# Quantification and scales in change

Edited by Remus Gergel Jonathan Watkins

### Topics at the Grammar-Discourse Interface

Editors: Philippa Cook (University of Göttingen), Anke Holler (University of Göttingen), Cathrine Fabricius-Hansen (University of Oslo)

In this series:


# Quantification and scales in change

Edited by

Remus Gergel Jonathan Watkins

Gergel, Remus & Jonathan Watkins (eds.). 2020. *Quantification and scales in change* (Topics at the Grammar-Discourse Interface 7). Berlin: Language Science Press.

This title can be downloaded at: http://langsci-press.org/catalog/book/252 © 2020, the authors Published under the Creative Commons Attribution 4.0 Licence (CC BY 4.0): http://creativecommons.org/licenses/by/4.0/ ISBN: 978-3-96110-265-5 (Digital) 978-3-96110-266-2 (Hardcover)

ISSN: 2567-3335 DOI:10.5281/zenodo.3929261 Source code available from www.github.com/langsci/252 Collaborative reading: paperhive.org/documents/remote?type=langsci&id=252

Cover and concept of design: Ulrike Harbort Typesetting: Felix Kopecky Proofreading: Ahmet Bilal Özdemir, Amir Ghorbanpour, Andreas Hölzl, Brett Reynolds, Carla Bombi Ferrer, Hella Olbertz, Janina Rado, Jeroen van de Weijer, Jean Nitzke Fonts: Libertinus, Arimo, DejaVu Sans Mono Typesetting software: XƎLATEX

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

Storage and cataloguing done by FU Berlin

# **Contents**


# **Acknowledgments**

The current volume profited from the insights of many, first and foremost of course from the contributing authors. Importantly, the articles included herein have also undergone a thorough peer-review process in two steps, first, for the abstracts submitted to the conference, second for the manuscripts submitted to the volume. All manuscripts have subsequently been revised by the authors based on input received from two or more reviewers for each manuscript. We would thus like to express our gratitude to the following colleagues for reviewing for the conference or the resulting volume, and in some cases for both occasions: Nadine Bade, Sigrid Beck, Nora Boneh, Marco Coniglio, Ashwini Deo, Regine Eckardt, Matthias Eitelmann, Kai von Fintel, Chiara Gianollo, Dag Haug, Agnes Jäger, Łukasz Jędrzejowski, Martin Kopf-Giammanco, Krzyztof Migdalski, Gerhard Schade, Alexandra Simonenko, Augustin Speyer, Igor Yanovich, and Hedde Zeijlstra.

After the submission of the second and revised version of her paper and before the publication of the volume, our appreciated colleague and contributor Edit Doron sadly passed away. We thank Malka Rappaport-Hovav for assisting us with the final editorial steps of the corresponding first chapter of this volume.

Finally, we wish to thank the series editors Philippa Cook and Anke Holler for their advice as well as Felix Kopecky and Sebastian Nordhoff for their technical support in publishing the book with Language Science Press.

# **Editors' preface**

Remus Gergel Universität des Saarlandes

### Jonathan Watkins

Universität des Saarlandes

The articles for the current volume have emerged from presentations delivered at the second edition of *Formal Diachronic Semantics*, held at Saarland University, Saarbrücken from November 21–22, 2017. The conference featured key-note addresses by Ashwini Deo, Chiara Gianollo, and an integrated invited student presentation by Verena Hehl along with a fair amount of high-quality scholarly work that was accepted for presentation.

The majority of the contributions delivered to the conference revolved around topics of quantification and scales in the process of semantic change. This is a nice coincidence which we strove to incorporate in a volume. At the same time, it was also quite clear from the onset that quantification and scalarity in naturally attested case studies rarely appear as clear-cut as in idealized textbook trajectories. This led to the current compendium with Language Science Press for which a subset of the papers presented at the conference was submitted and included in accordance with standard review and revision procedures. In the remainder of this preface, we offer a brief outlook on what the readers can expect from the contributions contained within. While the order of the articles in the volume is alphabetical, we introduce the papers from a thematic point of view.

The papers by Doron, Jędrzejowski, Kellert and Simonenko & Carlier address – to varying degrees – issues pertaining to phenomena from language change, which are standardly treated in terms of quantification and therefore are hoped to be of interest, either in terms of data or analysis, to researchers concerned with the respective sub-topics, of which we will give a slightly longer outlook below.

### Remus Gergel & Jonathan Watkins

By contrast, contributions by Gianollo, Harris, and Kopf-Giammanco are primarily concerned with scalar structures of different kinds and how they have evolved over time. Before introducing the individual articles, it seems of note to mention that all of the contributions presented contain a fair amount of discussion on the interfaces of interpretation – be it through structural facts, the pragmatic component (including e.g. information-structural factors), or important morphological factors.

Doron's article addresses some of the fundamental questions in the trajectories which arise in the domain of universal quantification (see von Fintel 1995; Haspelmath 1995; Beck 2017) with a specific case study conducted on Biblical Hebrew. The general trajectory of meanings is roughly paraphrasable as 'all/any/ every' (notice, however, that the original Hebrew noun *kol* had both similarities, but also key differences from English *all*) and it is couched in terms of a cyclical view of language change (cf. van Gelderen 2011, see e.g. Gergel 2016 for an application of cyclical theory to issues of interpretation). While Doron's paper concentrates on the development from collective to distributive readings, it also sheds light on the so-called *distributivity cycle* as a whole and its role in the history of a language which is proposed originally not to have had any distributive determiner of the 'every' type (but other alternative mechanisms to express such meanings).

While Doron's focus is on the development of universal interpretations, Simonenko & Carlier's contribution incorporates the interaction of what they regard as a non-presuppositional existential inference with variation in the constituent order facts in the history of French. A key component of the analysis is a version of \*New > Given principle of Kučerová (2012). The authors attribute the principle to a binding configuration, specifically to how situation binding operates in clauses. The resulting account and the quantitative data obtained should be of interest to researchers in the diachrony of French syntax and how it could be modeled in tandem with pragmatic factors such as givenness. The diachronic span covered ranges from the twelfth to the seventeenth century.

Kellert's paper relates to the large topic of indefinites of indifference. She investigates the linguistic item called *equis* (x) in Mexican Spanish by giving a synchronic description and offering a diachronic suggestion. While the original meaning of the expression is that of the letter x, a discourse-related function is claimed to have appeared very recently. According to the discourse function, *equis* is used to refer to some utterance from the discourse which denotes a proposition and the speaker expresses her indifference as to whether this proposition is true or not. Descriptively, *equis* is claimed to have developed into a discourse adverb. The key idea beyond the diachronic analysis is that the language has undergone a shift in the domain of indifference associated with the word under

### 1 Editors' preface

discussion, namely from indifference with respect to the identities of entities towards indifference with respect to answers to questions under discussion. The latter meaning is claimed to be lexicalized, while the initial is taken to have been pragmatic. The syntactic correlate is suggested to reside in the reanalysis from a nominal modifier to a sentential one.

Jędrzejowski addresses the topic of modality, another domain classically treated as quantificational, namely over possible worlds, in semantic theory. His focus in the paper is on the morphosyntactic facts related to the appearance of an interesting clausal evidential marker in the history of Polish. The key argument is that the word *jakoby* developed from an original complementizer, with the meaning 'as if', into a hearsay complementizer. The case study laid out in the paper offers empirical evidence supporting the idea that the process happened around the 1500s, i.e. in the late Old Polish period. Jędrzejowski claims that the original presence of what he takes to be an equative comparison and the counterfactual meaning were the decisive factors in realizing the semantic reanalysis.

At the center of Giannollo's paper is the topic of scalar alternatives. She focuses on the Latin focus-sensitive negation *nec* ('furthermore not'/'neither'/'not even') and suggests a trajectory from a discourse-structuring particle with an additive component to a new emphatic (scalar) negative polarity item, which is later reanalyzed as an element of negative concord. The larger question in the background has to do with the issues of scope and the cyclicity of semantic change; cf. Lehmann's parameters of grammaticalization, which Gianollo re-evaluates with respect to semantic changes. The key proposal is tied to the way alternatives are retrieved from the context and the idea that an increase in bondedness and a decrease in syntagmatic variability correlate with a change in the form taken by alternatives, which decrease in scope from discourse units to individual alternatives.

Harris directly addresses classical degree scales and the issues posed by establishing the precise type of scale structure when it comes to the application of the affix -*ish* in English. By applying Burnett's (2017) framework situated within delineation semantics (cf. Cobreros et al. 2012) she proposes to account for the distribution of the affix. While certain attested corpus examples observed by Harris are left for further research, her main claim is that the relevant scale structure is derived from the adjective's context-sensitivity and vagueness patterns.

Kopf-Giammanco's article combines the issues of degree-scales (including, but not restricted to, the temporal ones) with the topic of presuppositions by focusing on the semantics of present day German *noch* in comparative readings (cf. Beck 2019 for a recent overview and synchronic analysis). He presents both experimental synchronic and diachronic corpus-based evidence from Old High German suggesting a reanalysis of the particular reading of *noch* under investigation based on temporal readings.

The assorted treatment of quantification across Biblical Hebrew, Old French, Mexican Spanish, and Polish, to issues of scalarity in Latin, English and German offers (researchers) first and foremost a broad linguistic research palate. By delving into quantification and scales the contributors to this volume shed light on both specific subfields as well as on the way interpretations can change. By addressing quantification from universal interpretations to modality and scalarity from scalar alternatives to degree scales, the volume lends itself nicely to anyone with research interests in semantic and pragmatic change at the interfaces. The papers in this volume help us once more to deepen our understanding of those particular languages. They show the diversity of a growing field and at the same time offer perspective towards the more general enterprise of understanding mechanisms of semantic change.

### **References**


### **Chapter 1**

# **From a collective to a free choice determiner in Biblical Hebrew**

### Edit Doron

Hebrew University of Jerusalem

The paper is a diachronic study of the Hebrew universal determiner *kol*. In Biblical Hebrew (BH), *kol* was originally a noun meaning 'entirety' which grammaticalized as a collective determiner akin to *all*. *Kol* induces maximality, like the determiner *all*, but, unlike *all*, it is not quantificational, hence its maximality does not preclude homogeneity. Semantically, *kol NP* is interpreted as the plural property corresponding to NP. In argument position, the strongest interpretation of *kol NP* results from the application of the definite type-shift (the *iota* type-shift). But within the scope of certain modals and in downward entailing environments, the indefinite typeshift (existential closure) yields a stronger interpretation. This results in the free choice (FC)/negative polarity (NPI) *any* interpretation of *kol* in these environments. In post-Biblical times, the *any* interpretation evolved into the distributive interpretation *every*. The paper thus traces the development of *kol*'s extensive meaning variation 'all/any/every'.

### **1 Introduction**

How does universal quantification develop in a language? Haspelmath (1995)suggests that collective universal determiners (like English *all*) often originate in an adjective meaning *entire/whole*, and that distributive universal determiners (such as English *every*) have various sources – free choice (FC) determiners like *any*, or collective universal determiners like *all*. The [FC → distributive] development was elucidated by Beck (2017), and here I would like to describe the [collective → distributive] development. I claim that at least for some languages, the latter development is a cycle which includes the former, as shown in (1) below:<sup>1</sup>

<sup>1</sup>The cyclical nature of (1) is due to its reversibility (cf. van Gelderen 2011 on the pervasive nature of cyclical change). *Every* in present-day English has completed the Distributivity Cycle and is acquiring a collective interpretation, as in *Everyone gathered in the hall*, by re-entering the cycle.

Edit Doron. 2020. From a collective to a free choice determiner in Biblical Hebrew. In Remus Gergel & Jonathan Watkins (eds.), *Quantification and scales in change*, 1–31. Berlin: Language Science Press. DOI:10.5281/zenodo.3929237

### Edit Doron

(1) collective universal determiner → FC determiner → distributive universal determiner

In (2) I add the original first step, where an Adj/Noun meaning *entire(ty)* evolves into a collective universal determiner:

(2) The Distributivity Cycle

Adj/Noun *entire(ty)* <sup>I</sup> <sup>→</sup> collective univ. det. II <sup>→</sup> FC det. III <sup>→</sup> distributive univ. det.

I will motivate the Distributivity Cycle on the basis of the history of Hebrew. Steps I + II took place in Biblical Hebrew (BH): The BH noun *kol* 'entirety' grammaticalized as the collective determiner *all*, and did not have a distributive meaning other than as a free choice (FC)/negative polarity (NPI) determiner akin to *any*. Modern Hebrew (and probably much earlier) underwent step III, whereby FC *kol* also came to have the universal distributive meaning *every*. <sup>2</sup> The present analysis thus accounts for the surprising array of interpretations 'all/any/every' of *kol* in Modern Hebrew without alleging that *kol* is existential rather than universal (Bar-Lev & Margulis 2013).<sup>3</sup>

The structure of the paper is the following. Section 2 shows that BH should be classified as a NP (rather than a DP) language. Section 2.1 argues that BH had no definite (or indefinite) determiner. Section 2.2 demonstrates that the determiner *kol* was originally a noun – it had both the morphology and the distribution of other nouns in the language. *kol* was often found heading the pseudopartitive construction, and accordingly underwent an independent-to-functional meaning-shift which grammaticalized it as the determiner *all*. Section 3 discusses the semantic properties of the determiner *kol*. Section 3.1 shows that it was not

<sup>2</sup>Hence Hebrew conforms to Haspelmath's view on the direction of development from 'any' to 'every' rather than the other way round, despite his own assessment of Hebrew as a counterexample (Haspelmath 1997:156 fn.13).

<sup>3</sup>The existential analysis of *kol* in Modern Hebrew was applied to the structure *kol NP* with a predicate NP. The partitive *kol DP* is undisputedly universal in Modern Hebrew, casting doubt on the existential analysis of *kol*. I return to Modern Hebrew at the end of the article, in Section 6. For now, I note that the root *kll* of *kol* (and the related roots *klkl, kwl, kly*) derive a plethora of nouns and verbs denoting completeness, containment, inclusiveness and generality. Biblical Hebrew has *kālā* 'to complete (intrans.)', *killa* 'to complete (trans.)', *kalīl* 'completely', *hēḵīl* 'to contain', *klī* 'container', *kāl* 'to measure', *kilkēl* 'to contain/sustain'. Later periods innovated *klal* 'whole', *klali* 'general', *biḵlal* 'at all', *miḵlol* 'ensemble', *tḵula* 'content', *kalal* 'to include', *kolel* 'including', *hiḵlil* 'to generalize', *haḵlala* 'generalization'. Not a single noun or verb derived from *kll* in any period of Hebrew has an existential interpretation. These factors have motivated the analysis of *kol* as universal (Doron & Mittwoch 1986, Glinert 1989, Francez & Goldring 2012, Danon 2013).

### 1 From a collective to a free choice determiner in Biblical Hebrew

distributive – it was never interpreted as *every*. Sections 3.2 and 3.3 discuss maximality and homogeneity, and show that *kol*'s homogeneity did not result in the lack of maximality which would be expected by Križ (2016). Section 4 describes the operator *each* which was responsible for distributivity in BH. Section 5 discusses the emergence of the free choice (FC) interpretation of *kol* within the scope of certain modal operators. Section 6 briefly relates the post-Biblical development whereby the FC reading gave rise to a distributive reading. This development is not elaborated in the present paper, relying on Beck (2017). Section 7 is the conclusion.

### **2 Biblical Hebrew as a NP-language**

Biblical Hebrew (BH) did not have a distributive universal determiner. This has been claimed for other languages as well, e.g. Salish (Jelinek 1993; Davis 2010; Davis et al. 2014; von Fintel & Matthewson 2008; Matthewson 2001; 2014). Yet BH did not just lack a distributive universal determiner, but other determiners as well. According to the typology of Bošković (2008), BH is an NP-language (in contrast to DP-languages). To derive the interpretation of NPs in argument position, BH makes use of type-shifts, in particular the *definite type-shift* (the *iota* type-shift) and the *indefinite type-shift* (existential closure). This accords with the fact that BH is a language without either a definite determiner or an indefinite determiner, and hence relies on the corresponding type-shifts instead. This is the topic of the next subsection.

### **2.1 The BH definite article as an inflectional prefix**

As argued by Doron & Meir (2013; 2016), the Hebrew article *han-*, though glossed as *the-*, is historically not a D but a word-level inflectional prefix.<sup>4</sup> It does not mark definiteness – which is a phrase-level category, but *state* – which is a wordlevel category. The article marks nouns (and adjectives) as being in the *emphatic state*. The emphatic state alternates with the other two values of the state category: the unmarked *absolute state* and the *construct state*, which marks the noun as relational/possessee.<sup>5</sup> A noun in the emphatic state projects its emphaticity

<sup>4</sup> See Rubin (2005): 65 for the history of the article *han-*. Phonological processes delete its final /*n*/, resulting in the prefix *hā-*, or assimilate /*n*/ to the first consonant of the ensuing noun.

<sup>5</sup>The term *emphatic* in 'emphatic state' is a Semiticists' term, used mostly in descriptions of Aramaic, marking a particular value of the inflectional state of a noun and is unrelated both to the phonological term *emphatic* in the sense of *stressed* and to the phonetic term *emphatic* in the sense of *pharyngealized*. The *emphatic state* form of N will be glossed as 'the-N' in the examples below, and the *construct state* – as 'N(of)'.

### Edit Doron

value to containing NPs, and eventually results in its maximal NP projection being interpreted as definite, through the definite type-shift to x.⟦NP⟧(x).<sup>6</sup> In the simplest case, an unmodified emphatic N forms an emphatic NP by itself, and is interpreted as definite. For example the noun *water* in (3a) is also a maximal NP, hence its prefixation by *han-* is understood as definite: *the water*. On the other hand, the noun *water* in (3b) is not a maximal NP but part of a larger NP. Accordingly, its prefixation by *han-* marks it as emphatic, not as definite. It is its emphatic NP projection *well of water* which is interpreted as definite, not *a well of the water* but *the well of water*: 7

	- b. *hinnē* prstv *ʔānōḵī* I *niṣṣaḇ* stand.ptcp.ms *ʕal* at *ʕēn* well(of) *ham-māyim* the-water 'Behold, I stand by the well of water.' (Gen. 24:43)

In contrast, an absolute-state NP is unmarked for definiteness. It is typically interpreted as indefinite as in (4):

(4) *way.yēlɛḵ* and.went.3ms *way.yimṣāʔ-ēhū* and.met.3ms-acc.3ms *ʔaryē* lion.ms *b-ad-dɛrɛḵ* in-the-road *way.yəmīṯ-ēhū* and.killed.3ms-acc.3ms 'When he was gone, a lion met him on the road and killed him.' (1Kings 13:24)

The absolute-state subject *lion* of the main clause of (4) denotes the predicate λx.lion(x). This predicate can combine with the clause's predicate λx.P(x) by predicate modification: λx.lion(x) & P(x). The truth value of the sentence is calculated by applying the indefinite type-shift (existential closure): ∃x.lion(x) & P(x).

<sup>6</sup> x.P(x) is the maximal individual satisfying P, defined both for singular and plural predicates (Sharvy 1980).

<sup>7</sup>Unless stated otherwise, all Biblical translations are from the New King James Version (NKJV). The pairs of allophones *b-β, g-ɣ, d-ð, k-x, p-f, t-θ,* are transcribed according to the Hebraist transcription *b-ḇ, g-ḡ, d-ḏ, k-ḵ, p-p̄, t-ṯ.* Three vowel qualities are distinguished, in accordance with the Tiberian tradition, e.g. *ā* vs. *a* vs. epenthetic *ă*.

### 1 From a collective to a free choice determiner in Biblical Hebrew

But since an absolute-state NP is unmarked, it can on principle also be interpreted as definite. The definite interpretation is normally thwarted by the principle of *maximize presupposition* (Heim 1991), which would favour the use of an emphatic-state NP to indicate definiteness. Yet there are special cases. An absolute-state NP may be interpreted as definite when the property it denotes holds of a unique entity by virtue of its meaning. This is the case of kind-names (Doron 2003), as in (5), or NPs headed by *kol*, as in (6), to which we return in Section 3.

	- (2Sam. 6:1)

### **2.2 The BH pseudo-partitive construction**

Pseudo-partitives, also called measure constructions, denote an amount (a particular degree of a measure function) of some substance (Selkirk 1977). In Hebrew, the substance is denoted by an indefinite NP complement of the determiner. The indefinite substance-denoting NP may be in the absolute state (as in the (a) examples below) or in the emphatic state (as in the (b) examples below) since emphaticity does not mark the substance NP but the whole construction as definite. The head of the construction is a degree N which partitions the substance into portions (Schwarzschild 2002; Ruys 2017): (7) partitions days/commandments into groups of ten, (8) and (9) partition the substance into small/large groups respectively. (10) partitions the craftsmen into groups consisting of all the craftsmen;

### Edit Doron

since there is only one such group, the absolute version in (10a) and the emphatic version in (10b) both denote a unique group:8,9

	- b. *ʕăśɛrɛṯ* ten(of) *had-dəḇārim* the-commandments 'the ten commandments' (Exod. 34:28)
	- b. *məʕaṭ* few(of) *haṣ-ṣōn* the-sheep *hā-hēnnā* the-those 'those few sheep' (1Sam. 17:28)
	- b. *rōḇ* many(of) *ziḇħē-ḵɛm* sacrifices-poss.2mp 'the multitude of your sacrifices' (Isa. 1:11)

<sup>8</sup>Accordingly, *kol NP* is often overtly case-marked in object position by the accusative *ʔɛṯ* which marks definite direct objects, even when NP is headed by a noun in the absolute state. This was already shown in (6) above, and is shown again here in (i) and (ii):(i) *way.yōmɛr ʔɛlōhīm hinnē nāṯattī lāḵɛm ʔɛṯ kol ʕēśɛḇ zōrēaʕ zɛraʕ* and.said.3ms God prstv gave.1s to.2mp acc kol herb.ms seed.ptcp.ms seedAnd God said, See, I have given you every herb that yields seed. (Gen. 1:29)(ii) *way.yaħărīm ʔɛṯ kol nɛp̄ɛš ʔăšɛr bah* and-destroyed.3mp acc kol soul.fs that in.3fsand destroyed all the people who were in it (Josh. 10:39)

<sup>9</sup>The BH *kol NP* is indeed a pseudo-partitive rather than a partitive construction where NP denotes an individual. Though the complement may be a name, as in *kol yiśrāʔēl* 'all Israel' (1Kings 12:20), *kol miṣrāyim* 'all Egypt' (Gen. 41:55), the name in this position never denotes an individual but a set of people, i.e. 'all Israelites', 'all Egyptians'. To express the totality of the geographic entity, the name has to be explicitly modified so as to clarify what kind of portions are being measured: *kōl ʔɛrɛṣ yiśrāʔēl* 'all the land of Israel' (1Sam. 13:19), *kol ʔɛrɛṣ miṣrāyim* 'all the land of Egypt' (Ex. 9:9).

1 From a collective to a free choice determiner in Biblical Hebrew

b. *kol* kol(of) *hā-ħăḵāmīm* the-skilled.mp 'all the craftsmen' (Ex. 36:4)

### **3 The determiner** *kol*

As just shown in (10), *kol* functions as a degree N which heads the pseudo-partitive construction; it denotes the *entirety* degree. The distribution of *kol* indicates that it originally was a noun. Indeed, traditional grammars of the Bible describe *kol* as an "abstract substantive denoting totality" (Joüon 1923: §139e). It occurs in the Bible not only in the construct-state form as in (10) above, but also in the absolute and emphatic states, as in (11) and (12) below. In these forms, *kol*'s vowel is not shortened as it often is in the construct state (cf. *kol* in 10), but is rather a long /*ō*/, as in *kōl* in (11) and (12):

	- b. *kī* because *ħann-ani* favoured.3ms-acc.1s *ʔɛlōhīm* God *wə-ḵī* and-because *yɛš* exst *lī* to.1s *ḵōl* kol 'for God has been generous to me and I have all I need' (Gen. 33:11)
	- b. *wa-yhwh* and-Lord *bēraḵ* blessed.3ms *ʔɛṯ* acc *ʔaḇrāhām* Abraham *b-ak-kōl* in-the-kol 'and the Lord had blessed Abraham in all things' (Gen. 24:1)

The nominal origin of *kol* is also evident in examples where it is still interpreted as the noun 'totality', e.g. when it heads the event-nominalization *count* in (13):

(13) *kol* kol(of) *mispar* count(of) *rāšē* chiefs(of) *hā-ʔāḇōṯ* the-officers *…* … *ʔalp-ayim* thousand-dual *wə-šēš* and-six *mēʔ-ōṯ* hundred-pl 'The total number of chief officers … was two thousand six hundred.' (2Chr. 26:12)

### Edit Doron

I reiterate that the translations of the Biblical verses are not my own, but are received translations, mostly from the New King James Version (NKJV). The translations are faithful to the meaning of each verse as a whole, but cannot be used to gauge the various nuances of the meaning of *kol* or other lexical items.

### **3.1 Non-distributivity of** *kol*

The present subsection demonstrates that *kol NP* is not quantificational/distributive. It denotes the entirety of a (group) individual rather than quantifying over its members/parts.

The first piece of evidence for the non-quantificational nature of *kol NP* is the possibility of predicating cardinality of it, unlike the English *all NP*, of which cardinality cannot be predicated. *All NP* contrasts in this respect with definite *NP*s: *The apostles were twelve*/\**All the apostles were twelve* (Dowty 1987; Winter 2002). In BH we find cardinals predicated of *kol NP*: 10

(14) *kol* kol *han-nɛp̄ɛš* the-soul.fs *lə-ḇēṯ* of-house(of) *yaʕăqōḇ* Jacob *hab-bāʔā* the-go.ptcp.fs *miṣraym-ā* Egypt-ill *šiḇʕīm* seventy 'All the persons of the house of Jacob who went to Egypt were seventy.' (Gen. 46:27)

Second, as shown in (15), *kol NP* does no distribute over another argument in the clause. For example, (15a) is unlike English and other languages, where the universal subject scopes in two different ways relative to the object, yielding ambiguity in *All the artisans made ten curtains*.

	- b. *yōm* day *la-yhwh* to-Lord *ṣəḇāʔōṯ* Sabaoth *ʕal* for *kol* kol *gēʔɛ* proud *wā-rām* and-lofty 'The Lord Almighty has a day in store for all the proud and lofty.' (NIV, Isa. 2:12) (non-distributive only)

I am not aware of examples like (15) where *kol NP* distributes over another argument.

<sup>10</sup>For the sake of brevity I will henceforth mostly use the gloss kol rather than kol(of).

### 1 From a collective to a free choice determiner in Biblical Hebrew

Third, even when its complement NP is singular, *kol NP* denotes the entirety of a group and functions as subject of collective predicates, unlike other languages where NPsing only cooccurs with distributive *every*:

	- b. *way.yiqqāhălū* and.assembled.3mp *ʔɛl* to *ham-mɛlɛḵ* the-king *šəlōmō* Salomon *kol* kol *ʔīš* man(of) *yiśrāʔēl* Israel 'Therefore all the men of Israel assembled with King Solomon.' (1Kings 8:2)
	- c. *wə-ʔēlay* and-to.1s *yēʔāsp̄ū* congregated.3mp *kol* kol *ħārēḏ* tremble.ptcp.3ms *bə-ḏiḇrē* at-words(of) *ʔɛlōhē* God(of) *yiśrāʔēl* Israel 'Then everyone who trembled at the words of the God of Israel assembled to me.' (Ezra 9:4)

In other examples with NPsing, *kol NP* denotes the entirety of an individual: *the whole NP/all the NP.*<sup>11</sup>

	- b. *kol* kol *rōš* head *lā-ħŏlī* in-sickness *wə-ḵol* and-kol *lēḇāḇ* heart.ms *dawwāy* faint.ms 'The whole head is sick and the whole heart faints.' (Isa 1:5)

NP may also be an absolute-state mass term:<sup>12</sup>

(i) *ʔim* if *yittɛn* give.mod.3ms *lī* to.1s *ḇālāq* Balak *məlō* fullness(of) *ḇēṯ-ō* house-poss.3ms *kɛsɛp̄* silver *wə-zāhāḇ* and-gold 'If Balak gave me his house full of silver and gold …' (MEV, Num. 22:18)

<sup>11</sup>These examples argue against Naudé's (2011b) account of *kol*, which consists in translating *kol* as *every* with NP+count−def and as *all* with NP±count+def. Naudé's account is mistaken for (17). Moreover, it is incompatible with the lack of distributive interpretation of NP+count−def in (15) and (16): we would expect distributivity with *every*. Naudé's account ignores *kol* applied to NP−count−def as in (18), which Naudé claims does not exist (2011b: 418), and also ignores all examples where *kol* can be translated as neither *all* nor *every*, cf. Section 3.3 below.

<sup>12</sup>The nouns *gold* and *silver* are mass nouns in BH, just as they are in Modern Hebrew and in English, since they do not pluralize, and, though singular, denote space-filling substance:

### Edit Doron

(18) *wə-ḵōl* and-kol *kɛsɛp̄* silver.ms *wə-zāhāḇ* and-gold.ms *ū-ḵlē* and-vessels(of) *nəħōšɛṯ* bronze *u-ḇarzɛl* and-iron *qōḏɛš* sacred.ms *hū* pron.3ms *la-yhwh* to-Lord 'But all the silver and gold, and vessels of bronze and iron, are consecrated to the Lord.' (Josh. 6:19)

Fourth, verbal agreement provides additional evidence for the lack of distributivity of *kol NP*. If *kol* were distributive, we would expect *kol NP*sing to strictly agree in the singular like *every* and unlike *all* (which agrees either in the plural or the singular). Yet irrespective of the number marking of NP, verbal agreement is often plural, even for singular NP. Example (19) shows plural agreement when NP is plural, as is to be expected. (20) shows the same plural agreement when NP is singular. The relevant NPs are in the absolute state in the (a) examples, and in the emphatic state in the (b) examples:

	- b. *kol* kol *hā-ʔɛzrāħ* the-native.ms *bə-yiśrāʔēl* in-Israel *yēšḇū* sit.mod.3mpl *b-as-sukkōṯ* in-the-booths 'All who are native Israelites shall dwell in booths.' (Lev. 23:42)

If *kol* were distributive, it would be unexpected for *kol NP*sing to co-occur with Vpl in (20a), unless we think that Biblical subject-verb agreement is haphazard: there indeed are many other examples where *kol NP* cooccurs with Vsing. But in fact these are all *kol NP*sing. There are no examples where *kol NP*pl cooccurs with Vsing. <sup>13</sup> This agreement pattern is actually systematic under the assumption that *kol NP* is collective and may hence be marked as plural [kol NP]pl independently

<sup>13</sup>I exclude irrelevant examples such as left-conjunct agreement (Doron 2005), passive verbs, and verbs where the subject of Vsing is actually not *kol NP*pl but a null expletive as in (i):

### 1 From a collective to a free choice determiner in Biblical Hebrew

of the number feature of NP. Accordingly, Vsing only cooccurs with *kol NP*sing, whereas Vpl cooccurs both with [*kol* NPsing]pl and [*kol* NPpl]pl. 14

Lastly, it is important to distinguish distributivity from what has been called *lexical distributivity* (Winter 2000), which is due to the lexical nature of the predicate. E.g. *weeping* in (21) below can only be predicated of a group by attributing it to the individual members of the group.<sup>15</sup> Lexical distributivity does not induce scopal ambiguity (de Vries 2017) and is not mediated by quantifiers.

(21) *wə-ḡam* and-also *ham-mɛlɛḵ* the-king *wə-ḵol* and-kol *ʕăḇāḏ-āw* servants-poss.3ms *baḵū* wept.3mp *bəḵī* weeping *gāḏōl* big *məʔōḏ*

very

'Also the king and all his servants wept very bitterly.' (2Sam. 13:36)

I conclude that *kol* is not quantificational. Rather, *kol* applies to a NP which denotes substance, mass or count, singular or plural, and yields a portion of the NP substance that consists of the entirety of those individuals whose parts satisfy NP. Hence *kol* maps a predicate P to the set of all individuals, atoms or sums, satisfying \*P.16,17

(22) ⟦kol⟧ = P. x. \*P(x)

In argument position, the predicate *kol NP* is given a definite interpretation as the maximal individual ιx.*kol*⟦*NP*⟧(x) satisfying it.<sup>18</sup>

(i) *wa.yəhī* was.3ms *kol* kol *han-nō p̄līm* the-fall.ptcp.mp *b-ay-yōm* in-the-day *ha-hū* the-that *mē-ʔiš* of-man *wə-ʕaḏ* and-including *ʔiššā* woman *šənēm.ʕāśār* twelve *ʔālɛp̄* thousand 'So it was that all who fell that day, both men and women, were twelve thousand' (Josh. 8:25)

<sup>14</sup>Under Naudé's (2011b) account, the agreement pattern remains mysterious. <sup>15</sup>Lexical distributivity can be averted by the use of collective adverbs such as *together*, e.g.


<sup>17</sup>I assume that the absolute/emphatic *kōl* in (11)/(12) above combines with a null P which spans the entire relevant domain.

<sup>18</sup>In a downward entailing environment, the definite interpretation is disfavoured, as it is weaker than the indefinite (existential closure) interpretation. We return to this below in Section 3.3.

### Edit Doron

### **3.2 Maximality of** *kol*

We have seen that *kol* does not contribute distributivity. So what does it contribute? Why say 'all the men' rather than simply 'the men', if it is not for the purpose of allowing distributivity?

The answer seems to be that *kol NP* denotes the sum of *all* parts of NP, including absolutely all of them (Brisson 1997; 2003). This is illustrated by the following example, which demonstrates that tearing away the kingdom is compatible with not tearing away all the kingdom:

(23) *qārōaʕ* tear.inf *ʔɛqraʕ* tear.mod.1s *ʔɛṯ* acc *ham-mamlāḵā* the-kingdom *mē-ʕālɛḵā* from-over.2ms *…* … *raq* but *ʔɛṯ* acc *kol* kol *ham-mamlāḵā* the-kingdom *lō* neg *ʔɛqraʕ* tear.mod.1s *–* – *šēḇɛṭ* tribe.ms *ʔɛħāḏ* one.ms *ʔɛtēn* give.mod.1s *li-ḇn-ɛḵā* to-son-poss.2ms

'I will surely tear the kingdom away from you … However, I will not tear away the whole kingdom; I will give one tribe to your son.' (1Kings 11:13)

*kol* disallows the slack allowed by *thepl* (Krifka 2006; Lasersohn 1999; Schwarz 2013). Lasersohn characterizes *slack* as pragmatic looseness which involves approximation to the truth that does not affect truth conditions. When speaking loosely, the speaker takes it to be unlikely that the (possible) difference between the actual world and his assertion is relevant for present purposes. To adapt an example of Lauer (2012), *I live in Tel-Aviv* is true in a context where the speaker lives in Jaffa, which abuts Tel-Aviv, but is not part of it. Various expressions, such as *proper*, are seen as *slack regulators* in this respect. *I live in Tel-Aviv proper* cannot be used with slack: it is never appropriate if the speaker lives in Jaffa.

The plural definite *thepl* displays pragmatic slack: it makes a sentence such as *The boys smiled* true even if there are some exceptions, assuming those exceptions do not matter for present purposes. The role of *all*, on this view, is that of a slack regulator. It disallows the flexibility permitted by the plural definite *thepl*. This view integrates what Dowty (1987) called the 'maximizing effect' of *all. All the boys smiled* is interpreted maximally.

Winter (2001) attributes the maximality of *all* to its being quantificational. Winter shows that (24a) but not (24b) is entailed by (25):

	- b. All the members of the organizing committee met.

### 1 From a collective to a free choice determiner in Biblical Hebrew

(24a) has a reading equivalent to (25). Under this reading the denotation of the definite *the members of the organizing committee* is mapped to a group individual representing the committee itself. Such a process is impossible in (24b), where the only way to achieve collectivity is to use quantification which requires every committee member in (24b) to participate in the meeting.

In BH, the maximality of *kol* is not due to quantification over individuals, since *kol* is not quantificational. Rather, the maximality of *kol* is a consequence of measurement as expressed by the pseudo-partitive construction. Measuring an individual requires taking into account its full extent, preventing non-maximality.<sup>19</sup>

### **3.3 Homogeneity of** *kol*

In dictionaries and traditional grammars of Biblical Hebrew, *kol* is translated as *all* (sometimes as *every*, mistakenly in my view). But in addition, these sources mention that in combination with negation, *kol* is interpreted as *none at all* (rather than *not all*). Hence, it seems to exhibit what has been called *polarity* (Löbner 2000) or *homogeneity* (recently Križ 2016), which is surprising, since this phenomenon is said to be incompatible with the maximality of *all* (Križ 2016 argues that maximality is the by-product of lack of homogeneity).

### **3.3.1 The puzzle**

*Homogeneity* is a property of plural predication which requires that a plurality not be mixed with respect to the property predicated of it (or its negation). For (26a) below to be true, the subject must have reacted to all the external stimuli. For (26b) to be true, the subject must have reacted to none of the external stimuli. In mixed scenarios, where the subject reacted to some but not all of the stimuli,


<sup>19</sup>I therefore beg to differ from one passage in the medieval Rabbinic exegetical literature (Assaf 1929: 245), where the maximality of *kol* is disputed, and it is argued that *kol* only gives rise to an existential commitment. The problem is the apparent contradiction between two verses in Chapter 9 of the book of Exodus, the first describing the extinction of all Egyptian livestock by the plague, and the second – Moses' subsequent words to the Pharaoh, which presuppose that not all the livestock had perished.

### Edit Doron

neither (26a) nor (26b) is true. These scenarios are what Križ calls an "extension gap", where (26a) and (26b) are neither true nor false:

	- b. The subject did not react to the external stimuli.

Homogeneity is also found with measure phrases, as in the following English examples from the web.

	- b. It said it had friction modifier already in it so I didn't add the bottle of motorcraft modifier.

Homogeneity disappears in English in the presence of *all*. In (28), if the subject reacted to some but not all of the stimuli, (28a) is simply false and (28b) is true.

	- b. The subject did not react to all the external stimuli.

It is therefore surprising that in BH, sentences with *kol* do exhibit homogeneity. In BH, negating a sentence with *kol* does not yield 'not all' but 'none at all', i.e. 'not any'.

(29) a. *wə-ḵōl* and-kol *śīaħ* plant(of) *haś-śāḏɛ* the-field *ṭɛrɛm* still.not *yihǝyɛ* be.mod.3ms *ḇ-ā-ʔārɛṣ* in-the-earth *wə-ḵōl* and-kol *ʕēśɛḇ* herb(of) *haś-śāḏɛ* the-field *ṭɛrɛm* still.not *yiṣmāħ* grow.mod.3ms 'before any plant of the field was in the earth and before any herb of the field had grown' (Gen. 2:5) ≠ 'before all plants of the field were in the earth and before all herbs of the field had grown' b. *kol* kol *ʔăšɛr* that *lō* neg *yāḏəʕū* knew.3mp *ʔēṯ* acc *kol* kol *milħămoṯ* wars(of) *kənāʕan* Canaan 'all who had not experienced any of the wars in Canaan' (Judg. 3:1) ≠ 'all who had not experienced all of the wars in Canaan' c. *lō təḇaʕărū ʔēš bə-ḵōl mōšḇōṯ-ēḵɛm bə-yōm*

neg kindle.mod.2mp fire in-kol dwellings-poss.2mp on-day(of) *haš-šabbāṯ* the-Sabbath 'You shall kindle no fire throughout your dwellings on the Sabbath day. ' (Ex. 35:3) ≠ 'You shall not kindle fire throughout all your dwellings.'

### 1 From a collective to a free choice determiner in Biblical Hebrew

There is a well-known dialogue in the story of the Garden of Eden, where the snake queries Eve as in (30). Her answer starts by denying that she and Adam had been forbidden from eating any of the fruit of the garden, thus indicating that she interprets the snake's use of *kol* as involving homogeneity:<sup>20</sup>

(30) *ʔap̄* indeed *kī* indeed *ʔāmar* said.3ms *ʔɛlōhīm* God *lō* neg *tōḵlū* eat.mod.2mp *mik-kōl* from-kol *ʕēṣ* tree(of) *hag-gān* the-garden 'Has God indeed said "You shall not eat of any tree of the garden?"' (MEV, Gen. 3:1)

According to Križ (2016), maximality derives from lack of homogeneity, whereas here we see that the maximality of *kol* is compatible with its homogeneity. A parallel puzzle in English is mentioned by Križ (2016: 515), where maximality does not depend on lack of homogeneity. His example is of definite plurals with numerals. These plurals are homogeneous in English, but are only interpreted maximally, e.g. *The six professors smiled* requires all of them to have smiled. Interestingly, the syntax of such plurals in BH parallels that of *kol NP*. Both have the structure *N(of) NP* where N functions as a degree determiner and NP is indefinite irrespective of its emphatic marking (as emphaticity marks the whole construction as definite rather than the complement NP):

	- b. *kōl* kol(of) *yəmē* days(of) *ħayy-āw* life.pl-poss.3ms 'all the days of his life' (1Sam. 7:15)

The structure in (31) is that of the pseudo-partitive discussed above in Section 2.2. In English too, definite plurals with numerals are not interpreted like other definite plurals. A definite plural does not presuppose anything beyond existence; in particular it does not presuppose uniqueness. A definite plural with the numeral *six* presupposes contextual uniqueness of a group individual with the measure *six*. The phrase *the six working days* is interpreted as the unique individual in the context of a week which has measure *six* out of the substance

<sup>20</sup>Other translators, for example the NKJV, consider *kol* here to be focused, and hence translate 'Has God indeed said, "You shall not eat of *every* tree of the garden"?'.

### Edit Doron

*working days*. Accordingly, the English *the six working days* is a pseudo-partitive, i.e. a measure phrase, just like the BH (31a). The denotation of the relevant degree is given in (32), where #x denotes the number of atoms that the individual x consists of.

(32) ⟦*šēšɛṯ*⟧ = P. x. \*P(x) & #x = 6

As in the case of *kol*, measurement is what guarantees maximality despite homogeneity, both in Hebrew and in English. Unlike the case of *kol*, x.⟦*šēšɛṯ NP*⟧(x) is not necessarily defined (unless the cardinality of NP is 6).<sup>21</sup>

### **3.3.2 An account of homogeneity**

As was shown in Section 3.1, *kol NP* is a predicate, hence there are two ways of combining it with the sentence predicate VP which is of the same type. One way, represented in (33a), is to type-shift *kol NP* to type *e* by applying the definite type-shift. The other way, represented in (33b), involves combining *kol NP* with VP via predicate modification, followed by the application of the indefinite typeshift (existential closure).

In general, the stronger interpretation is the definite interpretation in (33a). But in a downward entailing environment, e.g. negation, (33b) is stronger. If no

(i) *šēšɛṯ* six(of) *yāmīn* days *taʕăḇōḏ* work.mod.2ms *u-ḇ-ay-yōm* and-in-the-day *haš-šəḇīʕī* the-seventh *tišbōṯ* rest.mod.2ms 'Do your work in six days and rest on the seventh day.' (CEV; Exod. 34:21)

<sup>21</sup>When the complement NP of the numeral is in the absolute state, the measure phrase is interpreted as indefinite. Moshavi & Rothstein (2018) attribute the "durational measuring phrase" interpretation of such phrases, e.g. (i) below, to indefiniteness. Yet definite measure phrases are also attested, such as (31a).

### 1 From a collective to a free choice determiner in Biblical Hebrew

element of NP satisfies VP, then neither does the maximal element. But if the maximal element of NP does not satisfy VP, this does not entail that no element of NP does.

The two type shifts available for the derivation of a sentence with *kol NP*, coupled with the *stronger meaning hypothesis*: Pick the stronger meaning (Dalrymple et al. 1994), predict the following:<sup>22</sup>

(34) a. Definite type-shift (33a) in non-downward-entailing environments b. Indefinite type-shift (33b) in downward-entailing environments

Indeed the indefinite type-shift is attested in downward entailing environments, including, besides negation, other downward entailing environments as well. Indefinite type-shifted *kol NP* can thus be interpreted as a negative polarity item (NPI).<sup>23</sup>

(35) negation


<sup>22</sup>The analysis follows Krifka (1996) (also Malamud 2012; Spector 2018), where plural referential expressions are interpreted as universal/existential on the basis of the stronger meaning hypothesis.

<sup>23</sup>Raising *kol NP* out of the downward-entailing environment and interpreting it by the definite type-shift (33a) does not yield the right truth conditions in the question example in (38), and is impossible in (41) because of the island nature of the conditional protasis. Hence there is no way of forgoing the indefinite type-shift (33b).

	- a. *nɛp̄ɛš* soul.fs *ʔăšɛr* that *tiggaʕ* touch.mod.3fs *bə-ḵol* at-kol *dāḇār* thing *ṭāmē* unclean *…* … *wə-hū* and-he *ṭāmē* unclean *wə-ʔāšēm* and-guilty

'a person who touches any unclean thing… he shall be unclean and guilty' (NET, Lev. 5:2)

≠ 'a person who touches all unclean things… he shall be unclean and guilty'


*kol* kol *nɛp̄ɛš* soul.fs *ʔăšɛr* that *tōḵal* eat.mod.3fs *kol* kol *dām* blood *wə.niḵrəṯā* and.will.be.cut.off.3fs *han-nɛp̄ɛš* the-soul.fs *ha-hī mē-ʕamm-ɛhā*

the-that.fs from-people-poss.3fs

'Whoever eats any blood – that person will be cut off from his people.' (Lev. 7:27)

≠ 'Whoever eats all the blood – that person will be cut off from his people.'

(38) question

*hinnē* prstv *ʔănī* I *yhwh* Lord *ʔɛlōhē* God(of) *kol* kol *bāśār* flesh *– – hā-mimɛnni* Q-from.1s *yippālē* be.beyond.ability.mod.3ms *kol* kol *dāḇār* thing 'Behold, I am the Lord... Is there anything too hard for Me?' (Jer. 32:27) ≠ 'Are all things too hard for Me?'

(39) complement of adversative verbs *wə-šōmēr* and-keep.ptcp.3ms *yāḏ-ō* hand-poss.3ms *mē-ʕăśōṯ* from-do.inf *kol* kol *rāʕ* evil '… and keeps his hand from doing any evil ' (Isa. 56:2) ≠ '… and keeps his hand from doing all evil things'

(40) *before*-PPs

*ʕōḏ-ɛnnū* still-3ms *ḇə-ʔibb-ō* in-green-poss.3ms *lō* neg *yiqqāṭēp̄* cut.mod.3ms *wə-lip̄nē* and-before *ḵol* kol *ħāṣīr* plant *yīḇāš* wither.mod.3ms 'While it is yet green and not cut down,it withers before any other plant.' (Job 8:12) ≠ 'While it is yet green and not cut down,it withers before all other plants.'

	- a. *ʔim* if *yiggaʕ* touch.mod.3ms *ṭəmē* unclean(of) *nɛpɛš* dead.body *bə-kol* at-kol *ʔēllɛ* these *ha-yiṭmā* Q-be.unclean.mod.3ms

'If one who is unclean touches any of these, will it be unclean?' (Hag. 2:13)

≠ 'If one who is unclean touches all of these, will it all be unclean?'

b. *kī* if *yištaħū* worship.mod.3ms *l-aš-šɛmɛš* to-the-sun *ʔō* or *l-ay-yārēaħ* to-the-moon *ʔō* or *lə-kol* to-kol *ṣḇā* host(of) *haš-šāmayīm …*

the-heavens …

'If [he] worships the sun or moon or any of the host of heaven …' (Deut. 17:3)

≠ 'If [he] worships the sun or moon or all the host of heaven …'

(42) comparative PPs

*wat.tērɛḇ* and.was.as.large.3fs *maśʔaṯ* serving.fs(of) *binyāmin* Benjamin *mim-maśʔōṯ* as-servings(of) *kull-ām* all-3mp *ħāmēš yāḏōṯ*

five portions

'but Benjamin's serving was five times as much as any of theirs' (Gen. 43:34)

≠ ? but Benjamin's serving was five times as much as all of theirs'

### Edit Doron

### **4 Distributivity in BH**

In English, *all* is quantificational, and may be interpreted distributively. As shown above, *kol* is a non-quantificational degree determiner in BH, and is not distributive. Distributivity is achieved in BH by other means. Various BH syntactic structures express distributivity through the LF application of the operator *each* (defined by Link 1987) to a property:

(43) ⟦each⟧ = λP.λx.∀y ≤ x [Atom(y) → P(y)]

We only expect the distributivity operator to modify VPs predicated of a subject *kol NP* if the latter is derived by the definite type shift (33a). Such *kol NP* denotes an individual, for which the ≤ part-of relation is defined. We indeed do not find the distributivity operator when *kol* is interpreted as *any*, by the application of the indefinite type shift (33b).

### **4.1 The lexical item** *ʔīš* **'each'**

In the simplest case, the distributivity operator is expressed by a VP-premodifier, the lexical item *ʔīš* 'each' (literally 'man'), sometimes reduplicated as in (45):

(44) a. *way.yaggīšū* and.brought.3mp *kol* kol *hā-ʕām* the-people.ms *ʔīš* each *šōr-ō* ox-poss.3ms 'So every one of the people brought his ox.' (1Sam. 14:34) b. *kī* for *kol* kol *hā-ʕammim* the-peoples *yēlḵū* walk.mod.3mp *ʔīš* each *bə-šēm* in-name(of) *ʔɛlōh-āw* God-poss.3ms *wa-ʔănaħnū* and-we *nēlēḵ* walk.mod.1p *bə-šēm* in-name(of) *yhwh* Lord *ʔɛlōh-ēnū* God-poss.1p 'For all people walk each in the name of his God, but we will walk in the name of the Lord our God.' (Mic. 4:5) (45) *way.yāḇōʔū* and.came.3mp *kol* kol *ha-ħăḵāmīm…* the-experts… *ʔīš* each *ʔīš* each *mim-məlaḵt-ō* from-work-*ʔăšɛr* poss.3ms *hēmmā* that *ʕōśīm* they do.ptcp.mp

'Then all the craftsmen … came each from the work he was doing.' (Ex. 36:4)

1 From a collective to a free choice determiner in Biblical Hebrew

### **4.2 Reduplication**

The distributivity operator can also be expressed by reduplicative adverbials, as shown by Beck & von Stechow (2006), Naudé & Miller-Naudé (2015):

	- take.impr.2mp to.2mp from the-people twelve men man *ʔɛħāḏ* one *ʔīš* man *ʔɛħāḏ* one *miš-šāḇɛṭ* from-tribe 'Take for yourselves twelve men from the people, one man from every tribe' (Josh. 4:2)
	- c. *middē* whenever *šānā* year *bə-šānā* in-year 'year after year' (1Sam. 7:16)
	- d. *bə-ḵol* in-kol *dōr* generation *wā-ḏōr* and-generation 'forever and ever ' (Ps. 45:17)
	- e. *wə.sāp̄əḏū* mourn.mod.3mp *… kol* kol *ham-mišpāħōṯ* the-families *han-nišʔārōṯ* the-remain.ptcp.mp *mišpāħōṯ* families *mišpāħōṯ* families *ləḇāḏ* alone 'all the families that remain shall mourn, every family by itself' (Zech. 12:14)

### **4.3 Floated** *kol*

Another VP-premodifier which is interpreted as *each* is the inflected *kol*. <sup>24</sup> In (47a), the subject is null and the predicate is modified by *kullām*, i.e. *kol* inflected in the plural. In (47b) the subject is a group individual, and the predicate is again modified by *kullām*.

<sup>24</sup>This modification has been called "quantifier float" by Shlonsky (1991) and Naudé (2011a).

### Edit Doron

	- b. *wə-kol* and-kol *ṣāray-iḵ* adversaries-poss.2fs *kull-ām* kol-poss.3pl *b-aš-šəḇī* in-the-captivity *yēlēḵū* go.mod.3mp 'And all your adversaries, every one of them, shall go into captivity. ' (Jer. 30:16)

*kol* may also be inflected in the singular, e.g. *ḵullō* in (48):

(48) *ʔɛp̄ɛs* edge(of) *qāṣē-hū* extremity-poss.3ms *ṯirʔɛ* see.mod.2ms *wə-ḵull-ō* and-kol-poss.3ms *lō* neg *ṯirʔɛ* see.mod.2ms

'You shall see the outer part of them [the nation], and shall not see every one of them.' (Num. 23:13)

### **4.4 Dependent relational nouns**

Relational nouns denoting e.g. body-parts, kinship and socially defined roles, or other relations which involve atomic individuals rather than groups, give rise to distributive interpretations when they depend on group nouns. Examples are shown in (49). Each example includes a dependent relational noun, where the dependence is marked by poss inflection, as in (49a–b), by the presence of a possessor which raises in LF to yield inverse scope readings, as in (49c–d), or by the presence of an implicit possessor, as in (49e–f):

(49) a. *kol* kol *šōməʕ-ō* hear.ptcp.ms-poss.3ms *təṣillɛnā* tingle.mod.3fp *štē* both.fp(of) *ʔozən-āw* ear.fp-poss.3ms 'Both ears of everyone who hears it will tingle.' (1Sam. 3:11) b. *kol* kol *hā-ʔănāšīm* the-men *hay-yōḏəʕīm* the-know.ptcp.mp *kī* that *məqaṭṭərōṯ* fume.ptcp.fp *nəšē-hɛm* wives-poss.3mp *lē-ʔlōhīm* to-gods *ʔăħērīm* other.mp 'all the men who knew that their wives had burned incense to other gods' (Jer. 44:15)

1 From a collective to a free choice determiner in Biblical Hebrew


The dependence of the relational noun on a group individual gives rise to the introduction of the distributivity operator at the predicate level (Winter 2000). The LFs of (49a–f) can be represented as (50a–f) respectively. The predicate abstracted from the clause which contains the bound x<sup>i</sup> is distributively predicated of the group subject:

	- b. [[all men] [each<sup>i</sup> [x<sup>i</sup> knew that x<sup>i</sup> 's wife had burned incense to other gods]]]
	- c. [[all gifted artisans] [each<sup>i</sup> [I have put wisdom in the heart of x<sup>i</sup> ]]]
	- d. [[all kinds] [each<sup>i</sup> [bring two of x<sup>i</sup> into the ark]]]
	- e. […[all of the nation] …] [each<sup>i</sup> [the mouth (of x<sup>i</sup> ) was speaking vile words]]
	- f. […[the inhabitants] …] [each<sup>i</sup> [the house (of x<sup>i</sup> ) is shut up so that no one enter it]]

### Edit Doron

### **5 Free choice**

Existential modals such as *may* give rise to what has been called *the distribution requirement* by Kratzer & Shimoyama (2002). This requirement results in a free choice (FC) reading of particular expressions under existential modals.<sup>25</sup> We find the same reading for *kol NP* in Hebrew. Under an existential modal, *kol NP* receives a FC reading, as in the following examples, where *kol NP* is satisfied by different individuals in different accessible worlds:

	- b. *wə-ħēlɛḇ* and-fat(of) *nəḇēlā* animal *…* … *yēʕāśɛ* be.used.mod.3ms *lə-ḵol* to-kol *məlāḵā* craft *wə-ʔāḵōl* and-eat.inf *lō* neg *ṯōḵəlu-hū* eat.mod.2mp-acc.3ms 'And the fat of an animal … may be used in any other way; but you shall by no means eat it.' (Lev. 7:24)

The FC reading is also available in the scope of imperative/commissive modal operators (cf. Dayal 2013) if *kol NP* is modified by a relative clause, as in (52) below. In such examples, *kol* is interpreted as *whatever/whoever* and receives a FC interpretation:

	- kol attack.ptcp.ms Jebusite in-the-first be.mod.3ms to-chief *ū-lə-śar* and-to-captain 'Whoever attacks the Jebusites first shall be chief and captain.' (1Chron. 11:6)

A minimal pair is shown in (53), where *kol+relative clause* has a FC interpretation in the commissive (53a), but merely a collective interpretation in the episodic (53b):

<sup>25</sup>According to Bar-Lev & Fox (2017), ♢(p ∨ q) excludes ♢(p & q) by exhaustivity, but includes ♢p & ♢q by innocent inclusion, hence implies FC.

1 From a collective to a free choice determiner in Biblical Hebrew

(53) a. *wə-ḵōl* and-kol *ʔăšɛr* that *tōmar* say.mod.2ms *ʔēlay* to.1s *ʔɛʕɛśɛ* do.mod.1s 'and I will do whatever you say to me' (Num. 22:17) b. *way.yaʕaś kōl ʔăšɛr ʔāmār*

and.did.3ms kol that said.3ms 'so [Moses] did all that he had said' (Ex. 18:24)

FC readings have been accounted for by the pervasive view (from Kadmon & Landman 1993 to Chierchia 2013) that FC items are existential.<sup>26</sup> In the case of *kol*, the FC interpretation is due to the indefinite type-shift in (51b) above. Under the present approach, the availability of this type-shift depends on its deriving a stronger reading than the competing definite type-shift. This indeed seems to be the case. If John or Mary may sign a check, then each of them may. But if John and Mary may sign the check, it is not clear they may each sign separately.<sup>27</sup>

I assume that the FC interpretation was eventually reanalyzed as part of the lexical meaning of *kol*. The change conforms to Eckardt's (2006: 236) notion of semantic reanalysis – the overall sentence meaning did not change, but there was redistribution of conceptual content: *kol* acquired FC interpretation in the environment of certain modal operators.

### **6 Beyond step II**

In Modern Hebrew (MH), we find that step III of the Distributivity Cycle has occurrred (perhaps as early as Rabbinic Hebrew). The universal determiner *kol* is now interpreted as the distributive *every* in addition to its categorization as *any*:

(54) *kol* NP = *every/any* NPet

I will not discuss step III in the present paper, and rely on Beck's (2017) account of the development from FC to distributive interpretations. Beck shows how conjunction of the alternative propositions which underlies FC readings develops into universal quantification over individuals.

Moreover, in post-Biblical Hebrew, definite noun phrases are not NPs but are all headed by D; as shown by Doron & Meir (2016), the article *ha-* was reanalyzed as a definite determiner of category D. Accordingly, when the complement of *kol* is definite, it is categorized as an individual DP rather than a predicate NP:

<sup>26</sup>Menéndez-Benito (2010) and Zimmermann (2008) treat FC items as universal, but this crucially depends on the distributive interpretation of the universal determiner, which *kol* does not have.

<sup>27</sup>In general, ♢P(a∨b) → ♢P(a) & ♢P(b), but ♢P(a⊕b) ↛ ♢P(a) & ♢P(b).

### Edit Doron

(55) *kol* DP = all DP<sup>e</sup>

The construction in (55) is definite due to its partitive structure. There isn't any longer an indefinite type-shift giving rise to NPI/FC interpretations, not even in downward entailing or modal environments, as shown in (56). (56) contrasts with parallel Biblical examples such as (29–30) or (41–42) above, which have a pseudo-partitive structure, and hence have NPI/FC interpretations.

	- b. *ha-hanhala* the-administration *hitħayḇa* commited *le-qabel* to-accept *et* acc *kol* kol *ha-tlunot* the-complaints The administration commited to accept all/\*any complaints.

However, in construction (54), we do find NPI/FC interpretations, as shown in (57):


The Biblical origins of the *any NP* construction in (54) are also manifested by the number feature of *any*'s complement within this construction. It is only within this construction that the complement of *kol* can be a plural NP in MH, just like in the Biblical (19a), (35b–c). The following are MH examples found on the web:


### 1 From a collective to a free choice determiner in Biblical Hebrew

We thus find remnants of Biblical syntax within the MH (54) construction where *kol* is interpreted as *any*, alongside the new *every* interpretation derived from it. The original definite interpretation of *kol* as *all* is now restricted to the partitive (55) structure. This completes the account of the full array of *kol*'s interpretations in MH.

### **7 Conclusion**

Hebrew originally lacked a distributive determiner *every*. Distributivity was achieved in Biblical Hebrew through operators applying to the sentence predicate, such as the distributivity operator *each.* Step I of the Distributivity Cycle consisted in the noun *kol* 'entirety' grammaticalizing as the collective determiner *all*. The determiner was not quantificational – its combination with a NP yielded the plural property corresponding to NP. In argument position, it was interpreted either by the definite or the indefinite type-shift, depending on which yielded a stronger reading. This gave rise to step II, where *kol* received NPI/FC interpretations in particular environments. It is only at step III that *kol* acquired a distributive interpretation. Modern Hebrew *kol* also retained its previous uses, which accounts for the extensive variation in its interpretations: 'all/any/every'. The paper has shown how these interpretations unfolded along the Distributivity Cycle.

### **Abbreviations**


### **Acknowledgements**

I am grateful to the following people for helpful comments and suggestions which have greatly benefited the article: Chanan Ariel, Bar Avineri, Moshe Elyashiv Bar-Lev, Ido Benbaji, Luka Crnič, Danny Fox, Itamar Francez, Kevin Grasso, Andreas Haida, Robert Holmstedt, Geoffrey Khan, Omri Mayraz, Wendy Sandler, Todd Snider, Ruth Stern, and Yoad Winter. This research has received funding from the Israel Science Foundation grant No. 1296/16 and from the European Research Council H2020 Framework Programme No. 741360.

### **References**


Kadmon, Nirit & Fred Landman. 1993. Any. 16(4). 353–422.


1 From a collective to a free choice determiner in Biblical Hebrew


### **Chapter 2**

# **Grammaticalization parameters and the retrieval of alternatives: Latin** *nec* **from discourse connector to uninterpretable feature**

### Chiara Gianollo

Università di Bologna

By means of the study of Latin focus-sensitive negation *nec* ('furthermore not'; 'neither'; 'not even'), I address a more general question on the scope and the cyclicity of semantic change. I review Lehmann's syntagmatic parameters of grammaticalization (structural scope, bondedness, syntagmatic variability) with the aim of evaluating to what extent they are reflected in some types of semantic change. With *nec* we see the evolution, from Latin to Romance, of a discourse-structuring particle with an additive component into the building block of new emphatic (scalar) negative polarity items, which in turn are later reanalyzed as elements of Negative Concord (endowed with uninterpretable formal features). I argue that an important aspect of this change concerns the way alternatives to the focused element are retrieved in the context. I propose that increase in bondedness and decrease in syntagmatic variability correlate with a change in the form taken by alternatives, which decrease in scope from discourse units to individual alternatives.

### **1 Introduction**

Recent formal research on semantic change has dealt in particular with change affecting elements of the functional lexicon, and has already provided a number of significant generalizations on the way diachronic phenomena of this kind are triggered and develop over time. Some of these generalizations confirm and sharpen observations that had previously emerged from typological research and, in particular, from the investigation of grammaticalization.

Chiara Gianollo. 2020. Grammaticalization parameters and the retrieval of alternatives: Latin *nec* from discourse connector to uninterpretable feature. In Remus Gergel & Jonathan Watkins (eds.), *Quantification and scales in change*, 33–65. Berlin: Language Science Press. DOI:10.5281/zenodo.3929239

### Chiara Gianollo

I adopt a very general, theory-neutral definition of grammaticalization as "a process which may not only change a lexical into a grammatical item, but may also shift an item "from a less grammatical to a more grammatical status", in Kuryłowicz's words" (Lehmann 2015: 13).

Grammaticalization provides important insights for formal approaches to diachronic semantics. This is due, on the one hand, to the fact that grammaticalization phenomena follow systematic trajectories and, thus, disclose regularities and general mechanisms of language change. On the other hand, grammaticalization phenomena are multidimensional, in the sense that they involve various linguistic levels and require the simultaneous consideration of morphosyntactic, semantic and pragmatic factors.

Research on grammaticalization unanimously acknowledges the existence of systematicity in grammaticalization phenomena, although the evaluation of the forms and extent of such systematicity vary considerably across frameworks, and often involve a radical discussion of the notion of grammaticalization itself (cf. the contributions in Campbell 2001 for discussion).

In typological research on grammaticalization, the way generalizations have been formulated is clearly influenced by the intrinsically multidimensional nature of grammaticalization phenomena: structural as well as semantic factors are encompassed, and often no clear-cut distinction is drawn between them. For formal approaches, this raises the question of how to distinguish which linguistic modules, and consequently which factors within them, are responsible for the observed regularities.

Structural factors have more readily lent themselves to individuation: Lehmann (2015) (whose first version appeared in 1982) singled out a number of paradigmatic and syntagmatic parameters of grammaticalization, and generative research uncovered recurrent mechanisms, such as the reanalysis of movement dependencies as local merge relations, or of phrasal elements as heads (cf. Roberts & Roussou 2003; van Gelderen 2004 for a comprehensive discussion).

A still open question concerns the possibility of singling out similar general mechanisms affecting the semantic component, in grammaticalization as well as in other phenomena, and of expressing them in a formal theory: Eckardt's (2006) seminal study has paved the way for this kind of research, which has already yielded significant results (cf. Eckardt 2012; Deo 2015; Gianollo et al. 2015 for an overview).

As has been the case with formal diachronic syntax, in order to reach an answer it is necessary to collect a substantial amount of cross-linguistic evidence by means of empirical research. The present study is an attempt in this direction: my aim is to provide an analysis of the diachronic path followed by the Latin

### 2 Grammaticalization parameters and the retrieval of alternatives

particle *nec* ('furthermore not'; 'neither'; 'not even') and its Romance continuations, and to compare the emerging generalizations about the involved semantic trajectory with those formulated with respect to the structural aspects of change. I analyze *nec* as a focus particle in all its functions, and I derive its different uses, and their diachronic distribution, from the way alternative meanings to the focus associate are computed and retrieved in context. I then propose that the format of the changes observed in this respect is comparable to Lehmann's (2015) syntagmatic parameters of grammaticalization, in this way capturing important parallels between the syntactic and the semantic side of context-dependence.

The discussion in this chapter largely abstracts away from the broader debate on the nature of grammaticalization, and focuses on well-attested systematic diachronic tendencies, which I regard as part of a grammaticalization process (but remain empirically valid even if they are not considered specific to grammaticalization), and which are argued to affect in a parallel fashion the morphosyntactic and the semantic-pragmatic components.

The structure of the chapter is as follows: in Section 2, I provide a first description of the functions and of the diachronic development of *nec*, and I single out the reasons why I believe this case study to be particularly relevant for our more general understanding of semantic change. Section 3 is dedicated to a more indepth analysis of the particle's various functions. In Section 4, an analysis of the steps involved in the grammaticalization path is provided. In Section 5, I compare the conclusions emerging from the case study to Lehmann's (2015) syntagmatic parameters for grammaticalization, showing the existence of clear correlations between structural and meaning change in grammaticalization, and I summarize the main conclusions reached.

### **2 Latin** *nec* **from discourse connector to uninterpretable feature**

Thanks to the uninterrupted and rich documentation on Latin and its Romance descendants, it is possible to follow the history of *nec* for millennia and to see how this element developed multiple functions: some of them coexist since the most ancient texts, some others represent later developments; at least one of these functions is uniformly continued in Romance, whereas others were lost in all or in some Romance languages. Section 2.1 gives a first overview of these functions and of their diachronic distribution. Section 2.2 comments on the theoretical relevance of the case study.

### Chiara Gianollo

### **2.1 A first overview**

The etymology of the Latin particle *nec* is straightforward: it derives from the Indo-European negative morpheme \**nĕ* and the enclitic conjunction -*que*, yielding *neque*. <sup>1</sup> The form *nec* is derived by apocope of the last syllable of *neque*: the two forms coexist and are functionally equivalent in Early and Classical Latin. For simplicity, I mostly refer to *nec* because it is the most pertinacious form from a diachronic point of view, *neque* becoming rarer in Late Latin texts and being continued only in Romanian.

The negative particle \**nĕ* is continued in Latin only in univerbation with other elements. It yields negative indefinites, such as *nemo* 'nobody' (< *nĕ* + *homo* 'man'); *nullus* 'no' (< *nĕ* + *ullus* diminutive of 'one'). It is also at the core of the negative marker *nōn* 'not' (< *nĕ* + *oenum* 'one'), originating through a process of reinforcement (Fruyt 2008; Gianollo 2018), and of other connectors, such as *nisi* 'if not', *ne...quidem* 'neither, not even'. It also appears as negative morpheme in verbs, such as a *nĕscio* 'ignore' (< *nĕ* + *scio* 'know'), *nolo* 'not-want' (< *nĕ* + *volo* 'want'), etc.<sup>2</sup>

The clitic conjunction -*que* is employed for the coordination of various types of constituents in Latin; its syntactic distribution is constrained by its postpositive nature, cf. (1).<sup>3</sup> The positive counterpart of *neque*/*nec* is represented by the pair *atque*/*ac* 'and (also)'.

(1) *terra* land:abl *marique* sea:abl.and 'on land and sea'

<sup>1</sup>The etymological facts are complicated only by the occurrence, in Archaic Latin, of *nec* in a usage that does not fall into the canonical functions of the particle and that disappears at later stages, namely the expression of plain sentential negation with no apparent correlative function. Scholars tend towards an explanation in terms of an etymologically different particle in these cases: Orlandini & Poccetti (2007: 29–30) defend a deictic origin, motivated as negation strengthening, for the element *-c* (<\**ke*, cf. Latin *hic* 'this') in this archaic particle.

<sup>2</sup>The particle *nē* 'lest', the negation used a.o. in prohibitions and as a negative complementizer in purpose clauses, has a different etymological origin, as evidenced by the long vowel that characterizes it. For discussion of its controversial etymology cf. de Vaan (2008).

<sup>3</sup> In the Latin examples, the glosses follow the Leipzig Glossing Rules and are limited to basic morphological information, for the sake of readability (case on nominals and number on verbs; for non-finite forms, information on mood is provided). The abbreviations for Latin authors and works follow the *Thesaurus Linguae Latinae* (http://www.thesaurus.badw.de/en/ user-tools/index/) Texts are cited according to the editions in Brepols' electronic corpus *Library of Latin Texts – Series A* (http://www.brepolis.net).

### 2 Grammaticalization parameters and the retrieval of alternatives

The particle *nec* itself is found in combinations with other particles, yielding complex elements such as *necdum* '(and) not yet', *necne* 'or not', *necnon* 'and also, and yet' (the latter yielding a positive meaning in conformity to the Double Negation system of Latin).

Latin *nec* is a multifunctional element (Orlandini 2001; Orlandini & Poccetti 2007). Three main functions can be singled out:

	- (i) Discourse-structuring connector 'and not'; 'furthermore, it is not the case that', at the beginning of new textual units;
	- (ii) Correlative particle 'neither'...'nor';
	- (iii) Stand-alone focus particle with an additive ('also not') or a scalar ('not even') interpretation.

The examples in (3–5) show *nec* in the functions listed in (2). Each function will be described in more detail in Section 3, where I will spell out the criteria for contextual disambiguation. For now, it is sufficient to remark that *nec* is always intrinsically negative. In the clearest examples that show its function as discoursestructuring connector, it performs a polarity switch with respect to a positive antecedent, cf. (3).

(3) (i) Discourse-structuring connector

*Accessum* approached:ptcp *est* be:3sg *ad* to *Britanniam* Britannia:acc *omnibus* all:abl *navibus* ships:abl *meridiano* midday:abl *fere* around *tempore,* time:abl, *neque* and.not *in* in *eo* that *loco* place *hostis* enemy:nom *est* be:3sg *visus.* seen:ptcp 'All ships got into Britain at around noon, **and no** enemy was spotted there.' (Caes. *Gall.* 5.8)

(4) (ii) Correlative particle

*nam* for *postquam* after *exercitui* army:dat *praeesse* command:inf *coeperat,* start:3sg *neque* and.not *terra* land:abl *neque* and.not *mari* sea:abl *hostes* enemy:nom *pares* equal:nom *esse* be:inf *potuerant.* can:3pl

'And after he (Alcibiades) started to be the army commander, the enemies could not compete, neither by land nor by sea.' (Nep. *Alc.* 6.2)

### Chiara Gianollo

(5) (iii) Stand-alone focus particle

*Veteres* ancient:nom *quattuor* four *omnino* altoghether *servavere* observe:3pl *per* for *totidem* as.many *mundi* world:gen *partes* part:acc *- ideo* therefore *nec* and.not *Homerus* Homer:nom *plures* more:acc *nominat.* mention:3sg 'The ancients reckoned only four (winds) corresponding to the four parts of the world – and also Homer does not mention more' (Plin. *nat.* 2.119)

Functions (i) and (ii) are historically primary, and are attested since the beginning of the textual tradition. Their respective fate is quite different. Function (i) is not productively continued in Romance (although it shows some signs of retention in Old Romance, this usage is infelicitous in Modern Romance). Function (ii), instead, is continued in all Standard Romance languages (e.g. French and Spanish *ni*, Italian *né*, Romanian *nici*). As for function (iii), it is only sporadically attested in Early and Classical Latin, and gains in significance only later (1st c. CE), first with an additive and then with a scalar meaning, which becomes very frequent in Late Latin (from the 3rd–4th c. CE).

Function (iii) is continued to various degrees in Romance. As a focus particle, *nec* typically undergoes a cycle of reinforcement of the additive/scalar component: cf. e.g. Spanish *ni siquiera* 'not even' (originally: 'not even if you wish'),<sup>4</sup> Portuguese *nem mesmo* 'not even' (originally: 'not even itself'), Italian *neppure, neanche* 'neither, not even' (*ne-* + multifunctional particle *pure* 'also, though' or *anche* 'also'), Romanian *nici macăr* 'not even' (originally: 'not even if you wish'); alternatively, it is substituted by another element (French *même pas* 'not even', originally 'itself not').

Function (iii) also motivates the use of *nec* as a negative morpheme in many newly grammaticalized Romance indefinites that become elements of Negative Concord (n-words), like e.g. Spanish *ninguno*, Portuguese *nenhum*, Old French *neuns*, Italian *nessuno* 'nobody'. In fact, if a Romance n-word is negatively marked, the negative morpheme always derives from *nec*. 5

This latter outcome is indicated with (iv) in Table 1, which provides an overview of the diachronic distribution of the various functions. A further, pervasive change, which is omitted from the table for readability, concerns the reanalysis of the negative feature carried by *nec*, which is reanalyzed from a semantic feature [Neg] in Latin into a formal uninterpretable feature [uNeg], according to

<sup>4</sup> Spanish *ni* can also be used by itself, without *siquiera*, cf. Aranovich (2006).

<sup>5</sup> For a more detailed analysis of the etymological origin of these indefinites, which sometimes contain further building blocks (e.g. *ipse* 'himself' in Italian *nessuno*) and vary in the retention of the velar component of *nec*, see Gianollo (2018: 225–228).

### 2 Grammaticalization parameters and the retrieval of alternatives

the general change from a Double Negation to a Negative Concord system from Latin to Old Romance (Gianollo 2018: chapters 4–5). The table just indicates that the newly grammaticalized indefinites containing *nec* are elements of Negative Concord (i.e., [uNeg] indefinites) since the beginning.

Table 1: *nec*: overview of the diachronic developments


As exemplification of the Modern Romance outcomes, consider the data from Italian in (6): from a morphological point of view, Latin *nec* is most directly continued by the correlative particle *né* in Italian. This particle is unacceptable as discourse connector (function i). It is typically used in correlative structures (function ii), but cannot be used in function (ii) to join two clauses of different polarity, i.e., to perform a polarity switch, unlike what happens in Latin. For function (iii) the reinforced form *neanche* may be used as a focus particle, and is ambiguous between an additive and a scalar reading.<sup>6</sup> Finally, a continuation of Latin *nec* is visible in the initial morpheme of the word for 'nobody', *nessuno*: in fact, the morphological makeup of the indefinite is not transparent for Modern Italian speakers, but the originally negative element can still be attributed the function of carrying a formal uninterpretable feature for negation, which allows it to enter Negative Concord (see further Section 4.5).

	- a. (i) Discourse particle

*Maria è andata al supermercato. \* Né ha ricordato di portare la borsa.* 'Maria went to the supermarket. Né she remembered to bring the bag.'

b. (ii) Correlative particle *Maria non ha comprato né il latte né il burro.* 'Maria bought neither milk nor butter.'

<sup>6</sup> For the additive value of *anche* cf. Franco et al. (2016); also further reinforced forms exist, e.g. *neppure*, *nemmeno*, whose diachronic development deserves to be studied more carefully in future research.

### Chiara Gianollo


### **2.2 Broader theoretical relevance of the case study**

Table 1 gives us a first impression of the remarkable diachronic path followed by *nec*. We see it starting as an element operating on discourse units, and ending up as a word-internal component (a morpheme and eventually a functional feature). In the development from Latin to Romance, *nec* turns out to be diachronically pertinacious, but at the same time it undergoes a wide-ranging grammaticalization process affecting its multifunctionality. This process can be understood as a form of *functional enrichment* that "depletes lexical items of their semantic and interpreted features and eventually reduces them to purely functional elements with only uninterpreted features" (Kiparsky 2015: 73). At the same time, however, we see some of the original functions coexisting with the newly developed ones.

The history of Latin *nec*, thus, raises a number of issues for our theoretical understanding of semantic change: how are the different functions related? What determines whether and how these functions coexist at a certain stage? Why are some functions lost and others newly developed? And does the shape of this grammaticalization process tell us something more general on possible formats of semantic change?

In the following sections I try to provide at least partial answers to these questions. First of all, I account for the multifunctionality of the particle: I analyze the functions of *nec* synchronically and diachronically, and, capitalizing on the bimorphemic nature (*ne-c*, *ne-que*) of the particle, I propose that, across functions, it shares a homogeneous internal syntactic structure, corresponding to its two basic semantic components: additivity and negation.

I further show how the various uses can be derived from the interaction between these two operators and the surrounding structure into which the particle is merged. The focus-sensitive nature of the particle, i.e., its sensitivity to alternatives, is held responsible for its multifunctionality: the structural position of the particle influences its pragmatic properties, in terms of the form of the evoked alternatives and the way they are retrieved.

On the one hand, the mechanism governing the retrieval of alternatives is involved in the development of a scalar reading for the particle (pragmatic enrich-

### 2 Grammaticalization parameters and the retrieval of alternatives

ment), which is an intrinsic possibility for additive particles but seems to gain ground in Latin only after a functional competitor, the particle *ne...quidem*, falls out of use.

On the other hand, increase in bondedness and decrease in syntagmatic variability witnessed in the diachrony of *nec* correlate with a change in the form taken by alternatives, which decrease in scope from discourse units to individual alternatives ordered on a scale. In the development of *nec* we see, thus, the semantic-pragmatic counterpart of Lehmann's syntagmatic parameters of grammaticalization, resulting in decrease or loss of discourse-dependence.

### **3 The functions of** *nec***: Distribution**

In this section we will have a closer look at the various functions of Latin *nec*. I will focus on data from Classical Latin (1st c. BCE–1st c. CE), but occasionally also Late Latin data (3rd–4th c. CE) will be taken into consideration, in order to show the functional extension that the particle undergoes already during the history of Latin. Note that, as seen in Section 2.1, the two forms of the particle, *neque* and *nec*, are functionally indistinguishable in the texts on which I base my conclusions.

### **3.1 Discourse-structuring connector**

As a discourse-structuring connector, *nec* introduces a full clause belonging to a new discourse unit, which may be connected in the discourse to a previous clause independent of the polarity of the latter. Latin is a Double Negation language: each negatively marked element introduces a semantic negative operator, independently of its position in the clause (Gianollo 2016). The particle *neque*/*nec* conforms to this system and typically suffices to negate a clause (or a smaller constituent) by itself.

In (7) the clause preceding the one introduced by *neque* has positive polarity. The particle marks the subsequent one for negative polarity. The discourse function of the clause introduced by the particle is to bring forward the narration in a temporal progression: *neque* therefore connects two clauses expressing a coordinating discourse relation according to Asher (1993).

(7) *Concurrunt* clash:3pl *equites* knight:nom *inter* between *se;* refl:acc *neque* and.not *vero* indeed *primum* first:acc *impetum* impact:acc *nostrorum* our:gen *Numidae* Numidian:nom *ferre* resist:inf *potuerunt,* can:3pl *sed* but

### Chiara Gianollo

*interfectis* killed:ptcp.abl *circiter* around *CXX* 120 *reliqui* remaining:nom *se* refl:acc *in* in *castra* camp:acc *ad* towards *oppidum* city:acc *receperunt* withdraw:3pl

'The respective cavalries clashed, but the Numidians were not able to withstand the initial impact of our men. Instead, after about a hundred and twenty were killed, the rest withdrew into the camp next to the city.' (Caes. *civ.* 2.25)

In some cases the demarcation between the two discourse units is even sharper, and is highlighted by modern editors through the insertion of specific punctuation or even paragraph breaks, as in (8). Often a contrastive flavor is present, also because of accompanying particles (*vero* 'in fact', *tamen* 'however'), as in both (7) and (8).

(8) [7.4] *Qua* which:abl *ex* from *re* thing:abl *creverat* grow:3sg *cum* and *fama* reputation:abl *tum* and *opibus,* power:abl *magnamque* close:acc *amicitiam* friendship:acc *sibi* refl:dat *cum* with *quibusdam* certain:abl *regibus* king:abl *Threciae* Thracia:gen *pepererat.* procure:3sg [8.1] *Neque* and.not *tamen* however *a* from *caritate* love:abl *patriae* fatherland:gen *potuit* can:3sg *recedere.* recede:inf

'Because of this he (Alcibiades) grew in reputation and power, and he procured close friendships with certain Thracian kings. Nonetheless he could never abandon the love for his own country.' (Nep. *Alc.* 7.4–8.1)

As a discourse particle, *nec* does not impose any constraint on the polarity of the previous unit, therefore it can perform a polarity switch. This possibility is still attested for *né* in Old Italian (9a), but is lost in Modern Italian (9b; cf. also 6a), cf. Zanuttini (2010) for Old Italian, and Doetjes (2005) for similar Old French uses:<sup>7</sup>

<sup>7</sup>The example in (9a) is a strict Negative Concord structure, where negation is expressed both on the particle *né* and on the negative marker *non*, yielding a single-negation reading. Similar structures are not grammatical in Modern Italian, but are in line with the grammar of Old Italian, which allowed for strict Negative Concord under certain conditions (Garzonio 2018; Gianollo 2018: chapter 5).

2 Grammaticalization parameters and the retrieval of alternatives

(9) a. *e* and *perció* thus *in* in *mezzo* middle *della* of.the *via* road *l'uccise;* him.killed *né* and.not *Catone* Cato *non* not *avea* had *podere* faculty *di* of *difenderlo* defend.him

'and thus he killed him in the middle of the road; and Catone did not have the faculty of defending him' (Old Italian, Brunetto Latini *Rett.* p. 115 l. 9–10)

b. *\*Francesco* Francesco *è* is *andato* gone *a* to *mensa* cafeteria *né* and.not *Giovanni* Giovanni *lo* him *ha* has *accompagnato.*

accompanied.

'Francesco went to the cafeteria and Giovanni did not go with him' (Modern Italian)

The impossibility for Italian *né* to perform a polarity switch amounts, in my framework, to the loss of the discourse-structuring function for the particle derived from *nec* in this language (as in the rest of Romance). The same-polarity requirement between the two discourse units, emerging in its diachronic development, results in a usage that is hardly distinguishable from the correlative one (where the conjuncts are subject to stricter parallelism constraints, encompassing polarity).

### **3.2 Correlative negation**

As correlative negation, *nec* co-occurs with another instance of *nec* or another negative element (e.g. the negative marker *non* or a negative indefinite) in the same syntactically complex discourse unit. Each negative element contributes a semantic negative operator, in compliance with the Double Negation nature of Latin: two or more propositions are at the same time coordinated and negated. In this use, *nec* can introduce clauses or smaller syntactic constituents.

Coordination of full-fledged clauses by means of correlative *nec* can be seen in the two passages in (10), with main and with subordinate clauses, respectively:

(10) a. *nec* and.not *satis* enough *exaudibam,* hear:1sg *nec* and.not *sermonis* conversation:gen *fallebar* miss:1sg.pass *tamen,* though *quae* which:nom *loquerentur* speak:3pl.pass 'I couldn't hear perfectly what was being said, but I didn't miss the general drift of their conversation.' (Plaut. *Epid.* 239–240)

### Chiara Gianollo

b. *animus* soul:nom *autem* however *solus* alone:nom *nec* and.not *cum* when *adest* be.present:3sg *nec* ad.not *cum* when *discessit* depart:3sg *apparet* appear:3sg 'only the soul remains unseen, both when it is present and when it departs' (Cic. *Cato* 80)

Often, *nec* apparently coordinates non-clausal constituents (cf. 11); however, in Section 4.3 we will see how in fact these structures can be analyzed as cases of ellipsis affecting clausal constituents.

(11) *neque* and.not *enim* indeed *obscuris* obscure:abl *personis* character:abl *nec* and.not *parvis* small:abl *in* in *causis* issue:abl *res* situation:nom *agetur* develop:3sg.pass 'for the persons involved are not obscure, nor are the issues trivial' (Cic. *fam.* 3.5.2)

In Latin the particles introducing each of the coordinated elements can be identical, like in the Romance languages and unlike in many other languages (e.g. English *neither...nor*, German *weder...noch*; cf. Bernini & Ramat 1996: 100–106 and Haspelmath 2007 for a typological overview). However, in Latin different negative elements correlating with *nec* are possible as well, cf. (12):<sup>8</sup>

(12) *branchiae* gill:nom *non* not *sunt* be:3pl *ballaenis* whale:dat *nec* and.not *delphinis* dolphin:dat 'neither whales nor dolphins have gills' (Plin. *nat.* 9.19)

Examples (10–11) are characterized by syntactic parallelism between the coordinated constituents, which also encompasses their (negative) polarity. However, there are also cases where polarity switch between conjuncts is attested, as in (13), similarly to what we saw with the discourse-structuring connector:

(13) *eius* he:gen *enim* in.fact *nomine,* name:abl *optimi* excellent:gen *viri* man:gen *nec* and.not *tibi* you:dat *ignoti,* unknown:gen *maledicebat* slander:3sg *tibi* you:dat 'he slandered you under the name of this man, an excellent person and not unknown to you' (Cic. *Deiot.* 33, from Pinkster 2015: 688)

<sup>8</sup>As an anonymous reviewer remarks, these examples often have the flavor of an afterthought: 'Whales don't have gills, and dolphins don't have them either'.

### 2 Grammaticalization parameters and the retrieval of alternatives

Polarity switch uses are often employed to create rhetorical effects: *nec ignotus* in (13), where the adjective is marked by the negative prefix *in*-, yields a litotes, in a structure of "asymmetric coordination" that is known as "epitaxis" and is used to add parenthetical comments (cf. Orlandini & Poccetti 2007 and references cited there).

In many cases of polarity switch a contrastive flavor can be detected, and in these contexts it is not easy to tell apart the discourse-structuring connector use from the correlative use: in annotating the examples, I decided for the latter when the correlated constituents are built in a syntactically parallel way, since this contextual condition ensures by itself cohesion among correlates, with no need for the particle to create such cohesion, as is instead the case with the discoursestructuring connector (see Orlandini & Poccetti 2007 for further discussion).

### **3.3 Focus particle**

In the stand-alone focus-particle use, no direct correlation with other negative constituents is present; rather, the particle attaches to a sub-clausal constituent (in the clearest cases, to a nominal element), and finds its antecedent not in a syntactically parallel structure, but in the broader discourse context. The particle requires for its interpretation that at least one alternative to the constituent in focus holds in the context (additive interpretation). In later texts a scalar reading emerges, which is typically dependent on general world knowledge (providing the scalar alternatives), rather than on specific contextual conditions.

In (14), *nec* attaches to the nominal phrase *cuniculos* 'rabbits'. The previous context provides the alternative: the land of Ebusus (Ibiza) is known for driving away snakes; a further blessing of this land is that it does not produce rabbits, which can compromise the harvest:

(14) *nec* and.not *cuniculos* rabbit:acc *Ebusus* Ebusus:nom *gignit,* generate:3sg *populantes* destroy:ptcp.nom *Baliarium* Balearic:gen *messes* harvest:acc 'Ebusus (the island of Ibiza) neither generates rabbits, which destroy the harvests of the Balearic islands' (Plin. *nat.* 3.78)

In (15), the focus associate is the nominal *Sirenes* 'sirens': here Pliny is discussing birds and is discarding a series of fabulous birds that were mentioned by previous authors; these birds represent the contextually provided alternative set to the focus associate introduced by *nec*. After saying that he considers birds like *pegasi*

### Chiara Gianollo

and gryphons as invented (*fabulosos reor* 'I consider them legendary'), Pliny adds a comment expressing similar scepticism about sirens:

(15) *nec* and.not *Sirenes* siren:nom *impetraverint* obtain:3pl *fidem* credit:acc 'the sirens also cannot obtain great credit with me' (Plin. *nat.* 10.136)

Cases where the element in focus is not a nominal sub-constituent of the clause, but the whole predication, are more difficult to distinguish from the correlative use, since typically the previous clause directly provides a symmetric alternative, which could be interpreted as a first conjunct. For instance, (16) seems an intermediate case:

(16) *quoniam* since *hoc* this:acc *solum* only:acc *animal* animal:acc *ex* from *marinis* marine:abl *non* not *percutiat,* strike:3sg *sicut* as *nec* and.not *e* from *volucribus* bird:abl *aquilam* eagle:acc 'since this is the only animal, among the marine ones, which it (the thunder) never strikes; similarly, neither (it strikes) the eagle, among birds' (Plin. *nat.* 2.146)

It seems that the use as stand-alone focus particle is closely related to the correlative one, but emerges in cases where the syntactic parallelism among alternatives is not obvious. This could of course have played a role in providing bridging contexts (in the sense of Heine 2002) during the diachronic development, since the focus particle function is attested later than the discourse-structuring and correlative ones.

The use as a focus particle is encountered very rarely in Early and Classical Latin texts, and becomes more frequent only from the Imperial age (1st. c. CE) on. Apart from sporadic attestations in early documentation (typically in cases where *nec* is accompanied by *saltem* 'at least'), the scalar reading of the focus particle emerges even later, in texts dating to the 3rd–4th c. CE. Due to the peculiarities of the Latin documentation, it is difficult to assess whether the focus particle use was a feature of the spoken language, which has a late attestation in our documents only due to prescriptive control during the Classical stage. What emerges quite clearly, though, is that the increase in the use of *nec* as focus particle correlates with the decrease of its main competitor in this function, the discontinuous particle *ne...quidem* 'neither', 'not even' (see Orlandini 2001: chapter 7; Gianollo 2017 for the functions of this particle and the possible causes for its demise).

An example of *nec* with a scalar intepretation is given in (17):

### 2 Grammaticalization parameters and the retrieval of alternatives

(17) *dico* say:1sg *autem* yet *vobis* you:dat *quoniam* that *nec* and.not *Salomon* Salomon:nom *in* in *omni* all:abl *gloria* splendor:abl *sua* his:abl *coopertus* dressed:ptcp *est* be:3sg *sicut* as *unum* one:nom *ex* from *istis* this:abl 'Yet I tell you that not even Solomon in all his splendor was dressed like one of these' (Vulg. *Matth.* 6.29)

Note that Spanish, unlike other Romance languages, can still reproduce this use with *ni*, the continuation of *nec*:

(18) Spanish translation of (17) (Nueva Biblia) *Pero les digo que ni Salomón en toda su gloria se vistió como uno de ellos.*

Many of the negative scalar focus particles seen in Indo-European languages are also used as correlative negation (cf. König 1991: chapter 4; Haspelmath 2007). As seen in Section 2.1, Romance languages typically use reinforced forms of the correlative particle in this function, pointing to a cyclical development in which the additive/scalar additive component is formally renewed.

### **3.4 Combination with 'one'**

The function of *nec* as a scalar focus particle that develops in Late Latin is at the core of its further development into a negative morpheme in the new Romance n-words (i.e., elements of Negative Concord). The combination of *nec* with the scalar endpoint represented by the cardinal numeral *unus* 'one' emerges in Late Latin as one of the ways to express emphatic negation. An example from a Christian author of the 5th c. CE is given in (19):

(19) *Quaesivi* look.for:1sg *consolantem,* comforter:acc *et* and *non* not *inveni;* find:1sg; *tot* so.many *milia* thousand:nom *saturati,* sated:nom *tot* so.many *milia* thousand:nom *salvati,* saved:nom *infiniti* endless:nom *edocti* instructed:nom *et* and *nec* and.not. *unus* one:nom *inventus* found:ptcp.nom *est* be:3sg *mihi* me:dat *consolator.* comforter:nom

'I looked for someone to comfort me, and I did not find him; so many thousands of them sated, so many thousands of them saved, endless ones instructed and not one comforter for me has been found.' (Arnob. Iun. *in psalm.* 68.35)

### Chiara Gianollo

It is very interesting to remark that in (19) *nec* co-occurs with the conjunction *et*, showing clearly that the particle in this use has lost its correlative function.

Another interesting fact observed in Late Latin texts is that sometimes, in less controlled registers, the use of the combination *nec unus* co-occurs with a further marker of negation in a single-negation reading (as in 20). These structures can be interpreted as a sign of the ongoing development of Negative Concord.

(20) *et* and *de* from *electis* chosen:abl *israhel* Israel:gen *non* not *dissonuit* be.dissonant:3sg *nec* and.not *unus* one 'and of the chosen ones of Israel not even one was dissonant' (Aug. *loc. hept.* 2.102)

Once the change from the Latin Double Negation system to the Romance Negative Concord ones is completed, we see the resulting indefinites, which have emerged through a process of univerbation (e.g. Portuguese *nenhum*, Spanish *ninguno*, Italian *nessuno*, Old French *neuns*, etc.), behave as n-words, i.e., as indefinites that can both express negation by themselves and co-occur with other negative elements in a single-negation reading. Romance n-words do not necessarily contain *nec* as their building block (cf. Spanish *nada*, French *personne*), but if they do contain a negative morpheme, this morpheme is invariably derived from Latin *nec*. This pan-Romance phenomenon hints at a pervasive use of the combination of *nec* and *unus* in the Late Latin varieties from which the Romance languages derive.

### **4 The functions of** *nec***: Analysis**

In the previous section I traced the evolution of a discourse-structuring particle into the building block of new emphatic (scalar) indefinites, which in Romance behave as elements of Negative Concord. The task of the present section is to provide an analysis of the functions of *nec* that accounts for the particle's multifunctionality in a parsimonious way, i.e., by assuming a common semantic and structural core and by deriving the various functions as an effect of contextual factors, either at the synchronic level (in the case of simultaneous availability of multiple functions) or at the diachronic level (in the case of reanalysis). The starting observation is the following: in all the functions surveyed above, *nec* contributes, besides negation, an additive component, whereby alternatives are provided by the broader surrounding discourse or by more local antecedents.

### 2 Grammaticalization parameters and the retrieval of alternatives

### **4.1 The common core**

The two basic semantic components of *nec*, additivity and negation, correspond to the two morphemes into which the particle can be analyzed: the negative morpheme *ne-* and the additive morpheme *-que*/*-c*. Assuming the two morphemes to be heads of their respective syntactic projection, I propose that *nec* is a syntactically complex lexical item, whose internal structure stays the same in all functions: it consists of the projection of a negative operator Op¬P, topped by the projection of an additive focus operator FocP (see Gianollo 2017 for a first version of this proposal).

The resulting structure is shown in (21). In (21), Op¬P is the syntactic projection of [Neg], the semantic feature proposed by Zeijlstra (2004; 2014) to characterize intrinsically negative items in Double Negation systems. The projection of Op¬P amounts to the instruction "Insert operator" for the interface, and has no further role in the syntactic computation (i.e., it does not enter into syntactic dependencies).

(21) Structure for *nec* in all its functions

### Chiara Gianollo

The FocP projection has a basic additive meaning: I adopt a presuppositional analysis, and assume accordingly that the particle contributes the presupposition that the predication about the element in focus *p* also holds of at least one of its alternatives *q* in context C:

(22) presuppositional analysis for additive particles *also* p: (1) *p*

(2) presupposition: ∃ *q* ∈ C ∧ *q* ≠ *p*

Polarity switch cases, where the negative polarity of the conjunct introduced by *nec* contrasts with the positive polarity of the antecedent conjunct, show that the negative operator contributed by *nec* only takes scope over the conjunct directly introduced by the particle. That is, the additive component outscopes the negation (Additive Focus > Negation).

Now, the surface order of the two elements in *neque*/*nec* is the mirror image of their scope relation. The reason resides in prosodic facts governing the distribution of enclitic *-que*/*-c*, and word formation in Latin more in general. Enclitic -*que* is phonologically defective; it is a syntactic head but it is not a phonological word, thus it does not properly align a phonological word with a syntactic head (Agbayani & Golston 2010). As a repair strategy, *ne-* is raised to the superordinate head (prosodic inversion); *ne-* is itself proclitic: together, the two elements form a prosodically acceptable unit for Latin.

FocP and Op¬P are syncategorematic functional shells: they do not select for a specific category, thus they may attach to elements of various semantic types and of different constituency; the focus associates remain transparent for c-selection (cf. Cinque 1999: 120–126 for the status of negation in this respect, and Biberauer et al. 2014: 199–203 for conjunctions and other particles).

Because of its focus-sensitivity, *nec* requires the consideration of alternatives in order to be interpreted. Following Katzir (2007) and Fox & Katzir (2011), I assume the generation of focus alternatives to be structure-based: alternatives are obtained by replacing the focused constituent with constituents that are at most as complex as the element in focus. The nature of alternatives therefore depends on the type of the constituent *nec* combines with. This naturally yields varying meaning effects in the case of a syncategorematic particle like *nec*.

The basic intuition on which my analysis rests is that the various functions of *nec* emerge from the interaction of the particle with the surrounding context (both in discourse and in structural terms). As is routinely assumed, contextual interaction determines how the alternatives to the ordinary semantic value are

### 2 Grammaticalization parameters and the retrieval of alternatives

retrieved: alternatives exploited for the interpretation of *nec* may be represented by preceding discourse units, in the case of the discourse-structuring function, or more locally by preceding clauses within the same discourse unit (sentence topic), in the case of the correlative function. In the stand-alone use as focus particle, no strict syntactic parallelism is required: the additive reading is dependent on the anaphoric retrieval of a suitable discourse referent in the broader discourse context; the scalar reading emerges when such an anaphoric link cannot be established and the alternatives are provided by a scale instead.

In what follows, I more closely review the contextual factors that trigger the various functions of *nec*.

### **4.2 Discourse-structuring connector**

In accordance with the structure-based mechanism of generation of alternatives, I propose that as a discourse-structuring connector *nec* 'furthermore not' takes a whole discourse unit as its complement. I tentatively assume that the highest projection above the CP is a DIS(course) projection (Giorgi 2015) where (some) discourse relations are syntactically represented, and that this projection is taken by the particle as one of its arguments.

The specifier of the additive Focus projection contains a phonetically null propositional anaphor, which represents the other argument of the operator expressed by *nec*. The silent anaphor ensures discourse cohesion by connecting the newly introduced clause to the previous context, thus satisfying the additive presupposition of the particle. A null propositional anaphor is similarly assumed by Poletto (2014: 22–27) in her analysis of Old Italian *e* 'and, thus' as discourse particle, and by Ahn (2015) in her analysis of *too* and *either* (cf. also Beck 2006 for the structural representation of an anaphoric element in the presupposition of *again*).

In my analysis, discourse-structuring *nec* is therefore considered as a focus particle operating at the discourse level. The salient alternative satisfying the additive presupposition is a previous discourse unit. No polarity requirement is imposed on it.<sup>9</sup>

Sentences introduced by *nec* in its use as discourse particle are never discourseinitial. In this, *nec* recalls the behavior of so-called one-place *and*, an adverbial connector according to Zeevat & Jasinskaja (2007):

<sup>9</sup> I leave aside the issue of how to properly derive the interpretation of negation with discoursestructuring *nec*: although it surfaces high in the structure (discourse- and sentence-initially), the negation is interpreted as plain propositional negation, i.e., it operates at the propositional level, unlike some known cases of 'high negation' with a denial interpretation, operating beyond the propositional level.

### Chiara Gianollo

(23) one-place *and*:

**And** John gave him a push (Zeevat & Jasinskaja 2007: their ex. 7)

Discourse-structuring *nec* introduces coordinating discourse relations (List, Narration, Result): like *and*, *nec* is used when "the sentence topic of the pivot is abandoned to start dealing with a new topic" (Zeevat & Jasinskaja 2007: 325); one-place *and* "seems to mark a distinct sentence topic under the continued discourse topic" (Zeevat & Jasinskaja 2007: 325). In (3) the general discourse topic is represented by the Romans' expedition in Britain; the sentence topic of the first clause is the arrival, the sentence topic of the second clause, introduced by *nec*, is the result of the first patrol. In (7) the general discourse topic is the battle; the sentence topic of the first clause is the clash between the cavalries; the sentence topic of the following clause, introduced by *neque*, is the result of the confrontation.

The fact that the units connected by *nec* share the same discourse topic shows that *nec* obeys the condition of "shared topicality" on additive particles discussed for *too* in Schwenter & Waltereit (2010). At the same time, discourse-structuring *nec* introduces a distinctness requirement (cf. 22), leading to an update of the Common Ground.

### **4.3 Correlative negation**

The common structural and semantic core proposed for *nec* in (21) can be maintained for correlative *nec* once the particle is also analyzed as focus-sensitive in this use. Correlative particles have been accounted for as focus-sensitive particles in a number of works: Hendriks (2004), followed by den Dikken (2006), proposes this analysis for English *either*, *neither*, *both*. Similarly, Wurmbrand (2008) treats *nor* in correlative structures as composed by an additive focus particle and a negation.

I thus analyze correlative *nec* as a focus particle introducing each of the conjuncts; in other words, also in the correlative function the morpheme -*que*/-*c* realizes an additive Focus operator, not a conjunction. Correlation between the conjuncts introduced by *nec* is analyzed as asyndetic coordination, adopting the structure for 'edge coordination' ('not only...but also') proposed by Bianchi & Zamparelli (2004). In the structure in (24), JP stands for Junction Phrase (cf. Munn 1993; den Dikken 2006; Szabolcsi 2013; Mitrović & Sauerland 2014), the structure responsible for the coordination, whose null head hosts the conjunction operator.

### 2 Grammaticalization parameters and the retrieval of alternatives

Differently from the discourse-structuring function, the sentence topic does not change across conjuncts. Moreover, the conjuncts in correlative structures are subject to a parallelism constraint: they are parallel in terms of their organization into foreground and background (i.e., they have parallel foci, cf. König 1991: 64) and, as we observed in Section 3.2, they are also parallel in terms of their syntactic structure.

That is, in the case of correlative negation, alternatives are provided locally by the correlative construction itself, and originate from the substitution of a subconstituent of the clause, i.e., the element in focus. In the correlative construction it is particularly clear that the alternatives relevant for the interpretation are generated structurally and obey the complexity constraint discussed by Katzir (2007) and Fox & Katzir (2011).

The parallel syntactic construction and the pragmatic role of focus are the factors licensing ellipsis within the conjuncts. In fact, correlates are arguably always propositional units, reduced by ellipsis; a proper treatment of the association of the semantic operators involved must eventually lead to such an analysis. For instance, TP-ellipsis would be involved in cases like (11), which is therefore only apparently an instance of constituent negation, as shown in (25):

(25) cf. (11), Cic. *fam.* 3.5.2:

*neque* and.not *enim* indeed *obscuris* obscure:abl *personis* character:abl [TP *res* situation:nom

### Chiara Gianollo

*agetur*] develop:3sg.pass *nec* and.not *parvis* small:abl *in* in *causis* issue:abl *res* situation:nom *agetur* develop:3sg.pass 'for the persons involved are not obscure, nor are the issues trivial'

Crucially, parallelism seems to extend to the polarity value of the conjunct, at least in the most obvious cases of correlation. Syntactically, this means that the complement of *nec* must be at least a Polarity Phrase. Same-polarity requirements have often been observed for polarity particles (cf. Wurmbrand 2008: 519– 520 for *nor*); they are also well known for being subject to diachronic change (cf. the observations on *nor* at earlier stages of English in Jespersen 1917: 114). They are usually treated as an additional presupposition encoded in the particle's lexical entry. It is not clear whether this would be the right analysis for Latin, though, since, as seen in Section 3.2, exceptions to the same-polarity requirement do arise in examples where *nec* has a special rhetorical effect. An alternative hypothesis, safeguarding the correspondence between the lexical entry of the discourse-structuring connector and of the correlative particle, would be not to incorporate the same-polarity presupposition in the lexical entry and to admit that correlative *nec* may express two different relations, Parallel or Contrast, in the sense of Asher (1993). With Parallel, alternatives are required to be of the same (negative) polarity, whereas with Contrast the polarity switch is at the core of the Contrast relation itself.

The incorporation of the same-polarity presupposition into the lexical entry may represent a later diachronic step, correlating with the general loss of polarity-switch uses, as we saw for Italian in (9a–9b).

### **4.4 Focus particle**

Since I treated correlative *nec* as a focus particle, the assumption that the standalone particle is structurally identical to the correlative particle naturally follows. The only difference concerns the syntactic context in which they are used: the stand-alone focus particle does not have an immediate overt syntactic correlate. Rather, it finds its alternatives in the broader context. In this, it is more similar to the discourse-structuring use; however, the form of the alternative(s) differs: while in the discourse-structuring use alternatives are provided by discourse units, in the focus-particle use alternatives are typically represented by propositional alternatives. For example, in (14), the proposition 'Ebusus does not generate rabbits' is interpreted in the context of its anaphorically available alternative 'Ebusus drives away snakes'.

### 2 Grammaticalization parameters and the retrieval of alternatives

I assume that this use is in principle always available for a correlative focus particle in virtue of its meaning, but that it may be blocked in some languages by the presence of a more suitable competitor. In Latin such a competitor is represented by the discontinuous particle *ne...quidem*. In Late Latin this particle loses productivity, thus opening an additional functional space for *nec* through loss of lexical blocking.

The ambiguity between an additive and a scalar reading for focus particles is cross-linguistically frequent (cf. König 1991: 158–159, Gast & van der Auwera 2011: 24–25): it is observed, for instance, with German *auch* and Italian *anche*, as shown in (26).

(26) a. German (König 1991: 62)

*Auch Riesen haben klein angefangen.*

'Even giants started from small beginnings.' (*auch* = *sogar*)

b. Italian *Anche i giganti hanno iniziato in piccolo.* 'Even giants started from small beginnings.' (*anche* = *perfino*)

The way alternatives are retrieved and the form the alternative set takes determine whether the reading for the particle is additive or scalar. In the scalar reading, alternatives ordered along a scale are evoked; the focus denotation is then the extreme of the scale. In the additive reading, instead, alternatives come in an unstructured set. Under the scalar reading, *nec* corresponds to the negative counterpart of *even*, whose contribution is schematically represented in (27), to be compared with (22):

	- *even* p:
	- (1) *p*

(2) presupposition: ∀ ∈ C [ ≠ → < ]

(3) alternatives come in an ordered set, where : contextually determined probability measure

Also in this case I adopt a presuppositional analysis, whereby scalarity originates from the presupposition that the alternative being predicated is striking with respect to some contextually established scale; in (27) I adopt a probability measure, but a scale of informational or pragmatic strength (cf. Gast & van der Auwera 2011 for discussion) would work equally well for my purposes. A more fundamental assumption concerns my choice of a scope-based analysis for *even* to account

### Chiara Gianollo

for the reading obtained when it interacts with negation.<sup>10</sup> The focus operator always takes wide scope with respect to the negative operator. This way scale reversal, operated by negation, obtains before focus applies, satisfying the scalar presupposition of the particle in situations where the complement of *nec* denotes the most probable (i.e., less striking) element to obtain (as is the case with minimizers or generalizers). In its scalar reading, *nec* means 'even [not *x*]': it is even the case that the most probable alternative does not hold.

As discussed in Section 3.3, the scalar reading emerges later than the additive one, pointing to the fact that a reanalysis takes place, whereby the lexical entry of the focus particle is enriched by the scalar presupposition (a form of pragmatic enrichment in the sense of Traugott & Dasher 2002). This would motivate the observed divergence between the fate of the correlative particle (continued by all Romance languages), and the fate of the scalar particle, which often undergoes reinforcement or lexical substitution.

To explain how a scalar reading for *nec* emerges and subsequently becomes conventionalized, it is important to consider the different way in which alternatives are retrieved in the additive reading and in the scalar reading. This, in turn, influences the structure that the set of alternatives has, as has been shown for Italian *neanche* 'neither, not even' by Tovena (2006). The use as an additive focus particle is possible only when suitable alternatives for the element in focus are explicitly provided in the context. This happens by means of correlation in the correlative use, and by anaphoric linking to an element in the broader previous discourse in the stand-alone focus particle use. No accommodation is possible (Zeevat 1992 and following): the alternative has to be explicitly available in the conversational background, and this explains why additive particles out of the blue are infelicitous:

### (28) # *John had dinner in New York too.*

That is, additive particles are strictly anaphoric and the lack of a proper antecedent leads to presupposition failure: presupposition accommodation with additive operators is impossible or highly restricted.

In the absence of these preconditions, only a scalar interpretation is possible: in that case alternatives can be accommodated by evoking a scale, whose dimension is usually suggested by the element in focus: in (17), 'Solomon', the element in focus, suggests a scale of people likely to be splendidly dressed: a king is the most probable option on this scale.

<sup>10</sup>For the debate on this issue and alternative, ambiguity-based analysis, see Rooth (1985) and the recent proposal by Collins (2016).

### 2 Grammaticalization parameters and the retrieval of alternatives

Tovena (2006) shows that this mechanism of accommodation regularly takes place with Italian *neanche*. <sup>11</sup> A process of presupposition accommodation by means of scale retrieval may have been responsible for the conventionalization of a scalar meaning for *nec*: hearer-based accommodation processes are costly and they have been argued to be a frequent trigger to processes of reanalysis driven by economy considerations (Traugott & Dasher 2002; Eckardt 2006; Schwenter & Waltereit 2010). It is plausible to assume that, once the competitor *ne...quidem* lost ground and the employ of *nec* as focus particle became more frequent in Late Latin, the scalar reading originally resulting from accommodation may have ended up being conventionalized, by incorporating a scalar presupposition into the lexical entry of *nec*.

### **4.5 Combination with 'one': the** *nec***-words**

The last step in the development of *nec* is represented by its recruitment as a morpheme of the newly created narrow-scope indefinites that I dubbed '*nec*-words' in previous work (Gianollo 2018: chapter 5). While these elements appear to be fully grammaticalized in the earliest Romance documents, in Late Latin we encounter, with increasing frequency, their syntactic source: this has to be identified in structures like (19–20), where *nec* syntactically combines with the cardinal numeral *unus* 'one'.

The frequent scalar use of Late Latin *nec* renders the particle a suitable item to strengthen negation, according to a crosslinguistically frequent pattern which witnesses 'even' as a component of polarity-sensitive quantificational expressions (Haspelmath 1997; Lahiri 1998; Watanabe 2004; Chierchia 2013). Therefore, this last step of the grammaticalization process finds its prerequisite in the expansion of scalar uses for *nec* in Late Latin. Being a natural scalar endpoint, *unus* 'one' represents an optimal strengthening strategy: it is logically entailed by all its alternatives, but, thanks to scale reversal under negation, it yields the most unlikely, informationally strongest proposition.

On the basis of Lahiri's (1998) analysis of Hindi *ek bhii* 'one-even', the meaning of emphatic 'even-one' NPIs is formalized by Chierchia (2013: 156) as follows, using the probability measure already seen in (27):

(29) a. 'even-one' NPIs, adapted from Chierchia (2013: 156): ∃ [one() ∧ () ∧ ()] ‖one‖ −alt = { ∃ [() ∧ () ∧ ()] ∶ ≥ one} b. alt() = ∧ ∀ ∈ alt[ < ]

<sup>11</sup>But see Umbach (2012) for the fact that not all scalar additives allow for accommodation, cf. German *noch*.

### Chiara Gianollo

The scalar alternatives () of *one* are strictly ordered; given the shape of the alternatives, in Chierchia's framework their exhaustification has to take place by means of the *E* operator, which corresponds to the meaning of 'even'. The result is felicitous only in NPI-licensing contexts.

The Romance version of the indefinite contains, in addition, a negative morpheme, which I assume receives narrow scope with respect to the scalar focus operator.

In Romance, new indefinites formed with *nec* appear invariably as elements of Negative Concord grammars, i.e., as n-words.<sup>12</sup> According to a prominent line of analysis (Zeijlstra 2004; 2014; Penka 2011), n-words are analyzed as bearers of a formal uninterpretable negative feature [uNeg], whose function is to trigger Agree operations, which result in the creation of syntactic dependencies.

Once the indefinite behaves as an n-word, it is clear that *ne(c)* has become an uninterpretable feature: it does not directly contribute a semantically interpretable negative operator, but only a morphosyntactic signal to insert one in the further derivation. Moreover, it is clear from the Early Romance data that the focus contribution has disappeared already at that stage: the *nec*-words behave as plain existentials and do not contribute scalarity.

When *nec* is grammaticalized as the component of an indefinite, the change involves a restriction in the particle's possibilities to select a focus semantic value: consequently, the alternatives must now be lexically selected by the predicate determining the restriction of the quantification (cf. discussion in Lahiri 1998, Chierchia 2013).

According to Chierchia (2013: 156–157), this explains why a sentence like (30) is ungrammatical in Hindi (where the focus particle is a morphological component of the indefinite *ek bhii* 'even one'), unlike its English counterpart:

	- a. *\*ek* one *bhii* even *aadmii* man *aayaa* came

'Any (=Even one) man came'

b. *Even one man came*

According to Chierchia, because of the univerbation, in Hindi alternatives are strictly specified by the lexical content of the nominal restriction: therefore, in

<sup>12</sup>These indefinites also show negative-polarity uses, where they have a 'positive' meaning in NPI-licensing contexts. I refer to Gianollo (2018: chapter 5) and the literature cited there for attempts to reconcile these uses with the n-word analysis.

### 2 Grammaticalization parameters and the retrieval of alternatives

(30a) they are lexically restricted to be men. This yields infelicity in an upwardentailing context, because a contradiction arises between the presupposition of the focus particle that the associate be the informationally strongest proposition and the fact that the lexically constrained alternatives ('two men came', 'three men came') are actually informationally stronger. In English, instead, alternatives are not lexically constrained: depending on the broader context, it is possible in principle to evoke a scale of alternative denotations to the element in focus, comprising less striking alternatives ('a woman came', 'a child came'); this way, the presupposition of *even* can be satisfied.

Assuming that an analogous mechanism is universally forced by univerbation, I conclude that the same applies to Latin *nec*-words, resulting in a maximal degree of bondedness of the particle and, consequently, in a context-independent, lexically restricted generation of alternatives.

### **5 Conclusions: The grammaticalization path**

In this final section I compare the conclusions emerging from the study of *nec* to the syntagmatic parameters for grammaticalization formulated by Lehmann (2015), with the aim of ascertaining whether what we learn from the diachrony of *nec* can provide generalizations on the nature of semantic change similar to those proposed for syntactic change.

Lehmann's (2015: 129–134; 152–170) syntagmatic parameters of grammaticalization capture the main effects of grammaticalization processes at the syntactic level. They are listed in (31):

	- a. *Structural scope*: the structural size of the construction which a grammatical item interacts with (Lehmann: 'helps to form'); it decreases during grammaticalization;
	- b. *Bondedness*: syntagmatic cohesion to another sign, varying from 'juxtaposition' to merger; it increases during grammaticalization;
	- c. *Syntagmatic variability*: degree of variation in combination, positioning, and syntactic dependence with respect to other phrases; it decreases during grammaticalization.

These syntactic parameters appear to have an interpretational correlate in the case of *nec*: the multiple functions of *nec* show varying degrees of discoursedependence, according to 'how far' the particle can look in order to retrieve a

### Chiara Gianollo

suitable alternative. Given the assumed structure-based mechanism for the generation of alternatives, the locus where the particle merges in the structure has a determining effect on the type of the alternatives. Increase in bondedness and decrease in syntagmatic variability with *nec* correlate with a change in the form taken by alternatives, which decrease in scope (i.e., size) from discourse units to (individual) scalar alternatives (ultimately lexically restricted once bondedness leads to univerbation).

Based on what was discussed in Section 4, we can formulate a cline for the retrieval of alternatives as in (32), which reflects the two parameters of structural scope and syntagmatic variability:

(32) Cline in the retrieval of alternatives:

across sentence topics > within the same sentence topic > within a scale introduced by the item in focus

When *nec* is a discourse-structuring connector, the alternatives are represented by the preceding discourse units across sentence topics. Alternatives to the additive stand-alone focus particle can also be found in the broader context, spanning across sentence topics.

In the correlative negation use, instead, the alternatives are found within the same sentence topic: they are explicitly listed as syntactic conjuncts, joined by means of the Junction Phrase, and are subject to an information-structural and syntactic parallelism condition. This step in the cline is reminiscent of what Zeevat & Jasinskaja (2007) observe for English *and*: they propose a uniform analysis of two-place and one-place *and* as an additive particle, and argue for a diachronic grammaticalization path from adverbial connector to clausal conjunction, consisting of the syntacticization of the retrieval of the alternative satisfying the particle's presupposition, which becomes fixed to the first conjunct. In Section 4.3 I proposed that the loss of polarity-switch uses with the correlative particle may be due to a strengthening of the parallelism requirement so as to encompass polarity. In turn, the loss of the discourse-structuring use attested in Romance may be diachronically connected to the loss of polarity switch with the correlative particle, due to a general stricter parallelism requirement on alternatives (that have to contain negation). The same-polarity requirement may be argued to lead to the loss of the contrastive value that the discourse connector could have.

In the innovative use as scalar particle, alternatives are not directly dependent on the surrounding discourse context, since a scale can be accommodated on the basis of world knowledge alone. The crucial ingredient of discourse-dependence is the anaphoric requirement imposed by the additive presupposition. Hence, the

### 2 Grammaticalization parameters and the retrieval of alternatives

loss of this anaphoric requirement when the particle gets reanalyzed as scalar amounts to a decrease in discourse-dependence.

We see, therefore, that the cline in (32) closely corresponds to a cline of discourse-dependence in the interpretation of the particle, as summarized in (33):

(33) Cline of discourse-dependence (from higher to lower): discourse-structuring > additive focus marking > correlation > scalar focus marking

The cline in (33) can in fact be argued to motivate the cline in (32): the loss of context-dependence lies at the core of the grammaticalization process, which in turn entails the decrease of the size of alternatives (decrease in scope and syntagmatic variability, and increase in bondedness).

The use of *nec* as a morpheme in the new Romance n-words shows a complete absence of discourse-dependence (in the relevant sense), representing the endpoint of the grammaticalization process: an extreme increase in bondedness concludes the process of functional enrichment, which causes the particle to become a mere morphosyntactic expression of uninterpretable formal features.

To conclude, the main generalization on semantic change emerging from this case study is that the loss of context-dependence is a determining factor in grammaticalization processes involving functional items. The main dimension of context-dependence considered here is the mechanism of retrieval of alternatives, which is dependent on the level of syntactic attachment of the particle. Different degrees of context-dependence depend on the portion of context within which the particle may look for a suitable antecedent. In the case of *nec*, the loss of the anaphoric requirement linked to its additive semantics emerges as the main trigger towards the decrease in the scope of the alternatives that is connected to the development of a scalar reading.

### **Acknowledgements**

I am grateful to the public of FoDS 2, the editors of this volume, three anonymous reviewers, Mira Ariel, Lieven Danckaert, Katja Jasinskaja, and Alan Munn for discussing the content of this chapter with me. Their constructive input helped improve and clarify many aspects of the presentation of data and of the analysis.

### **References**


### Chiara Gianollo


### **Chapter 3**

# **Vagueness, context-sensitivity and scale structure of four types of adjectives with the suffix** *-ish*

### Tabea Harris

Universität Mannheim

In English, the adjective-forming suffix *-ish* can be productively tacked onto relative adjectives (e.g. *tall-ish*) and total absolute adjectives (e.g. *dry-ish*), but not to most partial absolute adjectives (e.g. *?bent-ish*) or inherently non-scalar adjectives (e.g. *?pregnant-ish*). By applying Burnett's (2017) recent DelTCS framework situated in delineation semantics, which she enriched using the notions *tolerant, classical, strict*, first formulated by Cobreros et al. (2012), I will show why suffixal *-ish* is felicitous with the first two subtypes of adjectives, but not with the latter two. After a brief comparison with a similar framework from Lasersohn (1999), it will be shown that the scale structure in the DelTCS approach is derived from the adjective's context-sensitivity and vagueness patterns. Furthermore, the discussion will point to a few instances that do not neatly fit into the mold of current semantic analyses as well as some suggestions on how to obtain a clearer picture of the actual attested data.

### **1 Introduction**

The suffix *-ish* in Present-day English originates from the bound morpheme *-isc* in Old English and denotes associative meaning with nouns (i.e. *N.ish* 'having the character/nature of N', e.g. *baby.ish*) and approximative meaning mostly with adjectives and numerals as bases (i.e. *A.ish / Num.ish* 'approaching the quality of A/Num', e.g. *green.ish, 30-ish*) 1 (cf. the *Oxford English Dictionary*, Proffitt (2015)

<sup>1</sup>The notions *associative* and *approximative* in this context originate from Traugott & Trousdale (2013).

Tabea Harris. 2020. Vagueness, context-sensitivity and scale structure of four types of adjectives with the suffix *-ish*. In Remus Gergel & Jonathan Watkins (eds.), *Quantification and scales in change*, 67–84. Berlin: Language Science Press. DOI:10.5281/zenodo.3929241

### Tabea Harris

entry for *-ish*). Unlike nominal bases, which were present in Old English, adjectival bases started to appear in Middle English with color adjectives (e.g. *yellowish* in 1379, cf. Marchand (1969: 306).<sup>2</sup> Both uses are still highly productive, cf. Bauer et al. (2013: 305); Dixon (2014: 235) uses which are attested in several corpora and in many cases already listed in the OED. Thus as a suffix, *-ish* is quite prolific, attaching to numerous bases, including verbal (e.g. *snappish*, 1542, see the OED entry for *snappish*), adverbial (e.g. *nowish*; not listed in the OED yet, but attested in the COCA Davies (2008) once and 42 times in the corpus iWeb, featuring data from numerous websites), numeral bases (e.g. *elevenish* (1916), *fifty-five-ish* (1941), cf. OED, entry *-ish*, suffix1), and proper names (e.g. *Heine-ish* (1887), cf. OED, entry *-ish*, suffix1), as well as multi-word units such as compounds (e.g. *schoolgirlish* 1821, see corresponding OED entry), and phrases (e.g. *middle-of-thenightish)*<sup>3</sup> . The derivation of ethnic nouns and adjectives (e.g. *Engl.ishA/N*) is no longer productive and will be excluded from consideration here. Kuzmack (2007), in Traugott & Trousdale (2013: 234) observed that in the associative use, *-ish* denotes similarity to its base, while the approximative *-ish* stresses dissimilarity. The focus in this paper will be on the approximative use of *-ish* that most often occurs with adjectival bases. Consider examples (1–2) below:


<sup>2</sup>An anonymous reviewer refers to the results of a large-scale corpus study and states that some relative adjectives (e.g. *thinnish*, *thickish*) have started to appear as *-ish* derivatives at roughly the same time as color adjectives. Unfortunately I have no way of verifying this claim as the reviewer has not disclosed the corpus or corpora used. My own investigation of the *Penn-Helsinki parsed corpus of Middle English 2* (Kroch & Taylor 2000) has not turned up evidence of de-adjectival *-ish* adjectives in Middle English. After checking with the MEC (McSparran et al. 2000–2018), formerly *Middle English Dictionary* and the OED, it can be said that the dates of color adjectives and relative adjectives of the type mentioned above are indeed not far apart: for example, the earliest instances of *greenish*, *yellowish*, *reddish*, and *whitish* are dated in both dictionaries to 1398, while *blackish* and *brownish* first appear around the early 15th century (the OED dates *brownish* to 1555, however). Conversely, *thinnish* and *thickish* appear slightly later in the MEC (1425) and are dated to the middle of the 16th century in the OED. Thus, depending on the source used, the earliest dates of occurrence will slightly change. It is not wrong to say, however, that deadjectival *-ish* adjectives generally occurred with bases of color at one of the earliest stages.

<sup>3</sup>BNCweb entry Lehmann et al. (2000) BMS 1806, Fiction and Verse: *Gate-crashing the dream party*.

### 3 Vagueness, context-sensitivity and scale structure

In both examples the addition of *-ish* to the adjectival bases *tall* and *green*, respectively, explicitly marks that the standards set by the adjectives are approximated, but not reached completely.

The picture is complicated by the inherent vagueness of the relative adjective *tall* in (1) and, I propose, the total absolute adjective *green* in (2).<sup>4</sup> How can we know that a person counts as tall if we do not have a standard to which we can compare that person? And how do we determine such a standard? Where is the minimum threshold above which an object can be considered tall? These are questions frequently discussed in the literature about vagueness and vague adjectives in particular.<sup>5</sup> They address the central problem of vague predicates: determining borderline cases, fuzzy boundaries and the classical paradox of the Sorites, i.e. if we continuously add one centimeter to a building of average height, at one point we have to admit that it is tall. We do encounter the problem of not being able to say at which point exactly the building has reached the threshold and can unambiguously be considered tall, which is due to the incremental fashion of adding to the height of the building. Now consider examples (3–4) below:


Example (3) features a partial absolute adjective *wet*, which acquires the meaning 'less than fully wet' when *-ish* is attached. Similarly, in (4) the inherently non-scalar adjective *dead* is given a gradable meaning with *-ish* and denotes that the lawn is not yet totally beyond repair, but in a state that requires (professional) help. Note here that only 7 tokens were found for *wettish* in the *Corpus of Contemporary American English* (COCA) and only 2 tokens for *dead-ish* (and

<sup>4</sup>Whether color adjectives are actually classified as relative or as absolute adjectives is a matter of ongoing debate and the results thus far have eluded a clear picture (cf. Hansen & Chemla 2017 for an experimental approach). Burnett (2012a,b) considers them to fall into the relative camp due to the (syntactic) tests that are felicitous with them.

<sup>5</sup> For instance, concerning the determination of the standard degree of tallness, von Stechow (1984) proposes the positive operator *Pos* which aims at giving a unified treatment of polar opposites such as *tall-short*. I will not go into further detail here, the interested reader is kindly referred to von Stechow's work (e.g. 1984; 2009). See also Kennedy & McNally (2005) and Kennedy (2007) who employ the operator *pos* in their frameworks.

### Tabea Harris

none for *deadish*<sup>6</sup> ) in the corpus *Global web-based English* (GloWbE, Davies 2013). Both corpora are considered representative and balanced and feature 560 million words (COCA, Davies 2008) and a considerable 1.9 billion words (GloWbE), respectively. Compared to the adjectives *tallish* (43 tokens in GloWbE) and *greenish* (751 tokens in COCA), both *wettish* and *dead-ish* are virtually non-existent by comparison. Of course, the usual caveats for corpus-analytic studies apply. Since the aim here is not to provide a full-fledged corpus study, it will suffice to say that the preliminary results of the corpora indicate that *-ish* does not easily attach to adjectives that are partially absolute or non-scalar.<sup>7</sup>

The paper is structured as follows: Section 2 will introduce the four adjectival subtypes mentioned earlier, Section 3 will give an introduction to the *tolerant, classical, strict* framework that is employed in Burnett (2017). In this respect Section 4 features an analysis of the scale structure of the four different subtypes of adjectives with *-ish* and will encompass a discussion of why relative and total absolute adjectives are productive with the suffix, whereas partial absolute and non-scalar adjectives are infelicitous in most cases. Section 5 will conclude the findings and point to further areas of research.

### **2 Four subclasses of adjectives**

In the classification of four subgroups of adjectives, I follow Burnett (2017) and others who propose adjectives like *tall*, *expensive* and *cheap* as belonging to Relative Adjectives (RA), *empty*, *clean*, *straight* and others as being part of Total Absolute Adjectives (AA<sup>T</sup> ), *dirty*, *bent* and *wet* as included in the Partial Absolute Adjective class (AA<sup>P</sup> ) and, finally, adjectives such as *pregnant*, *dead*, or *hexagonal* which belong to the Non-Scalar adjective type (NS) (cf. Burnett 2017: 4.<sup>8</sup> The distinction into relative and absolute adjectives is nothing new, just as the

<sup>6</sup>The fact that the spelling of *dead-ish* contains a hyphen can be an indicator that its usage is unusual and marked for the speaker. It is often found in cases with hiatus (e.g. *country-ish*), when the base word consists of an abbreviation (e.g. *Espn-ish*, *Cia-ish*), after certain numerals (e.g. *23-ish*), and frequently after proper names (e.g. *Verne-ish*). One phonological reason why *-ish* nevertheless attaches to *dead* (but not to *pregnant*, *hexagonal*, *illegal*, etc., which are also all non-scalar) is that the suffix primarily selects monosyllabic bases (cf. Dixon 2014: 235). While it can easily be shown that this is not a constraint, preference of monosyllabic bases should be understood here in the sense of frequency.

<sup>7</sup> In order to conduct a proper corpus analysis, among other things, we would need to expand the class of adjectives to include an equal number of each subclass which is then compared in each of the corpora mentioned above.

<sup>8</sup>The distinction between total and partial adjectives is said to have originated from Yoon (1996) cf. Kennedy & McNally (2005: 355).

### 3 Vagueness, context-sensitivity and scale structure

observation that the absolute class incorporates two distinct subclasses is well established (see, for example, Rotstein & Winter 2004; Kennedy & McNally 2005; Kennedy 2007; Toledo & Sassoon 2011). For instance, Kennedy & McNally (2005) have investigated the RA-AA distinction with respect to the felicity of degree modifiers (e.g. *slightly*, *perfectly*, *completely*) and *for* phrases, the latter of which explicitly specifies the contextual domain that determines the standard degree and which is natural for relative adjectives, but odd for absolute adjectives. They follow Unger (1975), who claims that only relative adjectives are context-sensitive and vague. A different view is proposed by Rotstein & Winter (2004), who investigate the total-partial distinction for gradable adjectives and who contend that absolute adjectives can *also* be context-sensitive and vague when they are used in "loose talk" cf. Lewis (1979). In these cases, the standard for total/partial absolute adjectives is not necessarily the exact maximal or minimal degree on a scale. Evidence for this view is provided by the use of the degree modifier *completely*, which selects absolute standards. For instance, consider the "maximality modifier" *completely* as discussed in Kennedy & McNally (2005) and Rotstein & Winter (2004), among others. In (5) below, when *completely* modifies the AA<sup>T</sup> *clean*, the adjective conveys the maximal amount of cleanliness, where the floor is so immaculate as to be able to literally eat from it.

### (5) *The floor is not clean, it is completely clean.*

Toledo & Sassoon (2011) maintain that both accounts have their merits, but neither is entirely able to account for all the facts (cf. Toledo & Sassoon 2011: 140). They propose an approach which takes comparison classes (CCs) into account, an approach which also finds application in Burnett (2017) concerning comparisonclass based context sensitivity. Relative adjectives show a more substantial form of both context-sensitivity and vagueness than absolute adjectives in that they exhibit universal *and* existential context-sensitivity (AAs only show the latter) and in that they permit potentially vague positive *and* negative forms (AAs show a different distribution depending on the subtype), cf. Burnett (2017). The notions above require some explanation. Universal context-sensitivity, according to Burnett (2017: 41), corresponds to the adjective's ability to shift its thresholds in any comparison class. Kennedy (2007: 28) and others have suggested to employ a definite description test to distinguish whether an adjective is able to shift its standard of comparison:


### Tabea Harris

According to this test, the adjective in (6) would independently be considered true of two objects (e.g. watches) outside of the test, but has been shown to pertain to the *more* expensive one in the context of this utterance. In (7) however, when both objects (e.g. sweaters) are either both green or not green, uttering (7) becomes infelicitous. Absolute adjectives are therefore not considered to be universally context-sensitive, but rather existentially so. They appear to be able to shift their criteria of application in some comparison classes, but do so only when they appear in the context of "imprecision" cf. Burnett (2017: 42). Hence, when the absolute adjective *green* is used loosely, modification with a *for*-phrase becomes more acceptable:

### (8) *This lawn is green for midsummer in Texas.*

Example (8) is acceptable when we describe a lawn that only shows a few brown patches here and there, but is not completely dead. The same utterance would seem infelicitous in a context in which the lawn is completely lush green.

By using the notion *potentially vague*, Burnett (2017: 49) aims at showing that the vagueness of relative adjectives and the vagueness in the case of "loose" uses of absolute adjectives is due to a single underlying phenomenon. In her conception, vagueness is a stage-level property that is subject to contextual variation, and potential vagueness is defined as follows:

(9) An adjective *P* is potentially vague iff there is some context *c* such that *P* gives rise to a Soritical argument in *c* (Burnett (2017: 50), emphasis in original).

The potential vagueness described is not distributed symmetrically over absolute adjectives. While relative adjectives have potentially vague positive (P) and negative forms (¬P), total absolute adjectives only have positive potentially vague forms (P), and partial absolute adjectives display the opposite pattern, i.e. they only have negative potentially vague forms (¬P). In order to illuminate the relationship, consider the following examples:


### 3 Vagueness, context-sensitivity and scale structure


The given examples require some clarification. In order to illustrate (12), consider two containers, one of which is entirely empty (*x*), the other of which holds exactly one bean (*y*). In a tolerant use of the adjective, both containers would be judged empty. Conversely, container *x* in example (13) holds a single bean, whereas *y* contains no bean at all. Thus, container *y* is not judged ¬*empty* by comparison and the principle of tolerance does not hold in this case, cf. Burnett (2017: 51). Let us now turn to the examples illustrating AA<sup>P</sup> s, i.e. adjectives typically associated with scales that have minimal elements. (14) claims intolerance concerning the positive *dirty*: An object *y* that is completely clean will not be considered dirty, even if object *x* only differs in having one stain. The negative form *not dirty* satisfies tolerance however, i.e. it is P-vague. Consider a situation where Mary wants to paint her living room. In that case, choosing an outfit which has a speck of dirt on it will be considered as being not dirty, i.e. in the context of painting, a single stain on the chosen outfit will be perceived as irrelevant, cf. Burnett (2017: 52).

In sum, relative adjectives are symmetrically vague in that they do not discriminate between positive and negative forms (hence, they are potentially vague with either form of the predicate). Absolute adjectives, by contrast, exhibit an asymmetric distribution of vagueness. Total absolute adjectives are tolerant with positive forms, but intolerant when it comes to distinguishing individuals situated at the bottom end of some scale from those that are at the second to last degree: In Burnett's (2017: 52) example, *empty* is infelicitous with *x* containing a single spectator (in a theater, for instance), and *y* having no spectator at all. In other words, if *x* is *not empty* and differs by only one individual from *y*, *y* cannot be considered tolerantly true in this context. With partial absolute adjectives, the picture is reversed. They exhibit negative potential vagueness in that they are tolerant with respect to individuals at the lower endpoint of a scale (15), but display no potential vagueness with corresponding positive forms (14).

### Tabea Harris

Finally, non-scalar adjectives exhibit neither context-sensitivity nor potentially vague forms in their precise uses. Burnett however claims that they can be turned into scalar absolute adjectives when they assume gradable interpretations (2017: 44). In a "loose" use of *pregnant*, for instance, we can see the properties of a gradable, context-sensitive adjective:

### (16) *Sue is very pregnant, for being in the third month.*

(16) is only felicitous when we assume that Sue's pregnancy is already much more showing in her third month when compared to other women. Burnett (2017: 44) observes that the scalar modifier *very* facilitates this gradable use of the adjective. In assuming that the non-scalar adjective has been coerced into an absolute adjective, Burnett is able to keep the semantic class of non-scalar adjectives "pure" and can claim that as such they are not context-sensitive and non-gradable. She assumes that the distinction of AAs and NSs is of a pragmatic nature, i.e. a shifting operation in the level of precision with which the adjective is used (2017: 95). This assumption involves the view that both AAs and NSs have precise semantic denotations, but are variable with respect to their pragmatic denotations. Applying this line of reasoning to the workings of complex words addresses a desideratum for morphological theory formulated in Plag et al. (1999: 226). Specifically, it takes the study of pragmatics into account and investigates the effect context has on the use of complex words.<sup>9</sup> Exactly what this perspective entails will be the subject of the following sections.

### **3 The** *delineation tolerant, classical, strict* **framework**

Bochnak & Csipak (2014) discussed the semantics of *-ish* within a degree semantics framework. They observed that *-ish* is felicitous with adjectives containing an open scale or those exhibiting an upper bound (i.e. a maximal value), but are questionable with adjectives which contain a lower bounded scale (2014: 435– 436), i.e. relative adjectives, total absolute adjectives, and partial absolute adjectives, respectively. The few cases where *-ish* attaches to non-scalar adjectives like *dead* are not discussed in their framework.

In the present paper, I propose an analysis in an alternative framework based on Cobreros et al.'s (2012) notion of *tolerant, classical, strict* (henceforth TCS), which has been applied to vague adjectives in the recent framework of Burnett (2017). The idea to formalize vague predicates in a trivalent non-classical logic

<sup>9</sup> I thank the anonymous reviewer who pointed me to their study.

### 3 Vagueness, context-sensitivity and scale structure

stems from the fact that "the principle of tolerance gives rise to the sorites paradox" in classical logic (Cobreros et al. 2012: 348). That is, in order to allow a truth value which makes reference to tolerance as exemplified in (17) a third value has to be introduced.

(17) If some individual *x* is *P*, and *x* and *y* are only imperceptibly different in respects relevant for the application of the predicate *P*, then *y* is *P* as well. (Cobreros et al. 2012: 348)

This general idea has been implemented in different ways, as for example in the pragmatically oriented framework by Lasersohn (1999), which assigns a pragmatic halo around expressions such as the following:

### (18) *Mary arrived at three o'clock.*

The time expression is taken to be close enough to truth in case of Mary arriving 15 seconds after three o'clock, for instance. The expression in (18) is assigned a denotation under which it is true and additionally contains a set of times that "differ from the denotation only in some respect that is pragmatically ignorable in context" (Lasersohn 1999: 526). This set of times is then understood as the pragmatic halo of the expression in (18) which is at the halo's center. Burnett (2017: 29) notes that the *Halo* framework was not originally designed as a semantic theory of vagueness, but observes that the basic ideas are very similar to hers (and the model of TCS in general). In particular, the notion of *tolerant truth* is taken to be the equivalent to Lasersohn's *close enough to truth* (Burnett 2017: 32). Further, what is described as *pragmatically ignorable* in Lasersohn's framework is paralleled by a notion of *indifference* in TCS (Burnett 2017: 32). Thus, both frameworks include the "core intuition that at least one aspect of vagueness/pragmatic slack involves loosening the conditions of application of an expression with a precise semantic denotation to include other objects that are considered to differ in only *pragmatically ignorable* ways" (Burnett 2017: 32–33). However, even though the two frameworks may superficially be understood as mirror images of each other, Burnett suggests the TCS framework to constitute a refinement of pragmatic halos. Specifically, in TCS it is possible to derive non-classical denotations and orderings, which are simply given in the model by Lasersohn (1999) (cf. Burnett 2017: 33).

Burnett's *delineation tolerant, classical, strict* (DelTCS in short) is part of a class of comparison-class-based semantic frameworks going back to Klein (1980). The aim of her model is to provide a new relationship between the vagueness of adjectival predicates, the properties of context-sensitivity that these predicates

### Tabea Harris

involve as well as their corresponding scale structure. Her logical framework, based on indifference relations, "preserves the intuition that vague predicates are tolerant, but avoids the Sorites paradox" by a step-wise validation of tolerance (Burnett 2017: 28). Indifference relations refer to change that is marginal enough to not make a difference to categorization of a predicate (i.e. one Euro more or less will not make a watch expensive or not expensive as compared to another watch which exhibits the value P or ¬P, respectively) (cf. Burnett 2017: 1). Within the TCS extension of the delineation framework, she assumes that the classical semantic framework is enriched with tolerant/strict denotations, which are established as a second step (Burnett 2017: 72, cf. also Cobreros et al. 2012). In doing so, she adds the function ∼, which "maps a predicate and a comparison class to an indifference relation on the members of the class" (2017:72). Applied to the pattern of potential vagueness of the AA<sup>T</sup> *empty* (see examples 12 and 13 above) we obtain the following distribution (cf. Burnett 2017: 76):


The definition for a tolerant model according to Burnett (2017: 72) is given below:

(21) Tolerant model: For all P and all X ⊆ D, ∼ is a binary relation on X.

The novelty and difference to Cobreros et al. (2012) lies in the fact that a predicate *and* a CC are mapped to the indifference relation ∼ (cf. Burnett 2017: 72). She splits the denotational system in half, assuming the classical denotations to be semantic and the secondary tolerant and strict denotations to be pragmatic in nature, the latter of which are formally defined in (22) and (23) (cf. Burnett 2017: 73):

(22) Tolerant denotation

$$\begin{aligned} \left[ \begin{bmatrix} P \end{bmatrix} \right]\_X^t = \left\{ \mathbf{x} \, : \, \exists d \; \sim\_P^X \; \mathbf{x} \, : \, d \in \left[ P \right] \mathbf{x} \right\} \end{aligned} $$

(23) Strict denotation

$$\left\lbrack \left[ P \right] \right\rbrack\_X^s = \left\{ \mathbf{x} \; : \; \forall d \; \neg\_P^X \; \mathbf{x}, d \in \left[ P \right] \mathbf{x} \right\}$$

By adopting the view that tolerant and strict denotations are pragmatic in nature, she presents a solution to the paradox of absolute adjectives which become

### 3 Vagueness, context-sensitivity and scale structure

gradable through a derivational process from comparison-class-based (existential) context-sensitivity that is essentially pragmatic, not semantic as in the classical interpretation (2017:89).

The vague and context-sensitive properties that adjectival predicates possess are modeled by assuming a series of constraints that pertain to their different distributions. For the present purposes I will restrict the discussion of constraints to those that differ for the four subgroups of adjectives. Hence, given that relative adjectives are potentially vague with P and ¬P, they are symmetrical in their indifference relations, whereas absolute adjectives display an asymmetric distribution (see above), which is encoded into their indifference relations,<sup>10</sup> as shown below (cf. Burnett 2017: 77):


The axioms ensure that total absolute adjectives are identical in their classical and strict denotations, and partial absolute adjectives have identical classical and tolerant denotations. Since non-scalar adjectives do not show potential vagueness, it is assumed that P and ¬P are both precise, i.e. their classical (semantic) denotations coincide with their pragmatic denotations, which is ensured by the pragmatic constraint *Be precise* (cf. Burnett 2017: 77–78).

The assumptions for context-sensitivity patterns follow in a straightforward manner. Given that relative adjectives exhibit both universal and existential context-sensitivity, their denotations are much less restricted in variation depending on a comparison class. Thus, relative adjectives are consistently felicitous with *for* phrases. Consider the following example, which depicts the assessment to an exchange about the prices of two bottles of wine, one of which costs \$130 and another which is of a more affordable price of \$17. The *for* phrase characterizing the less expensive of the two is fully felicitous with the RA *cheap*:

### (26) Gifford: *Seventeen, that's cheap for a bottle...* (COCA, Spoken, NBC Today: *Today's talk*, 27.04.2011)

Absolute adjectives have been shown to only allow for existential contextsensitivity and their denotations vary according to subtype: total absolute ad-

<sup>10</sup>A note on vocabulary: a<sup>1</sup> , a<sup>2</sup> , a<sup>n</sup> refer to individual constants, Q refers to AA<sup>T</sup> , R to AA<sup>P</sup> . For a comprehensive vocabulary of Burnett's model, see her p. 56.

### Tabea Harris

jectives have context-sensitive tolerant denotations (the classical and strict denotations are identical across comparison class, see above), partial absolute adjectives involve corresponding strict denotations. Non-scalar adjectives do not show variation in their tolerant and strict denotations since they are subject to the pragmatic constraint *Be precise* (cf. Burnett 2017: 85). The denotations associated with these patterns are given below (cf. Burnett 2017: 85, slightly adapted):

(27) AA<sup>T</sup> : For all X ⊆ D, J T K <sup>=</sup> <sup>J</sup> T K (28) AA<sup>P</sup> : For all X ⊆ D, J P K <sup>=</sup> <sup>J</sup> P K

The patterns can give an explanation for why AAs become more acceptable with *for* phrases: the *for* phrase specifies explicitly the content of a given comparison class. Burnett (2017: 86) holds that the informational content CCs contribute to the interpretation of absolute adjectives is non-trivial and informative in their tolerant or strict uses. Thus while the semantic denotations remain fixed across contexts, sentences like (8) above improve when they are used loosely. Since the scale structure is derived from the context-sensitivity patterns associated with the different classes of adjectives, corresponding effects are predicted. The scale structure properties will be subject of the section below.

### **4 Scale structure and the suffix** *-ish* **with adjectives in DelTCS**

As we have seen in the examples at the beginning, the suffix *-ish* felicitously selects those adjectives that fall in the classes of RAs and AA<sup>T</sup> s, respectively (i.e. *tallish*, *greenish*), but is rather infrequent with the other two classes, i.e. AA<sup>P</sup> s and NSs, even if we find a number of attestations. In Bochnak & Csipak's (2014) framework, it was assumed that the reason why *-ish* was felicitous with openscale adjectives (i.e. RAs) and those that feature a maximum value (i.e. AA<sup>T</sup> s) was due to the fact that they pick out a degree that is slightly less than a contextually given standard (2014: 436). With adjectives that make reference to a lower-bounded scale, this option is not given, since they are already situated at the lower end of the scale and hence cannot be below the minimum standard.

In the DelTCS framework introduced above, scales are derived from the adjective's corresponding denotations. Relative adjectives show both universal and existential context-sensitivity and are potentially vague for the positive form of

### 3 Vagueness, context-sensitivity and scale structure

a predicate as well as for its negation. Their scales are assumed to be derived from their semantic denotations, i.e. they have neither maximal nor minimal elements (cf. Burnett 2017: 90, 107), which corresponds to Bochnak and Csipak's open scale. Total absolute adjectives were shown to be potentially vague for the positive form and to exhibit existential context-sensitivity. They are associated with scales that are derived from their tolerant denotations, i.e. their scales exhibit maximal elements, whereas the opposite is true for partial absolute adjectives (2017: 90, 106). AA<sup>P</sup> s exhibited an asymmetry in their potential vagueness pattern that showed the reverse, i.e. they were considered to be (potentially) vague for the negation of the predicate (¬P). Their scales are correspondingly derived from their strict denotations and thus contain minimal elements (2017: 90, 106). Since the pragmatic denotations (i.e. the tolerant and strict denotations) of non-scalar adjectives are identical with their semantic denotations, they will not be associated with any scales Burnett (2017: 90). Burnett however notes an exception to this observation. Hence, non-scalar adjectives can be coerced into scalar predicates, making them subject to the same constraints that hold for absolute adjectives. In that case, they do not follow the axiom *Be precise*, which has the consequence that they can be analyzed in the same way as absolute adjectives. Here, Burnett's approach is very different from other frameworks in that she assumes non-scalar adjectives to actually be absolute adjectives which are used with a higher degree of precision (2017: 97–98). In other words, the difference between AAs and NSs is manifested in their pragmatic denotations, not their semantic ones (cf. Burnett (2017: 98) and it depends on how an adjective is used in a particular context: In a precise use, a non-scalar adjective will not allow for variable meaning by the constraint *Be precise*, whereas in a context that allows for a lower level of precision, the NS is coerced and becomes context-sensitive, i.e. showing the characteristics of a vague and gradable predicate. For instance, example (16) above, which was used in a context in which a non-scalar adjective (i.e. *pregnant*) can turn into a gradable one with the help of a *for* phrase, shows that these adjectives can be used felicitously with a gradable meaning given that the context allows for a loose use. In these cases, NSs are also quite natural in comparative constructions (example from Burnett 2017: 96):

### (29) *Sarah is more pregnant than Sue; Sarah is showing more.*

What do these assumptions mean for the suffix *-ish* and its application? For RAs and both AAs the account can be laid out in a quite straightforward manner. The scale structures show the same characteristics as in Bochnak & Csipak (2014) and *-ish* targets these scales. In doing so, it lowers the precision with

### Tabea Harris

which the adjectives are used, making them available for "loose" use. The difference in Burnett's (2017) account is that the scales are derived from the adjective's context-sensitivity. Recall that with relative adjectives, both universal and existential context-sensitivity is possible. This allows RAs to be associated with an open scale that shows neither maximal nor minimal elements (as for instance with *tall*). With its approximative meaning, *-ish* can approach the quality instantiated by the adjective (e.g. *tall-ish*), but does not reach it in full measure. The predictions concerning potential vagueness can be explained by the indifference relations of relative adjectives, which were said to be symmetric, i.e. they are potentially vague with both P and ¬P. With *tallish* (i.e. P) and *not tallish* (i.e. ¬P), we can observe that the respective antonyms *shortish* and *not shortish* are also relative adjectives, i.e. the two adjectival types display a mirror image of each other.

Total absolute adjectives were shown to be associated with scales derived from their tolerant denotations, i.e. they exhibited maximal elements and only have potential vagueness with their positive forms. That is, the positive forms of *clean*, *dry*, or *straight* can be targeted by *-ish*, again with *-ish* adding the meaning of approximation to the positive form. Their antonyms *dirty*, *wet*, or *bent*, however are all partial absolute adjectives, which are associated with scales that have minimal elements which are derived from their strict denotations. These adjectives show potentially vague negative forms (¬P) and cannot be targeted as easily by *-ish* (cf. also Bochnak & Csipak 2014: 437. We can thus say that the applicability with *-ish* correlates with the type of adjective and their associated properties (with a few exceptions, see example 3 above).

Since non-scalar adjectives generally occur in contexts that favor precise uses (cf. Burnett 2017: 95–96), they are hardly found with *-ish*. However, as we have seen in (4) above, in some cases they can be turned into an absolute adjective when the standard of precision with which they are used is loosened. This is what happened with the NS *dead*. In example (4), *-ish* is able to attach to *dead* because the adjective's conditions of application have been loosened. In this case, it features an upper bound that is approached by *-ish*, i.e. it has a maximal element as is generally the case with total absolute adjectives. Burnett (2017: 112–113) notes however that many coerced non-scalar adjectives are able to be associated with both types of scale structure, i.e. those that have maximal and those that have minimal elements, which her examples show, given slightly adapted in (30) to (32), respectively.

(30) DEA agent 1: *Bring me up to speed on Tuco Salamanca.* DEA agent 2: *Dead.*

3 Vagueness, context-sensitivity and scale structure

DEA agent 1: *Still?* DEA agent 2: *Completely.* (*Breaking Bad* 2009. Season 2, episode 5, "Breakage.")


Examples (30) and (31) illustrate the coerced NS *dead* with a scale associated with maximal endpoints (i.e. characteristic of AA<sup>T</sup> s), whereas (32) is an example for a partial absolute adjective, indicated by the modifier *slightly*. Thus, *dead* may be coerced into either type of adjective, depending on context. The NS *dead* is not exclusive in showing both patterns. Burnett (2017: 113) gives ethnic adjectives such as *Canadian* and the non-scalar adjective *illegal* as further examples. The exact conditions that are responsible for an NS to select which end of a scale remain to be elucidated.

The findings of attachability for *-ish* should not be taken to be absolute. For example, we find cases of relative adjectives that so far have not been attested with *-ish* (e.g. *?intelligent-ish*), whereas some non-scalar adjectives occur quite freely with *-ish* (e.g. *squarish*, *dead-ish* to some extent). Thus, there has to be a further factor that plays a role in the applicability of *-ish* that has not yet been accounted for. One factor that immediately comes to mind is a non-semantic one. It has to do with the syllable structure of adjectives that are favored by *-ish*. As has been mentioned above, *-ish* preferably attaches to monosyllabic bases which is true for *squarish* and *dead-ish*, but not for *?intelligent-ish*. This might be a factor which rather concerns productivity of *-ish* derivatives however, and not so much whether they are generally well-formed and felicitous as we do of course find bases with more than one syllable to which *-ish* attaches. Thus, this factor alone will undoubtedly not be sufficient to account for the patterns we have encountered with *-ish*, but it could be seen as a contributing factor.

### **5 Conclusion**

This paper has investigated the English suffix *-ish* with the four subtypes of adjectives that are discussed in the literature. In doing so, the *delineation tolerant, classical, strict* framework recently implemented by Burnett (2017) was used. It has been found that the patterns *-ish* shows with different types of adjectives can felicitously be described in a framework in which the scale structure of adjectives is derived by the patterns of context-sensitivity they depict. Context-sensitivity

### Tabea Harris

thus correlates with the patterns found for potential vagueness and scales. By mapping tolerant and strict pragmatic denotations on basic classical (semantic) ones, the framework approaches issues of vagueness from a more pragmatic angle than other well-known frameworks such as degree semantics.

It has further been noted that suffixal *-ish* does not show an absolute fit concerning its productivity patterns with different adjectival types. Therefore, it has been suggested to look for further (e.g. phonological, but presumably also other semantic) factors in conjunction with the semantic ones introduced above to explain these divergent occurrences of *-ish*. To be sure, the general tendency of productive derivations with a certain type of base is not disputed, but the counterexamples mentioned above should nevertheless be accounted for in a semantic theory, even though they amount to only a few attestations. However, rather than dismissing them as rare and subscribing to the view that unless frequent (counter-)examples of a phenomenon are attested, the phenomenon is nonexistent, I take the attestations that deviate from the general pattern as being deemed acceptable enough by speakers in particular contexts. In order to make this claim stronger, a full-fledged corpus analysis could be devised in order to see which forms are actually attested with which frequencies and in what contexts they are found to occur.

### **References**


### Tabea Harris

Traugott, Elizabeth C. & Graeme Trousdale. 2013. *Constructionalization and constructional changes*. Oxford: Oxford University Press.

Unger, Peter. 1975. *Ignorance: A case for scepticism*. Oxford: Clarendon Press.


## **Chapter 4**

# **On the semantic change of evidential argument** *jakoby***-clauses in Polish**

Łukasz Jędrzejowski Universität zu Köln

> The main aim of this chapter is to examine the semantic change of evidential argument clauses headed by the complementizer *jakoby* in the history of Polish. Mainly, I argue that *jakoby* developed from a hypothetical comparative complementizer meaning 'as if' into a hearsay complementizer, and provide empirical evidence showing that this process happened in the late Old Polish period, i.e. around 1500. To begin with, I compare *jakoby*-clauses with complement *że*-clauses ('that' clauses) at the syntax-semantics interface, elaborate on their selected differences, and account for the source of these differences. Diachronically, I show that two factors in the lexical meaning of *jakoby* were responsible for the semantic change that it underwent: (i) equative comparison and (ii) counterfactual meaning. Both factors are taken to have paved the way for inferences from reportative or hearsay information and, simultaneously, for the compatibility with an informational conversational background.

### **1 The puzzle**

Compare the two following sentences from Polish introduced by the complementizer *jakoby*. Whereas the example given in (1a) is from Old Polish, (1b) illustrates how argument *jakoby*-clauses are mainly used in Present-day Polish:

(1) a. *iżeć* that *się* refl *jest* be.3sg *ludziem* people.dat *na* on *ziemi* earth.loc *tako* so *było* be.l-ptcp.sg.n *widziało,* seem.l-ptcp.sg.n *jakoby* jakoby *się* refl *ono* it *na* on *nie* them.acc *obalić* slay.inf

Łukasz Jędrzejowski. 2020. On the semantic change of evidential argument *jakoby*-clauses in Polish. In Remus Gergel & Jonathan Watkins (eds.), *Quantification and scales in change*, 85–130. Berlin: Language Science Press. DOI:10.5281/zenodo.3929243

Łukasz Jędrzejowski

> *było* be.l-ptcp.sg.n *chciało* want.l-ptcp.sg.n 'that it seemed to the people on earth as if it wanted to slay all of them' (KG, *Kazanie I: Na Boże Naordzenie*, 26–27) b. *Firma* company *zaprzeczyła,* deny.l-ptcp.sg.f *jakoby* jakoby *były* be.l-ptcp.n-vir.pl *zgłoszenia* reports *o* about *wadliwych* faulty *kartach.* cards.loc 'The company denied that there were supposedly reports about faulty prepaid cards.' (NKJP, *Dziennik Zachodni*, 27/9/2006)

In (1a), the dependent clause is introduced by the hypothetical comparative complementizer *jakoby* corresponding to the meaning of the English complex complementizer *as if*, as the English paraphrase of (1a) indicates, and it is embedded under the matrix verb *widzieć* 'seem'. In (1b), in turn, the *jakoby*-clause is embedded under the speech verb *zaprzeczać* 'deny'.<sup>1</sup> What both clauses have in common is that they occupy one of the argument positions of the matrix verb (= argument clauses). However, in (1b) *jakoby* itself does not render the meaning of what English *as if* expresses; instead it comprises the compositional meaning of a complementizer introducing a dependent declarative clause (= *that*) and, at the same time, of a hearsay adverb (e.g. *allegedly*, *supposedly* or *reportedly*), giving rise to a hearsay or a reportative interpretation. The meaning of *jakoby* must have changed because in Present-day Polish *jakoby*-clauses are unembeddable under verbs of seeming, as was the case in Old Polish, see (1a) above:

(2) \* *Firmie* company.dat *wydaje* seem.3sg *się,* refl *jakoby* jakoby *były* be.l-ptcp.n-vir.pl *zgłoszenia* reports *o* about *wadliwych* faulty *kartach.* cards.loc Intended meaning: 'It seems to the company as if there were any reports about faulty prepaid cards.'

1 *Jakoby* can also be used as a hearsay adverb:

(i) *Sąsiedzi* neighbors *kupili* buy.l-ptcp.vir.pl *jakoby* jakoby *nowy* new *samochód.* car 'Supposedly, our neighbors have bought a new car.'

I am not concerned with *jakoby* used as an adverb in this chapter; for more details see Jędrzejowski (2012), Socka (2010), Stępień (2008), Wiemer (2015), Wiemer & Socka (2017a, 2017b), Żabowska (2008), among many others.

### 4 On the semantic change of evidential argument jakoby-clauses in Polish

Remarkably, other West-Slavic languages like Czech have not experienced this change:

### (3) Czech, Radek Šimík (pc.):


The main objective of this study is to figure out to what extent and under what circumstances *jakoby* used as a complementizer changed in the history of Polish.

The structure of this chapter is as follows. Section 2 is concerned with the question of how argument *jakoby*-clauses are used in Present-day Polish. In this context, I will compare *jakoby*-clauses with canonical subordinate clauses headed by the complementizer *że* 'that', and point out several striking differences between both clause types at the syntax-semantics interface. In Section 3, I will give an overview of how *jakoby*-clauses could be used in older stages of Polish. A formal account of to what extent and under what circumstances *jakoby* changed is presented in Section 4. In modeling this change, I will make use of the possible worlds semantics initiated by Kratzer (1981; 1991; 2012) and developed further for evidential expressions by Faller (2002; 2011) and Matthewson et al. (2007). Finally, I conclude the findings in Section 5.

### **2** *Jakoby***-clauses in Present-day Polish**

In this section, I examine selected properties of *jakoby*-clauses in Present-day Polish at the syntax-semantics interface. In doing so, I focus first on syntactic peculiarities by comparing *jakoby*-clauses to canonical declarative *że*-clauses (= *that*clauses). Then, I account for where the differences between both clause types come from by decomposing the meaning of the complementizer *jakoby*.

### Łukasz Jędrzejowski

### **2.1 Licensing conditions**

Complement clauses in Polish are usually headed by the complementizer *że* 'that'.<sup>2</sup> In this connection, I propose the following descriptive condition: If a *jakoby*-clause occupies an argument slot of a clause-embedding predicate, it can always be replaced by a *że*-clause. Correspondingly, the embedded *jakoby*-clause given in (1b) – repeated here for convenience as (4a) – is supposed to be replaceable by a *że*-clause. This prediction is borne out:

(4) a. *Firma* company *zaprzeczyła,* deny.l-ptcp.sg.f *jakoby* jakoby *były* be.l-ptcp.n-vir.pl *zgłoszenia* reports *o* about *wadliwych* faulty *kartach.* cards.loc 'The company denied that there were supposedly reports about faulty prepaid cards.' b. *Firma* company *zaprzeczyła,* deny.l-ptcp.sg.f *że* that *były* be.l-ptcp.n-vir.pl *zgłoszenia* reports *o* about *wadliwych* faulty *kartach.* cards.loc

> 'The company denied that there were any reports about faulty prepaid cards.'

(i) *Każda* every *matka* mother *chce,* want.3sg *żeby* żeby *jej* her *syn* son *chodził* go.l-ptcp.sg.m *do* to *przedszkola.* kindergarten.gen 'Every mother wants her son to go to the kindergarten.'

Complements embedded under volitional or desiderative predicates require the presence of the complex complementizer *żeby*, consisting of the simple complementizer *że* 'that' and the conditional/subjunctive clitic *by*. The clitic has to occur adjacent to *że* and cannot be omitted:

(ii) \* *Każda* every *matka* mother *chce,* want.3sg *że* że *jej* her *syn* son *chodził* go.l-ptcp.sg.m *do* to *przedszkola.* kindergarten.gen Intended meaning: 'Every mother wants her son to go to the kindergarten.'

Following the generative mainstream literature on Polish complex clauses going back to Tajsner (1989), Willim (1989), Witkoś (1998), Bondaruk (2004), among many others, I take *żeby* to be a complex C-head. Alternatively, one could argue for a more fine-grained C-layer analysis along the lines of Rizzi (1997) and postulate two different structural positions - one for *że* and one for *by* - within the C-domain, as Szczegielniak (1999) does. Alternative analyses are offered by Migdalski (2006; 2010; 2016) and Tomaszewicz (2012). As nothing hinges on whether one compares *jakoby* with *że* or with *żeby*, I restrict myself to the former in the present chapter.

<sup>2</sup>Note that in some environments a more complex complementizer is required, i.e. *żeby*:

### 4 On the semantic change of evidential argument jakoby-clauses in Polish

However, not every *że*-clause can be replaced by a *jakoby*-clause. In other words, the condition proposed above is not bidirectional:

	- b. \* *Dziwi* be.amazed.3sg *mnie,* me.acc *jakoby* jakoby *były* be.l-ptcp.n-vir.pl *zgłoszenia* reports *o* about *wadliwych* faulty *kartach.* cards.loc

Based on this contrast, we observe that *jakoby*-clauses cannot be embedded under exclamative predicates like *dziwić* (*się*) 'be amazed'/'be surprised'. Such a restriction does not occur with regard to *że*-clauses. A similar conclusion can be drawn as to perception verbs being used metaphorically. Ibarretxe-Antuñano (1999) points out, based on Sweetser (1990), that olfactory verbs, e.g. *smell*, in English, Spanish and Basque can have a non-literal meaning that, in turn, depending on the language can be paraphrased as *trail something*, *suspect*, *guess* or *investigate*. They are often connoted with negative situations, as the following example illustrates:

(6) *I smell something fishy about this deal.* (Sweetser 1990: 37)

The Polish olfactory verb *czuć* 'smell' (lit. 'feel') behaves in a similar way. If it is used metaphorically, it means 'suspect' and can embed *że*-clauses:

(7) *Jeden* one *z* of:the *polityków* politicians *czuje,* feel.3sg *że* that *niebawem* soon *wybuchnie* break:out.3sg *wielki* huge *skandal* scandal *na* on *arenie* arena.loc *międzynarodowej.* international 'One of the politicians suspects that a huge scandal will soon break out in the international arena.'

Similar to the situation with exclamative predicates outlined above, the use of *jakoby*-clauses leads to ill-formed results:

### Łukasz Jędrzejowski

(8) \* *Jeden* one *z* of:the *polityków* politicians *czuje,* feel.3sg *jakoby* jakoby *niebawem* soon *wybuchnie* break:out.3sg *wielki* huge *skandal* scandal *na* on *arenie* arena.loc *międzynarodowej.* international

(8) appears to be appropriate only in a context in which the sentence subject, i.e., one of the politicians, literally uttered that a huge scandal will break out. The speaker wants to distance himself/herself from what the politician said by using the complementizer *jakoby*. On the other hand, (8) is infelicitous in the context in which the speaker describes what the politician might suspect without having written or said it. In other words, the content of the proposition must be known to the speaker from a foreign source. This also accounts for why (2) is ungrammatical: using verbs of seeming, the speaker mainly draws conclusions based on what (s)he has perceived, and not based on what (s)he has heard from others. As *jakoby*-clauses tend to occur in the context of speech/report expressions, they can disambiguate or specify the meaning of a clause-embedding predicate, cf. (9) below:

(9) *Niektóre* some *kluby* clubs *nie* neg *wiedzą,* know.3pl *jakoby* jakoby *zgłaszały* propose.l-ptcp.n-vir.pl *graczy.* players.acc 'Some sports clubs admit not knowing that they supposedly proposed players.'

(NKJP, *Gazeta Krakowska*, 25/6/2007; slightly modified by author: ŁJ)

The semi-factive matrix verb *wiedzieć* 'know' is usually used as a verb of retaining knowledge. In (9), the embedded *jakoby*-clause adds an additional layer of meaning to it, turning it into a verbum dicendi.<sup>3</sup> Accordingly, we have to conclude that *że* and *jakoby* as complementizers differ in meaning and that their licensing conditions depend on lexical properties of clause-embedding predicates. Following the well-known classification of embedding verbs proposed in Karttunen (1977), the most frequent *jakoby*-embedders are verbs of one-way communication, e.g. *twierdzić* 'claim', *zaprzeczać*, *dementować* both: 'deny', *powiedzieć* 'say' or *sugerować* 'suggest'.

A final note is in order here concerning the licensing conditions of *jakoby*clauses. Remarkably, they can also be attached to DPs:

<sup>3</sup>Reis (1977: 142–148) has already observed for German *wissen* 'know' that it can be used in a similar way.

### 4 On the semantic change of evidential argument jakoby-clauses in Polish

(10) *Absurdalne* absurd *jest* be.3sg [DP *twierdzenie*]*<sup>i</sup> ,* claim [*jakoby* jakoby *okulary przeciwsłoneczne* sunglasses *miały* have.l-ptcp.n-vir.pl *ograniczać* restrict.inf *widoczność*]*<sup>i</sup> .* visibility.acc 'The claim that sunglasses supposedly restrict visibility is absurd.' (NKJP, *Gazeta Ubezpieczeniowa*, 7/3/2006)

In (10), the DP *twierdzenie* 'claim' is derived from the verb *twierdzić* and its content is modified or specified by the following *jakoby*-clause. For the sake of convenience, I restrict myself in the present study to *jakoby*-clauses that are selected by verbs. Currently, there are different technical possibilities for how one could analyze examples as given in (10). For an overview the interested reader is referred to Moulton (2009), Haegeman & Ürögdi (2010), and de Cuba (2017), among many others.

### **2.2 Previous descriptions**

The view on licensing conditions presented in this subsection sharply contrasts with what Taborek (2008: 110–115, 156–157) claims about *jakoby*-clauses:

Als die letzte Kategorie gilt hier der mit der Subjunktion *jakoby* (und ihren Alternaten *jakby* und *jak gdyby*) eingeleitete Komplementsatz in der Subjektfunktion. Die *jakoby*-Sätze werden von Verben des Sagens selegiert und implizieren Zweifel des Sprechers.

'As the last category, one should mention here the subjunction *jakoby* (and its alternative subjunctions *jakby* and *jak gdyby*) introducing complement clauses in the subject position. The *jakoby*-clauses are selected by verbs of saying and imply speaker's doubts.' (my translation: ŁJ) Taborek (2008: 100–101)

Although Taborek (2008) correctly observes that *jakoby*-clauses are selected by verbs of saying, he does not discuss any appropriate examples from Present-day Polish. Instead, he cites examples from older stages of Polish with *jakoby*-clauses occurring after verbs of seeming. The second problem concerns the replaceability of *jakoby* by *jakby* and *jak gdyby*, both meaning 'as if'. As the following example illustrates, neither *jakby* nor *jak gdyby* can replace *jakoby*:

### Łukasz Jędrzejowski

(11) *Firma* company *zaprzeczyła,* deny.l-ptcp.sg.f *jakoby* jakoby / / *\*jakby* as if / / *\*jak gdyby* as if *były* be.l-ptcp.n-vir.pl *zgłoszenia* reports *o* about *wadliwych* faulty *kartach.* cards.loc Intended meaning: 'The company denied that there were supposedly any reports about faulty prepaid cards.'

If one continues (11) with *jakby* or *jak gdyby*, the dependent clause modifies the way the company denied (= adjunct clause), not what the company denied (= complement clause). In other words, the embedded clause headed by *jakby* or *jak gdyby* does not occupy the internal argument position of the matrix verb *zaprzeczać* 'deny'. Instead, it forms an A-bar dependency with the matrix clause, giving rise to a hypothetical comparative interpretation. Independent evidence for this argument follows from the observation that *jakby*- and *jak gdyby*-clauses (contrary to *jakoby*-clauses) cannot modify DPs derived from speech/report expressions:

(12) *Absurdalne* absurd *jest* be.3sg [DP *twierdzenie*]*<sup>i</sup> ,* claim [*jakoby* jakoby / / *\*jakby* as if / / *\*jak gdyby* as if *okulary przeciwsłoneczne* sunglasses *miały* have.l-ptcp.n-vir.pl *ograniczać* restrict.inf *widoczność*]*<sup>i</sup> .* visibility.acc Intended meaning: 'The claim that sunglasses supposedly restrict visibility is absurd.'

Likewise, Wiemer (2005) assumes *jakoby*-clauses to be still embeddable under verbs of seeming. Empirically, this view cannot be upheld, though. I was not able to find solid evidence from Present-day Polish in the *National Corpus of Polish* illustrating the usage of *jakoby*-clauses after verbs of seeming.<sup>4</sup> Based on Łojasiewicz (1992), Wiemer (2005) elaborates on the following example:

(i) *Zdaje* seem.3sg *mi* me.dat *się,* refl *jakoby* jakoby *Hobbit* Hobbit *uważał* think.l-ptcp.sg.m *inaczej.* differently 'It seems to me as if Hobbit would think differently.'

(NKJP, an internet forum, 19/8/1999)

Personally, I judge this example as ungrammatical and would use *jakby* instead of *jakoby*.

<sup>4</sup> I built queries looking for all morphological forms of both perfective and imperfective verbs of seeming; compare, for example, the aspectual pair *zdać się* vs. *zdawać się*. As verbs of seeming are reflexive in Polish, I also built queries with syntactic interveners between the verb and the reflexive pronoun *się*. One of such interveners is, for example, a DP argument marked for the Dative case and stemming from the matrix verb, giving rise to such results as *wydaje mi się* 'it seems to me'. I was able to find only one example from an internet forum:

### 4 On the semantic change of evidential argument jakoby-clauses in Polish

(13) *Zdaje* seem.3sg *mi* me.dat *się,* refl *jakobym* jakoby.1sg *słyszał* hear.l-ptcp.sg.m *jakieś* some *wołanie.* crying.acc 'It seems to me as if I heard someone crying.' (Łojasiewicz 1992: 105)

It is not clear, however, how old this example is. Moreover, I judge it as ungrammatical and would use the hypothetical comparative complementizer *jakby* 'as if' instead of *jakoby* in this context. In addition, Wiemer (2005: 122–124) notices that *jakoby* clauses can be embedded under speech verbs. However, he discusses only one example with the matrix verb *śnić się* 'dream':<sup>5</sup>

(14) *Przeszłej* last *nocy* night *śniło* dream.l-ptcp.sg.n *mu* him.dat *się,* refl *jakoby* jakoby *gruszki* pears.acc *z* from *drzewa rwał.*

tree.gen pluck.l-ptcp.sg.f

(i) 'Last night he dreamt as if he were plucking pears from a tree.'

(ii) 'Last night he dreamt that he was supposedly plucking pears from a tree.'

(iii) 'Last night he is supposed to have dreamt that he was plucking pears from a tree.' (Wiemer 2005: 123, ex. 22)

Three issues deserve to be addressed in connection with the example given in (14). Firstly, (14) is taken from the Positivist novel *Nad Niemnem* 'On the Niemen', which was written in the New Polish period in 1888 by Eliza Orzeszkowa. Secondly, *śnić się* 'dream' is not an inherent verb of saying. In essence, dream reports allow a multiplicity of readings. If someone dreams, (s)he can dream that (s)he is someone else. In this sense, one reports what (s)he dreamt about and VP denotes a set of situations in which someone had a dream/dreams. Though *śnić się* 'dream' does not necessarily involve a speech context (for more details on dream reports, see Shanon 1980, Percus & Sauerland 2003 or Kauf 2017). Thirdly, in my opinion (14) is ambiguous and has three different readings. *Jakoby* can be interpreted either as the hypothetical comparative complementizer 'as if' or as a reported speech complementizer in the Present-day Polish sense. In the former case, it is used because the matrix subject cannot remember what he exactly dreamt about. He has the impression that he were plucking pears from a tree, but he is not sure. In the latter case, two readings have to be distinguished. It can be either the subject himself who reports about his dreams or someone else who tries to render the content of subject's dreams. Both scenarios are imaginable; see also the discussion in Section 3.

<sup>5</sup>Glosses and English paraphrases are mine: ŁJ.

### Łukasz Jędrzejowski

### **2.3 Syntax**

If lexical licensing conditions of *jakoby*-clauses differ from those of *że*-clauses, there must also be syntactic differences between both clause types. Some of them are presented in this section.

### **2.3.1 Left periphery**

One of the differences between *że*- and *jakoby*-clauses refers to movement to the left periphery of the matrix clause. Consider the following pair:


b. *Dorota* Dorota *twierdziła,* claim.l-ptcp.sg.f *jakoby* jakoby *Jan* Jan *był* be.l-ptcp.sg.m *szczęśliwy.* happy 'Dorota claimed that Jan supposedly was happy.'

What distinguishes both clause types is that only *że*-clauses can be A-bar-moved to the left periphery. As the following contrast illustrates, movement of *jakoby*clauses is prohibited:

	- b. \* *Jakoby* jakoby *Jan* Jan *był* be.l-ptcp.sg.m *szczęśliwy,* happy *twierdziła* claim.l-ptcp.sg.f *Dorota.* Dorota Intended meaning: 'That supposedly Jan was happy, Dorota claimed.'

At this moment, I have no explanation for why *jakoby*-clauses are banned from a higher structural position in the Polish clause structure. There must be a conflict between the meaning of the complementizer and an information-structural movement.

### **2.3.2 Future tense form**

Another difference is connected to the use of the future auxiliary verb *będzie* 'will'; for its detailed analysis see in particular Błaszczak et al. (2014). Interestingly enough, *jakoby*-clauses cannot combine with *będzie*, whereas no such restrictions occur with regard to *że*-clauses:

	- b. \* *Dorota* Dorota *twierdziła,* claim.l-ptcp.sg.f *jakoby* jakoby *Jan* Jan *będzie* will.3sg *biegać* run.inf *codziennie.* daily Intended meaning: 'Dorota claimed that Jan will supposedly go jogging every day.'

The questionability of (17b) is surprising in the light of the rigid hierarchy of functional projections developed in Cinque (1999; 2006; 2017):

(18) [*frankly* Moodspeech act [*fortunately* Moodevaluative [*allegedly* **Moodevidential** [*probably* Modepistemic [*once* T(Past) [*then* **T(Future)** [*perhaps* Moodirrealis [*necessarily* Modnecessity [*possibly* Modpossibility [*usually* Asphabitual [*again* Asprepetitive(I) [*often* Aspfreuentative(I) [*intentionally* Modvolitional [*quickly* Aspcelerative(I) [*already* T(Anterior) [*no longer* Aspterminative [*still* Aspcontinuative [*always* Aspperfect [*just* Aspretrospective [*soon* Aspproximative [*briefly* Aspdurative [*characteristically* Aspgeneric/progressive [*almost* Aspprospective [*completely* AspSgCompletive(I) [*tutto* AspPlCompletive [*well* Voice [*fast/early* Aspcelerative(II) [*again* Asprepetitive(II) [*often* Aspfrequentative(II) [*completely* AspSgCompletive(II) ]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]

Accordingly, we expect *jakoby* as an evidential complementizer to merge as a functional head in Moodevidential, meaning that it should be able to take scope over all other functional material associated with lower functional projections including T(Future). This is not the case, though. It still needs to be accounted for why *będzie* is incompatible with *jakoby*-clauses.

### **2.3.3 Conditional mood**

In contrast to *że*-clauses, *jakoby*-clauses cannot contain a verbal head to which the conditional/subjunctive clitic *by* is attached, triggering a counterfactual interpretation of the embedded proposition:

(19) a. *Dorota* Dorota *twierdziła,* claim.l-ptcp.sg.f *że* that *Jan* Jan *poszedł-by* go.l-ptcp.sg.m-subj *do* to *kina.* cinema.gen 'Dorota claimed that Jan would have gone to the cinema.'

### Łukasz Jędrzejowski

b. \* *Dorota* Dorota *twierdziła,* claim.l-ptcp.sg.f *jakoby* jakoby *Jan* Jan *poszedł-by* go.l-ptcp.sg.m-subj *do* to *kina.* cinema.gen Intended meaning: 'Dorota claimed that Jan would supposedly have gone to the cinema.'

This difference might be due to the fact that *jakoby* as an evidential complementizer has not been fully bleached yet and that the clitic *by* still contributes to the compositional evidential meaning of what *jakoby* expresses in Present-day Polish. It straightforwardly follows that the second occurrence of *by* appears to be redundant in this context. I will come back to this issue later on.

### **2.3.4 The discourse particle** *chyba*

According to *Słownik Współczesnego Języka Polskiego [Dictionary of Modern Polish]* (1998), *chyba* 'presumably' is defined as follows:

*chyba*: tym słowem mówiący sygnalizuje, że nie wie czegoś dokładnie, nie jest czegoś pewien, ale decyduje się to powiedzieć, sądząc, że to prawda; przypuszczalnie; być może, prawdopodobnie, bodaj;

'*chyba*: using this word, the speaker signals that (s)he doesn't know something exactly, that (s)he is not certain about something, but at the same time (s)he decides to say it, claiming it is true; assumedly; maybe, probably, perhaps;' (my translation: ŁJ).

*Słownik Współczesnego Języka Polskiego [Dictionary of Modern Polish]* (1998: 117)

Consider the example given in (20), illustrating the use of *chyba* in a declarative clause:


Using the discourse particle *chyba* 'presumably', the speaker establishes a particular common ground relationship among discourse interlocutors. Concretely, the speaker indicates that her/his commitment towards the truth of what is embedded is speculative. Accordingly, I analyze *chyba* as a modifier of assertive

### 4 On the semantic change of evidential argument jakoby-clauses in Polish

speech acts, contributing to a weaker commitment of the speaker to the proposition, cf. Zimmermann (2004; 2011) for a similar analysis of the German discourse particle *wohl*.

(21) Meaning of *chyba*(p): ⟦*chyba* ⟧ = assume(, ), whereby = speaker

Usually, it is the speaker who is uncertain about the content of the embedded proposition using *chyba*:


However, in reported speech the attitude holder can be shifted to the clause subject itself (for more details on discourse particles in shifted contexts, see Döring 2013 and references cited therein):

(23) *Adam* Adam *twierdzi,* claim.3sg *że* that *piwo* beer *mu* him.dat *chyba chyba nie* neg *smakuje.* be:tasty.3sg 'Adam claims that he probably doesn't like the beer.'

What is interesting about *jakoby*-clauses is that they cannot license the discourse particle *chyba*, contrary to *że*-clauses:

	- b. \* *Dorota* Dorota *powiedziała,* say.l-ptcp.sg.f *jakoby* jakoby *chyba* chyba *pójdzie* go.3sg *do* to *kina.* cinema.gen Intended meaning: 'Dorota said that supposedly she presumably will go to the cinema.'

The speaker questions the truth value of the embedded proposition using *jakoby*. If we shift the attitude holder to the clause subject, it should be possible to combine *jakoby* and *chyba*, as the latter is not attributed to the speaker. (24b) is ruled out, though. A possible explanation comes from the fact that *chyba* as a speech

### Łukasz Jędrzejowski

act modifier takes a wider scope: It involves the matrix subject and its subjective attitude. *Jakoby*, in turn, does not take scope over the matrix subject leading to a clash. This is to be expected if we assume Moodevidential to outscope Modepistemic, see (18) above.

### **2.3.5 Modal verb** *musieć* **('must')**

It is a well-known fact that modal verbs can occur in embedded environments resulting in a shift of the attitude holder, cf. Hacquard (2006) and Hacquard & Wellwood (2012):

(25) *Dorota* Dorota *powiedziała,* say.l-ptcp.sg.f *że* jakoby *Jan* Jan *musi* must.3sg *być* be.inf *chory.* sick a) deontic: 'Dorota said that Jan has to be sick.' b) epistemic: 'Dorota said that Jan must be sick (now).'

In (25), the modal verb *musieć* can be interpreted in two different ways. Imagine a situation in which Dorota is a stage director of a play and determines how the stage play should be. According to this interpretation, *musieć* is evaluated against a bouletic modal base and narrowed down by a deontic conversational background. If, on the other hand, Dorota supposes Jan to be ill, but she is not sure about this, *musieć* is interpreted epistemically. In both cases, the attitude holder is the matrix subject, i.e. Dorota. *Jakoby*-clauses restrict the quantification domain of *musieć*:

	- a) deontic: 'Dorota said that supposedly Jan has to be sick.'
	- b) ?/\*epistemic: 'Dorota said that supposedly Jan must be sick (now).'

It is very hard to imagine a scenario in which *musieć* would be interpreted epistemically, even though the attitude holder has shifted to the matrix subject.<sup>6</sup> Remarkably, this problem disappears as soon as *musieć* is replaced by the existential modal verb *móc* 'can'/'may':

(i) ? *Dorota* Dorota *powiedziała,* say.l-ptcp.sg.f *jakoby* that *Jan* Jan *musiał* must.l-ptcp.sg.m *być* be.inf *chory.* sick Intended meaning: 'Dorota said that supposedly Jan must have been sick.'

Still, (i) sounds marked.

<sup>6</sup> Interestingly enough, this constraint is weakened as soon as the modal verb *musieć* occurs in a complex past tense structure:


It still needs to be figured out why the complementizer *jakoby* and the epistemic modal verb *musieć* cannot co-occur.

### **2.3.6 Matrix subject constraint**

If *jakoby*-clauses occupy one of the arguments of a clause-embedding predicate, the matrix subject usually occurs in the third person. 1st and 2nd person subjects, on the other hand, disprefer *jakoby*-clauses:


b. \* *Wczoraj* yesterday *powiedział-e-m,* say.l-ptcp.sg-m-1sg *jakoby* jakoby *pójdę* go.1sg *dzisiaj* today *do* to *kina.* cinema.gen Intended meaning: 'Yesterday I said that I will supposedly go to the cinema today.'

(28a) appears to be appropriate in one specific context. Let assume that A is the speaker, whereas B is the matrix subject. Imagine that B uttered *p* to C, i.e., to another discourse interlocutor, but not to A. It is natural to utter (28a) provided that C reported to A that B is supposed to have said *p*. The incompatibility of the 1st person with *jakoby*-clauses can, in turn, be accounted for by assuming that the speaker cannot question the truth value of what is embedded if *jakoby* presupposes the existence of a foreign information source and if (s)he herself/ himself is the information source (see also the discussion in Curnow 2002). No such restrictions occur with respect to *że*-clauses:

(29) a. *Wczoraj* yesterday *powiedział-e-ś,* say.l-ptcp.sg-m-2sg *że* that *pójdziesz* go.2sg *dzisiaj* today *do* to *kina.* cinema.gen 'Yesterday you said that you will go to the cinema today.'

### Łukasz Jędrzejowski

b. *Wczoraj* yesterday *powiedział-e-m,* say.l-ptcp.sg-m-1sg *że* that *pójdę* go.1sg *dzisiaj* today *do* to *kina.* cinema.gen 'Yesterday I said that I will go to the cinema today.'

Interestingly enough, this constraint is not absolute and depends on the semantics of the clause-embedding verb. It can be overwritten, as soon as the matrix verb is an inherent negative verb, e.g. *zaprzeczać* 'deny':

	- b. *Zaprzeczył-e-m,* deny.l-ptcp.sg-m-1sg *jakoby* jakoby *wygrał-e-m* win.l-ptcp.sg-m-1sg *w* in *lotka.* lottery 'I denied that I have supposedly won the lottery.'

The use of inherent negative verbs presupposes the existence of a covert negation resulting in ¬*p*. In this context, *p* is known to the speaker from hearsay. Using an inherent negative verb in combination with an *jakoby*-clause opens up the possibility for the speaker to question the validity of *p*.

A final note is in order here about the status of *jakoby* occurring as an evidential complementizer. One of the anonymous reviewers objects that *jakoby* as a complementizer can co-occur with other complementizers, e.g. with *że* 'that', posing a challenge for my account:

(31) *Mój* my *przyjaciel* friend *mówi,* say.3sg *że* comp *podobno* comp */* / *jakoby* comp */* / *rzekomo* comp *faszyści* fascists *zniszczyli* destroy.l-ptcp.vir.pl *jakieś* some *biblioteki.* libraries. 'My friend keeps saying that apparently / allegedly / reportedly fascists destroyed some libraries.'

The anonymous reviewer assumes (31) to be a case of complementizer doubling, a phenomenon which is taken to be absent in the grammar of Polish in general. I disagree with the view that (31) exemplifies complementizer doubling and analyze *jakoby* as an evidential adverb (see also footnote 1 above and references cited there). There are several arguments showing why *jakoby* 'supposedly' – as well as *podobno* 'apparently' and *rzekomo* 'reportedly' – in (31) cannot be analyzed as complementizers. In what follows, I discuss some of them.

Firstly, neither *podobno* 'apparently' nor *rzekomo* 'reportedly' can introduce embedded clauses:

	- b. \* *Mój* my *przyjaciel* friend *mówi,* say.3sg *rzekomo* comp *faszyści* fascists *zniszczyli* destroy.l-ptcp.vir.pl *jakieś* some *biblioteki.* libraries.

(32a) and (32b) are only well-formed when *podobno* and *rzekomo* are analyzed as evidential adverbs expressing the matrix subject's attitude towards what is embedded. In this case, direct speech complements are embedded, and not subordinate clauses. This mainly follows from concord relations:

	- c. \* *Świadek* witness *twierdzi,* claim.3sg *rzekomo* comp *morderca* murderer *był* be.l-ptcp.3sg.m *jakoby* allegedly *wysoki.* tall
	- d. \* *Świadek* witness *twierdzi,* claim.3sg *podobno* comp *morderca* murderer *był* be.l-ptcp.3sg.m *jakoby* allegedly *wysoki.* tall

If *jakoby* introduces evidential subordinate clauses as given in (33a) and (33b) taking a propositional scope, it is also possible to use additional evidential adverbs having a narrow scope.<sup>7</sup> Concretely, it is *rzekomo* 'reportedly' in (33a) and

<sup>7</sup>Appropriate prosodic contours are required for the concord reading.

### Łukasz Jędrzejowski

*podobno* 'apparently' in (33b) taking scope over the adjective *wysoki* 'tall'. I refer to such cases as evidential concord in the sense claimed by Schenner (2007). However, it is impossible to reverse the order of the evidential expressions. As (33c) and (33d) illustrate, *podobno* and *rzekomo* cannot be employed as complementizers and glossed as comp, as suggested by the reviewer. Correspondingly, I exclude *podobno* and *rzekomo* from further investigation here.

Secondly, as mentioned above *jakoby*-complements are banned from the matrix prefield position. If *że* 'that' precedes *jakoby*, the embedded clause can move though:

(34) *Że* that *jakoby* allegedly *Jan* Jan *był* be.l-ptcp.sg.m *szczęśliwy,* happy *twierdziła* claim.l-ptcp.sg.f *Dorota.* Dorota 'That supposedly Jan was happy, Dorota claimed.'

This clearly indicates that *jakoby* is an adverb, not a complementizer.

Thirdly, if a *że*-clause hosts *jakoby*, future reference in the embedded clause itself becomes possible:

	- b. *Dorota* Dorota *twierdziła,* claim.l-ptcp.sg.f *że* that *jakoby* allegedly *Jan* Jan *będzie* will.3sg *biegać* run.inf *codziennie.* daily

'Dorota claimed that allegedly Jan will go jogging every day.'

Furthermore, conditional mood is also allowed:

	- b. *Dorota* Dorota *twierdziła,* claim.l-ptcp.sg.f *że* that *jakoby* allegedly *Jan* Jan *poszedł-by* go.l-ptcp.sg.m-subj *do* to *kina.* cinema.gen

'Dorota claimed that allegedly Jan would have gone to the cinema.'

### 4 On the semantic change of evidential argument jakoby-clauses in Polish

Lastly, (28b) illustrates that evidential *jakoby*-complements cannot be embedded if the matrix verb is inflected for the first person. No such constraint occurs with regard to the combination of *że* 'that' and *jakoby* 'allegedly':

(37) *Wczoraj* yesterday *powiedział-e-m,* say.l-ptcp.sg-m-1sg *że* that *jakoby* allegedly *pójdę* go.1sg *dzisiaj* today *do* to *kina.* cinema.gen 'Yesterday I said that I will supposedly go to the cinema today.'

(37) convincingly demonstrates that *jakoby* as an evidential adverb can be in the scope of the declarative complementizer *że* 'that'.

Finally, the diachrony of Polish provides abundant evidence showing that *jakoby* 'supposedly' as an evidential adverb came into being in Middle Polish, whereas *jakoby* as a complementizer existed already in the early Old Polish period.

In other words, the co-occurrence of *że* and *jakoby* does not instantiate complementizer doubling. Instead, they ought to be analyzed as a declarative complementizer and an evidential adverb, respectively. In this context, the same reviewer asks what the difference is between evidential *jakoby*-complements, on the one hand, and complement clauses headed by the complementizer *że* 'that' and containing the evidential adverb *jakoby*, on the other hand. Importantly, the main difference refers to embedding restrictions and selection.<sup>8</sup> As illustrated in Section 2.1, *jakoby*-complements are not embeddable under, for example, exclamative verbs. This restriction disappears as soon as a *że*-complement clause contains the evidential adverb *jakoby* 'allegedly':

(38) a. \* *Dziwi* be.amazed.3sg *mnie,* me.acc *jakoby* jakoby *były* be.l-ptcp.n-vir.pl *zgłoszenia* reports *o* about *wadliwych* faulty *kartach.* cards.loc Intended meaning: 'I'm amazed that there supposedly were any reports about faulty prepaid cards.' b. *Dziwi* be.amazed.3sg *mnie,* me.acc *że* that *jakoby* allegedly *były* be.l-ptcp.n-vir.pl *zgłoszenia* reports *o wadliwych kartach.*

> about faulty cards.loc

'I'm amazed that there allegedly were any reports about faulty prepaid cards.'

<sup>8</sup>As there are many structural differences between *jakoby* used as a complementizer and as an adverb, it seems reasonable to assume the restrictions on the use as a complementizer to be syntactic by nature. I thank Todor Koev for drawing my attention to this issue.

### Łukasz Jędrzejowski

Based on the syntactic differences between *jakoby*- and *że*-clauses pointed out above, one needs to examine semantic properties of the complementizer *jakoby*.

### **2.4 Semantics**

### **2.4.1 Speaker commitment**

Cross-linguistically, there are two types of reportatives, depending on whether they involve some kind of speaker commitment to the reported proposition, cf. Faller (2011), Kratzer (2012), Murray (2017), among many others:

	- b. *According to the rumour, Roger must have been elected chief (but I wouldn't be surprised if he wasn't).* (Faller 2011: 679)

*Jakoby* clearly does not require any degree of speaker commitment (for a possible analysis of similar cases cross-linguistically, see AnderBois 2014):

(40) *Mówi* say.3sg *się,* refl *jakoby* jakoby *Jacek* Jacek *został* pass.aux.l-ptcp.sg.m *wybrany* elect.ptcp.m *na* on *naczelnika,* chief.acc *ale* but *ja* I *w* in *to* this *nie* neg *wierzę.* believe.1sg 'It is said that supposedly Jacek was elected chief, but I don't believe that.' (Jędrzejowski & Schenner 2013: 14)

In this respect, Polish *jakoby* patterns with the English phrase *according to* as well as with the reportative suffix *=si* in Cuzco Quechua. The speaker using the reportative morpheme *=si* has the possibility of not having any opinion on the truth of *p* (for more details see Faller 2011):<sup>9</sup>

(41) *Pay-kuna=s* (s)he-pl=rep *ñoqa-man=qa* I-illa=top *qulqi-ta* money-acc *muntu-ntin-pi* lot-incl-loc *saqiy-wa-n,* leave-1o-3 *mana-má* not-impr *riki* right *riku-sqa-yki* see-ptcp-2 *ni* not *un* one *sol-ta* Sol-acc *centavo-ta=pis* cent-acc=add *saqi-sha-wa-n=chu.* leave-prog-1o-3=neg 'They left me a lot of money, (but) that's not true, as you have seen, they didn't leave me one sol, not one cent.' (Faller 2011: 679, ex. 37)

<sup>9</sup>This sharply contrasts with the reportative morpheme *ku7* in St'át'imcets, as reported by Matthewson et al. (2007). Accordingly, *ku7* patterns with English *given that*.

### 4 On the semantic change of evidential argument jakoby-clauses in Polish

Following Kratzer (2012) and Faller (2011), I construct a modal base based on the contents of relevant reports giving rise to an *informational* conversational background. Such conversational backgrounds represent the information conveyed by reports and other sources of information:

(42) *f<sup>r</sup> (w)* = {*p* ∣ *p* is the content of what is said in *w*}

### **2.4.2 Dubitativity**

*Jakoby* contributes a dubitative component. There is a clear difference between *jakoby*-clauses and regular conditional/subjunctive *że*-clauses as complements to speech verbs. If the speaker wants to distance herself/himself from the content of the reported proposition, *jakoby* has to be used instead of a regular complement clause:

	- b. \* *Anna* Anna *twierdzi,* claim.3sg *że* that *wygrała-by* win.l-ptcp.sg.f-subj *w* in *lotka.* lottery Intended meaning: 'Anna claims that she would have won the lottery.'

### **2.4.3 Negation**

Similar to other evidential expressions attested cross-linguistically, *jakoby* cannot be under the scope of a negation marker. It takes a wide scope:

(44) *Firma* company *twierdziła,* claim.l-ptcp.sg.f *jakoby* jakoby *nie* neg *było* be.l-ptcp.sg.n *zgłoszeń* reports *o* about *wadliwych* faulty *kartach.* cards.loc

'The company claimed that there supposedly weren't any reports about faulty prepaid cards.'

a) The speaker has reportative evidence that there have not been any reports about faulty prepaid cards.

b) #The speaker does not have reportative evidence that there have not been any reports about faulty prepaid cards.

### Łukasz Jędrzejowski

In this regard, *jakoby* patterns with reportative expressions attested in Cheyenne, St'át'imcets or Cuzco Quechua; cf. (45) for Cheyenne:

	- 3-not-sing-negan-rep.3sg Annie
		- a) 'Annie didn't sing, they say.'
		- b) #'I didn't hear that Annie sang.'
		- c) #'Annie sang, they didn't say.' (Murray 2017: 29, ex. 2.56b)

### **2.5 Interim summary**

What we have seen so far is that *jakoby*-clauses radically differ from complement clauses introduced by the declarative complementizer *że* 'that' in Presentday Polish. The former are much more restricted than the latter, not only with respect to their licensing conditions but also with respect to their syntactic and semantic properties. As it turns out, these differences follow from the compositional meaning of the complementizers in question (cf. Moulton 2009). Table 1 furnishes the main differences between both clause types.



In what follows, I give an overview of the way *jakoby*-clauses could be used in older stages of Polish. Having described the usage and the distribution of *jakoby* in individual historical periods, I analyze its semantic change.

### 4 On the semantic change of evidential argument jakoby-clauses in Polish

### **3** *Jakoby***-clauses in the history of Polish**

Based on Klemensiewicz (2009), Walczak (1999), and Dziubalska-Kołaczyk & Walczak (2010), Table 2 distinguishes language stages in the history of Polish.


Table 2: Historical stages of Polish

Dziubalska-Kołaczyk & Walczak (2010: 823) summarize the most important reasons for assuming this classification as follows:

The Old Polish period is assumed to have terminated in 1543 with the publication of all the bills of a parliamentary session for the first time in Polish. Thus, the year 1543 marks the introduction of Polish as an official language of documents beside Latin. Additionally, it was in the same year that the first popular literary piece written in Polish was published. It was *Krótka rozprawa między trzema osobami: Panem, Wójtem i Plebanem* ('a short debate among three persons: a lord, a commune head and a pastor'), by Mikołaj Rej, who was the first Polish Renaissance writer writing exclusively in Polish. Middle Polish lasted till 1795 – the election year of king Stanislaus August Poniatowski and symbolic beginning of the period of Enlightenment. The outbreak of the World War II marks the end of the New Polish period and beginning of Modern Polish.

As it turns out, the proposed classification is to be traced back to historical events in the first instance. For major system-internal changes being distinctive of a particular language period, the interested reader is referred to the references cited above.

### **3.1 Etymology**

*Jakoby* is a typical example of head adjunction. Its origin is traced back to the preposition *jako* 'as' and the conditional/subjunctive clitic *by* ≈'would':

### Łukasz Jędrzejowski

(46) a. *Od* from *18* 18 *lat* years *pracuje* work.3sg *jako* as *księgowy.* public:servant 'Has has been working as public servant for 18 years.'

(NKJP, *Tygodnik Podhalański*, 31/1999)

b. *Zdecydowaliśmy,* decide.l-ptcp.vir.1pl *by* subj *zorganizować* organize.inf *akcję* action.acc *wśród* among *harcerzy.* scours 'We decided to organize an action among the scouts.'

(NKJP, *Dziennik Zachodni*, 17/8/2002)

The conditional/subjunctive clitic *by*, in turn, is traced back to *by*, i.e. 3rd person singular aorist of the Proto-Slavic predicate *byti* 'be'; for its diachrony, see in particular Migdalski (2006; 2010; 2016) and Willis (2000). I analyze it in (46b) as a complementizer.10,11

(i) *Zdecydowaliśmy,* decide.l-ptcp.vir.1pl *żeby* comp *zorganizować* organize.inf *akcję* action.acc *wśród* among *harcerzy.* scours 'We decided to organize an action among the scouts.'

This scenario is not surprising at all in the history of Polish because *by* incorporated into *jako* forming together the hearsay complementizer *jakoby* being a clear C-head. In other words, *by* is eligible for the C-head position. At this moment, I am not aware of any arguments speaking against *by* being base-generated as a C-head and establishing a subordinating relation between the matrix clause and the embedded clause. Notably, there is one strong counter argument against the view that *by* cannot be used as a complementizer. In complement clauses under desiderative/volitional predicates *by* has to occur adjacent to the declarative complementizer

<sup>10</sup>Berit Gehrke (p.c.) pointed out to me that *by* in Czech can never be used as a complementizer. This might explain why (3b) is ungrammatical. If neither *jako* nor *by* merge as C-heads, the development into a hearsay complementizer is blocked.

<sup>11</sup>One of the anonymous reviewers points out that "*by* is never a complementizer in Polish. It is a conditional/subjunctive auxiliary, and it may occur in the complementizer position only when it incorporates into true complementizers or conjunctions (e.g. *aby* or *żeby* 'that'). So it is not only Czech that does not use *by* as a complementizer, the same holds for Polish." It is not clear what syntactic position *by* occupies in (46b). Following Migdalski (2006), *by* originates in MoodP below TP. On the one hand, we can assume it to be base-generated in MoodP and to remain in-situ in (46b). But on the other hand, there is no evidence showing that *by* in (46b) cannot be associated with the CP layer occupying the C-head position. According to Migdalski (2016: 171), "[a]ll the examples that require encliticization of the auxiliary clitic *by*, which may occur in second position immediately following the complementizer express some kind of non-indicative Force-related meaning, such as hypothetical counterfactual conditionality, potentiality, or optative mood." Tomić (2000, 2001) treats such clitics as operator clitics, as they scope over the entire proposition. And this is what we observe in (46b), too. The embedded clause is a complement clause of the perfective verb *zdecydować* 'take a decision' expressing purposiveness. This indicates that the declarative complementizer *że* 'that' may have been dropped, that the clitic *by* took over its function and, finally, that it has frozen as a C-head:

### 4 On the semantic change of evidential argument jakoby-clauses in Polish

### **3.2 Old Polish (until 1543)**

Already in op, *jakoby*<sup>12</sup> fulfills miscellaneous functions. To determine its polyfunctionality, I extracted and analyzed 262 examples containing *jakoby* from the PolDi corpus.<sup>13</sup> Its distribution is given in Table 3.

Table 3: The use of *jakoby* in the PolDi corpus


The label *adverb* refers to all cases in which *jakoby* is used as an adverb, see also the example in (i, fn. 1) above. The question of whether it could have different meanings in op still needs to be addressed. I am not concerned with this use of *jakoby* in this chapter. In 93 cases *jakoby* combines and compares two phrases, for example two DPs, two PPs or a DP with a PP. In this function, *jakoby* is comparable with English *like*. The next example shows a combination of two DPs:


*że* 'that', i.e. it occurs within the CP-domain (see also footnote 2 above). What is interesting in this context is the fact that *że* 'that' can be deleted. It is then *by* which introduces the embedded clause and marks its illocutionary force as well its subordinate status:

(ii) *Każda* every *matka* mother *chce,* want.3sg *żeby* comp *jej* her *syn* son *chodził* go.l-ptcp.sg.m *do* to *przedszkola.* kindergarten.gen 'Every mother wants her son to go to the kindergarten.'

Concretely, the view that *by* is disallowed from being a C-head introducing embedded clauses in Polish is not correct.

<sup>12</sup>Two alternative orthographic variants of *jakoby* existed in older stages of Polish: (i) *jako by* and (ii) *kakoby*. For methodological reasons, I ignore both variants in this study.

<sup>13</sup>PolDi is a collection of texts from Polish language history. 40 texts, both from Old and Middle Polish, are supposed to be annotated and integrated into the ANNIS search engine. Unfortunately, I was not able to find any information about how large the corpus is in terms of word counts. According to my understanding, 22 texts are currently searchable. The 262 examples stem from these 22 texts. However, in this section I elaborate only on cases from op.

### Łukasz Jędrzejowski

pdp *jakoby* cannot compare one DP with another DP. Instead, *jakby* 'as if' has to be used:

(48) a. *Urządzili* set:up.l-ptcp.vir.3pl *tam* there *sobie* refl.dat [DP *coś*] something *jakby* jakby [DP *klub*]*.* club 'There, they have set up something like a club.' (NKJP, *Dziennik Zachodni*, 30/12/2009) b. \* *Urządzili* set:up.l-ptcp.vir.3pl *tam* there *sobie* refl.dat [DP *coś*] something *jakoby* jakoby [DP *klub*]*.* club Intended meaning:'There, they have set up something like a club.'

When and under what circumstances *jakby* replaced *jakoby* in the history of Polish still needs to be investigated. The next label – *DP complement* – includes all cases in which a DP is modified by a *jakoby*-clause. In all three attested cases the modified DP is related to a verb of speech: *wzmianka* 'mention', *rada* 'advise', and *krzyk* 'scream', see also the example given in (10) and the discussion in Section 2.1. The first example I came across includes the DP *wzmianka* 'mention' and stems from mp. I will not discuss it here. The last two examples come from late op (around 1500) from *Rozmyślania przemyskie* 'The Przemyśl Meditation':<sup>14</sup>

(49) a. *a* and *zatem* thus [DP *krzyk*] scream *wielki* huge *pobudził* wake.up.l-ptcp.sg.m *wszytek* all *dwor,* court *jakoby* jakoby *krol* king *jż* already *umarł* die.l-ptcp.sg.m 'and thus a loud scream woke up all the court that the king supposedly died already' (PolDi, *Rozmyślania przemyskie*, ≈1500, 92) b. *jako* as *licemiernicy* duplicitous *Żydowie* Jews *z* from *biskupy* bishop.gen *uczynili* do.l-ptcp.vir [DP *radę*]*,* advise.acc *jakoby* jakoby *umęczyli* harass.l-ptcp.vir *Jesukrysta* Jesus Christ 'as duplicitous Jews they followed the bishop's advice by supposedly killing Jesus Christ' (PolDi, *Rozmyślania przemyskie*, ≈1500, 298)

<sup>14</sup>The example (49a) is ambiguous. Out of the blue, it can be interpreted either as an adjunct clause or as an argument clause. A further context, however, disambiguates its interpretation.

### 4 On the semantic change of evidential argument jakoby-clauses in Polish

In addition, *jakoby* could also introduce comparative hypothetical adverbial clauses (= *adverbial clause* in Table 3):


As (50) exemplifies, op *jakoby*-clauses could modify the matrix clause without being an argument of the matrix verb. Concretely, they could merge as modal adjunct clauses being often linked with a degree correlate occurring in the matrix clause, cf. *tako* 'so' in (50). In pdp, this clause type is headed by the complementizer *jakby* 'as if':

(51) *Wszyscy* all *zachowują* behave.3pl *się* refl *tak,* so *jakby* as if *chodziło* go.l-ptcp.sg.n *o* about *napad* assault *na* on *bank.* bank 'Everyone is behaving as if it were a bank robbery.'

(NKJP, *Samo życie*, episode 237)

The use of *jakoby* in such contexts is not possible any longer:

(52) \* *Wszyscy* all *zachowują* behave.3pl *się* refl *tak,* so *jakoby* jakoby *chodziło* go.l-ptcp.sg.n *o* about *napad* assault *na* on *bank.* bank Intended meaning: 'Everyone is behaving as if it were a bank robbery.'

Finally, as has been illustrated in Section 1, *jakoby* could introduce comparative hypothetical argument clauses after verbs of seeming. For the sake of convenience, I repeat the example given in (1a) as (53) below:

(53) *iżeć* that *się* refl *jest* be.3sg *ludziem* people.dat *na* on *ziemi* earth.loc *tako* so *było* be.l-ptcp.sg.n *widziało,* seem.l-ptcp.sg.n *jakoby* jakoby *się* refl *ono* it *na* on *nie* them.acc *obalić* slay.inf *było* be.l-ptcp.sg.n

### Łukasz Jędrzejowski

*chciało* want.l-ptcp.sg.n 'that it seemed to the people on earth as if it wanted to slay all of them' (KG, *Kazanie I: Na Boże Naordzenie*, 26–27)

In Table 3, I refer to cases like in (53) as *argument clauses*. What is important here is that (53) is one of the oldest examples stemming from early op. In late op *jakoby*-clauses began to be embedded under other clause-embedding verbs. An overview is given in Table 4.

Table 4: The distribution of *jakoby*-clauses as argument clauses in op based on the data extracted from the PolDi corpus


The occurrences with verbs of seeming are the oldest ones. Around 1500, verbs of thinking and verbs of speech/report started to occur with *jakoby*-clauses:


	- they would supposedly kill him' (PolDi, *Rozmyślania przemyskie*, ≈1500, 379)

### 4 On the semantic change of evidential argument jakoby-clauses in Polish

Remarkably, in pdp *myśleć* 'think' is not inclined to occur with *jakoby*-clauses, as it is not a classical verb of speech. However, there is one specific context in which someone renders someone else's thoughts reporting on what other persons (might) think. Although I was not able to find any appropriate corpus example, the following sentence sounds well-formed but marked a bit:<sup>15</sup>

(56) ? *Myśli,* think.3sg *jakoby* jakoby *jest* be.3sg *najlepszy.* best 'He think that he would be the best.'

Another possibility to interpret the five cases with verbs of thinking would be to analyze them as verbs of seeming in a broader sense. This would explain the expansion of *jakoby*-clauses after verbs of seeming to other clause-embedding verb classes. To what extent both classes are related and whether this link is conceptually reasonable remains an open issue. What is more striking with regard to the development of *jakoby*-clauses is their use after verbs of speech, *powiadać* 'keep saying' in (55a) and *smawiać się* 'conspire' in (55b). In this respect, late op does not deviate from pdp. As it turns out, not much changed in mp.

### **3.3 Middle Polish (1543–1765)**

The situation in mp resembles the picture of how *jakoby* was used in op. In general, I extracted 162 cases from the KorBa corpus, also known as *The baroque corpus of Polish*. <sup>16</sup> An overview of how *jakoby* was used in mp is given in Table 5.

Table 5: The use of *jakoby* in the KorBa corpus


Two major language changes can be observed. In what follows, I briefly comment on them.

(i) *Myśli,* think.3sg *jakoby* jakoby *jest* be.3sg *najlepszy,* best *ale* but *ja* I *w* in *to* this *nie* neg *wierzę.* believe.1sg 'He thinks that he would be the best, but I don't believe this.'

<sup>15</sup>One of the anonymous reviewers remarks that (56) improves when the speaker objects to what the matrix subject claims:

I agree with this view and share the same intuition.

<sup>16</sup>KorBa contains historical texts from the 17th and 18th centuries, consists of 718 texts, counts over 10 million word forms, and is available for free.

### Łukasz Jędrzejowski

Firstly, the use of *jakoby* as a comparative particle decreases (37% in op vs. 17% in mp), whereas as an adverbial clause complementizer it is still often used. What should be kept in mind, though, is that *jakoby* does not always introduce hypothetical comparative clauses; in some cases, it can also introduce purpose clauses:

(57) *Tak* so *trzeba* need *Rzemień* belt.acc *ciągnąć* pull.inf / / *jakoby* jakoby *się* refl *nie* neg *zerwał* peter:away.l-ptcp.sg.m 'One needs to pull the belt in such a way as to not break it off.' (KorBa, *Proverbium polonicorum*, 1618)

I leave it as an open question here what kinds of adverbial clauses *jakoby* could introduce in older stages of Polish.

Secondly – and more importantly – the use of *jakoby*-clauses as argument clauses increases (4% in op vs. 12% in mp). Among 20 examples, different classes of clause-embedding verbs can be attested (Table 6).

Table 6: The distribution of *jakoby*-clauses as argument clauses in mp based on the data extracted from the KorBa corpus


Selected examples follow; (58a) for *zdać się* 'seem', (58b) for *myślić* 'think' and (58c) for *mniemać* 'suppose':


### 4 On the semantic change of evidential argument jakoby-clauses in Polish

c. *iż* that *mniemali* suppose.l-ptcp.vir */* / *jakoby* jakoby *Zona* Zona *torrida* torrida *miała* have.l-ptcp.sg.f *być* be.inf *dla* for *zbytniego* too:him *gorąca* hot 'that [they] supposed that supposedly Zona torrida would be too hot for him' (KorBa, *Relacje powszechne*, part I, 1609)

Similar to the situation in the late op period, *jakoby* can be used as a hearsay complementizer in mp. As for embedding verbs, verbs of speech or report definitely outnumber verbs of seeming. What is different in mp in comparison to what we have observed in op is the expansion of argument *jakoby*-clauses to other verb classes. In the next example, the internal argument of the transitive verb *czytać* 'read' is occupied by a *jakoby*-clause:

(59) *listy* letters (...)*,* (...) *w* in *których* which *czytał,* read.l-ptcp.sg.m *jakoby* jakoby (...) (...) *W.Ks.L.* W.Ks.L *miał* have.l-ptcp.sg.m *się* refl *już* already *ożenić* get:married.inf *w* in *Śląsku* Silesia 'letters in which he could read that supposedly W.Ks.L would have already gotten married in Silesia' (KorBa, *Pamiętnik z czasów Jana Sobieskiego*, between 1690 and 1696)

Uttering (59) the speaker is reporting on what the clause subject was reading. This context enables the speaker to turn the verb *czytać* 'read' into a verb of report. At the same time, the speaker may question either the claim that someone got married or the observation that the clause subject was reading this claim. Both interpretations are conceivable.

To test for statistical reliability, statistical tests were run. The two language change processes described above were analyzed by means of generalized linear modeling using the package *lme4* (Bates et al. 2012) in R (R Core Team 2012). Table 7 shows the results and the last column lists the *p*-values.

The relevant factors, i.e. language period as an independent variable and complement type as a dependent variable, were coded to test whether differences between both language periods are significant. As it turned out, the tests statistically confirmed the diachronic observations.<sup>17</sup>

<sup>17</sup>I thank Frederike Weeber who helped me with the statistics.

### Łukasz Jędrzejowski


Table 7: Summary of the relevant factors in the generalized linear model

### **3.4 New Polish (1765–1939)**

The use of *jakoby* in np remains constant. Its all main functions attested in op and mp still occur in the 19th century. I extracted and analyzed a sample of 85 *jakoby*-cases from NewCor, a corpus of 1830–1918 Polish. Table 8 portrays the picture of how *jakoby* is used.

Table 8: The use of *jakoby* in the NewCor corpus


Interestingly enough, *jakoby*-clauses modifying DPs dominate. They usually modify such DPs as *pogłoska* 'rumour', *wieść* 'news', *wiadomość* 'message', *twierdzenie* 'claim' , *mniemanie* 'opinion' or *zarzut* 'accusation'. All of the DPs are related to verbs of speech/report. *Jakoby* can still occur as a hypothetical comparative element, either comparing two phrases or introducing adverbial *as-if* clauses. In eight cases, *jakoby*-clauses occupy an argument of a clause-embedding predicate (Table 9).

Table 9: The distribution of *jakoby*-clauses as argument clauses in np based on the data extracted from the NewCor corpus


### 4 On the semantic change of evidential argument jakoby-clauses in Polish

I could not find any examples with verbs of thinking. Of course, more data needs to be analyzed in order to be able to exclude this class altogether. In two cases, the *jakoby*-clause is an argument of a *seem*-verb, as the next example shows:

(60) *zdaje* seem.3sg *się* refl *nam,* us.dat *jakoby* jakoby *wzory* patterns *te* these *były* be.l-ptcp.n-vir *mędrsze* smarter *od* from *nas* us 'it seems to us as if these patterns would be smarter than us' (NewCor, *O związku pomiędzy światłem i elektrycznością*, 1890)

The other cases illustrate the use of *jakoby*-clauses after verbs of speech/report, known from pdp:

(61) *i* and *nie* neg *można* can.pred *też* also *było* be.l-ptcp.sg.n *twierdzić,* claim.inf *jakoby* jakoby *łacińscy* Latin *biskupi* bishops *stróżami* guards.ins *byli* be.l-ptcp.vir *Kościoła* Church.gen *ruskiego* Ruthenian 'and one couldn't claim either that supposedly Latin bishops would have been guards of the Ruthenian Church'

(NewCor, *Sprawa ruska na Sejmie Czteroletnim*, 1884)

The availability of *jakoby*-clauses after verbs of seeming in np might account for why Łojasiewicz (1992), Wiemer (2005) and Taborek (2008) still cite their occurrence in pdp. Since their incompatibility appears to be a very young development in the history of Polish, one would not be surprised to come across similar examples from the beginning of the 20th century.

### **3.5 Interim summary**

In this section, we have seen that *jakoby* developed its main functions already during the op period. As far as argument *jakoby*-clauses are concerned, they started to occur after verbs of speech/report in late Old Polish and ceased to be used after verbs of seeming in Present-day Polish (Table 10).

Along with the latter change, *jakoby* also ceased to occur as a (hypothetical) comparative particle being replaced by *jakby* 'as if'. The question of how *jakoby* developed from a hypothetical comparative complementizer into a hearsay complementizer is addressed in the next section.

### Łukasz Jędrzejowski

Table 10: The development of *jakoby*-argument clauses in the history of Polish


### **4 Reanalysis**

The main objective of this section is to reanalyze the development of *jakoby* in the history of Polish. The main focus is on *jakoby*-clauses being used after verbs of seeming and after verbs of speech/report. I aim at identifying constant factors in the lexical meaning of *jakoby* over time and, at the same time, at locating the aspects responsible for the semantic change that *jakoby* underwent.

As detailed in Section 3, *jakoby* can be traced back to the fusion of the comparative preposition *jako* and the conditional/subjunctive clitic *by*. I argue that these components contributed two semantic seeds that determined the further development of *jakoby*: (i) equative comparison, (ii) non-factivity. I take *jako* 'as' to be a lexical anchor for an equivalence relation – along the lines proposed by Umbach & Gust (2014) – between the matrix clause and the embedded clause. The role of *by* is to mark non-factivity giving rise to a counterfactual reading, as defined in Bücking (2017: 988)<sup>18</sup> .

For Old Polish, the combination of these two elements is sufficient to explain the semantic contribution of *jakoby* itself. While component (i) enabled the use of *jakoby* in adjunct clauses, component (ii) paved the way for the dubitative meaning that *jakoby* contributes in complement clauses of verbs of speech/report.<sup>19</sup>

<sup>18</sup>Bücking (2017) examines hypothetical comparative clauses in German and distinguishes four different readings: i) extensional, ii) generic, iii) counterfactual, and iv) epistemic. All of them were available with *jakoby* in Old Polish, though it was the counterfactual reading that gave rise to the development of *jakoby* into a hearsay complementizer.

<sup>19</sup>One of the anonymous reviewers objects that the reanalysis concerns conditionality and does not involve subjunctive meaning as proposed here. As *by* can express both conditionality and

### 4 On the semantic change of evidential argument jakoby-clauses in Polish

Table 11: Etymological composition of *jakoby*


In early Old Polish, *jakoby* heads complement clauses of *seem*-type verbs that express indirect inferential evidence. The logical structure of these sentences is as follows, where *p* represents the proposition expressed by the embedded clause:

(62) [*seem* [*jakoby p*]]

The central question to be asked here is how these three elements, i.e. the clauseembedding verb, the complementizer, and the embedded proposition play together to yield the final meaning 'it seems as if *p*'. The clause-embedding verb *seem* expresses indirect evidence, indicating that the speaker has some body of evidence X from which it follows – or which at least strongly suggests – that *p* is true. The general idea for the case of *seem* can be thus expressed as follows:

(63) ⟦*seem*⟧ , = . speaker() has in inferential evidence that is true in

which can be modeled in the Kratzerian style along the lines of Faller (2011) as follows:

(64) ⟦*seem*⟧ , = . the content() provides a perceptual or epistemic modal base and a doxastic ordering source such that for all worlds in min()(∩()) it holds that is true in

subjunctive meaning, it is not surprising that the anonymous reviewer argues for one of the categories. What *by* does is that it introduces a set of alternative worlds, a hallmark of both conditionality and of subjunctive meaning. It is conditionality in Old Polish *jakoby*-complements embedded under verbs of seeming that is crucial for interpretative purposes (cf. Stalnaker 1968, Lewis 1973, von Fintel 2011, and in particular Bücking 2017). But if *jakoby*-clauses are complements to verbs of saying or reporting, it is rather a subjunctive meaning of *by* absorbing the illocutionary force in the sense claimed by Truckenbrodt (2006). It has been crosslinguistically observed that embedded clauses in reporting contexts are usually marked by subjunctive mood; for an overview, see Becker & Remberger (2010), Fabricius-Hansen & Sæbø (2004), Portner (1997, 2018), Sode (2014), among many others. *Jakoby*-complements in Presentday Polish ought to be treated as cases of reportive mood, and not as cases of conditionality.

### Łukasz Jędrzejowski

If the matrix verb already expresses indirect evidence, what is the contribution of *jakoby*? Confer the following examples:

	- b. *It seems that Donald is in Singapore.*
	- c. *It seems as if Donald is in Singapore.*
	- d. *It seems as if Donald were in Singapore.*

In a nutshell, the contribution of *jakoby* is to map (65b)-type meanings to (65d) type meanings, whereas (65d) uncovers the two original components of *jakoby* pointed out above, i.e. equative comparison and counter-factual meaning. The basic idea is stated as follows:

(66) ⟦*seem as if* ⟧ , = . the information (evidence) that speaker() has in is *just like* the information that speaker() *would have* if were the case

Let's make (66) more concrete by examining two explicit scenarios:

	- b. *I believe that if Donald is in Singapore, he is excited. Donald is flying to Helsinki to meet Vladimir. I believe Donald is bored. Donald is excited.*


Table 12: Modal bases and ordering sources for the two scenarios

In scenario 1 it is natural to assert (65a) or (65b). In scenario 2, in turn, it is natural to assert (65d). The latter case gives rise to conflicting beliefs and (65d) is one way to express a certain reluctance to embrace the proposition for which there is indirect evidence. Accordingly, *seem as if p* is used instead of *seem that p* if what the available evidence suggests is somehow in conflict with what the speaker (used to) believe. If one looks at the relevant properties of the actual reference world, one can see that they look the same as the properties of the possible

### 4 On the semantic change of evidential argument jakoby-clauses in Polish

worlds where Donald is in Singapore. To put it differently: *As if* introduces the accessibility relation by way of an explicit comparison between two classes of worlds. The accessibility relation simply relates two sets of words. What *jakoby* does after verbs of seeming is compare them, or rather expresses equivalence as to some relevant properties.<sup>20</sup> This corresponds to Bücking (2017: 988)'s counterfactual reading of hypothetical comparative clauses, according to which only those worlds are taken into account that are as similar as possible to the actual world, given of course that the conditional's antecedent is true.

In sum, the contribution of *jakoby* in op does not seem to be genuinely evidential. Rather, it arises from the meaning of the two elements it is composed of: equative comparison and counter-factual meaning. If this is this case, the following question automatically arises: How did the inferential meaning of *jakoby* change to a reportative one specified in (68)?

(68) ⟦*jakoby* ()⟧, = 1 iff there exists a non-empty reportative informational modal base () such that for all ′ ∈ ∩ (), ⟦⟧ ′ , = 1

Intuitively, *it seems that p* expresses that there is some body of information X which entails that *p* is the case. What kind of information is X? Verbs of seeming are surprisingly flexible and are definitely not limited to expressing inferential evidence:

	- b. from *conceptual* information X → infer p (= assumed);
	- c. from *reportative* → infer p (–).

The last case is usually not registered as an *inferential* evidential. However, in practice reportative strategies often involve a fair amount of inference from the original utterance to its reported version. De Haan (2007) and Grimm (2010) notice that English *seem* is capable of expressing both direct and indirect evidence. A similar observation has been made by Reis (2007) with respect to German *scheinen* 'seem'. Its Dutch counterpart *schijnen* developed into a marker of reportative evidence and is joined by *lijken* for expressing visual evidence, see Koring (2013). For Cuzco Quechua, Faller (2001: 53–55) claims that by using the reportative morpheme *=si*, the speaker does not necessarily deny having inferential evidentials.

Using the idea from Faller (2011) that inferential evidentials involve a nonempty ordering source whereas (informational) reportative evidentials make no

<sup>20</sup>I would like to thank Radek Šimík (pc.) for pointing this out to me.

### Łukasz Jędrzejowski

reference to an ordering source at all, we can picture the development of *jakoby* as in Table 13.

Table 13: Diachrony of *jakoby* in terms of admissible information types in the modal base


The semantic shift of *jakoby* involved two main developments. First, the meaning of *jakoby* was broadened to allow for inferences from reportative information (compatible with but not enforced by its*seem*-type embedding verbs). Second, the reportative flavor acquired by *jakoby* licensed its use in complements of speech/ report verbs. Since these new contexts were no longer compatible with the original inferential meaning, they ultimately lead to the inability to use *jakoby* in its original contexts.

### **5 Conclusion**

The main aim of this chapter has been to examine the development and the semantic change of the evidential complementizer *jakoby* in the history of Polish with the main focus on argument clauses. It has been shown that *jakoby* developed a hearsay meaning in the late Old Polish period (1450–1543) and that it ceased to be selected by verbs of seeming in Present-day Polish (1939–present). The semantic shift outlined above corresponds to the evidential hierarchy proposed by de Haan (1999) according to which inferential evidentials can give rise to reportative evidentials.

As for emergence scenarios of complementizers, Willis (2007: 433) argues that the emergence of a new complementizer may involve three scenarios: (i) reanalysis of main-clause phrasal elements as complementizer heads, (ii) reanalysis of main-clause heads (e.g. verbs, prepositions) as complementizer heads, (iii) reanalysis of embedded phrases (e.g. specifiers of CP) as complementizer heads. The development of *jakoby* instantiates a fourth scenario: reanalysis of a complementizer head as another complementizer head.

Finally, the question of where evidentials come from has been addressed in different studies so far, cf. Willett (1988), Lazard (2001), Aikhenvald (2004: 271–

### 4 On the semantic change of evidential argument jakoby-clauses in Polish

302), Aikhenvald (2011), Jalava (2017), Friedman (2018), to name but a few. Various development patterns have been attested cross-linguistically. Aikhenvald (2011) points out two major sources for the development of evidentials. They can either evolve from open classes (e.g. verbs) and from closed classes (e.g. pronouns) or emerge out of an evidential strategy as an inherent marker of the grammatical category of evidentiality. The case of *jakoby* illustrates the former scenario, in which a complementizer develops into another complementizer. However, not much attention has been paid to the pattern described in this chapter and finegrained analyses depicting individual micro-steps of how evidential expressions come into being and develop still require further research.

### **Abbreviations**


### **Acknowledgements**

This chapter emerged out of a collaboration with Mathias Schenner at the Leibniz-Center General Linguistics in Berlin (ZAS) in 2013, and resulted in a joint talk presented at the workshop *New insights into the syntax and semantics of complementation* at the 21st international conference on historical linguistics at the

### Łukasz Jędrzejowski

University of Oslo (September 2013), cf. Jędrzejowski & Schenner (2013). Mathias Schenner was responsible for the semantic analysis. I, in turn, was responsible for the syntactic analysis and provided substantial data from older stages of Polish. With Mathias' kind approval, I seized on selected issues we raised in our earlier work, developed them further, and presented new results at the colloquium *Semantics-pragmatics exchange* at the University of Düsseldorf (June 2017), at the conference *Formal diachronic semantics* 2 (FoDS 2) at the Saarland University (November 2017), at the conference *Formal description of Slavic languages* 12.5 (FDSL 12.5) at the University of Nova Gorica (December 2017), as well as at the colloquium *Slawistische Linguistik* at the Humboldt University of Berlin (June 2018). I am grateful to the audiences for the inspiring feedback. For thought-provoking questions and comments, I would like to thank in particular (in alphabetical order): Nora Boneh, Ashwini Deo, Edit Doron, Hana Filip, Berit Gehrke, Remus Gergel, Chiara Gianollo, Julie Goncharov, Wojciech Guz, Verena Hehl, Vera Hohaus, Uwe Junghanns, Todor Koev, Martin Kopf-Giammanco, Roland Meyer, Roumyana Pancheva, Andreas Pankau, Radek Šimík, Peter Sutton, Luka Szucsich, Carla Umbach, Jonathan Watkins, Frederike Weeber, Henk Zeevat, and Karolina Zuchewicz. I am also indebted to two anonymous referees for their helpful feedback and genuinely interesting suggestions. Last but not least, my special thanks go to Clare Patterson and Benjamin Lowell Sluckin for proofreading. Of course, all remaining errors are my own.

### **Primary sources**


4 On the semantic change of evidential argument jakoby-clauses in Polish

### **References**

Aikhenvald, Alexandra Y. 2004. *Evidentiality*. Oxford: Oxford University Press.


4 On the semantic change of evidential argument jakoby-clauses in Polish


### **Chapter 5**

# **Semantic and syntactic change of** *equis* **in Mexican Spanish**

### Olga Kellert

Georg-August-Universität Göttingen

In this paper, I investigate the synchrony and diachrony of the linguistic item *equis* in Mexican Spanish. The origin of this item is the letter *x* which stands for a variable. The item *equis* has been recently used to refer to an utterance in the previous discourse which denotes a proposition and the speaker expresses his indifference with respect to this proposition, e.g. A: *Es verdad.* 'It's true.' B: *Equis!* 'I don't care (whether it is true or not)'.

The diachronic change of *equis* from a variable use *x* into a discourse use is analyzed as a shift of indifference towards identities of atomic individuals into indifference towards truth values. This semantic shift is syntactically expressed as a shift from a nominal modifier into a sentence modifier. I will argue that only in the discourse use is the indifference lexically encoded.

### **1 Introduction**

In Mexican Spanish (Mex), the expression *equis* stands for the character *x* as in (1) which can be interpreted semantically as a variable 'x', if it is used as a nominal modifier as in (2); (henceforth, the variable use of *equis*):


*Imaginemos* let-us-imagine *un* a *planeta* planet *equis* equis *donde* on-which *viven* live *otros* other *seres* beings *humanos.* humans 'Let us imagine some planet x with human beings on it.'

Olga Kellert. 2020. Semantic and syntactic change of *equis* in Mexican Spanish. In Remus Gergel & Jonathan Watkins (eds.), *Quantification and scales in change*, 131–159. Berlin: Language Science Press. DOI:10.5281/zenodo.3929245

### Olga Kellert

*Equis* can also be used as a determiner (see also Diccionario del Español de México, DEM). In (3), *equis* expresses that the speaker should be informed no matter what the reason for the addressee's absence might be. In this context, *equis* can be replaced by a free choice (FC) indefinite determiner *cualquier* 'any' (see Alonso-Ovalle & Menéndez-Benito 2011 on *cualquier* in European Spanish):

(3) *Si* if *por* for *equis/cualquier* equis/any *cosa* thing *no* not *puedes* can *venir* come *mañana,* tomorrow *házmelo* let-me-it *saber.* know 'If for whatever reason you cannot come tomorrow, let me know.' (Mex)

Besides the determiner use of *equis*, it can also have a predicative function with an Evaluative Interpretation of 'unremarkable, unimportant' in Mexican Spanish:

(4) *Así* so *que* that *es* is *difícil* difficult *identificar* indentifying *algo* something *serio* serious *de* from *algo* something *equis.* equis 'It's difficult to distinguish something serious from something

unimportant.'

It can also be used as a sentence or discourse adverb signaling speaker's indifference with respect to the truth or falsehood of some proposition mentioned earlier in the discourse. The speaker B who utters *equis* in (5) does not consider it important that the name under discussion is Nora:

(5) A: *No* not *es* is *Dora,* Dora *Es* is *Nora.* Nora. 'A: It's not Dora, it's Nora.' B: *Equis!* equis 'B: I don't care!'

As will be shown in the section on diachrony of *equis*, these different uses did not arise simultaneously. The discourse adverb use appeared very late. The variable use appeared earlier than the discourse adverb use and it is restricted to written texts, whereas all other uses (determiner, predicative, discourse adverb use) are common in oral texts.

The main aim of this paper is to analyze the diachronic change of the linguistic item *equis* and to investigate the possibility of analyzing all uses of *equis*, or

### 5 Semantic and syntactic change of equis in Mexican Spanish

at least some of them, in a uniform way. This study will shed some light on the relation between synchrony and diachrony and it will provide one possible answer to the question of how different synchronic meanings of one lexical item may develop diachronically.

This paper is organized as follows. I will first analyze the distribution of *equis* and its functions on the basis of corpus data in synchrony (see Section 2). I will then suggest how to analyze different uses of *equis* in a uniform way in Section 3 and suggest a path about the diachronic development of different functions in Section 4. The suggested diachronic development will be then tested by diachronic data in Section 5. The summary and outlook will be presented in Section 6.

### **2 Syntactic and semantic functions of** *equis* **in modern Mexican Spanish**

This section shortly describes different syntactic and semantic functions of *equis* in Mexican Spanish according to the corpus data. Details will be presented in subsequent sections.

The synchronic data was taken from the *Corpus del Español: Web/dialects* (Davies 2017). 468 occurrences of *equis* have been extracted from the synchronic corpus and classified according to syntactic and semantic/pragmatic features.

The relative frequency per million shows that the number of occurrences of *equis* is higher in Mexican Spanish. (468 occ. per 260598272 occ. in total) than in other varieties of Latin American and European Spanish (for instance in Argentinian Spanish 218 occ. per 182704898 or in Europ. Sp. 418 occ. per 459312821).<sup>1</sup>


Table 1: Number of occurrences of *equis*

<sup>1</sup> It is impossible to use inferential statistics to compare the numbers in Table 1 because these numbers represent numbers of populations and not means of different groups.

### Olga Kellert

The higher frequency of *equis* in Mexican Spanish suggests that *equis* is used in more numerous contexts in Mexican Spanish than in other varieties of Spanish.<sup>2</sup> This suggestion is confirmed by the questionnaires with speaker's judgements according to which *equis* is used in more contexts in Mexican Spanish than European Spanish, e.g. the discourse adverb use in (6) or the predicative use in (7):

(6) A: *Oye,* hey *¿y* and *si* if *ya* now *no* not *te* you *llama?* call

'A: Ey, what if (s)he doesn't call you?'

B: *¡Ay,* oh-my *equis!* equis

'B: Oh my, whatever!/forget about it!'

(8 Mex. speakers use *equis* as discourse adverb, 5 Mex. speakers do not use it, but find the sentence grammatical, all 13 Europ. Spanish speakers find the sentence ungrammatical.)

(7) *Es algo equis.*

> is something equis

'It is something unimportant.' (used by Mex. speakers, not used by European Spanish speakers)

As one can see from the description of *equis* in Table 2, which does not include the function of *equis* as a proper noun (e.g. *Signor Equis, rayos equis* 'Mr. X, Xrays')<sup>3</sup> , *equis* is mostly used as a nominal modifier (among 128 occ. of nominal modifier *equis*, 105 occurrences as prenominal modifiers or determiners and 23 occurrences as postnominal modifiers). The second most frequent function is the discourse adverb function (49 occ.), followed by the predicative use of *equis* with copular(like) verbs (9 occ.).

In most cases, *equis* appears in sentences with verbs in present tense. The most frequent interpretation of *equis* is the expression of the speaker's ignorance (Ignor.) and/or indifference (Indif.) 'I don't care about x' (see Table 2).

The details of every syntactic and semantic function are discussed in the following subsections.

<sup>2</sup>Actually, a higher frequency in use of *equis* in Mexican Spanish does not entail necessarily that *equis* is used in more different contexts there. This is why I have tested speaker's judgements as well.

<sup>3</sup> I eliminated all occ. with *equis* as a proper noun (209 occ.).

### 5 Semantic and syntactic change of equis in Mexican Spanish


### Table 2: *Equis* in Mexican Spanish in CDE web/dialects

### **2.1** *Equis* **as prenominal modifier or determiner**

The prenominal modifier or determiner *equis* has the function of an existential quantifier 'some' when used with respect to the number or quantity of the noun it modifies (e.g. *equis pesos* 'some cents', *equis número de coches* 'some number of cars', *equis cantidad* 'some quantity', *equis años* 'some years'). There are no occurrences with the prenominal *equis* preceded by an article (e.g. *un, a* 'a'). This observation suggests that *equis* acts as a determiner in the prenominal position.

As will be shown in the following data, tense and modality matter for the interpretation of *equis*. If *equis* appears in FC contexts (e.g. generic sentences, conditional antecedent, comparatives, etc., see Aloni et al. 2010 for FC contexts), it has FC interpretation (i.e. every alternative is a possible option):<sup>4</sup>

<sup>4</sup> In order to be sure that *equis* has FC interpretation in FC contexts, native speakers of Mexican Spanish were asked if they could also use *cualquier* instead of *equis* in these contexts (for FC interpretation of cualquier, see Alonso-Ovalle & Menéndez-Benito 2011, among others).

### Olga Kellert


In episodic sentences, the determiner *equis* refers to some specific individual (e.g. some artist) with an ignorance or indifference inference, i.e. the speaker signals that he/she does not remember or know the name of the artist or does not find it important to mention it:

(10) *Pero* but *la* the *última* last *vez* time *que* that *estuve* was *allá,* there *fue* was *en* in *un* a *concierto* concert *de* of *equis* equis *artista* artist *donde* where *estábamos* were *yo* me *y* and *cuatro* four *personas* people *más.* more *Era* was *patético.* pathetic 'But the last time I was there, it was at a concert of some artist where it was me and four people more. It was pathetic.'

In sentences with negation as in (11), *equis* either refers to some reason with ignorance inference as in (10) or it has a universal interpretation as in (12) where the addressee is asked to inform the speaker for every reason there is that prevents the addressee from coming:

(11) *pero* but *en* in *definitiva* end *es* is *una* an *opción* option *para* for *las* the *familias* families *que* that *por* for *equis* equis *motivos* reason *no* not *tienen* have *oportunidad* oportunity *de* of *salir* going-out *de* of *vacaciones.* holidays 'But, in the end, it is an option for families that, for some reason, don't have the opportunity to go on holiday.'

5 Semantic and syntactic change of equis in Mexican Spanish

(12) *Si* if *por* for *equis* equis *razón* reason *no* not *vas* go *a* to *venir,* come *avisame.* inform-me 'If for whatever reason you cannot come, inform me.'

The determiner *equis* is much more flexible with interpretations than usual FC indefinites such as Sp. *cualquier* 'any'. It can have the meaning of 'much/a lot' which seems to be context dependent, especially in contexts that trigger the interpretation of long duration, e.g. *equis tiempo* 'long time' (see Rivero 2011 for a similar observation on *cualquier cuantidad* 'any quantity' or 'a lot of quantity'):

(13) *Sí,* yes *soy* am *divorciado.* divorced *– Sí,* yes *hace* ago *equis* equis *tiempo.* time 'Yes, I'm divorced. – Yes, a long time ago.'

It can also mean 'several' or 'many' when it appears with plural NPs:


'Or as if it were a subway station that has several stations and each one attends to its users.'

The meaning 'much/many/several' also has an ignorance or indifference interpretation, i.e. the speaker does not know the identity or the exact number of the variable x or the speaker does not find it important to mention the identity or the number of the variable x. To summarize this section about prenominal or determiner *equis*, it acts as an indefinite 'some' and depending on the modal or episodic context, it can either have FC interpretation or be interpreted as an indefinite with an ignorance/indifference interpretation. *Equis* can have the meaning of 'several' or 'many', if it modifies plural nouns.

### **2.2 Adjectival or predicative use of** *equis*

The corpus data contain examples with *equis* after copular(-like) verbs (*ser/estar* 'be-stative', *volver* 'become') with the meaning 'neither good nor bad/unremarkable' in (16) or 'unimportant' in (17):

### Olga Kellert

	- B: *Estuvo* was *equis,* equis *me* I *esperaba* expected *algo* something *mejor.* better 'The movie was unremarkable. I expected something better.'

The predicative *equis* can be modified by degree adverbs *muy, tan, bien, bastante, completamente* 'very/completely' usually denoting high degrees:


When *equis* is used under negation, the negation expresses that it is not the case that some individual holds an unremarkable or a bad property, but rather a good or a distinguished property.

(20) *No* not *está* is *equis.* equis *Es* is *una* a *galería* gallery *linda.* beautiful 'It [=the gallery] is not bad/ordinary. It is a beautiful gallery.'

### **2.3 Postnominal** *equis*

The postnominal *equis* is less frequent than the prenominal one (four times less). All occurrences found in the corpus with postnominal *equis* are restricted to indefinite or bare nouns (e.g. *motivos equis* 'reasons x', *una mujer equis* 'a woman x').

When postnominal *equis* appears after a copular(-like) verb, it is interpreted as a gradable adjective with the meaning 'ordinary, unimportant' and can be modified by degree adverbs like *muy, bastante*, 'very' (see also the adjectival use of *equis* in Section 2.2):

5 Semantic and syntactic change of equis in Mexican Spanish


When postnominal *equis* occurs in modal contexts (e.g. with modal verbs), it usually has the interpretation of an FCI 'any' (see also prenominal *equis* in Section 2.1):

(23) *Así,* so *en* in *un* a *momento* momento *dado,* given *un* a *país* country *equis* equis *puede* can *estar* be *generando* generating *ahorro,* savings *pero* but *este* this *puede* can *filtrarse* leak *hacia* towards *el* the *exterior;* exterior 'In that way, at one point, any country may save money, but this money can leak out.'

If postnominal *equis* appears in episodic contexts that refer to past events, postnominal *equis* can have the interpretation of a variable x as in (24) and (25). In (24), the speaker does not specify the exact reasons for why she stopped drinking and the hearer infers that these reasons are unimportant to be mentioned for the main argument of the discussion, namely that she lost weight. Note that in this context, degree modification of *equis* as with adjectival use of *equis* (see Section 2.2) would not make much sense, because the argument of the discussion is not about the speaker's evaluation of the reasons but about losing weight:

(24) *Estuve* was *tomando* drinking *por* for *espacio* space *de* of *dos* two *meses,* months *deje* quit *de* of *tomar* drinking *por* for *motivos* reasons *equis* equis *y* and *de* from *un* some *tiempo* time *a* to *esta* this *parte* side *fui* was *a* to *la* a *balanza* scale *y* and *hoy* today *estoy pesando 120 kilos.*

am weighing 120 kilos

'I was drinking for two months, I quit for reasons x and, for some time, I weighed myself on the scales and now I weigh 120 kilos.'

In the following context, the hearer infers from the variable use of *equis* that the speaker is ignorant or indifferent about the deasease with which every child was born. As in (24), degree modification does not make sense here either:

(25) *Uno* one *se* it *entera* learns *de* of *cada* each *caso* case *de* of *niños* children *que* that *nacieron* were-born *con* with *una* a *enfermad* disease *equis,* equis *a* a *destiempo,* wrong-time *de* of *bajo* little *peso,* weigh *que* that *se* them *tienen* have *que* to *quedar* stay *en* in *incubadora.*

incubator

'One learns about each case of children who are born *with a disease x*, at the wrong time, with little weigh, and they have to remain in the incubator.'

The variable use is always replacable by the character x or some other character (e.g. *una enfermedad x o z* 'a disease x or z').

To sum up, *equis N* and *un,a N equis* can have a similar meaning of 'some' or 'any' despite their different syntactic status. This is due to the possibility of interpreting *equis* in *un,a N equis* as 'some' in episodic contexts or as 'any' in modal contexts. Under copular(-like) verbs postnominal *equis* is always interpreted as a degree predicate with the meaning 'unremarkable, unimportant' which can be degree modified. In this case, the postnominal *equis* cannot be replaced by the character x.

(26) \* *es* is *una* a *fiesta* party *muy* very *x* x *(o* (or *z)* z)

To conclude, postnominal *equis* can have an adjectival function with an evaluative interpretation of 'unremarkable, unimportant' in predicative position or it can be interpreted as a variable x and additionally express speaker's indifference or ignorance about the value of x.

### **2.4 Discourse adverb use**

The discourse adverb *equis* is different from all other functions shown so far as the discourse adverb *equis* must not appear inside a clause and be adjacent to some category it modifies:

	- B: *Equis*
	- A: *Pues*

'A: It's not Dora, it is Nora, looser. - B: I don't care! - A: Well…'

5 Semantic and syntactic change of equis in Mexican Spanish

(28) *Sí* if *son* are *homosexuales,* homosexuals *equis.* equis *O* or *heterosexuales,* heterosexuals *equis,* equis *el* the *punto* key *es* is *respetarse* respecting *a* to *sí* one *mismo.* self 'If they are homosexual, I don't care, or heterosexual, I don't care, the most important is to respect oneself.'

As a discourse adverb, it can have several functions, one of which is to express that the speaker does not care about which alternative proposition is true. In (27) the speaker signals that he does not care whether the name of some girl is Dora or Nora. In (28), the speaker expresses his indifference as to which answer is true, i.e. the speaker is indifferent about whether they are homosexuals or not, it does not matter as long as everyone respects each other. To sum up, the discourse adverb *equis* expresses speaker's indifference with respect to the truth of alternative propositions.

There is no discourse adverb *equis* with degree modification in the corpus data in contrast to the predicative/adjectival use of *equis* in (Sections 2.2 and 2.3):

(29) A: Some utterance *x*

B: *Muty equis.* (0 occ. in CDE)

'I don't care (a lot) about utterance x.'

Moreover, the discourse adverb *equis* cannot be negated in contrast to the predicative use of *equis* (see e.g. 30):

	- B: *No equis.* (0 occ. in CDE)

'It's not true that I don't care about x.= I do care about x'

The discourse adverb *equis* can also appear as a reaction to a question being addressed by the interlocutors in a given discourse where the speaker signals that he is not interested in the right answer to the question being discussed, i.e. he is indifferent towards any answer to the current question under discussion (QUD; see Roberts 1996):

(31) *Porque* because *siento* feel *que* that *alguien* someone *más* more *lo* it *ha* has *dicho* said *mejor* better *que* than *yo,* me *pero* but *equis,* equis *si* if *me* me *pides* ask *mi* my *opinión,* opinión *yo* I *diría* would-say *que…* that 'Because I think somebody said it better than me, but I don't care. If you ask my opinion, I would say that…'

(32) *¿Cómo* how *estás?* are *Igual,* same *equis.* equis *Bien.* good *¿Cómo* how *están* are *tus* your *jefes?* bosses 'How are you? The same, but it's not important. Good. How are your parents?'

Besides expressing the indifference towards some (hidden) alternatives, it can have a discourse function signaling a topic change in the discussion. Usually, this function is expressed by *pero equis, ay equis, bueno, equis* 'But, well, anyway/whatever':


To sum up, *equis* as discourse adverb expresses speaker's indifference towards the right answer to the current QUD or towards whatever has been said in the previous discourse.

### **2.5 Summary of the data and further outlook**

According to the description of the synchronic data, *equis* as a determiner has the meaning 'some, certain, various' and as a postnominal modifier *equis* has a variable use x which is interpreted as 'some' with ignorance/indifference inference in episodic contexts (e.g. *motivos x* 'some reasons' and I don't know/care which ones). In modal contexts, it is interpreted as a free choice 'any'. In predicative

### 5 Semantic and syntactic change of equis in Mexican Spanish

position after copular(-like) verbs, *equis* has an adjectival use with the possibility of degree modification (e.g. *muy equis* 'very unremarkable/bad/unimportant'). As discourse adverb, *equis* expresses the speaker's indifference with respect to a proposition which corresponds to an answer to some QUD (see Table 3).


Table 3: Summary of contexts, syntactic categories, semantic/pragmatic functions

As one can easily see, the indifference inference appears almost everywhere. It is somewhat parasitic on the indefinite and variable use of *equis*. It is lexicalized as a degree predicate 'unimportant' under copular(-like) verbs and it is expressed by the meaning 'I don't care' of the discourse adverb. The main working hypothesis is thus that the indifference inference is the core property of (almost) every use of synchronic *equis*. It makes sense to suggest a unified analysis of *equis* based on the indifference inference and to explain the variation between different uses synchronically (Section 3) and diachronically (Section 4).

### Olga Kellert

### **3 Synchronic analysis of** *equis*

In Section 2, the data have shown that the notion of indifference is the common interpretation associated with many uses of *equis*. I will first introduce into the notion of von Fintel's indifference von Fintel (2000) in Section 3.1. I will then apply this notion to the variable use of *equis* in Section 3.2, the discourse adverb *equis* in Section 3.3 and the postnominal *equis* in Section 3.4.

### **3.1 Von Fintel's indifference**

One prominent analysis of indifference was suggested by von Fintel (2000) for the English indefinite *wh-ever* (see also Condoravdi 2005). The indifference inference in the following example is expressed as a presupposition according to which the agent Ed could have blurted out anything else:

(35) *In response, Ed blurted out whatever came to his mind first.* Assertion: In response, Ed blurted out the first thing that came to his mind.

Presupposition: Ed could well have blurted out anything else that came to his mind first.

Under von Fintel's (2000) analysis in (36), the presuppositional content of*whatever* makes reference to a contextually determined modal base F, a mapping from worlds to sets of worlds (see 36). F(w) is the set of worlds in which the speaker' s beliefs about w hold. Sim expresses similarity between worlds and maps a world w and a proposition p to a set of worlds maximally similar to w in which p is true. The presupposition triggered by *whatever came to his mind first* conveys that if the thing that came to his mind first had been different in all the counterfactual worlds, the truth of the proposition "Ed blurted out the thing that came to his mind first" would also hold in the counterfactual world (see Condoravdi 2005).

	- a. presupposes: (∀w' ∈ Sim (w, F(w)) ∩ ( w". ίx. P (x,w") ≠ ίx. P (x,w)) Q(w')(ίx. P (x,w')) = Q(w)(ίx. P (x,w)) b. asserts: Q(w)(ίx. P (x,w))

### 5 Semantic and syntactic change of equis in Mexican Spanish

On von Fintel's analysis, whatever-statements presuppose that the speaker cannot identify the referent of the free relative and assert that the referent – whatever it is – has the matrix property in all of the speaker's alternatives (see Condoravdi 2005). According to von Fintel (2000), ignorance and indifference inferences are the result of different modal bases: in case every alternative is a possible alternative according to the speaker's knowledge, the ignorance inference is triggered. In case every alternative is a possible alternative according to the speaker's preferences, the indifference inference is triggered.

Whether a *wh-ever* free relative has an external indifference or agent indifference interpretation depends on contextual factors, and is only an "epiphenomenal inference" that is drawn from the presupposition of variation, i.e. from the presupposition that every alternative is a possible option (Tredinnick 2005: 108). On Tredinnick's analysis whenever there is no overt expression of an agent, the indifference is speaker related (e.g. I blurted out whatever came to my mind first.). This is indeed what has been observed in the literature on Spanish FCI *cualquiera*. If *cualquiera* is embedded under volitional verbs that express some volitional agent, it expresses agent's indifference with respect to the choice of some alternative (see Choi & Romero 2008, Alonso-Ovalle & Menéndez-Benito 2011). In the following example, the agent's indifference refers to the choice of the book:

(37) *Juan* Juan *necesitaba* needed *un* a *pisapapeles,* paperweight *de* of *modo* way *que* that *cogió* he-took *un* a *libro* book *cualquiera.* cualquiera 'Juan needed a paperweight, so he took a random book.'

I suggest to derive the indifference of the variable use of *equis* from von Fintel's counterfactual inference (see variable use of *equis* in 3.2).

### **3.2 Variable use of** *equis*

I assume the following syntax of the variable use of *equis* represented as x in order to distinguish it from other uses such as the discourse adverb use. The postnominal x is represented as a reduced relative clause similar to postnominal adjectives in Romance and other languages (see Cinque 2010 and others) in Figure 1.

Olga Kellert

Figure 1: Syntax of the variable *equis*

The DP as represented in Figure 1 denotes some name. Additionally, the variable x adds the information that x is a specific name, say *Dora* or *Nora* (see 38b). I represent the requirement that x is a specific name as a set of domain alternatives which is exhaustified, i.e. only one alternative can hold (see 38c):

	- a. 〚*un nombre x*〛= P ∃x ∈ D [name(x) ∧ P(x)] in words: 'some name'
	- b. Domain Alternatives(DA) = {P ∃x ∈ D' [name(x) ∧ P(x)] | D' ⊆ D} in words : {Dora, Nora} where Dora, Nora are alternatives of 'some name'
	- c. Exhaustification of domain alternatives=x is either Dora or Nora.

If the denotation in (38) is applied to the rest of the whole sentence in (39a), the result is an assertion in (39b). (39c) represents a set of alternatives applied to propositions. The exhaustification in (39d) expresses that only one alternative proposition is true in the world of evaluation:

	- b. Assertion: ∃x [x is a name and she/he has some name(x)]

### 5 Semantic and syntactic change of equis in Mexican Spanish


I assume that (39a) expresses speaker's indifference with respect to the value of the variable x which follows from the counterfactual inference in (40), namely that if her name had been different in all the counterfactual worlds, the truth of the proposition that she has some name would still hold in the counterfactual worlds:

(40) Counterfactual inference associated with the variable x: (∀w' ∈ Sim (w, F(w)) ∩ (w". x: name (x,w") ≠ x: name (x,w)) ∃x.name (x,w') & be (her, x, w') = ∃x.name (x,w) & be (her, x, w) 'No matter what her actual name is, all names are equal.'

The question is where the counterfactual inference in (40) comes from. Is it lexically encoded in the variable use or does it follow from some pragmatic principle? I assume that it is a pragmatic inference that goes informally as follows. The hearer concludes from (39a). that it does not matter to the speaker which of the alternatives triggered by x holds in the actual world because if the speaker knew the name of the person he would have said so due to Gricean conversational maxims (roughly: be informative, say true things, be relevant) (see Aloni 2005 for formal analysis of Gricean implicatures). The indifference inference derived from the variable use of *equis* has a similar pragmatic status as disjunctions or plain indefinites (see Aloni 2005 for the indifference inference associated with these items).

(41) *Do you want coffee or tea?*

Pragmatic inference: The speaker does not have any preference as to whether I, hearer, should take coffee or tea.

(42) *Paul married some girl from Nebraska.*

Pragmatic inference: Hearer assumes that the speaker is ignorant about the identity of the girl or he does not consider it important to mention her name.

I assume that the indifference inference associated with the variable use x is not lexically encoded in the lexical entry of x itself but is derived as a pragmatic inference according to which every value for the variable x is a possible candidate for the truth of the proposition.

### Olga Kellert

### **3.3 Degree predicate analysis of** *equis*

In the following example the postnominal *equis* is neither interpreted as a variable nor as a discourse adverb *equis* but as a predicative adjective with the meaning 'unimportant' as it is contrasted to the meaning 'serious'.

(43) *Así* so *que* that *es* is *difícil* difficult *identificar* indentifying *algo* something *serio* serious *de* from *algo* something *equis.* equis 'It's difficult to distinguish something serious from something unimportant.'

The predicative *equis* can be degree modified:

(44) *Estuvo* was *bien* very *equis.* equis 'It was very ordinary/unimportant.'

The degree analysis of *equis* as 'unimportant' is represented as follows. The predicate *equis* takes a degree argument d, an individual x and an attitude holder s (usually the speaker) and states that x is unimportant or ordinary to degree d to s (see Figure 2):

(45) [Adv/Adj *equis*] = dx. *ordinary/unimportant to s to degree* (d)(x)

The degree predicate can be modified by degree adverbials such as *muy/tan/ bien* that denote high degrees:

(46) [DeP Deg° *muy/tan/bien* 'very' [Adv/Adj equis]] = x. very *ordinary/unimportant*' (x)

I assume that *algo equis* in (43) is represented as a small clause (PrP) that denotes a proposition that something is ordinary/unimportant to some degree d to some attitude holder s.

(47) [PrP algo equis] = d ∃x. ordinary/unimportant to s to degree (d) (x)

The syntax of the adjectival use of *equis* and the variable x (see Figure 1) are almost identical. They both represent predicative elements. However, they are semantically different. In the degree predicate use, the term *equis* is interpreted as a degree predicate. In the variable use, *equis* is a predicate that is identical with a specific element of the set denoted by the noun.

### 5 Semantic and syntactic change of equis in Mexican Spanish

Figure 2: Syntax of the degree predicate *equis*

### **3.4 Discourse use of** *equis*

In order to represent the syntactic structure of the discourse adverb *equis* in (48), I assume that *equis* adjoins to the sentence that serves as an answer to the QUD in (49) (i.e. *what is her name?*):


*Equis* can be used as a sentential adverb that selects a sentence with a focus feature. The focus feature generates alternatives (i.e. her name is Nora or some other name):

(50) Alt(S) = {[w: Her name is Nora in w]; [w: Her name is Dora in w]}

The function of *equis* is to express indifference with respect to these alternatives, i.e. the speaker is indifferent as to which alternative is the right answer to the current QUD (i.e. *what is her name*).

The discourse adverb *equis* has the following denotation. It takes a context variable (c), a world-time variable (i), a set of alternative propositions that represent some QUD, a speaker (s) and an addressee (a). The discourse adverb expresses speaker's indifference at some context c and world-time (i) to the addressee (a) about every proposition of QUD:

(51) ∥Discourse Adverb Equis∥ (c)(i) (QUD<<st>t>) (s)(a) = at time-world index i and in a context c, the speaker s expresses his indifference to the addressee a about ∀q<st> ∈ QUD<<st>t>

### Olga Kellert

If the denotation in (51) is applied to the sentence in (48), the result is the following denotation which expresses speaker's indifference wrt alternative proposition represented in (50):

(52) ∥Discourse Adverb Equis in (48) ∥ (c)(i) (QUD) (s)(a) = at time-world index i and in a context c, the speaker s expresses his indifference to the addressee a about every proposition in {her name is Dora, her name is Nora}

In the following example, *equis* is used with *yes-no* questions:

(53) Oye, ¿y si ya no te llama? – ¡Ay, *equis*! 'What if he doesn't call you? – Oh my, I don't care!'

Here I assume that *equis* scopes over the if-clause which generates two alternative propositions p; ¬p:

(54) *Equis* equis *[si* if *[me* me *llama]]* calls 'I don't care whether he will call me (or not).'

The following figure shows that the sentence adverb *equis* adjoins to the whole CP representing the clause *si me llama*. The whole clause is represented as a DiscourseP because the CP *si me llama* refers to some previous utterance in the discourse.

Figure 3: *Equis* in si-sentences

The focus sensitive operator *if* generates the two alternatives he either will call me or he won't and *equis* operates on these alternatives in that it expresses indifference whether p or ¬p is true:

5 Semantic and syntactic change of equis in Mexican Spanish

(55) Alt(S) = {[w: he will call me in w]; [w: ¬ he will call me in w]}

The denotation of *equis* expresses indifference with respect to these alternatives: i.e. he will call me or he won't call me. There are other uses of *equis* where the QUD is not overtly expressed and the alternatives have to be recovered from the discourse or the common ground as in the following example:

(56) *bueno* well *otro* another *dia* day *sigo* keep *con* with *el* the *tercer* third *paso.* step *Ay* oh *ya* now *equis,* equis *que* that *pase* passes *lo* that *que* which *tenga* has *que* to *pasar* pass 'Well, I'll continue next day with the third step. Anyway, come what may'

The analysis suggested for the discourse use of *equis* can be extended to these examples as well as long as it is possible to recover some QUD that triggers alternatives and the speaker signals his indifference with respect to these alternatives. A detailed analysis of such cases awaits future research. Now that different uses of *equis* in Modern Mexican Spanish have been analyzed, the question arises when these uses emerged in the diachrony.

### **4 The diachronic change**

### **4.1 From variable x into degree predicate** *equis*

I assume that the indifference inference of the variable use x has been lexicalized into the meaning of the evaluative predicate *equis* 'unimportant'. The variable interpretation is repeated as follows:

(57) *Sea* be-subj-it *un* a *día* day *x* equis *(o* (or *z).* z) 'Be it a day x (or z).' Assertion: ◇ ∃x ∈ D [day(x) ∧ be (it, x)] 'It can be some day' Alternatives = It can be day a ∨ It can be day b ∨ It can be day c

The pragmatic inference that the hearer derives via conversational maxims is that the speaker is ignorant or indifferent about the identity of the day x. Thus, the variable is not instantiated by some specific day (say Monday):

(58) Pragmatic inference in (57): For every x and x is a day, the identity of (x) is not important to the speaker.

### Olga Kellert

I assume that the change from the variable use into the predicate use is triggered by the lexicalization of the pragmatic inference:

(59) Lexicalization of the pragmatic inference: *equis*<e,t> = 'unimportant to the speaker'

In the next step, the lexicalized predicate *equis* turns into degree predicate:

(60) *equis*<e,d> = 'unimportant to the speaker to degree d'

The next step represents a shift from the speaker's indifference to indifference of any other attitude holder (e.g. the addressee's indifference in (62) or 3rd person's indifference in (61)):


The diachronic change of the variable use into degree predicate is thus the result of lexicalization of the pragmatic inference and the shift of speaker's indifference to indifference of any other attitude holder.

### **4.2 From predicate** *equis* **into discourse adverb** *equis*

The diachronic change of *equis* into a discourse adverb is not a very big step. It is just a shift in scope of indifference, i.e. from indifference over identities to indifference over propositions:

	- a. All *entities* are possible options according to speaker's preferences (diachronic step 1)
	- b. All *answers* are possible options according to speaker's preferences (diachronic step 2)

This shift in scope of indifference has been lexicalized into a new use of *equis*, namely sentence adverb with a different feature makeup, i.e. *equis* selecting a CP (Section 3.3). I represent the diachronic change of *equis* schematically as follows. The first diachronic step shows that *equis* has been used as a nominal modifier which has been reanalyzed as a sentence modifier in the second diachronic step (see Figure 4):

	- a. *Equis* used as a nominal modifier (1st diachronic step)
	- b. *Equis* used as a sentence modifier (2nd diachronic step)

Figure 4: Diachronic shift of discourse adverb *equis*

### **4.3 Summing up and formulating predictions for corpus analysis**

For a diachronic analysis, I have assumed that the variable use of *equis* is the trigger for other uses, namely the discourse adverb use and the predicative/adjectival use. The discourse adverb use is explained as the result of scope shift of speaker's indifference from nominal to sentential domain. The indifference of the variable x is derived from Gricean conversational maxims. The predictions for the distribution of diachronic data my analysis makes are the following:

### Olga Kellert


The synchronic data meet predictions 2 and 3 as has been shown in Section 2 (see examples 6–7). Other predictions are tested in the following section.

### **5 Testing the diachronic analysis on diachronic corpus data**

In this section, I will discuss the diachronic distribution of *equis* and investigate the question of when the different synchronic uses of *equis* discussed in Section 2 appeared in diachrony and test the predictions derived from my analysis in Section 4.3.

The diachronic data was taken from the corpus CDE. I have extracted 163 occurrences of *equis* and classified them according to syntactic and semantic/pragmatic function as for the synchronic *equis* (see Section 2).

*Equis* increases in relative frequency the in 20th century (see Table 4, rel. freq. value 5.43 in 20th c. against all other values below 1 before the 20th c.). Most occurrences of *equis* in the 20th century can be found in the text genre that represents oral speech (see freq. value of 24.57 in ORAL against other values in other registers below rel. freq. 3 in Table 4).

Before the 20th century, *equis* appears in Latin texts with the meaning 'horses' (39 occ. out of 42). Before 20th c., the variable or character use 'x' is very infrequent (3 occ. out of 42) in Spanish texts.

### 5 Semantic and syntactic change of equis in Mexican Spanish

 th c. 20th c. Context 13th 14th 15th 16th 17th 18th 19th 20th Acad. News Fict. Oral Per mil. 1.63 0.15 2.33 0.35 0.41 0.10 0.36 5.43 0.60 0.60 2.94 24.57 Total 163 1 19 6 5 1 7 124 3 3 14 104

Table 4: Diachronic distribution of *equis* in CDE subcorpus diachr.


'He made many Xs in the Semontano.' (18th c.)

	- the cases

'For some or other reasons, the man won't be against that because he was a real and generous gentleman and he knew how to react in every case.' (18th c.)

No discourse adverb use of *equis* was observed until the 20th century.<sup>5</sup> New functions and meanings appear in all varieties in the 20th century: prenominal and postnominal modifier and predicative function of *equis* with the speaker's ignorance (almost all in the oral register), e.g. *equis razón* 'some reason but I don't know which one' or modifier of quantifier (*cada equis años* 'every certain

<sup>5</sup>As a reviewer has pointed out, the fact that *equis* was not found with a certain function in diachrony does not necessarily mean that this function did not exist at that time, given that the corpus under investigation is relatively small. Therefore, the conclusion about the absence of some function can only be tentative. I agree with this conclusion and suggest drawing a stronger parallel to English *whatever!/whatevs!* in future research, which shows a similar behavior and offers more diachronic data.

### Olga Kellert

year'), *una suma de dinero equis al mes* 'some amount of money X per month', *las agencias son equis* 'the agencies are X'. *Equis* developed a discourse adverb use with the speaker's indifference in the 20th century (see Section 2).

The distribution of *equis* is summarized below (see Table 5) in order to capture different uses of *equis* and their diachronic distribution.


Table 5: Summary of the diachronic data for the analysis

The diachronic data shows that prediction 1 and prediction 5 are borne out empirically, i.e. the variable use precedes other uses (predicate and discourse use). The text genre prediction is also met, i.e. *equis* appears more often in the oral speech genre. The diachronic data needs to be tested with respect to other predictions in future research, especially prediction 4 about the shift in attitudeholders.

Finally, I would like to address the question whether the shift from variable x into discourse adverb *equis* can be considered as an instance of "subjectification" described in the literature on semantic change (Traugott 1995, Company Company 2003, among others). Subjectification is defined as a pragmatic-semantic process whereby "meanings become increasingly based in the speaker's subjective belief state/attitude toward the proposition", i.e. towards what the speaker is talking about (Traugott 1995: 31). The diachronic data has shown that the item *equis* was used to mean simply a variable in written texts that do not contain any dialogs or overt linguistic expressions of speaker's beliefs or attitudes. This restriction changes in 20th century where *equis* starts to be used in oral speech in dialogs that express speaker's attitudes such as indifference. This change seems to match the subjectification hypothesis. However, I doubt that the speaker's indifference appeared out of the blue with the use of *equis* in oral speech as the

### 5 Semantic and syntactic change of equis in Mexican Spanish

subjectification hypothesis might suggest. The indifference is also present in the variable use of *equis* but in contrast to the discourse use, the indifference is just a pragmatic inference and as such not lexically encoded. I thus assume that the subjectification hypothesis does not fully capture the diachronic process of *equis*.

### **6 Summary and outlook**

In this article, I have shown different functions of Mexican Spanish *equis* and suggested to analyze them uniformly on the basis of an indifference inference. I started my analysis with von Fintel's (2000) notion of indifference derived from the counterfactual inference expressed in English *whatever* free relatives. On my analysis, the variable *equis* or x can also be analyzed in a similar way expressing counterfactual inference, i.e. the speaker is indifferent about which of the values of the variable *equis* or x holds in the actual world. This indifference inference is not lexically encoded in the variable use itself but follows from pragmatic principles. Later in time, the indifference inference shifts towards truth values of propositions and it lexicalizes in the discourse adverb *equis*. The degree predicate *equis* also encodes the indifference inference of the variable *equis* as a lexical property.

I believe that there is a parallel between the diachronic change of *equis* and the shift from *whatever* as an indefinite pronoun (e.g. I grabbed whatever tool was in front of me) to *whatever/whatevs* as discourse adverb in American English which expresses roughly that the speaker does not find it important whether the proposition mentioned previously in the discourse is true or not:

### (69) A: *This is true.* B: *Yeah, whatever.*

I will investigate the parallel between the discourse adverb *equis* and *whatever/whatevs* in future research. The diachronic data is taken from the diachronic subcorpus CDE, which does not distinguish between Spanish varieties. In future, it is necessary to look into diachronic corpora that distinguish between varieties.<sup>6</sup> In future research, one could also investigate whether *equis* as degree predicate can be modified by other degree modifiers such as modifiers used in comparative clauses. I found some examples on the Internet that need to be checked with native speakers:

<sup>6</sup> I briefly investigated *equis* in the *Corpus Diacrónico y Diatópico del Español de América* (COR-DIAM). It contains the total amount of 6,435,906 words from 19 Latin-American countries from 1494 to 1905, with documents from archives, journals and literature. I did not find any use of the discourse adverb or the evaluative predicate as expected. Instead, I found the variable use x referring to some date or number, e.g. *México, x de noviembre, 1578* 'Mexico, x/10 of November 1578'.

### Olga Kellert

(70) *sin* without *duda* doubt *una* one *de* of *las* the *cosas* things *más* more *equis* equis *que* that *he* have *visto.* seen 'No doubt, it's the most unimportant/unremarkable/worst thing that I've ever seen.' (Internet)

Furthermore, it is necessary to investigate whether degree modification is impossible/ungrammatical with discourse adverb *equis*:

	- B: *(Muy) equis.*

'I don't care about utterance x.' (to be checked in future)

If degree modification is grammatical with discourse adverb *equis*, it should be possible to unify the predicative *equis* and the discourse adverb *equis*. Moreover, I need to account for the meaning 'several, many' of *equis* combined with plural nouns (see Section 2.1). As *equis* is mostly used in informal speech and many of its uses represent a recent phenomenon, I expect to find speaker variation with the uses of *equis*. Thus, the data need to be checked systematically for speaker variation, too.

### **References**


5 Semantic and syntactic change of equis in Mexican Spanish


### **Chapter 6**

# **German** *noch* **under reanalysis**

### Martin Kopf-Giammanco

Universität des Saarlandes

This paper investigates (i) the semantics of present day German *noch* in comparative readings. In doing so it (ii) presents experimental work on comparative *noch*'s presuppositional meaning component. In the second part, I will (iii) provide a survey of diachronic data from the Old German period and (iv) propose a process of reanalysis for the comparative reading of *noch* based on its temporal reading.

### **1 Introduction**

There is a good amount of synchronic work on German *noch* ('still/even/yet') and its various readings, uses and its logical equivalents and counterparts (e.g. König 1977; Löbner 1989; Ippolito 2007; Umbach 2009b,a; 2012; Beck 2016b,a). The major readings are temporal, additive, marginal, and comparative. By and large these categories are clear. However, there are a few blurred lines, inconsistencies and overlaps across the literature. What is missing is diachronic work tracing the development *noch* has undergone and how the various readings have come about. In this paper, I want to address the diachrony of the comparative reading of *noch* (*noch*comp), more specifically, what its origins might be. After a brief introduction to the major uses on *noch* in Section 2, I will discuss the main contributions to the semantics of *noch* in Section 3. In Section 4, I will report on an experiment geared towards identifying the presuppositional properties of *noch*comp which, in turn, will inform the discussion on the semantics of *noch*comp in Section 5. Section 6 will give an overview of the diachronic data, which is the basis for the discussion of an analysis of diachronic change in Section 7. The discussion on diachronic change is based on systematic semantic and pragmatic annotation of corpus data (cf. e.g. Gergel, Kopf-Giammanco & Watkins 2017; Gergel et al. 2016; Gergel, Kopf-Giammanco & Masloh 2017). At the core of the proposal, *noch* is undergoing a shift of scales – from a scale of times to a scale of degrees.

Martin Kopf-Giammanco. 2020. German *noch* under reanalysis. In Remus Gergel & Jonathan Watkins (eds.), *Quantification and scales in change*, 161– 198. Berlin: Language Science Press. DOI:10.5281/zenodo.3929247

Martin Kopf-Giammanco

### **2 Uses of** *noch* **in Present-day German**

In this section I want to briefly revisit the major uses of present day *noch*:

	- b. (presupposition) PSP: Peter is at the office at a relevant earlier time t\* which immediately precedes t.

The example in (1) shows the temporal use of *noch* (*noch*temp). Its semantics will play a central role in the discussion below and I will go into depth there. Sentence (3) shows the comparative use of *noch*. It is equally important for this paper. Its semantics and diachronic development will be discussed in depth. It has been suggested that the presuppositional contribution of *noch*comp is a condition on the context to the effect that the comparison base exceeds a contextually given standard (e.g. Hofstetter 2013):


	- b. ?PSP: The standard term of comparison, Peter's height, is relatively high.

The following use is the marginal reading of *noch*. The basic idea is, for e.g. (5), that, out of all places that are in Austria, Salzburg is a marginal case.

(5) *Salzburg* Salzburg *ist* is *noch* still *in* in *Österreich.* Austria intended: 'Salzburg is in Austria but just barely (since it's so close to the border)' → marginal reading

(6) is an example for the temporal, subconstituent reading of *noch*. One could argue that, for (6), out of all the times in the morning the time that Lydia left is a marginal time.

6 German noch under reanalysis

(6) (Adapted from Beck 2016b: ex. 27–28) *Lydia* Lydia *ist* is *noch* still *am* in the *Vormittag* morning *abgereist.* departed. Intended: 'It was still morning when Lydia left.' → temporal, subconstituent reading

The last reading to be introduced here is the additive use of *noch*:

(7) *(Felix* Felix *hatte* had *(schon)* already *drei* three *Bier.)* beers *Jetzt* now *trinkt* drinks *er* he *noch* still *ein* a *Bier.* beer 'Felix (already) had three beers. Now he is having another beer.'

→ additive reading

### **3** *Noch***comp**

Before entering the discussion on *noch*comp's diachronic development, we need to put the semantics for present-day *noch*comp in place. The following is a review of the literature on *noch*comp with a focus on the two most recent analyses of *noch*comp (Umbach 2009b; Hofstetter 2013), followed up by the report on an experiment which looked into the presuppositional meaning component of *noch*comp.

The major contributors to the semantics of the comparative reading of *noch* are König (1977) and Umbach (2009b) as well as Hofstetter (2013)<sup>1</sup> . König (1977) analyses *noch*comp from a marginality point of view, i.e. sentences like (8) – in König's words – "imply a second comparison involving Peter" (ibid., p. 189) based on the positive form of the adjective in *Peter is tall*:


The implicit comparison in *Peter is tall* compares *Peter* to a standard degree of tallness (i.e. average body height) and places Peter's height above that standard. Out of all individuals that are ranked on the scale of degrees of tallness, Peter is a marginal case (König 1977). Umbach (2009b), commenting on König 1977: there is a 'reversal of roles' when we compare this analysis (8 and 9) to König's analysis of a prototypical marginal reading of *noch* (*noch*marg), cf. (10) and (11):

<sup>1</sup>Hofstetter (2013) has a focus on the Turkish evaluative intensifier *daha* which, especially in its use in comparatives, shares crucial properties with German *noch*.

### Martin Kopf-Giammanco


Umbach (2009b) fleshes out König's (1977) proposal and concludes that the role reversal is due to different syntactic structures. In a comparative reading *noch* combines with an AP (3) and in a marginality reading *noch* combines with a DegP (3).


Umbach's (2009b) criticism of König's (1977) proposal is that it does not explain why a "comparative may trigger norm-relatedness when combined with comparative *noch*" (cf, Section 3.1, below). What is more, the diachronic data does not support a trajectory based on König's analysis. The comparative use of *noch* is attested considerably sooner than the marginal reading – at least as far as *noch*marg operating on a scale of degrees or paths is concerned.

### **3.1 Umbach's (2009b) analysis**

The core of Umbach's (2009b) proposal is that *noch*comp is anaphoric and, thus, relates to a preceding comparison. Her discussion is based on anaphoricity and norm-relatedness which is entailed in some but not all contexts that *noch*comp can occur in, cf. (12) to (14); with +/− *NR* indicating norm relatedness arising (+) or not arising (−).

	- a. *Adam ist größer als Chris. Aber Berta ist noch größer (als Adam).* −NR 'Adam is taller than Chris. But Berta is still taller (than Adam).'
	- b. *Adam ist größer als 1,80m. Aber Berta ist noch größer (als Adam).* −NR 'Adam is taller than 1.80m. But Berta is still taller (than Adam).'
	- 'Adam is not small. But Berta is still taller (than Adam).'

6 German noch under reanalysis

### (14) *Berta ist noch größer als Adam.* +NR 'Berta is still taller than Adam.'

According to Umbach (2009b), neither (12a), (12b) nor (13b) entail that Berta is taller than the norm. However, (13a) does entail norm-relatedness due the antecedent comparison involving the positive form of the same adjective as in the *noch*-sentence. This suggests that norm-relatedness is triggered by *noch*comp "if and only if the comparison base of the antecedent statement is given by the norm of the adjective in the *noch* comparative" (Umbach 2009b: 10). In other words, the antecedent comparison needs to contain (i) the same adjective as the *noch*sentence and (ii) the adjective must be in the positive form and (iii) provide a standard degree of tallness which (iv) serves as the comparison base of the antecedent comparison. These criteria do not hold for (12a) and (12b), where the comparison base of the antecedent is provided by the height of a third individual (*Chris*) or a measure phrase (*1.80m*), and for (13b), where a different norm is introduced by *klein* 'small'.

*Noch*comp occurring in the third type of context ("out of the blue"), shown in (14), entail that both Adam and Berta are tall. Umbach suggests to analyze (14) along the lines of (13a) and take the antecedent to be accommodated. The accommodated antecedent will be of the form *Adam is taller than the tallness norm*, i.e. composed of the comparison base of the *noch*-sentence and the norm of the adjective.

Umbach's (2009b) conclusion is that comparative *noch*, in some but not all contexts, entailing norm-relatedness is a consequence of *noch*comp being "anaphoric requiring an antecedent comparison" (Umbach 2009b: 10). It is precisely the anaphoricity for an antecedent comparison that is in contrast to König's (1977) proposal which suggests that an existential presupposition of an additional individual is the contribution of *noch*comp. Umbach's (2009b) point of view is that there is an antecedent comparison, not an antecedent individual, with the comparison consisting of a pair in a degree-relation.

In formalizing the semantics of her analysis, Umbach cites van der Sandt (1992) in following the "presupposition-as-anaphors paradigm" (Umbach 2009b: 11) and arrives at the interpretation of *noch*comp in (15). The underlined part is the presupposition, where is provided by the standard term of comparison and is a free variable bound by the antecedent comparison:

(15) (Umbach 2009b: ex. 24, emphasis in the original) [[ [AP *noch* [AP *größer* ]] ]] = y x.: ht(y) > d. ht(x) > ht(y)

### Martin Kopf-Giammanco

(15) applied to (16a) would yield (16b). The free variable can then be bound to one of the contexts in (12) and (13) which provide the degrees in (17): ht(chris), 1.80m, dS-tall, dS-small.

	- a. *Berta is noch größer als Adam.* 'Berta is still taller than Adam.'
	- b. ht(adam) > d. ht(berta) > ht(adam)


Consequentially, according to Umbach, it will be entailed that Berta is taller than Chris, taller than 1.80m, taller than the tall-standard, or taller than the smallstandard. However, that Berta is tall is only entailed by (17c) – since Adam is tall and it is asserted that Berta is taller than Adam.

With regard to Umbach's interpretation of *noch*comp in (15), she points out a particular shortcoming when compared to König's (1977) proposal, namely the lack of "order – of time or marginality – which is commonly regarded as essential for the meaning of *noch*" (Umbach 2009b: 12). Furthermore, additive *noch* (*noch*add) as well as the temporal and marginality readings of *noch* relate to a scale, with *noch*temp relating to the order of times, *noch*marg relating to the order of marginality (or inverse prototypicality) and *noch*add relating to the order of mentioning. This order of mentioning is "frequently aligned with a contextually given 'semantic' scale, for example, time in narratives" (Umbach 2009b: 12). And further:

Comparative *noch* requires an antecedent. This is what makes it additive. The related scale is, first of all, to [sic!] the order of mentioning. But the order of mentioning is aligned to the order of degrees given by the adjective of the *noch*-comparative such that the latter preserves the former: if comparison1 one [sic!] precedes comparison2 in mentioning, the comparison subject of comparison1 has to precede the comparison subject and the comparison base of comparison2 with respect to the order of degrees.

(Umbach 2009b: 13)

### 6 German noch under reanalysis

Essentially, Umbach states that all uses of *noch* are scalar, with the additive use of *noch* relating to the order of mention and the comparative use of *noch* being "subsumed as a particular instance of the additive reading relating primarily to the order of mention and secondarily to the degrees given by the adjective" (Umbach 2009b: 14).

### **3.2 Hofstetter's (2013) analysis of** *noch***comp**

For the following discussion, I turn back to example (8), repeated as (18). Hofstetter (2013) assumes that the PSP for *noch*comp demands that Peter's height is relatively tall, i.e. exceeds a contextually given standard, regardless of what the context is:

	- b. PSP: The standard term of comparison, Peter's height, is relatively high.

The underlined part in (20) points to the PSP that the comparison base of the *noch*-comparison, d' exceeds a contextually given standard. In other words, there is no norm-relatedness involved in Hofstetter's semantics for *noch*comp and not the same anaphoricity as in Umbach's (2009b) analysis.

Hofstetter applies the S-family test (Kadmon 2001) for presupposition but does so only for English *still* in an exemplary fashion and concludes that the test "clearly reveals that all members of the family directly presuppose that Peter is comparatively tall". Unfortunately, Hofstetter does not provide any introspective reasoning as to the projection behavior of the proposed PSP.<sup>3</sup>

<sup>2</sup>Hofstetter writes this as ⟦stillevaluative⟧. However, he states that German *noch* and English *still* share the same properties and are equivalent (Hofstetter 2013: 31).

<sup>3</sup> It seems odd to rely on English *still* as an equivalent for the German *noch*comp since American English speakers report that for translations of sentences like (18) they immediately get a temporal reading/a temporal reading is salient for them. It seems to be British English that allows *still* as an equivalent for *noch* in comparative uses. Speakers of American English seem to prefer *even* which, in turn, translates into German as *sogar*. In conclusion and in search of a "better equivalent", I will rely on *still/even* for the translations in this paper – for now.

### Martin Kopf-Giammanco

What Hofstetter does provide is judgment on the following sentence when testing if the meaning component in question in cancelable:

(21) (Adapted from Hofstetter 2013: 27, ex. 2/49) \* *Paul* Paul *ist* is *noch* still *größer* tall.COMP *als* than *Peter,* Peter *aber* but *Peter* Peter *ist* is *nicht* not *groß.* tall Intended as: 'Paul is still taller than Peter, but Peter is not tall.'

The judgement in (21), (\*), is in line with Umbach's (2009b) "out of the blue" example (14). It presupposes an antecedent comparison of the form *Peter is tall* and NRness arises. Consequentially, Hofstetter (2013) considers his intuition confirmed since it is one of the hallmark criteria for PSP that they are not cancelable. If we provide antecedents along the lines of Umbach (cf. 12–14), we can see that the PSP does not arise/can be canceled:

	- b. *Peter ist größer als 1,80m. Paul ist noch größer als Peter, aber Peter ist nicht groß.*

'Peter is taller than 1.80m. Paul is still/even taller than Peter, but Peter is not tall.'


'Peter is not short. Paul is still/even taller than Peter, but Peter is not tall.'

For all examples in (22), we have Hofstetter's sentence from (21) paired with an antecedent sentence fashioned after Umbach's design. All of these utterances are good and felicitous – except (22c), where the assertion in the antecedent sentence is contradicted by the final clause *...aber Peter ist nicht groß* ('...but Peter is not tall'). Conversely, contradicting a PSP in the other utterances (22a, 22b, 22d) should not be possible. Looking at the individual utterances in turn reveals that none of these entail that Peter (or Paul) are tall. These bits of introspective data indicate that Hofstetter's entry for *noch*comp is too restrictive regarding its PSP-component.

6 German noch under reanalysis

### **4 Experiment: Norm-relatedness vs. PSP**

### **4.1 Overview and material**

An experiment was conducted in order to get a clearer picture. At the heart of the experiment, Hofstetter's (2013) analysis, i.e. German *noch*comp triggers the presupposition that the standard term of comparison is taller than a contextually given standard, and Umbach's (2009b) analysis based on norm-relatedness (NRness) were tested against one another.

Table 3 shows one out of 16 token sets. Every token set consists of four target items which, in Table 3, are spread out across the four lines/conditions (for details on the conditions, cf. Section 4.2). Every target item consists of both *condition* and *continuation*. The continuation is the same across all conditions.<sup>4</sup> 16 such token sets were created (cf. appendix, p. 194f. for an overview).


Table 1: Four conditions per token set

The token sets were based on 16 predicative adjectives, thus, in total there were 64 target items. The 16 token sets were split into 8 antonym pairs (*groß– klein*, 'tall–short' etc.) which shared contexts when possible. Differing contexts were created when necessary. Female and male names were counterbalanced (3 female, 3 male), the remaining items are inanimate and unnamed individuals.

The 64 target items were split into eight questionnaire groups<sup>5</sup> which was done in order to prevent response fatigue and reduce questionnaire duration. Every participant rated eight different target items – two from every condition and,

<sup>4</sup> In the questionnaires, condition and continuation were presented as one string, without the gaps in Table 3. They are included here for ease of representation.

<sup>5</sup>They are not to be confused with "groups", i.e. specific groups completing specific conditions.

### Martin Kopf-Giammanco

at the same time, two from every token set. The 64 items were rotated among the questionnaire groups, for details I would like to refer to the appendix, specifically Table 12 on page 198.

In addition to the target items, 16 fillers were created which were the same across all questionnaire groups, i.a. across all participants. The fillers were designed based on the following criteria. They were made to "look" the same; i.e. they consisted of two sentences, the first of which consisting of two clauses. No item was to contain (any use of) *noch*. Moreover, the design required to avoid comparatives and predicative adjectives. There were two very bad fillers in order to prevent response fatigue and test for subject attention. German *auch* ('also/ too') was used as a distractor; ten filler items contained *auch*, six did not. Male and female names were counterbalanced (8 & 8). The filler items were based on parallel/similar contexts as the test items – as far as possible; for "good" fillers – contrasting contexts were created (*to like/dislike; to play an instrument well/awfully, ...*).

### **4.2 Experimental design, methods and participants**

The experiment was based on a two by two design, that is two factors with two levels each. The first factor was the proposition *A is tall* being asserted in the first clause (level 1, ass) or not (level 2, com, i.e. for *comparative* instead of assertion). The second factor was *noch* being absent (level 1, -n) or present (level 2, +n). This resulted in four conditions as shown in Table 2. For ease of representation and readability, I will use conditions 1–4 rather than the factor-level combinations for the discussion below. The four conditions amount to four minimal pairs. The numbering of the four conditions (1–4) and their vertical representation in the above table does not indicate any ranking as to the predictions for experimental ratings by either Umbach or Hofstetter.

Table 2: 2×2 design → 4 conditions


### 6 German noch under reanalysis

Subjects were presented with the respective target item. They were instructed to imagine that the first sentence (condition) and the second sentence (continuation) are uttered by one person in one situation. Their task was described as to judge whether both sentences can be true in one and the same situation. For every item the prompt was *Können beide Sätze als wahr geäußert werden?* ('Can both sentences be uttered as true?'). Subjects had a 6-point scale at their disposal ranging from *Nein, ganz sicher nicht.* ('No, definitely not'; 1 point) to *Ja, ganz sicher.* ('Yes, definitely'; 6 points), with these two as the only labels, at both ends of the scale. In the following, I will refer to high ratings of (close to) 6 points as "good" rating and vice versa to low ratings as "bad".

### **4.3 Predictions**

For conditions 1 and 2, both Hofstetter's and Umbach's predictions are that they are rated as bad since the (identical) continuations contradict the assertions.

Condition 3 is the critical condition. Hofstetter's prediction here is that participants would rate it as bad since the continuation should contradict the PSP that *A is tall*. This is due to A's height being presupposed as exceeding a contextual standard (cf. 20 on page 167). Arguably, following Hofstetter, one might expect ratings similar to condition 1 where the proposition *A is tall* is asserted and then contradicted in the continuation. Umbach's prediction for condition 3 is that it should be rated as good since norm relatedness (and the inference that C or A are tall) should not arise here and, thus, there is no contradiction. This is due to the free variable (cf. 15) being bound to an antecedent comparison of the form in (17a).

For condition 4, both Hofstetter and Umbach predict good ratings – *A is tall* is not asserted (factor 1, level 2), hence no contradiction with the continuation, and *noch* is absent (factor 2, level 2), hence no PSP can be triggered (for Hofstetter) or norm relatedness cannot arise (for Umbach).

As mentioned, condition 3 is the critical condition where Hofstetter's (2013) analysis, and Umbach's (2009b) analysis have differing predictions:

Table 3 sums up the structure of items in all conditions and the respective predictions in terms of ratings.

A final note on experimental design and the decisions made along the way: the experiment underwent a number of developmental stages and updates due to test runs yielding inconclusive results. For example, items fashioned after other examples from the existing literature were considered (e.g. 21 with adversative *aber* 'but') as well as weaker formulations in the prompts were considered instead of asking for truth judgments (i.e. tapping into participants' logical/

### Martin Kopf-Giammanco

Table 3: Experimental design; NR-ness; 4 conditions, 2×2


*<sup>a</sup>*Hofstetter's prediction

*<sup>b</sup>*Umbach's prediction

structural thinking). The latter decision was made in order to avoid issues of (non-)accommodation and processing effects.

The questionnaires were compiled and published on *SoSci: Der Online-Fragebogen* which provides a singly survey link and randomly selects questionnaires if anybody enters the study via the survey link. The survey link was shared on SurveyCircle (Johé 2019) and various social media accounts.

### **4.4 Participants**

123 participants completed the study. The following meta-data are reported as available: participants' ages ranged from 18 to 72 years old at an average age of 26.4 years. 74 identified as female, 35 identified as male, 14 did not identify. In terms of country of origin ("where did you grow up?"), 81 participants were from Germany, 18 from Austria, one from Switzerland, and one from Italy. The rest of the participants did not disclose that information.

### **4.5 Data processing**

Starting with 123 responses, I excluded subjects (i) whose native language wasn't German (10 participants did not disclose their native language at all and were excluded), (ii) who did not give positive consent to use their responses, (iii) who indicated negative overall commitment to the experiment, (iv) who indicated that

### 6 German noch under reanalysis

their responses should not be considered meaningful responses and (v) who admitted to having been distracted multiple times throughout the questionnaire. This resulted in 95 admissible participants. Disregarding filler items, each participant rated 8 items (2 from each of the 4 conditions), resulting in 760 data points overall, with 190 data points for every condition.

### **4.6 Results**

### **4.6.1 Descriptive statistics**

The following provides a first look at the results in terms of descriptive statistics. By and large, the results seem to support Umbach's (2009b) analysis. As expected without any bias for or against any of the analyses, conditions 1 (ass\_+n) and 2 (ass\_-n), where the assertion that e.g. *x ist groß* ('x is tall') is contradicted by the continuation, received low ratings when asked if both sentences can be uttered as true – the medians for both conditions are 1.0, cf. Table 4 and Figure 1, below. However, conditions 3 (com\_+n) and 4 (com\_-n), received quite high ratings with both their medians at 5.0. For more descriptive statistics see Figures 1 and 2 – for box plots and histograms respectively. See Section 4.6.2 for a more detailed discussion of the results based on more detailed statistical analysis.


Table 4: Descriptive statistics for the 4 conditions

### **4.6.2 Linear mixed effect model**

I built linear mixed effects models for my data with R (R Core Team 2019) <sup>6</sup> with the lme4-package<sup>7</sup> (Bates et al. 2015). The ratings (1–6) were z-transformed into norm scores. That is, for every participant I calculated means (part\_mean) and

<sup>6</sup>R version 3.6.0 (2019-04-26)

<sup>7</sup> lme4 version 1.1-12

Martin Kopf-Giammanco

Figure 1: Ratings (left) and norm scores (right) over 4 conditions

Figure 2: Histograms of 4 conditions (2 factors, 2 levels each)

### 6 German noch under reanalysis

standard deviations (part\_sd) and then for all eight ratings per participant (cf. Figure 1 for boxplots of ratings and normscores):

$$\text{(23)}\quad\text{norm score} = \frac{\text{rating} - \text{part\\_sd}}{\text{part\\_mean}}$$

For the further discussion on statistic modelling, the terms *norm score* and *score* will be used synonymously. For model 1, score is taken as a function of an interaction of the fixed effects (factor 1: assertion of *x is tall* (level 1) or not (level 2); factor 2: *noch* absent (level 1) or present (level 2)) and model 2 is a reduced version of model 1, i.e. without the interaction as in model\_1. Both models accounts for participants and contexts as random effects assigning random intercepts. I was able to also include random slopes for both factors correlated with the respective two random intercepts, resulting in a maximally random effects structure for my models. Find the outline of the basic structure for model 1 in (24) and for model 2 in (25):


Find the output for model 1 in Table 5. Note that I used the lmerTest package<sup>8</sup> (Kuznetsova et al. 2017) to add -values to the lmer-summary.


Table 5: R output for lmer() call on model 1 (24)

Factor 1, level 2 has a significant effect with a -value at 15.264 and a -value below 2 × 10−16. Most importantly, there is no significant effect of factor 2, level 2 with a -value of 0.819, cf. Table 5. Moreover, there is not interaction between factor 1, level 2 and factor 2, level 2. To be sure and test specifically for an interaction between factors 1 and 2 I calculated model 2. Comparing models 1 and 2 (in R with anova()) gives the output in Table 6 (cf. Winter 2013).

<sup>8</sup> lmerTest version 3.1-0

### Martin Kopf-Giammanco

Table 6: R output for anova() call on models 1 and 2


Model 2 has a slightly lower AIC and in the comparison of the two models the difference comes out as not significant. This suggests that there is no interaction between factors 1 and 2 ( 2 (1) = 0.7649, = 0.3818). There is an online repository<sup>9</sup> containing the experimental data and statistics scripts.

### **4.7 Conclusions**

At first glance, the results seem to support Umbach's (2009b) analysis. It seems that the non-asserted proposition, i.e. that the standard term of comparison in a *noch*-comparative exceeds a contextually given standard degree, can be canceled and may, therefore, not be regarded presuppositional.

As pointed out by an anonymous reviewer, there may be flaws inherent to the experimental design to the effect that, in line with Hofstetter's analysis, the PSP of *noch* in condition 3 is unmet, remains non-accommodated and nothing should be there to contradict/cancel by the continuation. Among other things, it was exactly this point that we attempted to address by asking participants to judge the compatibility of the truth of two sentences. Nevertheless, the issue may remain.

A few of the desiderata in retrospect is the lack of judgments for data like Hofstetter's (21) (repeated here as 26). What is it that makes this sentence seemingly infelicitous and under what circumstance could this sentence be felicitous? Would an antecedent comparison as in (27) (cf. conditions 3 and 4 above) make (26) felicitous? It is possible – in this instance I do not have reliable introspective judgments and conclude that more experimental work is required taking a different approach in eliciting judgments.

(26) (Adapted from Hofstetter 2013: p. 27, ex. 2/49) *\* Paul* Paul *ist* is *noch* still *größer* tall.COMP *als* than *Peter,* Peter *aber* but *Peter* Peter *ist* is *nicht* not *groß.* tall

<sup>9</sup>https://github.com/M-K-G/noch\_FoDS2

6 German noch under reanalysis

(27) ? *Peter* Peter *ist* is *größer* tall.COMP *als* than *Kurt.* Kurt *Paul* Paul *ist* is *noch* still *größer* tall.COMP *als* than *Peter,* Peter *aber* but *Peter* Peter *ist* is *nicht* not *groß.* tall 'Peter is taller than Kurt. Paul is still taller than Peter but Peter is not tall.'

With these caveats in mind, and accepting the results of the above experiment, I will turn back to the semantics of *noch* in the next section.

### **5 Updating the semantics of** *noch***comp**

Based on the above findings, I propose to update the lexical entry for *noch*comp:

(28) ⟦nochcomp⟧ = d\* ∈ D<sup>d</sup> .Co ∈ D⟨⟨d,t⟩,⟨⟨d,t⟩,t⟩⟩.D<sup>1</sup> ∈ D⟨d,t⟩ .D<sup>2</sup> ∈ D⟨d,t⟩ ∶ d\* ≤ max(D<sup>1</sup> ). CO max(D<sup>1</sup> ) max(D<sup>2</sup> ), where d\* is a free variable to be bound by the context and ranked lower than the max-degree of the comparison base and Co is the comparative operator of the *noch*-comparison.

I assume clausal comparison with the comparison operator of type (cf. Beck 2011):

⟨⟨d, t⟩, ⟨⟨d, t⟩, t⟩⟩

and the lexical entry in (29). The logical form for sentence (30a) (=14) is in (30b) where one can see that quantifier raising solves the problem of the type mismatch of the DegP and adjective (for both clauses). Via predicate abstraction in (31a) and intermediate steps in (32) (relying on the lexical entry 29), the LF in (30b) yields (33) (relying on the lexical entry for *noch* in 28), cf. LF in Figure 3.

	- b. [ noch d\* [-er than [2[Adam ist [AP t2 groß]]] [1 [ Berta ist [AP t1 groß]]]]]

### Martin Kopf-Giammanco

(34) (30a) is defined only if Adam is taller than something else relevant, i.e. the degree d\*, provided by the context, and it is true if and only if Berta is taller than Adam.

Figure 3: LF for (30a)

There is no condition on the context that Adam is taller than a contextually given standard. Thus, norm relatedness does not arise (cf. Umbach 2009b and Section 3.1), which seems to be the desired situation given the experiment reported above. In essence, the free degree variable *d\** is what norm relatedness hinges on. Depending on what in the context d\* refers to, *noch* will give rise to norm relatedness. Let's assume the context provides a proposition along the lines of condition 1 (cf. Table 1, page 169), e.g. *Adam ist groß* ('Adam is tall') preceding (30a). In such a case, the maximum degree to which *Adam is tall* (i.e. comparison base of *noch*-comparative) is equal or higher than the degree of the height of Adam as per the proposition from the context – which comes with a positive operator and puts Adam's height above a standard of height. Umbach (2009b) points to this aspect of meaning of positive degree adjectives in the context of *noch*comp. So do, more generally, von Stechow (1984; 2006) and Beck (2011): the idea is that there is a scale *S* introduced by an adjective, e.g. tall, and there is a function *N* that yields the neutral part of the scale and the neutral part *N(S)* contains all elements that are neutral in terms of tallness. The positive operator

### 6 German noch under reanalysis

universally quantifies over all the degrees contained in the neutral part of the scale (von Stechow 2006).

(35) (Adapted from von Stechow 2006: ex. 3-1–3-3) S |--------[-------------------]--------→|

. short Neutral tall


With regard to the semantics of *noch*, if there is no such proposition in the context but e.g. a comparison as in condition 3 (cf. Table 1, p. 169), then there is no Pos-operator involved (von Stechow 1984: 62) and, in Umbach's words, normrelatedness does not arise.

(38) *Berta ist noch größer als Adam.*

The question that has remained unanswered is what happens in out-of-theblue *noch*-comparatives – with no overt antecedent. Umbach argues that for (30a) (repeated here as (38)) an antecedent gets accommodated as being "composed out of the comparison base of the *noch* comparative and the norm of the adjective (with respect to the comparison class)" (Umbach 2009b: 10). Essentially, this accommodated antecedent is of the form 'Adam is tall', i.e. it features von Stechow's (2006) Pos-operator.

### **6 Diachronic data**

The following discussion mainly relies on data from the DDD corpora of Old German (OG, Donhauser et al. n.d.) via ANNIS (Krause & Zeldes 2016). Other supplementary sources include *Kali Korpus* (n.d.), and the TITUS project (Gippert et al. n.d.).

The OG period was split into subperiods:

*OG0* pre-750 *OG1* 750–850 *OG2* 850–950 *OG3* 950–1050

### Martin Kopf-Giammanco

The existing part-of-speech (POS) annotation was supplemented with annotations specifically geared towards occurrences of *noh*, the OG form of ModGerman *noch* (in the following, I will use the two forms interchangeably). The existing (DDD) POS tags for *noh* are ADV (adverbial) and KON (conjunctive). For this paper, I ignore conjunctive uses of *noh* such as in (39). The ADV occurrences of *noh* were – on top of the existing annotation – annotated for temporal, additive and comparative uses, i.e. *noch*temp, *noch*add, and *noch*comp.

(39) *Nist* Not-is *thes* of.this *gisceid* boundary *noh* **nor** *giuuant,* measure, *uuio* how *er* he *girrit* confuses *thaz* the *lant,* land, *uuio* how *er* he *iz* it *allaz* all *uuirrit,* stirs, *ioh* and *thesa* this *uuorolt* world *merrit.* injures.

'There is neither boundary nor measure to which he disturbs the country as he causes it trouble and to the entire world.'

(Otfried DDD\_O\_Otfr.Ev.4.20, edition 278–289 via ANNIS)

Table 7 gives an overview of occurrences of *noh/noch* in the diachronic data available in the DDD corpus. Table 8 shows the frequencies of all occurrences of adverbial uses of *noh* across the subperiods OG1–OG3 in all of the text of the DDD corpus.

Regarding OG1, the Heliand text had to be excluded since its periodisation is unclear with two different sources having found their way into the corpus texts.

Otfried is the only major text available from OG2 (and unfortunately in verse) and, therefore, was included. The additional material available from OG2 are minor hits from the *Smaller Old High German language monuments* ('Kleinere Althochdeutsche Sprachdenkmäler') which, in turn, are difficult to pin down in terms of periodisation as a whole. Hits from single texts were considered for annotation. The numbers in Table 9 are based on the final selection of corpus text considered.

A full annotation of 214 tokens from the OG3 (Notker) texts is incomplete as of yet. The numbers in the above table are based on 76 OG3-tokens annotated in detail and are to be taken to be representative for the entire subperiod. Among the 76 tokens categorized, there was one *noh* with an unambiguously comparative reading. The annotation of 76 tokens was supplemented with targeted corpus searches for *no(c)h*comp uses, with various queries among all 214 uses of *noh* in the Notker texts which yielded two more hits of *no(c)h*comp, bringing the total for OG3 to 78 tokens.

In the following, I want to discuss the most important aspects and examples of the diachronic data. The most problematic bit of diachronic data is (40).

### 6 German noch under reanalysis


Table 7: Subperiods and sources; KON vs. ADV

Table 8: Frequencies of *noh* (ADV), based on no. of tokens in OG subperiods


*<sup>a</sup>*The frequency for OG3 is is an approximation since the corpora's word counts include the parts in Latin.

Table 9: Subperiods and readings of *noh*.


*a* read as e.g.: in OG2, 2 *noh* are comparative, out of which (2) are ambiguous to other readings

### Martin Kopf-Giammanco

(40) *Ibu* If *auuar* but *in* in *aftrun* back *steti* place *ga sizzis* you.sit, *enti* and *quuimit* comes *dir* you *otlihhero* 'lower', *qui dit* says, *daer* who *dih* you *za* to *demo* the *naht muose* dinner *la dota,* invited, *sizzi.2SG.IMP* sit *NOH* still/even *hohoro.COMP* higher *baz.COMP* better, *enti* and *ist* is *dir* you *danne* then *guot lihhora;* honorable 'But if you sit down somewhere in the less prominent places and the person who invited you for dinner tells you to sit in a more prominent place, then it is better to sit still higher and that is then honorable.' (MonsF-1,M.XIV, edition 141–152)

(40) is problematic for a number of reasons. The most striking problem is that it could be a very early instance of *noh*comp. Example (40) is the reason that, in Table 9, <sup>10</sup> the first line for *noh*comp shows 1 (1/0?). Let us look at it in more detail: the wider context is about humility and humbleness. (40) is part of an allegory and the allegorical context is limited in potential to disambiguate. The preceding context talks about how shameful it is to take a prominent seat at a table when invited to dinner and then being told to take a less prominent seat.

The comparative reading doesn't have strong support as the expected action to attain humility (in a Christian world view) would be to turn down an offer to sit higher/take a more prominent seat. A temporal (further-to) interpretation runs into the problem that this is a hypothetical situation and there is no detectable temporal sequence aside from the salient time of mentioning. Moreover, in contrast to the majority of early uses of *noch*, it lacks a temporal particle adjacent to it. A conjunctive, coordinating interpretation would require another negative constituent to be coordinated with. The example is from the Mondsee Fragments, it is in Bavarian dialect and dates from the early 9th century (~810AD) (Krause & Zeldes 2016). Thus, if (40) constitutes an instance of *noch*comp, it would (i) indicate that Southern dialects of German might have been more innovative and (ii) mean that the comparative reading has been available relatively soon.

Let us turn to more examples of diachronic data, especially ones ambiguous between temporal and comparative readings (both 41 and 42 are from Otfrid, OG2):

(41) *Ladotun* invited *auur* but *tho* then *then* the *man,* man, *ther* who *thes* of.the *gisiunes* seeing *biquam,* became, *quadun,* said,

<sup>10</sup>These numbers do not straight forwardly match numbers in Table 7 since they account for ambiguities, only show *noch* labelled as ADV, and are based on a smaller set of texts.

6 German noch under reanalysis

*sih* himself *thera.GEN* of.the *dati.GEN* deed *noh* still/even *tho* there *baz* better *biknati.SUBJ.PAST.3.SG.* appraise. 'They then requested of the man who gained eyesight to appraise/evaluate his action still/even there more thoroughly.' (1.OG2.OtfEbKell.202.105)

The context for (41) is a story of Jesus giving a blind man eyesight. The miracle was worked on a Sabbath, which is the reason for public outcry. The formerly blind man is being questioned by the people and by the local high council about the events and about his opinion of Jesus – for the third time in (41). The criticism Jesus faces is rooted not only in breaking Sabbath but more importantly that he claims to be god's son, which, in turn, allowed him to do as he wished on a Sabbath.

This sentence is ambiguous to a temporal and a comparative reading. Both, the PSPs for the temporal and the comparative interpretations are satisfied in the context. The preceding context features two instances of the formerly blind man stating his opinion of Jesus. Moreover, there is a (locative/temporal cf. e.g. Petrova 2011) particle adjacent to *noh*. The comparative interpretation is supported by the fact, that the man has stated his opinion of Jesus twice before and, moreover, the statements regarding Jesus have changed in degree ('to the better') – in the eye of the public: at first, the man called Jesus 'the savior'; at the second time, he called him 'a friend of god ... a divine prophet'. Thus, he lessened the degree to which Jesus was stated to be akin to god. Another argument for a comparative interpretation is that, arguably, the finite verb *biknati* (*biknaen*, 'to confess, appraise, declare') is an atelic verb ('hold a belief/attitude') rather than an accomplishment ('declare your attitude/make a statement'). The lack of a direct (accusative) object would support that view. In conclusion, I argue that the comparative interpretation is salient.

As noted, (42) is ambiguous to a temporal and a comparative reading:

(42) *Thar* There *uuarun* were *mit* with *githuinge* violence *thie* the *iungoron* apostles *noh* still *tho* there *inne,* inside, *sie* they *scolta* should *ruaren* move *NOH* still/even *tho* there *mer* more *thaz* the *selba* same *uuoroltlicha* earthly *ser.* suffering. 'The apostles were still inside with violence, they should continue to move/stir the earthly suffering.' (1.OG2.OtfEbKell.351.12)

### Martin Kopf-Giammanco

The example in (42) is set in the context of an allegory with the apostles fishing on – and Jesus on the shores of – the lake Sea of Galilee. The story states that Jesus is not with the apostles anymore and they now have to continue their work without him. Thus, they are situated in the rough waters of the lake (= out in the world) whereas Jesus is on the calm and dry shore (= dead; in heaven). (42) is ambiguous to a temporal and comparative reading. I will discuss it in more detail in the following section.

The following bits of data can be straightforwardly interpreted as comparative uses of *noch*. They all date from the OG3 period, indicating that during this time (950–1050) the comparative reading of *noh* is available.

(43) *Úbe* if *árg* evil.ACC *uuéllen* to.want *uuêlih* bad *íst,* is, *árg* evil *kemúgen,* be.able.to.do, *dáz* that *íst* is *nóh* still/even *uuêlichera.* worse. 'If it is bad to want evil things, then to be able to do evil things is still/even worse.' (Notker.Boeth-DeConPhil.III.201)

(43) is unambiguously comparative. There is no temporal sequence available and there is no temporal particle adjacent to *noh*. In the comparative interpretation, the comparison base (wanting evil) is in the same token. Similarly, there is no temporal sequence discernible in (44):

(44) *Ér* He *íst* is *tero* of.the *góto* gods *chúnnigosto .* most.knowledgeable. *nóh* Still/even *tánne* then/there *bíst* are *tû* you *chúnnigora .* more.knowledgeable, *uuánda* because *ratio* reason *gemág* can.do *mêr* more *dánne* than *sermo .* the.conversation. 'He is the most knowledgeable about the gods. But you are still/even more knowledgeable because reason can achieve more than conversation.'

(1.OG3.N:Mart.Cap.II.111-121.J)

There is no temporal sequence that would support a temporal interpretation. While it is odd that the first clause has the superlative form of the adjective *chunnig* ('knowledgeable'), the Latin gloss does not feature superlative, and I assume that the superlative in the OG version is there for rhetorical reasons. The comparative reading is salient – in both the Latin and the OG versions. In (45) (from OG3) the *noh* is adjacent to *mêrun* ('more'), there is no temporal sequence and (45) can unambiguously be interpreted as comparative:

6 German noch under reanalysis

(45) *Michel* Great *ist* is *íro* her *guôllichi* glory *an* in *dînemo* your *haltâre* savior *christo.* christ. *[Lat.]* [Lat.] *Ímo* him *selbemo* self *gíbest* give *du* you *noh* still/even *mêrun* more *guôllichi.* glory. *unde* and *mêrun* more *ziêreda.* adornment. *sô* as *dû* you *in* him *gesezzest* set *ad* to\_LAT *dexteram* right\_LAT *tuam.* your\_LAT. 'Great is the glory of the church in your savior christ. You give him still/even more glory and more adornment by setting him at your right

side.' (1.OG3.N:Ps:20.61–63)

The major conclusions to be drawn from the data (Table 9) are (i) that the comparative reading of *noch* developed within the OG period and (ii) that the (unambiguously identifiable) additive reading became available alongside the comparative reading. Umbach (2009b) stresses that *noch*comp shares a number of properties with *noch*add, i.e. "patterns with the additive reading of *noch*" (Umbach 2009b: 9). Moreover, while Beck does not state so, her (2016b) analyses of the continuative, the subconstituent reading, and the further-to reading of *noch*temp seem a convincing trajectory from a "standard" continuative reading towards an additive reading. Both, Umbach's (2009b) and Beck's (2016b) analyses and views combined make for a compelling argument to assume that *noch*comp developed based on *noch*add. However, the mere observation that *noch*comp shares similarities with *noch*add does not justify the assumption that the former is derived from the latter – those similarities may well be due to the common origin in *noch*temp. While the diachronic, empirical basis – despite considerable efforts – is admittedly rather weak, I argue that the early ambiguous cases (*noch*comp and *noch*temp) should weigh more heavily. Both (41) and (42) and their contexts license a temporal reading (especially when excluding the comparative operator for the sake of contrasting the involved meaning components as minimal pairs introspectively). The fact that this ambiguity with a temporal interpretation exists among the earliest uses of comparative *noch* in those contexts leads me to propose an analysis of *noch*comp being derived from *noch*temp in the next section. With regard to example (40), as problematic as it is for the overall timeline I am suggesting, (40) could provide support for my proposal as a shift of scales (temporal to degrees): if (40) is indeed an instance of *noh*comp, then (allowing to some degree for the innovativeness in Southern dialects of German) a process of reanalysis from *noch*add to *noch*comp is arguably even less likely the case.

### Martin Kopf-Giammanco

### **7 Diachronic change: From** *noch***temp to** *noch***comp**

The comparative reading of *noch* is the direct offspring of the original temporal reading of *noch* through a process of reanalysis, from operating on a scale of times to a scale of degrees.

### **7.1 Stage 1 (pre-reanalysis)**

*Noh* has a standard temporal reading of *noch*. There is the presupposition of t\* (a free variable to be bound by context and left-abutting reference time) and a predicate P (a property of times, type ⟨i,t⟩, that holds of reference time t as per the assertion) holds for t\*.

(46) *ther* the *heilant* savor *...* ... *uuas* was *noh* still *thanne* then *in* in *theru* the *steti...* place... 'The savior was still in the place...' (1.OG1.TatianEvHarm.135.18)

Figure 4: LF for (46); cf. (Beck 2016b)

### **7.2 Stage 2**

Let's turn back to example (47) (=42) for the following discussion:

(47) *Thar* There *uuarun* were *mit* with *githuinge* violence *thie* the *iungoron* apostles *noh* still *tho* there *inne,* inside, *sie* they *scolta* should *ruaren* move *NOH* still/even *tho* there *mer* more *thaz* the *selba* same *uuoroltlicha* earthly *ser.* suffering. 'The apostles were still inside with violence, they should continue to move/stir the earthly suffering.' (1.OG2.OtfEbKell.351.12)

### 6 German noch under reanalysis

The *noh* in sentence (47) is ambiguous to a temporal and a comparative interpretation. When excluding the comparative operator *mer* ('more') from the interpretation, the temporal continuative interpretation arises with the predicate ('they move/stir the earthly suffering') being true at reference time and a presupposed earlier time (which can be inferred from the context and is overtly satisfied in previous chapters of the stories – albeit not necessarily in the words of the allegory). In this regard, this is a perfect example since the entire allegory is about the contrast between the earlier time (when Jesus was with the apostles) and the later (reference) time (when Jesus has left the apostles).

The comparative operator in example (47) has the effect of comparing the maximum of a property of degrees (subject of comparison/comparee term) to another maximum of a property of degrees (object/comparison base) – both of type ⟨d,t⟩ – with the standard term of comparison being temporally located before reference time.<sup>11</sup> The two different points in time are provided by the context since the *than*-clause is covert.

The temporal reading of *noh* puts a condition on the context that at an earlier, presupposed time t\* 'the apostles move the earthly suffering' and it asserts that 'the apostles move the earthly suffering' at reference time t, cf. LF in Figure 5 (cf. also Beck 2016b). With the comparative operator having scope over the entire structure, the assertion has to be something like 'the apostles move the earthly suffering more than at an earlier time'. Thus, there is a conflict: on the one hand, the temporal *noch* requires a predicate to be true at an earlier time and at reference time and, on the other hand, the comparative requires that the predicate for reference time and an earlier time differs with regard to degrees.

This type of context represents a critical context, i.e. there is an ambiguity and at the same time one reading fits the context better than the other. In Eckardt's (2011) words, this constitutes a bridging context. In her discussion of reanalysis, she mentions "precarious uses" and notes that the criteria for what constitutes a precarious use are manifold – among other things, they "can challenge the hearer by pragmatic infelicities" (Eckardt 2011: 44).

### **7.3 Stage 3**

The time interval at reference time becomes reanalyzed as the interval of degrees to which the comparee term and the comparison base differ. As a consequence,

<sup>11</sup>I assume gradable predicates here via a degree argument slot in an adverbial phrase, cf. Figure 5. I will not go into details as to whether or not (certain) verbs have a degree argument slot or where the degree argument is originating from; for discussion see Piñón (2008), Rett (2013), Kennedy & McNally (2005), Kennedy (2012) and references therein.

Martin Kopf-Giammanco

Figure 5: LF for *nochtemp* for (47); cf. (Beck 2016b)

the presupposed left-abutting time t\* (for which P(t\*) = 1) becomes reanalyzed as an interval of degrees (on the scale introduced in the matrix clause comparison) to which the comparison base and another, presupposed degree d\* differ. The degree d\* serves as the lower bound of this second interval of degrees. While *noch*temp's t\* is placed at a temporal location lower and relative to reference time (t\*<t), *noch*comp's presupposed d\* is located lower on a scale of degrees relative to the standard term of comparison (d\* < max(D⟨d,t⟩ )). In other words, the interval t\* presupposed by *noch*temp corresponds to the interval of degrees presupposed by *noch*comp, cf. below, (48), and Figure 6 for the post-reanalysis LF. The time variable for reference time is interpreted together with the matrix clause, i.e. in the LF below the comparison, as tense. This is necessary since the *than*-clause will have a different tense due to a different temporal location.<sup>12</sup> Thus, P(t) from the temporal interpretation corresponds to the property of degrees (at present

<sup>12</sup>Data like (i) suggests that aspect and tense need to be interpretable below comparison with both clauses having different tenses and aspect. See also (von Stechow 2006).

<sup>(</sup>i) *This time our guests are staying longer than they stayed last time.*

### 6 German noch under reanalysis

tense) in (47) in the comparative interpretation. It may be argued that the task of pointing to an earlier time is taken over by the comparison which might facilitate for t\* to be reanalyzed as the lower bound of another comparison, i.e. another difference in degrees.

A question that has remained unaddressed is, what happens to the rather strong condition that the presupposed time left-abuts reference time (t\* ∝ t). I argue that it remains intact in the sense that two areas of a scale of degrees are still ordered and adjacent, with the degree of the standard term of comparison being the marker at the boundary between the two different intervals.

Another argument that may be raised is that data like (47) say more about future times (times following reference time), rather than reference time or a preceding time t\* and, therefore, an analysis of diachronic change should take e.g. Beck's (2016b) further-to analysis of *noch*temp as a starting point. Here I argue that only due to the fact that the comparative is present some speakers may get this "forward-directedness". If (47) did not feature a comparative operator, the continuative reading of *noch*temp would give the right predictions and be perfectly satisfied by the context.

### **7.4 Stage 4**

Unambiguous *noch*comp is available as early as OG3 (950–1050). The context for (49) is that God took Jesus to him when he died among the humans. After that the christian church/religion is endowed with glory (since Jesus has lifted all sins from the humans) and Jesus is also endowed with (even more) glory because he sits next to God for eternity. (49) does not license a temporal reading. As with the previous example, there is an antecedent comparison where the degree to which the church has glory is compared to a standard degree of glory.

### Martin Kopf-Giammanco

Figure 6: LF for *nochcomp* for (47); cf. (Beck 2016b)

(49) *Michel* great *ist* is *íro* her *guôllichi* glory *an* in *dînemo* your *haltâre* savior *christo.* christ *[Lat.]* [Lat.] *Ímo* him *selbemo* self *gíbest* give *du* you *noh* still/even *mêrun* more *guôllichi.* glory. *unde* and *mêrun* more *ziêreda.* adornment. *sô* as *dû* you *in* him *gesezzest* set *ad* to\_LAT *dexteram* right\_LAT *tuam.* your\_LAT. 'Great is the glory of the church in your savior christ. You give him

still/even more glory and more adornment by setting him at your right side.' (1.OG3.N:Ps:20.61-63)

In the OG3 subperiod, the modern German interpretation of *noch*comp is fully available. The *no(c)h* in (49) together with its context does not allow a temporal reading any longer. This is in contrast to the earlier examples from OG2.

### **8 Conclusion**

The above aims to contribute to the understanding of the semantics of *noch*comp. I reported on an experiment investigating the PSP component of this use of *noch* and attempted to consolidate the findings with existing contributions to its semantics. Moreover, the experiment has informed the diachronic discussion of *noch*comp in the sense that it provided direction for *noch*'s diachronic trajectory. For this trajectory, I have proposed a process of reanalysis for the comparative reading of *noch* as having developed from the temporal reading of *noch*. The

### 6 German noch under reanalysis

Figure 7: LF for (49); cf. (Beck 2016b)

above proposal is motivated by the ambiguity between temporal and comparative readings of the earliest examples comparative readings of *no(c)h* available. Despite the empirical evidence being rather limited, there are a number of aspects that support the above proposal. Many things need further investigation, among others: the additive use of *noch* and its diachrony need detailed corpus based research in order to (i) better understand when and how it arose and (ii) its possible entanglement with the development of the comparative use of *noch*. Beck's (2016b) discussion of a variety of temporal readings leading up to additivity of *noch* provides a plausible diachronic trajectory for the development *noch*add which, in turn, (iii) requires a more thorough look at the data on *noch*temp. Furthermore, *noch*comp needs investigating in later periods as well.

### **References**


6 German noch under reanalysis


### Martin Kopf-Giammanco

### **Appendix**

Table 10 contains all the condition-3-target items for all 16 token sets. For space constraints I can only include one condition. However, based on Tables 10 and 1 (p. 169) it is straightforward to reconstruct the remaining conditions as Table 11 exemplifies by means of the first token set in line no. 01 in Table 10.

> Table 10: Experimental design; condition-3-target items (fac 1, lev 2 & fac 2, lev 2) for 16 token sets



### Martin Kopf-Giammanco


the t-shirt. And yet the t-shirt is not ugly.'

The following shows the combinatorics behind the compilation of the questionnaires (A – H). The goal was to minimize response fatigue and reduce questionnaire duration. Therefore, I ended up with 8 questionnaires, each containing 8 target items and 16 fillers. The 64 target items were rotated/pseudo-randomized among the questionnaire groups, cf. Table 12, below. This was done to ensure that every participant had to rate 8 items with the conditions: (i) never seeing any token set more than once, (ii) rating every condition twice, (iii) at least one item from every antonymous token set pair (i.e.: token set 1 – *tall* & token set 2 – *short*).



### 6 German noch under reanalysis


Table 12: Combination of token sets and conditions into questionnaires. quest. = questionnaire

## **Chapter 7**

# **Givenness marking in a mixed system: Constituent order vs. determiners**

### Alexandra Simonenko

Research Foundation Flanders & Ghent University

Anna Carlier

Université de Lille

This paper investigates the interaction between constituent order and the use of determiners as means of marking givenness, understood here as a non-presuppositional existential inference that arises as a result of interpreting a predicate with respect to a context-specified situation, in light of (a version of) \*New ≻ Given principle of Kučerová (2012). We attribute the principle to how situation binding operates in clauses, instead of postulating a presupposition-introducing operator and test it on new quantitative data from Medieval French, a system employing both determiners and constituent order for information structuring. Our results show that the constraint in question is respected across the board except for the cases when it is obviated by the presence of a morphological trigger of existential presupposition. We also show that a game-theoretic simulation incorporating this constraint matches very closely historical French data.

### **1 Introduction**

This chapter focuses on the interactions between determiner types and constituent order in the marking of givenness in the history of French, on the basis of data from the twelfth to the seventeenth century. We understand givenness here in a weak sense of an existential inference that emerges when a nominal predicate is interpreted relative to a particular, discourse-specified situation. Medieval French is commonly assumed to have employed syntactic means for the expression of information structure. Because of the absence of native speaker

Alexandra Simonenko & Anna Carlier. 2020. Givenness marking in a mixed system: Constituent order vs. determiners. In Remus Gergel & Jonathan Watkins (eds.), *Quantification and scales in change*, 199–228. Berlin: Language Science Press. DOI:10.5281/zenodo.3929249

### Alexandra Simonenko & Anna Carlier

judgements and speech recordings for historical data, identifying the exact information structural import of syntactic configurations no longer available in Modern French, such as the clause with a preverbal object in (1), is a complex task. However, the consensus states that the variable placement of arguments had mostly to do with the way their denotation related to the contextual information.

(1) *Iceles* these *miracles* miracles *vit* saw *li* the *pelerins.* pilgrim 'The pilgrim saw these miracles.' (1210-BORON-PENN-P,32.441)

Marchello-Nizia (1995), who was among the first to investigate the tendency of Old French to have the verb immediately follow the first constituent (i.e. the so called V2-order, explored in a long series of works starting with Skårup 1975), suggested that the first preverbal position was reserved for elements establishing a link with the previous discourse (for a similar intuition see Vennemann 1974 and Harris 1978). Rinke & Meisel (2009: 117) argue that "the pre-verbal position correlates with a topic-interpretation and the post-verbal position with a non-topic interpretation". Kaiser & Zimmermann (2011: 24) propose that "the positioning of one or more non-subject constituents to the left of the finite verb in declarative root clauses directly correlates with their discourse status, i.e. with their interpretation as either topicalized or focalized constituents". They assume a split CP involving Topic and Focus projections. Based on an extensive corpus data analysis, Labelle & Hirschbühler (2018) conclude that although the distribution is not categorical, the initial constituent in V2 configurations tend to be topical. V2 with non-subject preverbal constituents progressively becomes more rare until the constituent order in declarative sentences effectively converges onto SVO.

At the same time, French already has *le*/*la*/*les* determiners, the frequency of which will increase over the course of history. These determiners have to be analysed as definiteness markers. The two series of phenomena, constituent order and determiners are closely related since both are crucially involved in structuring the propositional content with respect to the background information. The idea that flexible word order and determiners stand in some sort of a "tradeoff" relation with each other has emerged in the literature on historical Romance languages on multiple occasions (e.g. Vincent 1997; Ledgeway 2012).

In this paper, we do not commit ourselves to any claims about causes of word order changes in French. Rather, we assume that word order changes happen for independent prosodic and/or morphological reasons (such as the loss of case

marking). We also, strictly speaking, do not present evidence that there is a causal relation between word order changes and changes in the use of determiners. Rather, we provide supporting arguments for such claims.

Definite, possessive, and demonstrative determiners are assumed in the Frege/ Strawson tradition to function as existential presupposition triggers since their felicitous use requires the background to entail the existence of an individual or entity with certain properties. The English utterance in (2) where the subject DP is headed by a definite determiner is felicitous just in case the existence of a dog in some relevant domain is part of the participants shared knowledge.<sup>1</sup>

(2) *The dog is barking.*

The increase in the frequency of definite determiners is closely followed by the increase in frequency of indefinite determiners, as discussed in Carlier (2013), which signal that a novel referent is being introduced and that a definite determiner could not have been used (Heim 1982; 1991).<sup>2</sup>

Conditions on the use of definite and indefinite determiners in English in some other languages partially correspond to the conditions on argument ordering. For instance, this is the case in Russian (Titov 2012). Consider (3–4), where a clauseinitial argument is likely to be interpreted as denoting an entity whose existence is part of the background information, whereas a postverbal argument is likely to be interpreted as introducing a novel referent.


Kučerová (2012) argues that in Czech object scrambling is a means of aligning the syntactic structure with the (default) Given ≻ New order, where a constituent is considered as given if it has an antecedent in the preceding discourse and if

<sup>1</sup>Even in light of the analyses that deny the definite determiner an existential presupposition, such as Coppock & Beaver (2015), as will be discussed below, it is enough for our purposes that in most cases they give rise to an existential inference as a result of the nominal predicate being interpreted with respect to a contextually provided situation.

<sup>2</sup> Simonenko & Carlier (to appear in the Canadian Journal of Linguistics) give quantitative data on the changes in the determiner system in French over the course of time.

### Alexandra Simonenko & Anna Carlier

the context entails the existence of an entity with the property denoted by this constituent. On this proposal, the two grammatical phenomena related to existential presupposition marking – syntactic and morphological (i.e. via pronominalization or a determiner) – can substitute for each other. The evolutionary trajectory of diachronic French data makes it an ideal test-case for this hypothesis since for several centuries the available texts feature both a very flexible and evidently information structure-driven constituent order *and* emerging definite determiners.

We will claim that medieval French data corroborate (an amended version of) Kučerová's (2012) model which predicts the infelicity of \*New ≻ Given order within a propositional domain. We will show that all bare noun configurations involve Given ≻ New sequence, and that determiners have an obviating effect on this principle in that New ≻ Given is possible if the second argument involves a presupposition-triggering determiner.

Based on this proposal, we build a game-theoretic simulation of the interpretation of a class of utterances and show that the results of the simulation are almost identical to the empirically observed picture in historical French.

We also show that the \*New ≻ Given makes a correct prediction with regard to the relative frequencies of different constituent orders. Finally, we use determiner distribution patterns to identify the information structural import of the OVS order, the only non-marginal configuration involving O ≻ S. We show that objects in OVS occur exceptionally frequently with demonstratives, which we analyse as signalling their status as shifted topics. Our quantitative analysis is based on the treebanks MCVF and Penn supplement to MCVF.<sup>3</sup>

In the next section we discuss Kučerova's model, propose an amendment and lay out the predictions the amended model makes for the historical French data. In Section 3 we show these predictions to be borne out. Section 4 presents our rational speech act model-based simulations and compares its predictions to the historical French data. In Section 5 we discuss the explanatory potential of our model of givenness marking for the constituent order frequencies. Section 6 concludes.

### **2 Marking givenness**

Using Modern Czech as an empirical base, Kučerová (2012) formulates the information flow principle in (5), which states that a constituent interpreted as con-

<sup>3</sup>MCVF and Penn supplement to MCVF with about 1.5 million words is the largest treebank for French diachrony to date.

veying new information cannot precede a constituent interpreted as conveying given information.

(5) Generalization \*New ≻ Given Within a domain [Dom Y ... X], if X is given, so is Y. (Kučerová 2012: 14)

The generalization captures the range of possible interpretations for utterances in terms of the sequences of new and given information. For instance, for (3), it captures the unavailability of the interpretation whereby *chlapec* ('boy') is interpreted as new and *lízátko* ('lillipop') as given. It also captures the fact that if *lízátko* ('lillipop') is made to precede *chlapec* ('boy'), the given interpretation becomes available for the former, as (4) shows.

Kučerová (2012: 18) assumes the notion of givenness as spelled out by Schwarzschild (1999: 151), as in (6).<sup>4</sup>

	- a. if U's type is e, then A and U corefer;
	- b. otherwise, modulo ∃-type shifting, A entails the existential F-closure of U.<sup>5</sup>

Kučerová (2012: 14) derives this generalisation from the mechanism of givenness marking in natural language. Specifically, she proposes that a givenness presupposition can be triggered by a syncategorematic G(ivenness) operator that can be applied anywhere in a propositional domain, dividing the domain into given (higher) and new (lower) parts, as illustrated in Figure 1 from Kučerová (2012: 3).<sup>6</sup>

Kučerová (2012) also assumes that the insertion of such an operator is necessary if the presupposition is satisfied in a given context and if there are no morphological means of marking it in the numeration (in the Minimalist sense). The latter part is based on the maximize presupposition principle of Heim (1991) supplemented with an assumption that the competition takes places between structures generated from the same numeration.

The predicted infelicity of New ≻ Given sequences corresponds to a presupposition failure since the operator G has the effect that all constituents to its

<sup>4</sup>Kučerová (2012: 18) notes that she follows Sauerland 2005 in assuming that "givenness gives rise to an existential presupposition", without spelling out the details".

<sup>5</sup>Existential F-closure involves replacing focused expression by existentially closed variables. 6 For the technical details of the recursive application of the G-operator which introduces restrictions on arguments' domains we refer the reader to Kučerová (2012) and (in even greater detail) Šimík & Wierzba (2015).

Alexandra Simonenko & Anna Carlier

Figure 1: Partitioning of a clause by the Givenness operator

left carry a givenness presupposition. One consequence of this proposal is that configurations where new material linearly precedes old in a given domain are only felicitous if the given material is morphologically marked as such. Kučerová (2012) assumes that once the givenness presupposition is morphologically marked, the G operator is not inserted. Morphological triggers of givenness presupposition involve proper nouns and personal pronouns.

Based on experimental results for Czech, Šimík & Wierzba (2015) replace the \*New ≻ Given principle with a \*Non-presupposed ≻ Presupposed contraint. They note, however, that the constraint is not absolute in that its violation does not result in the same degree of infelicity as the use of *a* instead of *the* in English in a context suitable for the latter.

We build on this version, proposing that the relevant notion is existential nonpresuppositional inference rather than a hard presupposition and that its violation causes a downstep in acceptability rather than strong infelicity. We also add to it an obviation condition that morphological triggers of existential presupposition, such as personal pronouns, proper names, demonstrative, definite, and possessive determiners, are exempt from the constraint. In addition to arguments

that involve morphological markers of existential presupposition, we note that an argument can be exempt from existential opposition altogether, as in the case of bare nouns forming complex predicates with finite verbs. We assume that this applies to idiomatic expressions or light verb constructions such as *avoir nom* ('to be called'), *avoir peur* ('to be afraid'), *avoir cure* ('to need'), *faire mal* ('to hurt'), where the bare noun cannot be interpreted as a referential expression. An example of such construction is given in (7).

(7) *out* had *numobj* name *cilsbj* this *Nabal* Nabal 'this (man) was called Nabal' (1150-QUATRELIVRE-PENN-P,49.1841)

An amended version of the constraint is given in (8) where the relevant notion of givenness is defined as in (9).

(8) Generalisation #New ≻ Given

Within a domain [Dom Y ... X], if X is given, so is Y unless X involves a morphological trigger of existential presupposition or unless the non-presupposed/presupposed opposition does not apply to one of the arguments,

(9) A constituent *C* of an utterance *u* in a context *c* (in Stalnaker's sense) interpreted with respect to a situation *s* is considered given if *c* entails the non-emptiness of the extension of *C* in *s*.

Instead of Kučerova's syncategorematic introducer of domain restrictions (i.e. presupposition trigger) G, we assume that the relevant operator is a situation binder ΣG, which binds situation variables of predicates to a topic situation down to a point where there is another binder. That is, Σ<sup>G</sup> binds all unbound situation variables. Following Kratzer (2017) and Schwarz (2009: 127–133), we assume that topic situations can be derived from questions under discussion (QUD, Roberts 1996; Büring 2003). Specifically, a topic situation is a minimal situation that exemplifies the set of situations in which the answers to QUD are the same as in the actual world.<sup>7</sup>

The relevant existential inference, which constitutes the content of givenness on our account, is an inference that the extensions of predicates interpreted relative to a topic situation are non-empty. The inference arises because the topic

<sup>7</sup>Definitions of exemplification and minimality from Kratzer (2017): A situation *s* exemplifies a proposition *p* iff whenever there is a part of *s* in which *p* is not true, then *s* is a minimal situation in which *p* is true. A situation is a minimal situation in which a proposition *p* is true iff it has no proper parts in which *p* is true.

### Alexandra Simonenko & Anna Carlier

situation is a situation whose contents are likely known to the speech act participants. That the relevant existential inference in question is non-presuppositional on our account matches the conclusion reached in Šimík & Wierzba (2015) that their \*Non-presupposed ≻ Presupposed constraint is relatively mild. If we assume that interpreting a predicate relative to a topic situation gives rise to an existential inference because the speech act participants expect the contents of the topic situation to be known, the inference is cancellable to the extent that this expectation can prove wrong, that is, that in some cases a topic situation does involve entities whose existence is not part of the common ground.

In the absence of any other sources of values for situation variables, this derives, in particular, the #NP1New ≻ NP2Given constraint, since if the situation argument of NP2 is bound, that of NP1 is bound as well (and is therefore interpreted relative to the topic situation, giving rise to an existential inference), just because the binding operators in an across-the-board fashion from top to bottom.

With regard to the rationale behind the obviation conditions, the presence of a determiner introducing its own resource situation pronoun and triggering existential presupposition relative to that situation (as, we assume, possessive, demonstrative, and definite determiners do) makes Σ<sup>G</sup> qua the binding mechanism unnecessary (although the determiner's situation pronoun may co-refer with the topic situation, cf. Schwarz 2009). As a result, a noun phrase interpreted with respect to some other situation may precede a noun phase with a presupposition-triggering determiner. This is illustrated in Figure 2 where stopic is a topic situation pronoun, s<sup>r</sup> is a resource situation pronoun associated with a determiner, -s stands for a situation argument of a nominal predicate, and Σ<sup>G</sup> is the situation binder in question. In this configuration, the two upper NPs are interpreted with respect to the topic situation, whereas the situation argument of the lowest NP is valued by a separate resource situation introduced by a determiner.

### **2.1 Morphological triggers of existential presupposition**

We assume the logical forms (LFs) of demonstrative, definite, and possessive determiners involve a resource situation pronoun which "stops" the binding triggered by ΣG. The LFs and lexical entries for definite and demonstrative determiners are based on Heim (2011) Elbourne (2008), and Schwarz (2009). The entry for possessives is based on Simonenko & Carlier (2019). All these are given in (10–15) for the sake of concreteness.

Figure 2: Givenness operator as a situation binder


Because of the existential presupposition they involve, these entries, if used in a felicitous sentence, give rise to an existential inference. The resource pronoun in their logical forms, in the absence of external binders, does not propagate its Alexandra Simonenko & Anna Carlier

value beyond the local DP, which predicts the felicity of NP1New ≻ NP2Given if the existential inference results from the use of a determiner. In particular, it correctly predicts the felicity of the Czech example in (16), from Kučerová (2012).

(16) *Chlapec* boy.nom *našel* found *ten* this *lízátko.* lollipop.acc 'A boy found this lollipop.'

### **2.2 Predictions**

With respect to its constituent order flexibility, Old French is more similar to Modern Slavic languages than to Modern French. For instance, a transitive clause with a nominal subject and object can have any of the 6 possible constituent orders: SOV, SVO, VSO, OSV, OVS, VOS. Relative frequencies of different orders are given in Table 1.

The counts in the table, which we extracted from MCVF (2010) and Penn supplement to MCVF (Kroch & Santorini 2010), include all finite transitive clauses with nominal subjects and objects.<sup>8</sup> We excluded all cases of pronominalization, first, because of their often restricted syntactic distribution in comparison with nominal arguments and, second, because pronouns either trigger existential presupposition or are explicitly incompatible with it and therefore will not help us evaluate the #New ≻ Given principle.


Table 1: Constituent order in transitive clauses

We are specifically interested in the orderings between subjects and objects. Ignoring verbal position, we give relevant counts and relative frequencies in Table 2.

<sup>8</sup>The relevant corpora are described, in particular, in Martineau (2008) and in Simonenko et al. (2018).


Table 2: Nominal argument order in transitive clauses

The conditional version of the #New ≻ Given principle in (8) makes a number of non-trivial predictions. Specifically, given a transitive clause with overt nominal subject and object, we expect to find the orders in (17) but not in (18), where det stands for a morphological trigger of existential presupposition.

(17) Predicted licit patterns for sequences involving new and old material:


(18) Predicted illicit patterns for sequences involving new and old material:

B1 Snew Ogiven B2 Onew Sgiven

### **3 Testing the predictions**

In the previous section we outlined the predictions made by \*New ≻ Given supplemented with a proviso about the obviating effect of morphological presupposition triggers or arguments to which new/given distinction does not apply, such as incorporated nominals. These predictions are testable in a corpus to the extent that it is representative and that we can approximate infelicity/ungrammaticality of a pattern by the absence thereof in a sufficiently large dataset. Assuming the #New ≻ Given principle, in historical French we expect not to find any patterns where an argument denoting new information linearly precedes an argument denoting given information, unless the second argument features a morphological presupposition trigger or one of the arguments is exempt from new/given

### Alexandra Simonenko & Anna Carlier

opposition. Below we report the results of querying for the licit and illicit configurations listed above.<sup>9</sup>

### **3.1 SO**

We first check all the SO clauses which can potentially contain an illicit string B1, that is, finite clauses where both subject and object are bare nouns. There are 283 such clauses in the corpus. Examples (19–23) illustrate SO strings involving arguments without morphological triggers of existential presupposition.


In the corpus we found no SO strings violating \*New ≻ Given (i.e. pattern B1 in 18).

On the basis of (19–23), one could argue that the absence of New ≻ Given sequences is an epiphenomenon of the sample involving bare nouns only. Namely, it is theoretically conceivable that in Medieval French all noun phrases without

<sup>9</sup>We considered transitive clauses with nominal arguments that are not preceded by any of the following: definite, demonstrative, indefinite, possessive or partitive determiner.

<sup>10</sup>(1155-ENEAS1-BFM-R,87.1898)

<sup>11</sup>(1183-ADGAR-BFM-R,265.3473)

<sup>12</sup>(1185-COBE-BFM-R,3.28)

<sup>13</sup>(1173-BECKET-BFM-R,74.2014)

<sup>14</sup>(1128-BESTIAIRE-BFM-R,43.564)

determiners either denote abstract notions (as in 19, 20), receive kind interpretation (as in 23) or are used in generic statements (as in 20 and 21). We assume that these interpretations inherently involve an existential presupposition.<sup>15</sup> This means that SO strings involving only those cannot in principle feature New ≻ Given sequence and therefore such a sample cannot be used to evaluate the relevant predictions. However, although there are indeed many bare arguments involving abstract, kind denoting, and generic NPs, there are also cases of SO with bare arguments denoting individuals, as (24–26) illustrate.

In (24) the subject *Osbercs e helmes* ('hauberks and helmets') is part of the given information, that is, the context entails that the exists individual with the relevant properties in a prominent situation.


In (25) the subject is a specific indefinite (the narrator is talking here about Saint Mary who restores humanity to life through Jesus Christ and about Eve, who brings death through sin), and the object is arguably indefinite as well (to be understood as 'a new life').

(25) *Fame<sup>S</sup>* woman *vie<sup>O</sup>* life *nous* to.us *restora* restores 'A woman restored us to life.'<sup>17</sup> A1

Example 26 speaks about Christians in some prominent situation.

(26) *De* from *vus* you *unt* have *crestiens<sup>S</sup>* Christians *cumfortO.* comfort 'Christians receive comfort from you.'<sup>18</sup> A1

Examples in (27–30) are cases of obviation where the presence of a morphological presupposition trigger arguably obviates the #New ≻ Given principle, as predicted in (17).

<sup>15</sup>We assume that generic statements involve an implicit quantifier over situations coupled with a presupposition of the non-emptyness of its domain, that is, that there exist situations accessible from the evaluation situation in which there are individuals with the nominal property (Lee 1995; von Fintel 1996).

<sup>16</sup>(1100-ROLAND-V,137.1820)

<sup>17</sup>(1190-BORON-PENN-R,27.431)

<sup>18</sup>(1183-ADGAR-BFM-R,181.1981)

Alexandra Simonenko & Anna Carlier


### **3.2 OS**

There are 37 finite clauses with OS order where both subject and object are nouns without determiners (but possibly with quantifiers or modifiers) in the corpus we used. A configuration involving Object ≻ Subject violates \*New ≻ Given if the Object is new and Subject is given. The example in (31) may look like a potential violation of #New ≻ Given as the object *duel* 'sorrow' precedes a clearly given subject *pere et mere* 'father and mother'.


However, *duel* is mentioned just a couple clauses before and, hence, cannot be considered as a evoking a new referent. The relevant clause is given in (36).

<sup>19</sup>(1150-QUATRELIVRE-PENN-P,5.47)

<sup>20</sup>(ID 1200-SERMMADN-BFM-P,18.136)

<sup>21</sup>(ID 1100-ROLAND-V,112.1445)

<sup>22</sup>(1376-FROISSART-7-P,763.1956)

<sup>23</sup>(1155-ENEAS2-BFM-R,12.199)


Like *avoir faim* ('to be hungry', literally, 'to have hunger'), *avoir deuil* ('to suffer') is a light verb construction combining a copula and a noun that denotes an event or a state. We assume that for these cases the notion of givenness or existential presupposition is undefined and, consequently, the principle does not apply.

Another example of OS with bare arguments is given in (33), where both NPs are given.

(33) *Force* force *de* of *deïté<sup>O</sup>* divinity *Demustre* shows *piz quaré<sup>S</sup>* forequarters 'The forequarters (of a lion) symbolize the divine power.'<sup>25</sup> A2

We find the same configuration in (34), where the denotations of both the subject and the object belong to the given information since the relevant passage describes an army being set in motion.

(34) *E* and *destendent* take.down *acubes<sup>O</sup>* tentes *serjant* sergeants *e* and *escuier<sup>S</sup> .* esquires 'And the sergeants and esquires take down the tents.'<sup>26</sup>

The #New ≻ Given principle predicts an obviation for OS strings where a new object precedes a given subject with a presuppositional determiner (pattern A4 in 17). This case is illustrated by in (31–33).


<sup>24</sup>(ID 1155-ENEAS2-BFM-R,12.197))

<sup>25</sup>(1128-BESTIAIRE-BFM-R,3.31)

<sup>26</sup>(1175-FANTOSME-BFM-R,48.514)

<sup>27</sup>(1100-ROLAND-V,112.1449)

<sup>28</sup>(1120-BRENDAN-R,59.776)

Alexandra Simonenko & Anna Carlier

(37) *Onbre<sup>O</sup>* shadow *li* him *fet* did *li* det *plus* most *biax* beautiful *arbres<sup>S</sup>* tree *c'* that *onques* ever *poïst* could *former* form *Nature.* nature 'That most beautiful tree that that the Nature coud form gave him shadow.'<sup>29</sup> A4

As an interim conclusion, in a transitive clauses with bare arguments we found no cases violating #New ≻ Given (with obviation conditions), that is, involving an argument associated with given information following an argument not associated with given information. These results are to be only taken as suggestive since the absence of a pattern in a limited sample cannot be straightforwardly interpreted as signalling ungrammaticality. However, the fact that among 320 clauses with bare arguments there are no instances violating #New ≻ Given is likely not a matter of chance. That is, we found no patterns B1 and B2. To test this, we compared the number New ≻ Given information states among SO with bare argument with the number of such information states among SO strings where the object has a morphological presupposition trigger (which makes them exempt from the #New ≻ Given constraint). As Table 3 shows, there are 69 instances of such information state, which means that a non-given constituent preceding a given one is not a very rare information state in general.

Table 3: Rate of New ≻ Given in finite clauses


### **4 Simulating a mixed system: Rational speech act model**

The information flow principle in (8) relates constituent order and morphological existential presupposition triggers as alternative markers of givenness in a type of a tradeoff relation. If a determiner is used, then the order of NPs does not matter for new/given encoding, and, conversely, if NPs meant to be interpreted as given precede NPs meant to be interpreted as new, determiners need not be used.

<sup>29</sup>(1170-YVAIN-R,12.383)

In terms of how they convey information structure, however, syntactic and morphological are not equivalent in so far as a constituent order NP<sup>1</sup> NP<sup>2</sup> , incompatible with New ≻ Given interpretation, is compatible with Given ≻ New, Given ≻ Given, and New ≻New interpretations, whereas a sequence NP<sup>1</sup> detNP<sup>2</sup> is compatible with fewer information states: New ≻ Given and Given ≻ Given. Assuming that language users are aware of this, we can try to simulate the use of the two types of markers in a mixed system and compare the results with the quantitative data from diachronic French.

To do the simulations we use rational speech act model (RSA, Frank & Goodman 2012). RSA assumes Bayesian reasoning on the part of the speech act participants. Specifically, the beliefs of the Speaker and Listener are represented as probabilities they associate with different states of affairs. Probabilities that the Listener has before an act of communication are called prior probabilities. An utterance used in an act of communication is considered to be data that allow the Listener to update their knowledge by inferring posterior probabilities of the states of affairs. Interpretation (or probability update) at the so called literal listener level, is based solely on the literal meaning of the utterances (a pre-set relation between utterances and states of affairs). Then at the so called pragmatic speaker level the model takes into account the properties of the literal listener and an assumption that the speaker wants to maximize their chances to be understood (for the listener to infer from the utterances the state of affairs that the speaker means). It is this level that we use in our model of interaction of constituent order and morphological presupposition triggers.

In our simulation, we assume the states of affairs as in Table 4 and possible utterances and correspondences between the two (literal meanings) as in Table 5. We use det here as a cover label for definite, demonstrative, possessive, and partitive determiners.

Table 4: States in RSA simulation

state Given ≻ New New ≻ Given Given ≻ Given New ≻ New

For the moment we assume that a priori all states of affairs and all utterances are equally likely. This means that, for instance, upon hearing "S O" a Literal Listener will end up with a uniform probability distribution over the states Given ≻ Given, Given ≻ New, and New ≻ New, as illustrated in Figure 3.

A Pragmatic Speaker model generates inferences about what constituent orders a speaker is likely to use in order to convey a certain target information


Table 5: Literal meaning: utterances and corresponding information states

Figure 3: Literal Listener with uniform priors interpreting "S O"

state given the assumptions of the Literal Listener and with the goal of maximizing the chances for the target state to be recovered. Figures 4–6 illustrate the inferences of the Pragmatic Speaker with regard to the Given ≻ Given, Given ≻ New, and New ≻ New states, respectively. For instance, the model predicts that in order to convey Given ≻ Given a speaker is most likely to use the order "det S det O" or "det O det S". If we look again at Table 5, we will see that among all the seven configurations eligible to convey a given Given ≻ Given information flow, these two are the least ambiguous in that they are associated with only one information state.

Figure 4: Pragmatic Speaker with uniform priors conveying Given ≻ Given

The New ≻ Given state can only be conveyed by one configuration, S det O (O det S order is not attested in the corpus and therefore is not part of our model).

A model of a Pragmatic Listener involves inferences with respect to the performance of a Pragmatic Speaker. That is, given a particular constituent order, this model makes inferences about most likely interpretations. Inferences for the "S O" configuration are shown in Figure 7.

We see that a Pragmatic Listener model predicts that "S O" configuration is most likely interpreted as conveying a New ≻ New information state. As Figures 4–6 show, to convey Given ≻ Given or Given ≻ New, there are better candidates than "S O", namely, "det S det O" or "det O det S" and "det S O" or "det O S", respectively. That the model predicts "S O" to be most likely interpreted as New ≻ New corresponds to our intuition that the listener expects the speaker to use "det S det O" or "det O det S" and "det S O" or "det O S" for conveying the two other possible states. Since the Speaker visibly did not use either of those, the most likely interpretation is New ≻ New, for which there is no better option than "S O".

Figure 5: Pragmatic Speaker with uniform priors conveying Given ≻ New

Figure 6: Pragmatic Speaker with uniform priors conveying New ≻ New

Figure 7: Pragmatic Listener with uniform priors interpreting "S O"

Let us now see how the Pragmatic Listener model fares compared to the historical French data. We classified all bare noun "S O" configurations in the corpus as New ≻ New, New ≻ Given or Given ≻ New (recall that we did not find any bare noun "S O" corresponding to New ≻ Given information state). Figure 8 shows the distribution of information states among "S O" sequences in the corpus.

Comparing these results with the predictions of our RSA Pragmatic Listener model in Figure 7, we see that in the actual data the frequency of Given ≻ New is higher than predicted, while the frequency of New ≻ New is lower.

Now, in our model we assumed that apriori all information states are equally likely (they had uniform priors). This is, however, most likely not the case (see Birner 2012 for references). We therefore need to make our information state priors more realistic. In order to do that, we used data from a syntactically annotated subcorpus of the Russian National Corpus, *Russian National Corpus* (2019). We classified 430 Russian transitive sentences with bare (i.e. without demonstrative or possessive determiners) nominal arguments (both "S O" and "O S") according to their information state. The obtained distribution is plotted in Figure 9.

We used these frequencies to set the priors for the information states in our RSA model. That is, instead of assuming that information states Given ≻ Given,

Figure 8: Distribution of informations states among "S O" in French, X-XVI c.

Figure 9: Distribution of informations states among transitive clauses with bare nominal arguments in the Russian National Corpus

Given ≻ New, New ≻ Given, and New ≻ New are equally likely, we set their probabilities to 0.35, 0.46, 0.02, and 0.16, respectively. We then reran our Pragmatic Listener model, which now generates inferences for interpreting "S O" configuration as in Figure 10, where it is plotted against the historical French data.

Figure 10: Distribution of informations states among "S O" as predicted by Pragmatic Listener model and in historical French corpus

As the figure shows, the shapes of the two distributions are remarkably similar, which means that our RSA Pragmatic Listener is a successful simulation of the pragmatic reasoning behind the historical French data.

The core assumptions of the simulation are encoded in the morphological entries in Table 5, where the strings with presupposition-triggering determiners are less ambiguous than strings with bare NPs only and where New ≻ Given information state cannot be conveyed by utterances involving bare nouns. This has the effect of predicting that, first, whenever there is a choice, speakers will be more likely to use strings with determiners than strings without, as this maximizes their chance to be understood (see Figures 4 and 5), and, second, that pragmatically reasoning listeners will tend to interpret bare nouns as conveying information states which could not have been conveyed using presupposition-triggering determiners, such as New ≻ New (see Figure 7). The simulation results match historical French data very closely, while they contrast with the data we took from the Russian National Corpus where Given ≻ Given is the second frequent information state of a transitive clause with bare nouns (see Figure 9). We suggest

that the difference is due precisely to the lack of definite determiners in Russian, which means that there is no better alternative for conveying Given ≻ Given than bare NPs, while in French "det S det O" is the best option (see Figure 4).

### **5 Givenness marking and constituent order frequencies**

In this section, we explore a connection between givenness marking and constituent order frequencies in historical French. Let us take another look at the constituent order distribution in Table 1. Orders involving O ≻ S are markedly more rare than those with S ≻ O, with one exception, namely, the OVS configuration. First, we suggest that the rarity of O ≻ S in medieval French, and thus the rarity of OSV and VOS, is a consequence of #New ≻ Given on the assumption that subjects denote given information more frequently than objects. This assumption can be tested, at least at a first approximation, by looking at the distribution of determiners with subjects and objects. The rates of definite and possessive determiners and demonstratives with subjects and objects will be indicative of their respective tendencies to be associated with existential presupposition. Figure 11 shows the determiner distribution with subjects and objects per century.

Figure 11: Determiner distribution: subjects vs. objects

Based on this approximation, we can estimate that during all periods subjects are at least 2–3 times more likely than objects to occur with a definite, possessive or demonstrative determiner, which indicates that subjects are much more likely to satisfy the conditions on the use of presupposition-triggering determiners, namely, to denote an individual whose existence is entailed by the common ground.

Extrapolating this conclusion onto clauses with bare arguments, we expect that subject noun phrases denote properties whose extension is entailed by the common ground to be non-empty much more frequently than objects. This, in turn, means that the order S ≻ O is expected to align with the (licit) information state Given ≻ New much more frequently than the order O ≻ S. We suggest that this is at least in part responsible for the very low frequency of OSV and VOS orders in historical French.<sup>30</sup>

We also observe that OVS order is more frequent than OSV and VOS. We suggest that OVS corresponds to a configuration of topic (situation) shift, where the preverbal position is associated with prosodic prominence. To probe into the properties of OVS, in Figure 12 we plotted distributions of determiners in the object position in finite clauses with different constituent order. We take all the clauses with nominal objects and any type of subject (i.e. either nominal or pronominal or null).

Figure 12: Determiner distribution in object position

Excluding from consideration numerically marginal (see Table 1) and therefore highly erratic OSV and VOS patterns, we observe a similarity between object determiner distributions in SOV, SVO, and VSO configurations.

<sup>30</sup>According to Dryer (2013), in a sample of 1188 languages where a dominant constituent order can be established, there are only 40 languages (or ≈ 3%) where the dominant order involves O ≻ S.

### Alexandra Simonenko & Anna Carlier

OVS stands out by an exceptionally high proportion of demonstratives in object position. To understand what that means for the status of OVS, let us consider the role of demonstratives in the information structure in modern languages.

The most notable feature of demonstratives is the requirement to have an antecedent (or an element in the extralinguistic reality serving as a referent). This property has been captured by assuming a silent individual pronoun in the structure of demonstrative phrases (Nunberg 1993; Elbourne 2008). On the view that pronominals are variables which get their value based on a context-determined mapping, in order for such structure to be interpretable, the context must involve a salient individual to which the assignment function will map the pronominal index.

Another potentially relevant fact is that the antecedent of a demonstrative is normally available in the immediately preceding context. According to Zulaica-Hernández & Gutiérrez-Rexach (2011: 180), in Spanish, 80% of demonstratives have their antecedents in the immediately preceding utterance. Stevens & Light (2013: 204) report that in English 78.08% of demonstratives have antecedents that are discourse-new in the context immediately preceding the relevant demonstrative.

In addition, demonstratives, in contrast to definite determiners, are characterized by the requirement that the nominal predicate do not denote a singleton (relative to a certain domain, Corblin 1987). This is illustrated by the infelicity of (38) and (39) with demonstratives in the contexts implying uniqueness and by their felicity in contexts involving more than one individual with the relevant nominal property, as in (40).


These three facts mean that an object noun phrase with a demonstrative requires an immediately preceding antecedent and that it also requires that the extension of the nominal predicate in the relevant situation do not correspond to a unique entity. An antecedent for a demonstrative must introduces a new entity, since an entity which had been introduced before would normally be realized as a pronoun or a noun phrase with a definite determiner, which is incompatible with the non-uniqueness requirement. In this respect, consider examples in (41) and (42).


Based on such considerations, Bosch et al. (2003) formulate a complementary hypothesis, which states that personal pronouns pick up discourse topics as referents, while demonstratives prefer non-topical referents. Furthermore, Zulaica-Hernández & Gutiérrez-Rexach (2011: 175) argues for Spanish that "speakers use demonstratives to mark topic or subtopic shifts". We thus conclude that the high rate of demonstratives with objects in OVS configurations indicates that the preverbal position was frequently used to indicate a shift in topic situation.

### **6 Conclusions**

In this paper we explored organization of the information in a clause in light of Kučerová (2012)'s proposal that givenness, when not expressed by dedicated morphemes, is monotonically marked from left to right. We proposed an amended version of the constraint that involves a non-presuppositional existential inference as the definition of givenness and tested it on a historical corpus of French, since the historical stages of French have both syntactic and morphological means of marking givenness. Our results show that the principle is borne out in the French historical data: in a corpus of 1.5 million words, we did not find any cases of New ≻ Given in clauses with bare arguments. The data also bear out the prediction that in case Given is marked morphologically, the left-to-right monotonicity requirement does not apply.

We also built the principle into a game-theoretic simulation of the use of constituent order and presupposition-triggering determiners to convey an information state. The results of our simulation come very close to the empirical historical French data, suggesting that it is viable component of a model of pragmatic language use.

Finally, we also showed that \*New ≻ Given may provide insight into relative frequencies of various constituent orders.

### **References**

Birner, Betty. 2012. *Introduction to pragmatics*. Vol. 38. John Wiley & Sons.

Bosch, Peter, Tom Rozario & Yufan Zhao. 2003. Demonstrative pronouns and personal pronouns. German *der* vs. *er*. In *In proceedings of the eacl 2003. Budapest. workshop on the computational treatment of anaphora*, 61–68.

Alexandra Simonenko & Anna Carlier



### Alexandra Simonenko & Anna Carlier

possessives. Slides for a talk given at the workshop *Sorting out Definiteness* at DGfS 2019. University of Bremen. 6–8/03/2019.


Agbayani, Brian, 50 Ahn, Dorothy, 51 Aikhenvald, Alexandra Y., 122, 123 Aloni, Maria, 135, 147 Alonso-Ovalle, Luis, 132, 135, 145 AnderBois, Scott, 104 Aranovich, Raúl, 38 Asher, Nicholas, 41, 54 Assaf, Simha, 13 Bar-Lev, Moshe E., 2, 24 Bates, Douglas M., 115, 173 Bauer, Laurie, 68 Beaver, David, 201 Beck, Sigrid, vi, vii, 1, 3, 21, 25, 51, 161, 163, 177, 178, 185–191 Becker, Martin G., 119 Bernini, Giuliano, 44 Bianchi, Valentina, 52 Biberauer, Theresa, 50 Birner, Betty, 219 Bochnak, M. Ryan, 74, 79, 80 Bondaruk, Anna, 88 Bosch, Peter, 225 Bošković, Željko, 3 Brisson, Christine, 12 Bücking, Sebastian, 118, 119, 121 Büring, Daniel, 205 Burnett, Heather, vii, 69–81 Błaszczak, Joanna, 94

Campbell, Lyle, 34

Carlier, Anne, 201, 206 Chemla, Emmanuel, 69 Chierchia, Gennaro, 25, 57, 58 Choi, Jinyoung, 145 Cinque, Guglielmo, 50, 95, 145 Cobreros, Pablo, vii, 67, 74–76 Collins, Chris, 56 Company Company, Concepción, 156 Condoravdi, Cleo, 144, 145 Coppock, Elizabeth, 201 Corblin, Francis, 224 Csipak, Eva, 74, 79, 80 Curnow, Timothy Jowan, 99 Dalrymple, Mary, 17 Danon, Gabi, 2 Dasher, Richard B., 56, 57 Davies, Mark, 68, 70, 133 Davis, Henry, 3 Dayal, Veneeta, 24 de Cuba, Carlos, 91 de Vaan, Michiel, 36 de Vries, Hanna, 11 den Dikken, Marcel, 52 Deo, Ashwini, 34 Dixon, Robert M. W., 68, 70 Doetjes, Jenny, 42 Donhauser, Karin, 179 Döring, Sophia, 97 Doron, Edit, 2, 3, 5, 10, 25 Dowty, David, 8, 12

Dryer, Matthew S., 223 Dziubalska-Kołaczyk, Katarzyna, 107 Eckardt, Regine, 34, 57, 187 Elbourne, Paul, 206, 224 Fabricius-Hansen, Cathrine, 119 Faller, Martina, 87, 104, 105, 119, 121 Fox, Danny, 24, 50, 53 Francez, Itamar, 2 Franco, Irene, 39 Frank, Michael, 215 Friedman, Victor, 123 Fruyt, Michèle, 36 Garzonio, Jacopo, 42 Gast, Volker, 55 Gergel, Remus, vi, 161 Gianollo, Chiara, 34, 36, 38, 39, 41, 42, 46, 49, 57, 58 Giorgi, Alessandra, 51 Gippert, Jost, 179 Glinert, Lewis, 2 Goldring, Katja, 2 Golston, Chris, 50 Goodman, Noah, 215 Grimm, Scott, 121 Gust, Helmar, 118 Gutiérrez-Rexach, Javier, 224, 225 Haan, Ferdinand de, 121, 122 Hacquard, Valentine, 98 Haegeman, Liliane, 91 Hansen, Nat, 69 Harris, Martin, 200 Haspelmath, Martin, vi, 1, 2, 44, 47, 57 Heim, Irene, 5, 201, 203, 206 Heine, Bernd, 46

Hendriks, Petra, 52 Hirschbühler, Paul, 200 Hofstetter, Stefan, 162, 163, 167–171, 176 Ibarretxe-Antuñano, Iraide, 89 Ippolito, Michela, 161 Jalava, Lotta, 123 Jasinskaja, Katja, 51, 52, 60 Jędrzejowski, Łukasz, 86, 104, 124 Jelinek, Eloise, 3 Jespersen, Otto, 54 Johé, Jonas, 172 Joüon, Paul, 7 Kadmon, Nirit, 25, 167 Kaiser, Georg, 200 Karttunen, Lauri, 90 Katzir, Roni, 50, 53 Kauf, Carina, 93 Kennedy, Christopher, 69–71, 187 Kiparsky, Paul, 40 Klein, Ewan, 75 Klemensiewicz, Zenon, 107 König, Ekkehard, 47, 53, 55, 161, 163– 166 Kopf-Giammanco, Martin, 161 Koring, Los, 121 Kratzer, Angelika, 24, 87, 104, 105, 205 Krause, Thomas, 179, 182 Krifka, Manfred, 11, 12, 17 Križ, Manuel, 3, 13, 15 Kroch, Anthony, 68, 208 Kuzmack, Stephanie, 68 Kuznetsova, Alexandra, 175 Kučerová, Ivona, vi, 199, 201–204, 208, 225

Labelle, Marie, 200 Lahiri, Utpal, 57, 58 Landman, Fred, 25 Lasersohn, Peter, 12, 67, 75 Lauer, Sven, 12 Lazard, Gilbert, 122 Ledgeway, Adam, 200 Lee, Ik-Hwan, 211 Lehmann, Christian, 34, 59 Lehmann, Hans-Martin, 68 Lewis, David, 71, 119 Light, Caitlin, 224 Link, Godehard, 20 Löbner, Sebastian, 13, 161 Łojasiewicz, Anna, 92, 93, 117 Malamud, Sophia, 17 Marchand, Hans, 68 Marchello-Nizia, Christiane, 200 Margulis, Daniel, 2 Martineau, France, 208 Masloh, Julia, 161 Matthewson, Lisa, 3, 87, 104 McNally, Louise, 69–71, 187 McSparran, Frances, 68 Meir, Irit, 3, 25 Meisel, Jürgen M., 200 Menéndez-Benito, Paula, 25,132,135, 145 Migdalski, Krzysztof, 88, 108 Miller-Naudé, Cynthia, 21 Mitrović, Moreno, 52 Mittwoch, Anita, 2 Moshavi, Adina, 16 Moulton, Keir, 91, 106 Munn, Alan, 52 Murray, Sarah E., 104, 106 Naudé, Jacobus A., 11, 21

Nunberg, Geoffrey, 224 Orlandini, Anna, 36, 37, 45, 46 Penka, Doris, 58 Percus, Orin, 93 Petrova, Svetlana, 183 Pinkster, Harm, 44 Piñón, Christopher, 187 Plag, Ingo, 74 Poccetti, Paolo, 36, 37, 45 Poletto, Cecilia, 51 Portner, Paul, 119 Proffitt, Michael, 67 Ramat, Paolo, 44 Reis, Marga, 90, 121 Remberger, Eva-Maria, 119 Rett, Jessica, 187 Rinke, Esther, 200 Rivero, Maria, 137 Rizzi, Luigi, 88 Roberts, Craige, 141, 205, 224 Roberts, Ian, 34 Romero, Maribel, 145 Rooth, Mats, 56 Rothstein, Susan, 16 Rotstein, Carmen, 71 Roussou, Anna, 34 Rubin, Aaron D., 3 Ruys, Eddy G., 5 Sæbø, Kjell Johan, 119 Santorini, Beatrice, 208 Sassoon, Galit, 71 Sauerland, Uli, 52, 93, 203 Schenner, Matthias, 102, 104, 124 Schwarz, Florian, 12, 205, 206 Schwarzschild, Roger, 5, 203 Schwenter, Scott A., 52, 57

Selkirk, Lisa, 5 Shanon, Benny, 93 Sharvy, Richard, 4 Shimoyama, Junko, 24 Shlonsky, Ur, 21 Šimík, Radek, 203, 204, 206 Simonenko, Alexandra, 201, 206, 208 Skårup, P., 200 Socka, Anna, 86 Sode, Frank, 119 Spector, Benjamin, 17 Stalnaker, Robert, 119 Stępień, Marzena, 86 Stevens, Jon, 224 Sweetser, Eve, 89 Szabolcsi, Anna, 52 Szczegielniak, Adam, 88 Taborek, Janusz, 91, 117 Tajsner, Przemysław, 88 Taylor, Ann, 68 Titov, Elena, 201 Toledo, Assaf, 71 Tomaszewicz, Barbara, 88 Tomić, Olga Mišeska, 108 Tovena, Lucia, 56, 57 Traugott, Elizabeth C., 56, 57, 67, 68, 156 Tredinnick, Victoria, 145 Trousdale, Graeme, 67, 68 Truckenbrodt, Hubert, 119 Umbach, Carla, 57, 118, 161, 163–171, 173, 176, 178, 179, 185 Unger, Peter, 71 Ürögdi, Barbara, 91 van der Auwera, Johan, 55 van der Sandt, Rob A., 165

van Gelderen, Elly, vi, 1, 34 Vennemann, Theo, 200 Vincent, Nigel, 200 von Fintel, Kai, vi, 3, 119, 144, 145, 157, 211 von Stechow, Arnim, 21, 69, 178, 179, 188 Walczak, Bogdan, 107 Waltereit, Richard, 52, 57 Watanabe, Akira, 57 Watkins, Jonathan, 161 Wellwood, Alexis, 98 Wiemer, Björn, 86, 92, 93, 117 Wierzba, Marta, 203, 204, 206 Willett, Thomas, 122 Willim, Ewa, 88 Willis, David, 108, 122 Winter, Bodo, 175 Winter, Yoad, 8, 11, 12, 23, 71 Witkoś, Jacek, 88 Wolter, Lynsey Kay, 224 Wurmbrand, Susi, 52, 54 Yoon, Youngeun, 70 Żabowska, Magdalena, 86 Zamparelli, Roberto, 52 Zanuttini, Raffaella, 42 Zeevat, Henk, 51, 52, 56, 60 Zeijlstra, Hedde, 49, 58 Zeldes, Amir, 179, 182 Zimmermann, Malte, 25, 97 Zimmermann, Michael, 200 Zulaica-Hernández, Iker, 224, 225

# Did you like this book?

This book was brought to you for free

Please help us in providing free access to linguistic research worldwide. Visit http://www.langsci-press.org/donate to provide financial support or register as a community proofreader or typesetter at http://www.langsci-press.org/register.

# Quantification and scales in change

This volume contains thematic papers on semantic change which emerged from the second edition of Formal Diachronic Semantics held at Saarland University. Its authorship ranges from established scholars in the field of language change to advanced PhD students whose contributions have equally qualified and have been selected after a two-step peer-review process.

The key foci are variablity and diachronic trajectories in scale structures and quantification, but readers will also find a variety of further (and clearly non-disjoint) issues covered including reference, modality, givenness, presuppositions, alternatives in language change, temporality, epistemic indefiniteness, as well as – in more general terms – the interfaces of semantics with syntax, pragmatics and morphology.

Given the nature of the field, the contributions are primarily based on original corpus studies (in one case also on synchronic experimental data) and present a series of new findings and theoretical analyses of several languages, first and foremost from the Germanic, Romance, and Slavic subbranches of Indo-European (English, French, German, Italian, Polish, Spanish) and from Semitic (with an analysis of universal quantification in Biblical Hebrew).