the pathway of a multifunctional lexical item into grammar, i.e. into a number of individual paradigms.

My approach combines synchronic and diachronic investigations on electronic corpora. Each paradigm presents the actual synchronic status of diachronic grammaticalization processes. By distinguishing the different contexts (labelled domains) in which the given forms appear, and state which semantic fields they cover (labelled frames), I can generate synchronic paradigms of which the grammatical entities are members. I aim to demonstrate that synchronic paradigms provide a precise and relatively simple presentation of what otherwise would seem utterly diverse usages of a lexical entity.

Kirsten Jeppesen Kragh. 2022. The importance of paradigmatic analyses: From one lexical input into multiple grammatical paradigms. In Gabriele Diewald & Katja Politt (eds.), *Paradigms regained: Theoretical and empirical arguments for the reassessment of the notion of paradigm*, 319–342. Berlin: Language Science Press. DOI: 10.5281/zenodo. 5675859

### Kirsten Jeppesen Kragh

## **1 Introduction**

I 1 take as my starting point that when lexical entities grammaticalize, they enter pre-existing or new grammatical paradigms. Therefore, paradigms are important for the understanding of the reanalyses leading to grammaticalization.

In the line of Henning Andersen's thinking I propose to conceive of grammar as composed of sets of paradigms (Nørgård-Sørensen et al. 2011). I here use the term *paradigm* not in the narrow sense of inflectional paradigm, nor entirely in the line of the "classical" grammaticalization approach of Lehmann (1985), but in the more general sense of a selectional set, composed of marked and unmarked members (Andersen 2008: 19). In previous studies (Kragh & Schøsler 2014, 2015, 2016, 2019, 2021) I have shown that the notion of a paradigm is useful for the understanding of grammatical structure.

In order to show the pathway of a lexical item into grammar, i.e. into a number of different paradigms, I will use the lexical French verb *voir* 'to see' and the derived phrases, *voici* and *voilà* to illustrate my point. When aiming at analysing a polysemous and multifunctional lexical entity like this verb, the researcher can choose between a polysemic or a monosemic approach, see Waltereit (2002, 2006). But when studying how a lexical item grammaticalizes, I do not consider this discussion to be the essential one. Rather, I think that the most interesting point is how the lexical item enters different grammatical paradigms.

My approach combines synchronic and diachronic investigations on electronic corpora. Each paradigm shown in the following presents the synchronic results of diachronic grammaticalization processes, based on synchronic paradigmatic analyses of very different functions. The paradigms comprise the following grammatical categories: verbal categories: tense, aspect and mood, i.e. progression (*je le vois qui arrive* 'I see him arriving'), voice (*il se voit refuser l'accès* 'he is refused entrance'), pragmatic categories: presentatives (*voilà le bateau* 'here is the boat'), focus constructions (*voici le bateau qui arrive* 'here is the boat arriving'), and discourse markers (*le bateau arrive, tu vois* 'the boat is arriving, in fact'), and the category of particles, i.e. the preposition *vu* 'considering' and the subordinate conjunction *vu que* 'considering that'. I shall illustrate my paradigmatic approach by means of three cases: the progressive paradigm, the presentative paradigm, and the focalization paradigm. I aim to show that the progressive and the presentative paradigms are preconditions for *voir*'s entrance in the focalization paradigm. By distinguishing the different contexts (labelled *syntactic domains*) in which the

<sup>1</sup>An extensive part of this work has been done in collaboration with Lene Schøsler. I am deeply grateful for her valuable suggestions and comments on this article.

### 11 The importance of paradigmatic analyses

forms appear, and state which semantic fields they cover (labelled *frames*), I identify synchronic paradigms of which the above exemplified grammatical entities are members. I claim that synchronic paradigms provide a precise and relatively simple presentation of what otherwise would seem utterly diverse usages of a lexical entity.<sup>2</sup>

## **2 Definition of paradigmatic level**

The grammatical paradigm can be identified through a set of five defining features (Nørgård-Sørensen et al. 2011: 5–6).<sup>3</sup>

First, the grammatical paradigm is a *closed set of items*, the number of members being fixed at a given language stage.<sup>4</sup> In Modern French, for instance, the category of tense, aspect, and mood (TAM) comprises the simple verb forms present, past, future tenses: *voit*, *vit*/*voyait*, *verra/verrait*, and a number of analytical forms.

Secondly, for every paradigm, the *syntactic domain*, i.e. the syntagmatic context in which it applies must be specified. Thus, in Modern French, the domain of the category TAM is the simple or composed finite verb form of a given sentence.

Thirdly, a paradigm has a *semantic frame*, i.e. a common semantic denominator, here tense, aspect and mood, within which the content of its members is defined in opposition to one another. Semantic frames are language specific and cannot be taken to be equivalent to the cognitive networks of Croft (2001) and Croft & Cruse (2004).

Fourthly, the choice between the members is *obligatory*, in the sense that in the given syntagmatic context that defines the domain of the paradigm, speakers cannot avoid selecting one of the members; they must choose for example one verbal form – simple or analytical, the present, the past, the future, etc. – to express the temporal and aspectual dimension of a given utterance.

Fifthly, a paradigm is *asymmetric*, distinguishing between marked and unmarked members, possibly in a hierarchical structure. In Modern French, the

<sup>2</sup>An alternative way of presenting a selection of diverse usages is the lexical approach provided in dictionaries. Please, see https://robert-correcteur.lerobert.com for an illustration of a rather unsystematic presentation with a mix of lexical and grammatical information on the different functions of *voir* in the digital dictionary Robert Connecteur.

<sup>3</sup>This section is a revised version of section 2.1 in Kragh & Schøsler (2015).

<sup>4</sup> In fact, over time most paradigms change their member list, so a *given language stage* is an abstract notion to be defined for each paradigm at a specifically defined synchrony. Since languages are always changing, it is not an easy task to identify the relevant synchronic stages permitting the establishment of a paradigm, without the risk of circular argumentation.

### Kirsten Jeppesen Kragh

present tense is the unmarked form, because it appears in both perfective and imperfective contexts and displays such values as present, habitual, progressive, recent past, near future, etc. Compared to the present tense, the simple past, the imperfect, the future, and the conditional are all marked, i.e. restricted, both in regard to the type of context in which they appear, and in regard to their temporal and aspectual values.

Whereas *the domain* refers to the syntagmatic delimitation of the paradigm on the expression level, *the frame* is its counterpart on the content level. A paradigm is not a pure expression system, but a sign system with the domain-and-frame pair constituting a unity of expression and content (according to the terminology of the Danish Functional tradition, see Engberg-Pedersen et al. 1996).

## **3 Lexical level: Level 0**

Before studying the processes of grammaticalization, we must start at the lexical level, i.e. the valency level with its possible constituents in free use. It is crucial here to distinguish the lexical level from the construction level; it is the former that provides the input to a grammaticalization process.

*Voir* is a highly polysemous verb, characterized by its frequency in usage. Such high frequency verbs are likely to grammaticalize or pragmaticalize (cf. Bolly 2010: 674–675).

This makes *voir* an obvious choice as object in an attempt to demonstrate how one single lexical item tends to enter a number of new constructions, thus becoming member of a wide range of grammatical paradigms. Each new usage of *voir* being routinized is a candidate for entering a new paradigm. Craig (1991) introduces the term *polygrammaticalization* to refer to this phenomenon where one particular lexical entity is the source of multiple grammaticalization chains.

As I see it, the function of a given lexical item in a given context triggers the specific meaning of this lexical item. Thus, when *voir* is used in new contexts, it is ascribed new functions (e.g. presentation, focus, progression, passive voice, discourse marker, etc.); they provide new meanings. I have shown in previous studies that secondary features of the original construction turn into primary features during the process of grammaticalization (cf. for instance Kragh & Schøsler 2015). In what follows, I will first present the lexical level (level 0), and subsequently a number of paradigms into which forms of the verb *voir* have entered as a result of the grammaticalizations.

As indicated in Figure 1, level 0 is lexical, it comprises the valency-bound constituents that combine with the verb *voir*, e.g. noun phrases (1), possibly with

### 11 The importance of paradigmatic analyses

Figure 1: Lexical level (level 0)

a subordinate relative clause (2), complement clauses (3); a number of nexus constructions,<sup>5</sup> with infinitive (4), with deictic relative clause (5), with present or perfect participles, (6) and (7), with adjectives (8), with nouns (9), or with a prepositional phrase as object complement, (10) and (11):


<sup>5</sup>The term *nexus* describes the relation of interdependency with the antecedent/referent in contrast e.g. to subordinate relative clauses. This implies that the antecedent/referent cannot be omitted, e.g. \**Je vois perdu*/*président* etc.

### Kirsten Jeppesen Kragh


No other verb of visual perception has such a large number of possible constituents (Willems & Defrancq 2000: 10).

We now proceed from the lexical level to the levels of grammatical paradigms.

## **4 Paradigms: Levels 1 to 3**

## **4.1 Level 1: Reanalysis of the subordinate relative clause into the deictic relative clause**

Levels 1 and 2 and later on also level 3 illustrate the use of *voir* in contexts where the original meaning of the verb *voir* is bleached due to a number of reanalyses and grammaticalizations.

Level 1 comprises the initial reanalysis and grammaticalization that are a prerequisite of the grammaticalizations at level 2 and 3.

At level 1 I find the grammaticalization of the deictic relative clause due to a reanalysis of the relative subordinate (see Kragh & Schøsler 2014, 2015), i.e. reanalysis of level 0. This is an important step for the subsequent grammaticalizations that involve *voir* in a progressive context (level 2).

Figure 2: From lexical level to grammatical level

### 11 The importance of paradigmatic analyses

In example (2), we had a clear subordinate relative clause which was part of the NP. In certain contexts, such subordinate clauses are reanalysed. I have found bridging examples in which the hearer may interpret the message of the relative clause in two different ways, see example (12):

(12) S'i erent venu apoier;/quant le cuens vit son escuier/qui sor le noir destrier estoit 'they have come to rest there/when the count saw his horseman/who was sitting on the black horse' (Les romans de Chrétien de Troyes, Erec et Enide, ca. 1213, p. 98, vers 3207–3210, Frantext)

The interpretation of this example is twofold. The point may be that the count firstly catches sight of his horseman and subsequently discovers him sitting on the black horse. However, another interpretation is also possible, providing a bridging or critical context, which permits reanalysis because of the ambiguity (see Diewald 2002: 117, Heine 2002), namely a holistic (progressive) perception of the horseman sitting on the horse.

I will consider the second interpretation of example (12) to be the result of the speaker reanalysing the subordinate type of relative clause, in the following way: A (subordinate relative clause specifying an NP) > B (deictic relative construction), i.e. into a new type of verbal complementation, without immediate change of the surface manifestation. This implies that the construction has acquired not only a) a different function, which is not a subordinate, but a nexus relation, but also b) a different meaning. This meaning has been described tentatively in terms of a holistic vision. Moreover, this vision is progressive, by which term I refer to an ongoing process performed by the referent of the direct object of the verb of perception.<sup>6</sup>

### **4.1.1 From level 1 to level 2, paradigm 1**

As the deictic relative construction (B) is accepted in the speech community and increases its use, it is *embedded,* i.e. integrated into grammar (Herzog et al., 1968: 185). Once embedded, it can be considered as yet another way of expressing progressivity: (*Je vois*) *Pierre qui chante*, 'I see Peter singing', cf. Figure 3.

Paradigm 1 shows the inventory of the progressive constructions in French in form of a paradigm. Please note that this paradigm has a diachronic dimension.

<sup>6</sup> For a more detailed description of the origin of the deictic relative, I refer to Kragh & Schøsler (2014: 178–182).

Figure 3: From level 0 to paradigm 1

From the early times until the end of the 17th century, *Pierre va chantant* is the unmarked form, also diatopically. *Pierre est à chanter* and *Pierre est après chanter* are diatopically and diastratically marked, whereas *Pierre est en train de chanter* turns into the unmarked form of progression from the 19th century. The type *Je vois Pierre qui chante* is a marked member of the paradigm from the 17th century: They all mean 'Peter is singing'. It is the only member that provides a holistic perception of the activity, perceived in its progression, and which has a different referent for the subjects of the two verbs (S1 ≠ S2).

## **4.1.2 From level 1 to level 2, Paradigm 2**

Let us now consider another paradigm, that of the fossilized imperative form of *voir, voici*/*voilà*, reanalyzed as member of the presentation paradigm.

French has a number of ways of expressing presentation: *c'est X*, *il y a X*, *il est X*, *voici/voilà X*, *X* being the presented entity. These constructions have been examined in Kragh & Schøsler (2019), where the inventory of presentation is discussed and a list has been established in accordance with Lambrecht (2000, 2001) and Riegel et al. (2009 [1994]):

*C'est X:*

(13) Ce n'est pas eux 'It's not them'


Table 1: Progressivity in French, progressive constructions (Kragh & Schøsler, 2015: 290)

### *<sup>a</sup>*→1700

*b* 1800→

## *Il y a X:*

(14) Il y a quelqu'un 'There is someone'

## *Il est X:*

(15) Il était une fois une belle princesse 'Once upon a time there was a beautiful princess'

## *Voici/voilà:*


### Kirsten Jeppesen Kragh

These presentatives can be listed in a paradigm (see Paradigm 2). In the following, I will focus on the presentatives *voici* and *voilà*, derived from the verb *voir*.

In this construction, *Voici*/*Voilà X* presents a referent *X*, known or unknown to the hearer, examples (16) and (17). This kind of constructions is also called *neutral focus structures*. *X* can be an NP, (16) and (17), possibly with a subordinate relative clause (18), a complement clause (19), or an interrogative clause (20).

It is characteristic for such constructions that they address explicitly the hearer and thus have the feature of +deixis, since they presuppose the presence of the hearer in the factual or fictive conversation space. In Modern French, *Voilà X* is more frequent than *Voici X*, and the latter is marked, since it has reduced possibilities of usage. The constructions are especially frequent in oral, informal communication, in accordance with the deictic character of the forms.

I consider the presentative construction exemplified in (16) to (20) a reanalysis of the lexical usage of the verb *voir*, level 0. This reanalysis is illustrated by a bridging example (21) in which the hearer may interpret the message of the utterance in two different ways. Example (21) is from Chanson de Roland, 1100 (cited in Oppermann-Marsaux 2006: 79):

(21) Dreiz emperere, veiz me ci en present 'Rightful Emperor, see me present here' (interpretation 1) 'Rightful Emperor, here I am before you' (interpretation 2)

According to interpretation 1, the speaker addresses the Emperor by saying 'see me being present here'; thus, *veiz* expresses visual perception. However, another interpretation is also possible, interpretation 2, which is a way of attracting the attention of the emperor. Thus, example (21) provides a bridging or critical context which permits reanalysis because of the ambiguity (cf. Diewald 2002, Heine 2002), namely an intention of attracting attention, i.e. a purely pragmatic function. I will consider the second interpretation to be the result of the speaker reanalysing the imperative form of *voir* followed by the particle *ci*, in the following way: A (imperative form of the verb of perception *voir* followed by a locative particle *ci*, meaning 'see here') > B (presentative), i.e. into a new way of attracting attention to a given item, without immediate change of the surface manifestation. This implies that the construction has acquired not only a) a different function. Thus, it is no longer a finite verb + a particle, but it is reanalysed as a fixed form with a particle, with b) a different meaning, i.e. presentation or attracting attention. Consequently, the new function is pragmatic.

This process of reanalysis and grammaticalization of the imperative verbal form in the second person singular and merge with the two adverbs *-ci* and *-là*,

### 11 The importance of paradigmatic analyses

losing its full lexical meaning, is schematized in Figure 4. I refer to Kragh & Strudsholm (2013: 212–216) and Kragh & Schøsler (2014: 190–191) for a detailed account of the reanalysis from level 0 to level 1 and to Oppermann-Marsaux (2006) for an account of the initial steps of this process. In the course of the subsequent reanalysis from level 1 to level 2, *voici*/*voilà X* is reanalysed as a member of the presentative paradigm (Figure 4)

Figure 4: From level 0 to paradigm 2

The presentative paradigm is listed in paradigm 2 and contains the constructions exemplified above in (13) to (20).

The syntactic domain, i.e. the syntagmatic context, for the presentative paradigm (S)VX is a verb with or without a subject, followed by X as the presented entity. The semantic frame is *presentation*, in the sense of an introduction of important and new information to the hearer about a new or already known entity. With respect to the content, the members of the paradigm are, in addition to *presentation*, characterized by two features, namely the option of identification/opposition and deixis. *Identification* refers to the designation of a referent, possibly combined with the designation in opposition to one of more other potential referents (*opposition*). The feature of *deixis* refers to the possibility of addressing explicitly a hearer and it presupposes the presence of the hearer in the factual or the fictive conversation room.

Having the least restrictions of the four members of the paradigm, *c'est X* is the unmarked member (Kragh & Schøsler 2019). In addition to presentation, *c'est X* is characterized by its capacity of identifying, possibly with specification of opposition. Deixis is possible. *Il y a* can only mean presentation, not identification,


Table 2: Presentative paradigm. <sup>m</sup>: marked member of paradigm; <sup>u</sup> : unmarked member of paradigm

*a* Fossilized variant of *il y a*.

and deixis is not required. *Il est* is the fossilized variant of the productive *il y a*; it has the same content as *il y a*, but is mainly used in introductions of fairy tales or to express time. *Voici*/*voilà* expresses presentation, not identification, but does, on the other hand, express deixis. The three latter are thus marked in proportion to *c'est X*.

### **4.1.3 From level 2 to level 3, Paradigm 3**

Members of the presentative paradigm can occur with a subordinate or deictic relative clause and thereby enter a paradigm of focalization: *c'est X qui/que*…, *il y a X qui/que*…, *voici/voilà X qui/que*… In addition to these, I have focalization constructions which are not derived from a presentative construction: *il a X qui/que*…, *X est là qui/que*…

The focalization paradigm comprises the following members:


Il était une fois une belle pricesse qui vivait dans un vieux château 'Once upon a time there was a beautiful princess who lived in an old castle'

### 11 The importance of paradigmatic analyses


As illustrated in example (25), the presentative *voici*/*voilà* can occur with a deictic relative. This is a new reanalysis which presupposes a number of previous reanalyses, presented in the preceding sections. I hypothesize that the reanalyses at the constructional level 1 from subordinate to deictic relative, and the grammaticalization of *voici*/*voilà* are more or less parallel processes during the Middle Ages, because I have no textual evidence that one should precede the other. The subsequent reanalyses as members of paradigms, Paradigm 1 and 2 (level 2), respectively, take place from the 16th century onwards (Oppermann-Marsaux 2006, Kragh & Schøsler 2015).

This means that both the ideas of progression and simultaneity expressed in the deictic relative, and presentation expressed by the grammaticalized form *voici*/*voilà* are preconditions for the grammaticalization of the focus construction composed of *voici*/*voilà* and a deictic relative.

*Voici/voilà* are deictic expressions rooted in the time of the utterance, and in order to express simultaneity, typical of holistic constructions, the verb of the deictic relative must be in the present tense. This is in accordance not only with my analysis of *voici/voilà X* and a deictic relative, where they are part of a progressive, holistic and deictic construction (Kragh & Schøsler 2014), but also with the view of Lambrecht (2000: 50–51), who states that the function of this type of focus construction is to present an entity and to express new information about it at the same time. Furthermore, the structure presupposes a known referent (Riegel et al. 2009 [1994]: 456).

Examples (28) to (30) illustrate the difference between presentation and focalization. Example (28) shows a presentative construction with a pronominalization of the object *ma sœur*. Example (29) is also a presentative construction where the NP is specified by a subordinate relative, and the NP including the relative is pronominalized in the object pronoun *la*.

<sup>7</sup>Example citated from Conti (2010).

<sup>8</sup>Example citated from Furukawa (2000: 104).

### Kirsten Jeppesen Kragh

In contrast, example (30) illustrates grammaticalized focalisation. Here I find a dislocation of the object *ma sœur* specified not by a subordinate relative, but by a deictic relative being part of a nexus construction. In this case, pronominalization by means of *la* of the antecedent *ma sœur* concerns only the NP *ma sœur*, not the deictic relative.


The possibility of a personal pronoun as antecedent is a specific feature of a deictic relative; this possibility does not exist for subordinate relative clauses, cf. example (29).

Summing up the reanalyses leading to the grammaticalized focalization paradigm: this paradigm is the result of reanalyses of two constructions at a higher level, each with its own paradigmatic structure (Paradigm 1 and Paradigm 2). This implies that characteristic features of Paradigm 1, e.g. progression, and of Paradigm 2, e.g. that *voici* and *voilà* function as presentatives are inherited in Paradigm 3, as schematized in Figure 5.

The focalization paradigm is presented in Paradigm 3 and is composed of the constructions exemplified above (examples 22–27).

The *syntactic domain* of the paradigm of grammaticalized focalization differs from that of presentative (neutral focus) because it contains an obligatory *qui*/ *que*-clause, which is presented in the figure as follows: (S)VX *qui/que*-clause. The *semantic frame*, i.e. the grammatical meaning, is focalization in the sense that the structures provide important and new information to the interlocutor on new or already known entities.

With respect to the content, the members of the paradigm are, in addition to *focalization*, characterized by four features, namely the option of expressing *opposition*, further information by means of a subordinate relative vs. a nexus construction introduced by *qui/que* expressing progressivity, the option of deictic concord, and the option of deixis. The feature of *deictic concord* refers to coincidence with respect to person, time and place between the matrix and the subordinate or deictic relative. Thus, in example (25) and (30) I find deictic concord between *Voilà* and *qui arrive*, because of the simultaneity of the perception (*voilà*) and the perceived object (*ma sœur qui arrive*). This concept should not

### 11 The importance of paradigmatic analyses

Figure 5: From Paradigm 1 and 2 to Paradigm 3

be confused with *deixis*, which refers to the possibility of addressing a hearer explicitly and presupposes the presence of the hearer in the factual or the fictive conversation room. Thus, there is *deixis* in example (25) and (30), because the speaker is addressing the hearer explicitly, which presupposes the presence of the hearer in the factual or the fictive conversation space.

The structures identified as focus constructions have been characterized by means of the following criteria: focalization ±opposition, subordination *versus* nexus relation, ±deictic concord between the matrix and the relative clause, with restrictions on tense, mood, etc., and ±deixis in the construction. According to these criteria, *C'est X qui*/*que* clause is the unmarked member of the paradigm; since it has fewest restrictions. It expresses focalization with or without opposition to another referent and can have deictic concord between the matrix and the relative, but has no restrictions on tense and mood, etc., and it has no deixis in the construction. The relation between the relative clause and the antecedent is a relation of subordination. The other structures are opposed to this unmarked construction as marked members. Among the marked members, the *il y a qui/que* structure is less marked than the *voici*/*voilà qui, il a X qui*, and *il est là qui* structures because it has fewer restrictions on tense and mood. It does not express

### Kirsten Jeppesen Kragh


Table 3: Paradigm of focalization (Kragh & Schøsler 2019). <sup>m</sup>: marked member of paradigm; <sup>u</sup> : unmarked member of paradigm.

### 11 The importance of paradigmatic analyses

deixis. The three last mentioned structures share the following criteria: like *il y a qui/que* they focus without indication of opposition and they form a nexus construction. In contrast with *il y a X qui/que*, they have deictic concord, but only *voici/voilà X qui* has deixis, i.e. presupposes the presence of the hearer in the same factual or fictive conversation room as the speaker.

As illustrated in Figure 5, level 3 presupposes level 2, and not vice-versa. In chronological terms this relation of precondition is confirmed by my corpus investigations, which show that level 2 can be found from the 13th century, whereas level 3 occurs by the end of the 17th century, spreading in the 19th century. Therefore, it is reasonable to conclude that grammaticalized focalization is the result of a reanalysis of the presentative structure, with the consequence of focalization being clearly marked.

It is my hypothesis that other usages of the verb *voir* have led to pathways similar to the ones presented so far and into other grammatical paradigms as suggested in Section 1.

I shall now give a brief overview of four of these processes of reanalyses and grammaticalizations leading to *tu vois* as a discourse marker (Section 4.2), *se voir* as a member of the voice paradigm (Section 4.3), *vu* as preposition, and *vu que* as member of the paradigm of subordinate conjunctions (Section 4.3.2), in order to show further perspectives of my approach.

## **4.2 Level 1: Reanalyses to constructions and sub-levels**

The origin of verbal discourse markers like *Tu vois, mon bonheur passe* 'You know, my happiness is waning' is widely debated (Kragh 2021). My hypothesis is that this type of discourse markers originates as a reanalysis of a complex phrase of the type *Tu vois que mon bonheur passe* → *tu vois, mon bonheur passe* (level 1).

Based on this first reanalysis, I have observed a subsequent reanalysis of *tu vois, vous voyez, voyons* and *voilà* leading to the creation of a discourse marker paradigm. For a detailed account of this process, I refer to Kragh (2021, forthcoming).

## **4.3 Level 1: Reanalyses to constructions and sub-levels**

As illustrated in Figure 7, I assume that different reanalyses also precede the voice paradigm (Paradigm 5), and the preposition and conjunction paradigms (Paradigms 6 and 7), respectively.

Figure 6: From constructional level to paradigmatic level, Paradigm 4

Figure 7: From constructional level to paradigmatic level, Paradigms 5, 6, and 7

11 The importance of paradigmatic analyses

### **4.3.1 Level 2, Paradigm 5**

As part of the voice paradigm, the reflexive form of *voir* (see example 31) competes not only with the active construction (*Le gouvernement augmente les prix* 'The government raises the prices'), but also with other ways of expressing a state of affairs without an agent or an active subject, e.g. the passive construction (*Les prix sont augmentés* 'The prices have gone up'), the unaccusative construction (*Les prix augmentent* 'The prices go up'), the reflexive unaccusative construction (*Les prix s'augmentent* 'The prices go up'), and the deontic reflexive passive (*Le vin blanc se boit frais* 'White wine should be served chilled'), and a number of impersonal constructions (e.g. *On augmente les prix* 'The prices have gone up'). Periphrastic reflexive passives, typically with an affected person as subject are found with the verbs *faire* ('to make') and *voir*, and express mainly activities that are adverse to an affected person:

(31) Pierre se voit refuser l'accès 'Peter is denied entrance…'

Thus, they differ from both active, unaccusative, and passive constructions with regard to types of agent and patient, and to the event described. It should be noted that *se voir* has been included among passive constructions by a few other scholars, e.g. by Grevisse & Goosse (2008: 1051) where it is called *auxiliaire du passif* and classified among the semi-auxiliaries, whereas Defrancq (2000: 188) uses the term *passif de l'objet prépositionnel*.

### **4.3.2 Level 2, Paradigms 6 and 7**

Paradigms 6 and 7 have in common that they are formed from the past participle of *voir*. Preliminary results indicate that *vu* as a preposition occurs from the 14th century (Rey & Rey-Debove 1986), and suggest that this precedes the conjunction *vu que*, of which the first occurrences found are from the 15th century (Frantext). Both the preposition *vu* and the conjunction *vu que* are grammatical entities; they have no lexical, but only grammatical meaning. Therefore, I do not consider them to be cases of lexicalization. Whether *vu que* is the result of a regrammation of the preposition *vu* or rather of the participle (as in *Il a vu que…*) remains to be investigated.

Paradigm 6 comprises the paradigm of prepositions. This paradigm includes simple forms like *à*, *de*, *en*, *dans*, *pour*, etc., complex formations such as *à côté de*, *au lieu de*, *pour cause de*, etc., and forms derived from past participles like *vu*, *attendu*, *exepté*, *compris*, *hormis*, as well as present participles such as *suivant*, *durant*, *moyennant*, etc. An exemple is:

### Kirsten Jeppesen Kragh

(32) *Vu* la situation actuelle, il faut partir au plus vite '*Considering* the actual situation, we must leave as quickly as possible'

Paradigm 7 is the paradigm of subordinate conjunctions. It includes the simple conjunctions *que*, *si*, *comme*, *quand*, etc., complex conjunctions like *à mesure que*, *avant que*, *dès que*, *bien que*, *à la condition que*, etc., and conjunctions derived from verbal forms *excepté que*, *vu que*, *suppose que,* and *suivant que*, *pourvu que*, *attendu que*:

(33) *Vu que* le texte de la recommendation n'est pas encore prêt, il est assez difficile de poursuivre l'analyse '*As* the text of the recommendation is not yet ready, it is rather difficult to undertake further analysis'

## **5 Conclusions**

As stated at the beginning of this paper, I think that when lexical entities grammaticalize, they enter pre-existing or new grammatical paradigms, and that therefore the concept of paradigm is important if we wish to understand the reanalyses that lead to grammaticalization. I claim that grammar is composed of sets of paradigms in the general sense of selectional sets, composed of marked and unmarked members, cf. Figure 8 (page 339).

An approach of grammar as sets of paradigms provides a precise and relatively straight forward presentation of what otherwise would seem utterly diverse usages of a lexical entity, see the illustration of a lexical approach in the the digital dictionary Robert Connecteur<sup>9</sup> , which confuses very different levels of usage, lexical, semi-grammatical, and grammatical.

I hope to have provided convincing evidence in favour of this claim.

## **References**

Andersen, Henning. 2008. Grammaticalization in a speaker-oriented theory of change. In Thórhallur Eythórsson (ed.), *Grammatical change and linguistic theory: The rosendal papers*, vol. 113 (Linguistik Aktuell/Linguistics Today), 11–44. Amsterdam, Philadelphia: John Benjamins.

<sup>9</sup>https://dictionnaire.lerobert.com/

### Kirsten Jeppesen Kragh


11 The importance of paradigmatic analyses


### Kirsten Jeppesen Kragh


Aaron, Jessi Elana, 269 Ackerman, Farrell, 2, 157, 298 Aikhenvald, Alexandra Y., 130 Albright, Adam, 145 Allen, Cynthia L., 203 Almasi, Oswald, 150 Amit, Daniel J., 269 Andersen, Henning, 2, 15, 30, 31, 72, 173, 177, 183–185, 189, 196, 224, 319, 320 Anderson, Stephen R., 146 Anttila, Raimo, 15, 41, 224, 290 Aronoff, Mark, 134, 153, 154, 156, 160 Askedal, John Ole, 182 Audring, Jenny, 142, 154, 268 Bacanlı, Eyüp, 130 Baechler, Raffaela, 182 Baerman, Matthew, 155 Bale, Alan, 154 Barber, Charles, 181 Barðdal, Jóhanna, 203, 204 Barner, David, 154 Batiukova, Olga, 136 Bauer, Laurie, 194 Behrens, Leila, 299, 305 Bennett, Charles E., 133 Bermúdez-Otero, Ricardo, 163 Bisang, Walter, 3 Bjerrum, Anders, 177, 179, 210 Blake, Barry J., 184 Błaszczak, Joanna, 80

Blevins, James P., 2, 95–97, 114–117, 247 Bloch, Bernard, 16 Bloomfield, Leonard, 12, 41 Boas, Hans C., 140, 268, 271 Bobaljik, Jonathan D., 128, 130, 143, 145–148 Boden, T. F., 268 Bolly, Catherine, 322 Bom, Kaj, 208 Bonami, Olivier, 142 Booij, Geert, 3, 55, 91–93, 116, 127, 128, 133, 137, 141, 143, 150– 152, 155, 158, 159, 270 Borthen, Kaja, 306, 307, 309 Boye, Kaspar, 272 Brattico, Pauli, 163 Brems, Lieselotte, 273, 279, 283 Breu, Walter, 70, 87, 90 Brown, Dunstan, 71, 150, 152, 156, 158 Buscha, Joachim, 138, 144, 147 Butt, John, 134, 136 Bybee, Joan L., 3, 93, 259, 277, 290 Byloo, Pieter, 248, 255, 258, 259, 261 Caers, Wim, 249, 252–255, 263 Carlson, Gregory N., 299, 306, 308, 310 Chierchia, Gunnaro, 299, 308 Chomsky, Noam, 51 Christensen, Tanya Karoli, 202 Colleman, Timothy, 273

Collins, John, 299, 300 Conti, Virginie, 331 Craig, Colette, 322 Crippen, James A., 149 Croft, William, 37–40, 204, 321 Cruse, D. Allan, 321 Crysmann, Berthold, 142 Dąbrowska, Ewa, 268, 271 De Clerck, Bernard, 273, 283 de Saussure, Ferdinand, 12, 268 De Smet, Hendrik, 269, 276, 289 De Vries, Matthias, 249, 250 Defrancq, Bart, 324, 337 Delorge, Martine, 289 Dickey, Stephen M., 107 Diderichsen, Paul, 185–187, 206, 235 Diepeveen, Janneke, 261 Diessel, Holger, 40, 46, 268, 277 Diewald, Gabriele, 2, 3, 94, 98, 100, 159, 160, 191, 247, 270, 272, 276, 289, 297–299, 325, 328 Dik, Simon C., 206 Dirven, René, 297, 299, 300, 302, 310– 312, 315 Divjak, Dagmar, 81 Dixon, R. M. W., 130 Dressler, Wolfgang U., 130 Duinhoven, A. M., 248, 259 Dyvik, Helge, 211 Embick, David, 130, 143, 147, 148, 156, 162 Engberg-Pedersen, Elisabeth, 204, 322 Enger, Hans-Olav, 156, 182 Eyþórsson, Þórhallur, 203, 204 Faarlund, Jan Terje, 203, 206, 213

Fábregas, Antonio, 94 Fabri, Ray, 1 Falk, Hjalmar, 181 Fennell, Trevor G., 135 Ferraresi, Gisella, 98 Fillmore, Charles J., 38, 49 Finkel, Raphael A., 23, 33 Fischer, Olga, 269, 289 Furukawa, Naoyo, 331 Fuß, Eric, 145 Gaglia, Sascha, 158 Gelman, Susan A., 300 Gelsen, Henry, 135 Geyken, Alexander, 275 Gil, David, 153 Givón, Talmy, 272 Gloning, Thomas, 275 Goldberg, Adele E., 37–40, 62, 91, 117, 127, 133, 140, 155, 158, 268, 270 Goosse, André, 337 Gorbova, Elena V., 73 Greenough, James B., 133 Grevisse, Maurice, 21, 337 Gunnerud, Sverre Martin, 305 Gurevich, Olga I., 128, 133, 141, 142 Hakulinen, Auli, 55 Halle, Morris, 128, 143 Halliday, Michael A. K., 193, 203, 226 Halmøy, Madeleine, 304, 307, 309 Hamp, Eric P., 12 Hankamer, Jorge, 130, 131 Hansen, Aage, 236 Hansen, Bjarne Simmelkjær Sandgaard, 174, 184, 185 Hansen, Erik, 225, 226

Harder, Peter, 272 Harley, Heidi, 128, 143, 146, 148 Hartmann, Stefan, 273, 275, 277–279 Haspelmath, Martin, 3, 88, 93, 94, 100, 114, 117, 127–130, 133, 139, 141, 143, 149–154, 160, 164, 184, 253, 290 Hathout, Nabil, 128 Haugen, Einar, 181 Haugen, Jason D., 153 Heinämäki, Orvokki, 41 Heine, Bernd, 69, 325, 328 Helasvuo, Marja-Liisa, 54 Helbig, Gerhard, 138, 144, 147 Heltoft, Lars, 173, 177, 183, 184, 190, 191, 193, 195–197, 204, 206– 208, 211, 213, 225, 226, 228, 230, 232, 234–237 Herce, Borja, 155, 157, 158 Hilpert, Martin, 91, 277 Himmelmann, Nikolaus P., 3, 69, 287 Hinzelin, Marc-Olivier, 158 Hippisley, Andrew, 150, 152, 156, 158 Hjelmslev, Louis, 177 Hockett, Charles F., 33 Hodge, Carleton T, 32 Hoffmann, Sebastian, 279 Hoffmann, Thomas, 136, 158 Hopper, Paul J., 183, 250, 269, 288 Houwer, Jan De, 272 Hrafnbjargarson, Gunnar Hrafn, 204 Igartua, Iván, 32 Isačenko, Aleksandr V., 29 Itkonen, Esa, 41 Ivanova, Elena Ju, 81

Jackendoff, Ray, 1, 142, 154

Jakobson, Roman, 12, 27, 29 Janda, Laura A., 29, 73, 74, 89 Jensen, Eva Skafte, 174, 177, 178, 182– 184, 234 Johnson, Tamar, 157 Jørgensen, Henrik, 234 Juul Nielsen, Peter, 202 Kabatek, Johannes, 316 Karker, Allan, 205 Karlsson, Fred, 41 Katz, Jerrold J., 154 Kauppinen, Anneli, 62 Kay, Paul, 38, 40, 49, 50, 52 Keenan, Edward L., 225 Keibel, Holger, 275 Ketrez, F. Nihan, 132 Kiparsky, Paul, 203–205 Kornfilt, Jaklin, 131, 132, 163 Kragh, Kirsten Jeppesen, 320–322, 324–326, 329, 331, 334, 335 Kramer, Ruth, 136, 147, 149, 155 Krug, Manfred G., 277 Kupietz, Marc, 275 Kuteva, Tania, 69 Lambrecht, Knud, 40, 91, 326, 331 Langacker, Ronald W., 38, 116, 288 Łaziński, Marek, 74 Lehmann, Christian, 3, 69, 94, 159, 183, 247, 268, 271, 276, 290, 291, 299, 302, 319, 320 Lehmann, Volkmar, 83, 86, 90 Lehmann, Winfred P., 202 Leino, Jaakko, 5, 37–41, 58, 60, 109, 270 Leino, Pentti, 41 Leiss, Elisabeth, 147 Leslie, Sarah-Jane, 300, 312

Lieb, Hans-Heinrich, 2 Liebesman, David, 299 Logan, Gordon D., 272 Lönnrot, Elias, 56 Loporcaro, Michele, 182 Luís, Ana, 153, 163 Lyons, John, 304 Maiden, Martin, 134, 156, 157 Malouf, Robert, 157 Manninen, Satu, 54 Mantlik, Annette, 271 Marantz, Alec, 128, 143 Marzi, Claudia, 152 Masini, Francesca, 154, 270 Matthews, P. H., 131 McGinnis-Archibald, Martha, 155 Meillet, Antoine, 181 Meinunger, André, 147 Mel'čuk, Igor A., 142, 206, 225, 228 Mende, Julia, 83, 89 Michaelis, Laura, 40, 91 Milin, Petar, 114 Moors, Agnes, 272 Morris, Caroline, 268 Namer, Fiammetta, 128 Neels, Jakob, 273, 275, 277–279, 290 Nelson, Diane, 54 Nielsen, Peter Juul, 184, 185, 208, 237 Nikiforidou, Kiki, 93 Norde, Muriel, 173, 177, 181, 184, 190, 196, 252, 268, 270 Nørgård-Sørensen, Jens, 2, 173, 183, 190, 191, 193, 195–197, 202, 207, 222, 228, 247, 298, 319– 321 Noyer, Rolf, 128, 143, 148 Nuyts, Jan, 248–250, 252–261, 263

Oppermann-Marsaux, Evelyne, 328, 329, 331 Östman, Jan-Ola, 5, 37, 39–41 Padučeva, Elena V., 80, 81 Pagliuca, William, 259 Palmer, F. R., 16, 17 Panhuis, Dirk, 132 Panov, Vladimir, 98 Parrott, Jeffrey Keith, 205 Pelletier, Francis Jeffry, 306, 308, 310 Perek, Florent, 268, 271, 287 Petersen, Kathrine Thisted, 183 Petruxina, Elena V., 83 Pettersson, Thore, 297–300, 302– 304, 308, 310, 311 Pijnenburg, Wilhelmus, 249, 259, 260 Pijpops, Dirk, 268 Plank, Frans, 142 Plungjan, Vladimir A., 68, 89, 96, 97, 100 Politt, Katja, 2, 3, 128 Pomino, Natascha, 129, 153 Popova, Gergana, 100 Pröll, Simon, 182 Radden, Günter, 297, 299–302, 310– 312, 315 Reiner, Tabea, 150, 154, 161 Remberger, Eva-Maria, 129, 153 Rey, Alain, 337 Rey-Debove, Josette, 337 Riegel, Martin, 326, 331 Ringgaard, Kristian, 182 Roberge, Paul T., 137 Rostila, Jouni, 52, 132 Rothstein, Björn, 163 Rumshisky, Anna, 136

Sapir, Edward, 15 Scalise, Sergio, 94 Schmid, Hans-Jörg, 268, 271 Schneider, Walter, 271 Schøsler, Lene, 320–322, 324–326, 329, 331, 334 Seržant, Ilja A., 69, 70 Shao, Bin, 273 Shapiro, Michael, 15 Shiffrin, Richard M., 271 Shore, Susanna, 54 Sigurðsson, Halldór Ármann, 204, 205, 218 Sihler, Andrew L., 27 Sims, Andrea D., 93, 94, 114, 117 Siro, Paavo, 56 Skautrup, Peter, 181 Šluinskij, Andrej B., 76 Smirnova, Elena, 3, 94, 100, 191, 247, 268, 272, 276, 289, 297, 299 Sommerer, Lotte, 268, 272 Spencer, Andrew, 96, 100, 133–137, 139–141, 143, 144, 153, 156, 158, 162 Stewart, Thomas W., 128, 140 Strudsholm, Erling, 329 Stump, Gregory T., 23, 33, 128, 134, 152 Tabor, Whitney, 41 Tardif, Twila, 300 Tatevosov, Sergej G., 76, 77, 89, 112 Te Winkel, Lamert Allard, 249, 250 Tiefenbach, Heinrich, 279 Todorović, Nataša, 81 Tomasello, Michael, 38 Torp, Alf, 181 Traugott, Elizabeth C., 7, 183, 204, 250, 267, 269, 273, 276, 277,

Trommer, Jochen, 149, 153, 157 Trousdale, Graeme,136,158, 204, 270, 289 Tuomikoski, Risto, 53 Van de Velde, Freek, 268 van den Toorn, Maarten C., 249 Van der Voort, Hein, 162 Van Valin, Robert D., 160 Verdam, J., 249 Verveckken, Katrien Dora, 269, 273 Verwijs, E., 249 Veselinova, Ljuba N., 84 Vilkuna, Maria, 43 Waltereit, Richard, 320 Werner, Martina, 161 Werner, Otmar, 2, 32 Wessén, Elias, 181, 183 Wiemer, Björn, 3, 28, 69, 70, 75, 81, 84, 107, 111, 112 Willems, Dominique, 324 Wood, Esther Jane, 76 Wurzel, Wolfgang U., 2 Zaliznjak, Andrej A., 86, 87

287, 289

Zúñiga, Fernando, 220 Zwitserlood, Pienie, 154

## Paradigms regained

The volume discusses the breadth of applications for an extended notion of paradigm. Paradigms in this sense are not only tools of morphological description but constitute the inherent structure of grammar. Grammatical paradigms are structural sets forming holistic, semiotic structures with an informational value of their own. We argue that as such, paradigms are a part of speaker knowledge and provide necessary structuring for grammaticalization processes. The papers discuss theoretical as well as conceptual questions and explore different domains of grammatical phenomena, ranging from grammaticalization, morphology, and cognitive semantics to modality, aiming to illustrate what the concept of grammatical paradigms can and cannot (yet) explain.