Is Negation Negative? (And a Discussion of Negative Concord in SOV Languages)

Abstract

Is negation negative? For some authors, in some languages, it is not. This is the case for so-called strict negative concord languages (e.g., Russian), in which negation is taken to be non-negative, following the cross-linguistic analysis for negative concord systems proposed by Hedde Zeijlstra’s work “Sentential negation and negative concord”. However, this analysis is focused on languages with SVO word order. In this paper, I propose to reconsider the typology of negative concord by zooming out of the focus on SVO languages that current literature has relied on. I discuss the case of SOV languages where observing a strict NC pattern leads to weaker conclusions about the nature of negation than for SVO languages with strict negative concord, leaving the negativity status of negation in those languages underdetermined. I then take a look at Turkish, an SOV language with three sentential negation markers: plain sentential negation -mA, copular negation değil, and existential negation yok. Evidence from the interaction of these markers with neither..nor phrases suggests that değil and yok, in contrast with -mA, are non-negative for some speakers. In order to explain the variation, I put forward a hypothesis about the learning process, in which there is sometimes insufficient evidence in the input to determine whether değil and yok are negative, and learners choose between two conflicting heuristics that result in the negativity or non-negativity of these markers.

1. Introduction

In negative concord (NC) languages, like Turkish, we observe sentences that contain several negative elements that give rise to an interpretation with only one semantic negation.

(1)	Hiçkimse gel  -*(me)  -di.
	nobody  come -*(neg) -past
	Nobody came.

If there is only one semantic negation, which of the several negative elements is it the meaning of, if any? A popular class of analyses of NC explains it to be a result of syntactic agreement between negative elements (Deal, 2022; Penka, 2011; Zeijlstra, 2004, 2008), which allows there to be only one interpreted negation even in the presence of several negative elements. As seminally proposed by Zeijlstra (2004, 2008), NC is an Agree operation that occurs between a negative operator with interpretable negative features, and one or several negative concord items (NCIs), which carry uninterpretable negative features. Typically, negative indefinites like hiçkimse (‘nobody’) are considered NCIs, which means they are assigned non-negative semantics and correspondingly carry uninterpretable negative features. These need to be satisfied by a c-commanding negative operator with matching interpretable features present in a local domain of agreement. Given that sentential negation is (almost) always present in a negative sentence, and that the negative indefinites have uninterpretable negative features and no negative semantics, it is tempting to view the sentential negation marker like Turkish -mA as the carrier of interpretable features and the locus of semantic negation. However, this is not always what ends up being claimed.

In particular, Zeijlstra (2004) proposes that some NC languages have a negative negation and others do not (I use the term ‘negation’ to refer to a sentential negation morpheme like Turkish -mA or English not, and ‘negative’ to qualify an element whose semantics is negative and, in all the cases we discuss in this paper, carries interpretable negative features). Whether negation is negative or not affects the surface NC pattern. Negative negation is proposed for ‘non-strict’ NC languages, where it licenses all NCIs in its c-command domain (i.e., post-verbal ones), but not those outside of it (i.e., preverbal ones), which are licensed by a last resort covert negation (see (2)-(3)). Non-negative negation is proposed for ‘strict’ NC languages, because it is observed to appear with all NCIs, even those outside its scope, and therefore, it cannot be responsible for licensing those (see (4)-(5)).

However, Zeijlstra’s analysis of this typology is implicitly restricted to languages with Subject–Verb–Object word order. It turns out that it is misleading to assign the labels of ‘strict’ and ‘non-strict’ to Subject–Object–Verb languages, and inappropriate to conclude the negativity status of the negation marker based on Zeijlstra’s conclusions for SVO languages. The reason is simple: in SVO languages, ‘preverbal’ directly implies a superior hierarchical relation to the verb (and a vP-level negation marker), whereas in SOV languages, it does not. As a result, we cannot conclude anything about whether the negation marker in SOV languages can license preverbal NCIs from surface facts.

In this paper, we begin with a discussion of Zeijlstra’s NC system extended to SOV languages and a discussion of what conditions must be met to be able to conclude whether negation is negative or not. We then look at the NC SOV language Turkish as a case study. We conclude that evidence from word order and arguments from previous work are compatible with Turkish negation markers being negative, but not sufficient to unilaterally claim that they must be.

I then put forth a novel diagnostic to probe the negativity of a sentential negation marker: whether it can result in a single negation reading in the scope of a ne..ne (neither..nor) phrase. If it can, I argue that it is strong evidence that the negation marker is non-negative. I test the negativity of the three sentential negation markers present in Turkish: plain negation -mA, copular negation değil and existential negation yok. The results of a questionnaire sent to 23 Turkish speakers reveal that while the behavior of -mA is relatively stable, there is interspeaker variation as to the behavior of değil and yok: on the one hand, -mA yields obligatory double negation readings in the scope of ne..ne, while değil and yok in the same position, for some speakers, allow for single negation readings.

I conclude that at least for some speakers, değil and yok can be non-negative. I discuss the variation and what it can tell us about how learners determine the link between a semantic operator and its exponent(s).

2. The Negativity of Negation in SOV Languages with NC

2.1. Background on NC as Syntactic Agreement and the Typology of NC Languages

In this paper, I discuss NC within Zeijlstra’s (2004; 2008) analysis of NC as a syntactic Agree relation (Chomsky, 1995, 2001) between a single interpretable negative feature [iNeg] and one or more uninterpretable negative features [uNeg], using Multiple Agree (Hiraiwa, 2001). Negative feature checking is assumed to be upwards and phase-bound: probes bearing [uNeg] search for a c-commanding phasemate goal bearing [iNeg].1

I illustrate how this mechanism works for the two main types of NC languages that Zeijlstra identifies (following Giannakidou, 1997): non-strict and strict. On the one hand, there are non-strict NC languages, like Spanish, Italian, and Portuguese, in which, descriptively, post-verbal negative indefinites (NIs) engage in NC with sentential negation (2-a), but preverbal ones do not (3-b).

(2)	a.	María *(no) llamó a  nadie.	b.	Nadie    (#no)  llamó.
		María neg   called to nobody		nobody (#neg) called
		María didn’t call anyone.		Nobody called.

For these languages, Zeijlstra (2004) proposed that there are two semantically negative operators able to license features on NIs: the sentential negation marker and a covert operator. The sentential negation is in the NegP, merged right above the vP. It can license negative items in positions it c-commands, namely post-verbal negative indefinites, as in (3-a). For preverbal NIs, namely subjects that have moved to spec-TP, the sentential negation is not in a c-commanding configuration, and therefore cannot be the licensor. In these cases, a higher covert negative operator is called for, as shown in (3-b).

(3)	a.	[TP María [NegP no[iNeg] [vP llamó a nadie[uNeg] ]]]
	b.	[Op¬[iNeg] [TP nadie[uNeg] [vP llamó ]]]

We obtain as a result the non-strict pattern, where preverbal NIs, licensed by the covert operator, appear without a sentential negation marker, while postverbal NIs are licensed by and thus must appear with the overt sentential negation.

On the other hand, there are strict NC languages, where any NI must appear with sentential negation, whether it appears after or before the verb, as shown for Russian below.

(4)	a.	Masha *(ne) zvonila nikomu.	b.	Nikto   *(ne) zvonil.
		Masha neg   called    nobody		nobody neg   called
		Masha didn’t call anybody.		Nobody called.

Zeijlstra’s (2004) proposal for these languages is that the sentential negation marker has uninterpretable negative features just like NIs, and a high covert negative operator, merged above the subject position, licenses both the sentential negation marker and the NIs.

(5)	a.	[Op¬[iNeg] [TP Masha [NegP ne[uNeg] [vP zvonila nikomu[uNeg] ]]]]
	b.	[Op¬[iNeg] [TP nikto[uNeg] [NegP ne[uNeg] [vP zvonil ] ] ] ]

The requirement for the sentential negation marker to appear with NIs is not discussed by Zeijlstra himself, but one could argue that it is due to a requirement on the licensing of a covert negative operator: it can only appear if matching uninterpretable features are present on the clausal spine (a stipulation adopted in Jeretič, 2022)—since sentential negation is on the clausal spine, but NIs are not, sentential negation is obligatory to invoke the covert negative operator.

2.2. The Typology of NC Extended to SOV Languages

In Zeijlstra’s system, the negativity status of negation is a factor that determines whether a NC language has a strict or non-strict pattern: a negative negation marker results in a non-strict pattern, while a non-negative one results in a strict pattern. However, this explanation does not straightforwardly extend to SOV languages.

Unlike in SVO languages, being preverbal in an SOV language does not imply being hierarchically superior to the verb, which determines whether an element can be licensed by a vP-level negation marker. We therefore cannot reach the same conclusions about the negativity of negation in a language with SOV word order based on observing a strict or non-strict NC pattern, whose definitions rely on the notion of ‘preverbal’ elements. Indeed, objects in SOV languages are preverbal but are assumed to merge below the verb and, in simple cases, not move above it2 (just like they do in SVO languages). Therefore, object NCIs can be licensed by a negation that merges above the verb, despite being preverbal. What about subjects? Just like objects, the hierarchical relation between a subject and a head negation is not reflected by linear order, because the subject, as a specifier, is in an initial position, and the head negation takes a final position. The question of the movement of subjects from their standardly assumed base-merge position in spec-vP to a vP-external position above the verb (typically spec-TP) in SOV languages is subject to debate. In fact, the answer might vary across languages and configurations. Diagnostics other than linear order are needed to determine the position of the subject; we discuss this question for Turkish in the Section 2.4. Now, we lay out the expected typology of the NC patterns of SOV languages whose negation marker merges right above the vP, depending on (a) whether (NCI) subjects move out of the vP, and (b) whether negation is negative.

If subjects are in spec-TP, they are located above the position of negation. Among SOV languages with subjects in spec-TP, we might expect to observe the following typological divide determined by the negativity status of negation: if negation has interpretable negative features, subject NCIs do not co-occur with it, as shown in (6-a) (akin to non-strict NC SVO languages); if it does not, then subject NCIs must co-occur with it, as shown in (6-b) (akin to strict NC SVO languages). I assume objects stay in the vP and thus can be licensed by the negative operator, whether it is low or high.

(6)	Expected types of SOV languages with subject NCIs above NegP:
	a.	Negative negation.
		(i)	Subject NCI: [TP nobody[uNeg] [NegP [vP arrive ] Neg¬[iNeg] ]]
		(ii)	Object NCI: [TP Mary [NegP [vP nobody[uNeg] see ] Neg¬[iNeg] ]]
	b.	Non-negative negation.
		(i)	Subject NCI: [[TP nobody[uNeg] [NegP [vP arrive ] Neg[uNeg] ]] Op¬[iNeg]]
		(ii)	Object NCI: [[TP Mary [NegP [vP nobody[uNeg] see ] Neg[uNeg] ]] Op¬[iNeg]]

As far as the literature on NC goes, no language has been claimed to have the surface pattern in (6-a), where high NCIs in SOV languages do not require sentential negation to be licensed. This may be considered an accident or have a principled reason (e.g., a result of a learning bias).

In contrast, if subject NCIs stay in spec-vP, then they can always be licensed by a negation marker, whether that marker is negative or not, and we should expect all NC languages with such low subjects to have a strict pattern, as shown below in the structures in (7-a(i)) and (7-b(i)). Object NCIs are also in the vP and again licensed by sentential negation regardless of its position.

(7)	Expected types of SOV languages with subjects below NegP:
	a.	Negative negation.
		(i)	Subject NCI: [NegP [vP nobody[uNeg] arrive ] Neg¬[iNeg] ]
		(ii)	Object NCI: [NegP [vP Mary nobody[uNeg] see ] Neg¬[iNeg] ]
	b.	Non-negative negation.
		(i)	Subject NCI: [[NegP [vP nobody[uNeg] arrive ] Neg[uNeg] ] Op¬[iNeg]]
		(ii)	Object NCI: [[NegP [vP Mary nobody[uNeg] see ] Neg[uNeg] ] Op¬[iNeg]]

Because these two types of languages are not different on the surface, at least in these simple cases, the status of negation in strict NC SOV languages with subjects in spec-vP is underdetermined.

Therefore, if we consider the variation in the position of subjects and the negativity of negation, we have three possible grammars for SOV languages with a surface strict NC pattern: (6-b) (high NCI subjects, non-negative negation), (7-a) (low NCI subjects, negative negation) and (7-b) (low NCI subjects, non-negative negation). Only one of them, (7-a), has a negative negation; hence, SOV languages indeed break the correlation between the negativity of negation and the strictness of the NC pattern argued by Zeijlstra.

Note there is an additional relevant point of variation within grammars of the (7-b) type: the position of the covert operator. It may be found merged below the TP (and right above the Neg head, e.g., spec-NegP) or above the TP. This means that if in a given language, the locus of interpretable negation is found to always be below the TP, its grammar is still compatible with (7-a) and (7-b). This point becomes relevant when we look at Turkish.

The multiplicity of logical possibilities for a grammar of an SOV language with a strict NC pattern begs the questions: Are all attested, or just a subset thereof? And how does the learner decide on one of these grammars? While the basic structures presented above converge on the surface, there may be other structures that do not, thus allowing the learner to determine the grammar of their input. But it also may be that there are no such structures, or insufficient evidence in the input that differentiates between the options. In such cases, the learner might decide on a grammar randomly, which therefore could be different from that of their input, or follow some learning heuristic that will favor one grammar over the other. For instance, we could imagine a preference for the sentential negation marker to be negative, as it results in a simpler and more transparent grammar than if it were not.

In this paper, we look for evidence to determine which of these grammars corresponds to the one(s) spoken by Turkish speakers. In order to do so, we first discuss the position of NCI subjects, together with that of semantic negation. After showing the basic facts that make Turkish a strict NC language in Section 2.3, we argue in Section 2.4 that Turkish is likely to be a language in which NCI subjects remain in spec-vP and are furthermore licensed by a negation found below the TP.

This finding narrows down the possible grammars of Turkish to the ones listed in (7). However, it does not decide on the negativity status of negation: either the negation marker itself carries semantic negation, as in (7-a), or we have a covert negative operator merged right above the Neg head and below the TP, as in (7-b).

We therefore need further investigation to determine the status of the negativity of the negation markers in Turkish. We do so in Section 3, which reveals a heterogeneous picture across Turkish speakers.

2.3. Turkish Strict NC

Turkish has been described as a strict NC language (Görgülü, 2020; Jeretič, 2018, 2022; Kelepir, 2001; Zeijlstra, 2004, 2008) because of the obligatory presence of a sentential negation marker in the presence of any NI. I show examples below of the three types of negation markers (see Section 3.1 for more information on these markers), together with the NI hiçkimse (‘nobody’).3

(8)	a.	Hiçkimse gel-*(me)-di.
		nobody come-neg-pst
		Nobody came.
	b.	Hiçkimse hasta *(değil).
		nobody sick neg.cop
		Nobody is sick.
	c.	Hiçkimse-nin para-sı   {yok,  *var}.
		nobody-gen   money-poss neg.exist exist
		Nobody has money.

In addition to the obligatory presence of a negative marker, no double negation reading is available in any of these sentences.

A similar requirement arises for any negative indefinite expression, namely with quantifiers like hiçkimse (“nobody”), hiçbirşey (“nothing”), hiçbir zaman (“never”), for any noun X, hiçbir X (“no X”), asla (“never”) … (see Kelepir, 2001; Özyıldız, 2017 for overviews on negative quantifiers in Turkish).

Thus, the surface pattern of Turkish is ‘strict’ according to the definition Zeijlstra proposes. Our goal is now to determine which of the three possible grammars laid out above in 2.2 it corresponds to. The first step is to establish the position of NCI subjects relative to the negative marker.

2.4. The Position of NCI Subjects and Semantic Negation in Turkish

Authors are not in full agreement on the position of subjects in Turkish, namely whether they are in spec-vP or spec-TP. Traditionally, it is assumed that subjects in Turkish move to spec-TP to satisfy the EPP and receive case (Aygen, 2002; Kelepir, 2001; Kornfilt, 1984; Kural, 1993, a.o.). In fact, this assumption has driven authors like Kelepir (2001) to propose the existence of a covert semantic negation above the position of the subject, so as to allow for licensing of subject NCIs (just as Zeijlstra proposed for strict NC SVO language, and the option laid out in (6-b)).

However, there is a body of work that challenges the idea that all subjects must move to spec-TP (İşsever, 2008; Kornfilt, 2020; Öztürk, 2002, 2004), and instead, in some cases, subjects can stay in spec-vP (the EPP then being satisfied by movement of v to T, à la Alexiadou & Anagnostopoulou, 1998), and only raise to spec-TP to achieve wide scope (e.g., specific interpretations for indefinites).

Indeed, quantifier subjects have ambiguous scope with respect to negation (some speakers disallow $\forall >\neg$, for others it is simply marked; other quantifier expressions, e.g., numerals, proportion expressions, are more robustly ambiguous between the two scopes).

(9)	a.	Her çocuk gel-me-di.
		all child  come-neg-pst
		All children did not come.            $\neg >\forall {,}^{\%} \forall >\neg$
	b.	Üç  çocuk gel-me-di.
		three child  come-neg-pst
		Three children did not come.           $\neg >3, 3>\neg$

This ambiguity shows that quantifiers occur below negation in their base position, and are able to raise above it. The question remains: do they raise above it and then reconstruct at LF? Or do they have the option to stay below and raise only for interpretative purposes?

The low position of subjects has been proposed by Öztürk (2002, 2005), who posits that subjects in Turkish generally do not move to spec-TP, unless it is to achieve a higher semantic scope. Öztürk’s main argument has to do with the fact that scope relations with universal quantifiers correlate with agreement on the verb: in the absence of plural agreement on the verb, universal quantifiers like bütün scope low with respect to negation, while when plural agreement is present, they take wide scope (some speakers allow wide scope of bütün in (9-a) and narrow scope in (9-b); in both these cases, wide scope bütün is stressed (Jaklin Kornfilt, p.c.)).

(10)	a.	Bütün çocuklar o   test-e  gir-me-di.
		all  children that test-dat take-neg-pst
		All children didn’t take that test.	(%all>not, not>all)
	b.	Bütün çocuklar o   test-e  gir-me-di-ler.
		all  children that test-dat take-neg-pst-pl
		All children didn’t take that test.	(all>not, %not>all)

The wide scope of the universal quantifier shows that it has moved above the semantic position of negation. According to Öztürk, the plural agreement on the tensed verb is a morphological reflex of the movement of the subject to spec-TP.4 Hence the unavailability of the narrow scope of the universal quantifier. If Öztürk’s analysis is right, it shows several things: first, that the subject need not move to spec-TP, second, if it does, it must scope above the semantic position of negation, which therefore must be merged between the vP and TP, and third, if it does move to spec-TP, it cannot reconstruct below negation, at least when it triggers plural agreement on the verb.

Since semantic negation is above the vP but not the TP, subject NCIs can only be licensed by it in their vP-internal position. They furthermore must stay there, because they are interpreted there (as expected from their nature as existential quantifiers, needing to scope below negation), and we assume they cannot, like subjects that trigger plural agreement on the verb, reconstruct.

We provide an additional argument for both Öztürk’s claim that plural agreement correlates with scope with respect to negation, as well as a low position for NCI subjects: NCI subjects are incompatible with plural agreement on the verb, as shown below.

(11)	Hiçbir-imiz-in  arkadaş-lar-ı gel-me-di-(*ler).
	none-poss.1pl-gen friend-pl-acc   come-neg-pst-(*pl)
	None of our friends came. (lit. friends of none of us came)

Note that it was crucial to construct the sentence with an embedded NCI, here in a possessive structure, because a bare NCI would not be expected to trigger plural agreement in the first place, due to its singular number marking with bir (‘one’). I further note that the possessive structure is not the culprit for the lack of plural marking, as we can see in the following example (note: the judgments are subtle for some speakers, but everyone agrees that (11) is worse than the sentences in (12)).

(12)	a.	Öğrenci-nin arkadaş-lar-ı gel-me-di-(ler).
		student-gen friend-pl-acc   come-neg-pst-(pl)
		The students’ friends didn’t come.
	b.	Hep-imiz-in  arkadaş-lar-ı gel-di-(ler).
		all-poss.1pl-gen friend-pl-acc come-pst-(pl)
		All of our friends came.

This result therefore supports an analysis of Turkish of one of the types in (7), in which subject NCIs must stay in spec-vP, to be licensed by a semantic negation located either at the negation marker -ma, which is found between the verb and tense, or right above it (but below the TP).

Note: the claim that NCI subjects must stay in the vP domain opens up questions about case-licensing. While in matrix clauses nominative case is null, nominalizations mark subjects with genitive marking, at least when they are specific. And it is natural to assume that genitive marking, usually correlated with specificity, is received in spec-TP (Kornfilt, 2020, a.o.). However, an NCI subject can receive genitive case in a nominalization, as shown in (13).

(13)	Hiçkimse-nin	köy-ü	bas-ma-dığ-ın-ı	duy-du-m.
	nobody-gen	village-acc	raid-neg-nmz-3s-acc	hear-pst-1s
	I heard that nobody raided the village.

One possibility would be to allow case-marking at a distance, or to allow non-interpretable movement of the NCI for it to receive case, after its negative features have been checked in spec-vP. This could also cover cases in which genitive plural subjects receive non-specific interpretations (as discussed in Kornfilt, 2020). I leave this open question to future work and assume that NCIs must remain in spec-vP.

We now move to determining the negativity status of negation with an independent diagnostic: we will create configurations in which the negation markers are in the scope of another negative operator, thus giving them the chance, if they are indeed non-negative, to be licensed not by the covert operator that typically licenses them, but by another negative operator. If they are negative, they should unambiguously yield double negation readings.

3. The Variable Behavior of the Three Negation Markers Under Neither..nor

In this section, I present evidence for whether the three negation markers in Turkish are negative or not. I do so by testing the interpretations that arise in the scope of ne..ne (‘neither..nor’) phrases. In Section 3.1, I present the three negation markers that exist in Turkish. In Section 3.2, I present a background on ne..ne phrases in Turkish, namely their peculiar NC behavior and an analysis of it. In Section 3.3, I present the scope of ne..ne phrases as a negativity diagnostic for sentential negation markers. In Section 3.4, I show results of a questionnaire testing the three negation markers in the scope of ne..ne phrases, which reveal that they sometimes engage in NC for some speakers, but not all.

3.1. Turkish Sentential Negation Markers

There are three sentential negation markers in Turkish (Göksel & Kerslake, 2005; Kornfilt, 1997). The first is the bound verbal morpheme -mA,5 shown in (14-b), which contributes a plain negative meaning. In copular sentences, the negative copula değil is used, see (15-b), and in existential sentences, we find the negative existential yok, see (16-b), which is the negative version of existential verb var, shown in (16-a); both değil and yok are free morphemes, in contrast with -mA.

(14)	a.	Duygu gel-di.
		duygu  come-pst
		Duygu came.
	b.	Duygu  gel-me-di.
		duygu   come-neg-pst
		Duygu didn’t come.

(15)	a.	Duygu   hasta.
		duygu    sick
		Duygu is sick.
	b.	Duygu    hasta  değil.
		duygu     sick    neg.cop
		Duygu is not sick.

(16)	a.	Duygu-nun para-sı    var.
		duygu-gen  money-poss exist
		Duygu has money.
	b.	Duygu-nun para-sı    yok.
		duygu-gen  money-poss neg.exist
		Duygu has no money.

3.2. Turkish ne..ne (‘neither..nor’) Phrases: Background

Before showing how Turkish ne..ne phrases can be used as a diagnostic for non-negativity of sentential negation markers in Section 3.3, I present some background on the construction and its general interaction with the NC system of Turkish.

3.2.1. Optional NC with ne..ne as a Result of Structural Ambiguity

Recall from Section 2.3 that Turkish can be described as a ‘strict NC’ language, i.e., one in which all of its NCIs must co-occur with sentential negation. However, there is one apparent exception: ne..ne (‘neither..nor’) phrases, for which these markers are generally optional (to yield the same meaning), as noted by Gencan (1979); Göksel (1987); Göksel and Kerslake (2005); Jeretič (2018), 2022); Şener and İşsever (2003).6 I show this below for the three negation markers.

(17)	a.	Ne Devrim ne Umut gel-(me)-di.
		ne  devrim  ne umut come-neg-pst
		Neither Devrim nor Umut came.
	b.	Ne Devrim ne Umut hasta (değil).
		ne  devrim  ne umut sick neg.cop
		Neither Devrim nor Umut is sick.
	c.	Ne Devrim  ne Umut’un  para-sı    {yok,  var}.
		ne  devrim  ne umut-gen money-poss neg.exist exist
		Neither Devrim nor Umut has money.

Furthermore, when they do appear, in addition to an NC reading (where only one negation is interpreted), they can yield a double negation reading, in contrast with typical NCIs in (8), in which only a single negation reading is available. They therefore exhibit an “optional NC” pattern.

Jeretič (2018, 2022), followed by Gračanin-Yüksek (2023) (in this special issue), shows that the optionality of NC appears to be an effect of the structural ambiguity in the size of the coordination: when the structure is underlyingly clausal, there is no NC, as in (18-a), and when ne..ne coordinates DPs, NC is obligatory, as schematized below.

(18)	a.	Ne [Devrim gel-di]   ne [Umut gel-di].
		ne  devrim   come-pst ne umut   come-pst
		Neither Devrim nor Umut came.
	b.	[Ne Devrim ne Umut] gel-me-di.
		ne   devrim  ne umut   come-neg-pst
		Neither Devrim nor Umut came.

Evidence in support of this analysis can be found in ne..ne coordinations that are unambiguously clausal or non-clausal. For instance, when ne..ne coordinates fully overt clauses negated with -mA, NC is no longer possible, as we can see in (19).

(19)	a.	Ne [Deniz dans   et-me-di] ne [Tunç şarkı söyle-me-di]. (Jeretič, 2022)
		ne  Deniz  dance do-neg-pst  ne Tunç  song say-neg-pst
		Deniz didn’t not dance, nor did Tunç not sing. (=Deniz danced and Tunç sang)
	b.	Ne [Deniz dans   et-me-di]   ne [Tunç dans et-me-di].   (Jeretič, 2022)
		ne  Deniz  dance do-neg-pst ne Tunç  dance do-neg-pst
		Deniz didn’t not dance, nor did Tunç. (=Both Deniz and Tunç danced)

In (19-a), we have a ne..ne phrase coordinating two negated clauses. In (19-b), we observe what has been described as a forward gapping structure (Hankamer, 1971; Kornfilt, 2012), where a full clause is overtly present in the first junct while the second junct only realizes a piece of a smaller constituent. This structure clearly suggests ellipsis of the rest of the clausal material in the second junct, leaving only the contrasting material. Thus, this is still a case of unambiguous clausal coordination, as argued in Jeretič (2022).

We can also construct cases of unambiguous DP coordination, as in cases of right-dislocation, shown in (20).

(20)	a.	Gel-me-di    ne Ali ne Beste.
		come-neg-pst ne Ali ne Beste
		They didn’t come, neither Ali nor Beste.
	b.	*Gel-di, ne Ali ne Beste.
		come-pst ne Ali ne Beste

With these basic cases as a starting point, Jeretič (2022) gives a number of additional arguments for the observed pattern of optional NC with ne..ne phrases as a direct result of the structural ambiguity of coordination structures. Gračanin-Yüksek (2023) provides additional support for this analysis from agreement patterns.

3.2.2. Ne..ne Phrases as Simple NCIs

The empirical generalization above allows for a natural account embedded in Zeijlstra’s (2004; 2008) Agree-based framework for NC of ne..ne phrases as basic NCIs. The apparent optionality turns out to be only due to the particular syntactic behavior of coordinations, and nothing special needs to be said about the NC system.

In Jeretič (2022), they are treated as disjunctions that are required to be in the scope of negation thanks to [uNeg] features, just like NCI indefinites. The negation morpheme -mA is a head merged above the vP and is assumed to carry [iNeg], and in addition, there is a higher null negation NegOp, an [iNeg]-carrying head merged above the CP, whose presence I assume is licensed only if [uNeg] are present on the clausal spine.

The optional NC pattern with ne..ne phrases is captured by proposing that the [uNeg] of the ne operator must be licensed by the closest [iNeg] available. When ne..ne coordinates DPs, it behaves exactly like an indefinite NCI like hiçkimse: it stays in its argument position in the vP, and thus must be licensed by the c-commanding Neg head -mA that carries [iNeg], as shown in (21). For this reason, ne..ne must be a disjunction, in order for the right meaning to be delivered in the scope of negation.

(21)	[Ne	Ali	ne	Beste]	gel-me-di.
	ne	Ali	ne	Beste	come-neg-pst
	Neither Ali nor Beste came.

When ne..ne coordinates clauses, its [uNeg] cannot be licensed by -mA, because -mA is inside the scope of the coordination and thus not in an appropriate c-commanding relationship to license the [uNeg] on ne. In this case, a higher covert [iNeg] must come to rescue ne’s [uNeg], resulting in the lack of -mA on the verb.7

(22)	Ne	Ali	<gel-di>	ne	Beste	gel-di.
	ne	Ali	come-pst	ne	Beste	come-pst
	Neither Ali nor Beste came.

In the example above, there is ellipsis in the first junct (which gives a string that is ambiguous between a TP and a DP coordination). There can also be ellipsis in the second junct, and the underlying tree is identical to the structure given above.

In a clausal coordination structure, if -mA is used, upward agreement between the [uNeg] from the ne operator and the [iNeg] of the negation morpheme -mA is not possible, because -mA is not c-commanding ne. Therefore, the same strategy as above is used, namely, a negative operator is inserted above the NeP. However, this produces double negation, as both -mA and the covert NegOp are semantically negative.

(23)	Ne	Ali	gel-me-di	ne	Beste	<gel-me-di>.
	ne	Ali	come-neg-pst	ne	Beste	come-neg-pst
	Neither Ali nor Beste didn’t come.

3.3. The Scope of ‘Neither..nor’ as a Negativity Test

I now show that we can use ne..ne phrases to test the negativity of negation markers. In order to diagnose whether each sentential negation marker contributes its own semantic negation, I investigate sentences in which it is found in the scope of ne..ne phrases and test whether it can engage in concord with them.

Indeed, the non-negativity of a negative concord item can be diagnosed if it occurs in the scope of another negative element, but yields only one sentential negation. There are not many negative environments that can embed a sentential negation. One can of course consider a biclausal environment, where a negative clause is embedded under a negative matrix environment (e.g., ‘I don’t think she didn’t leave’). However, these cases cannot yield NC due to their biclausal nature, NC being clause-bounded (see Jeretič, 2022; Kayabaşı & Özgen, 2018; Kornfilt, 1984, 1997; Şener, 2007 for Turkish, although this is a more general phenomenon).

There are arguably two such environments that do not introduce a new clause, but nevertheless can introduce a negative operator that outscopes a sentential negation marker: denial negation and clausal neither..nor phrases. In Turkish, it is possible to use değil as a denial negation that can negate full TPs. However, these cases do not yield negative concord.

(24)	a.	Duygu gel-me-di  değil.
		duygu  come-neg-pst neg.cop
		Duygu didn’t not come.
	b.	Duygu  hasta değil  değil.
		duygu   sick   neg.cop neg.cop
		Duygu is not not sick.
	c.	Duygu-nun para-sı   yok    değil.
		duygu-gen money-poss neg.exist neg.cop
		Duygu doesn’t not have money.

We should not conclude from this that the negation markers are necessarily negative: there could be locality constraints or pragmatic reasons to force the interpretation of both negations here.

We thus turn to the ne..ne phrase test. It was first observed in Jeretič (2022) that having değil in the scope of ne..ne phrases could yield an interpretation with only one semantic negation, for some speakers. This is crucially in contrast with -mA, as mentioned in the above Section 3.2, which was claimed to obligatorily yield double negation readings when inside the scope of ne..ne.

(25)	Ne [hasta değil-im]  ne [yorgun değil-im].
	ne  sick     ex.neg-1s  ne tired    ex.neg-1s
	a.  I am neither sick nor tired.
	b.  I am neither not sick nor not tired. (=I am both sick and tired)

In sentence (25), some speakers allow for a reading in which negation is interpreted only once, despite it being represented both by the ne particles and the sentential negation markers in each junct. I take this data to reflect the fact that for these speakers, değil is non-negative. This conclusion is based on the following assumptions: (1) the single negation readings are a result of real grammatical NC, and not some illusion of it; (2) these configurations are ones in which the sentential negation is syntactically in the scope of the ne..ne phrase; and (3) NC is an operation with a single direction: uninterpretable features search upwards to find their goal, a c-commanding interpretable operator, and never the other way around.

We now review the plausibility of each of these assumptions. The first one is of course difficult to evaluate, especially given the nature of the processing difficulty of double negation, that here might be all the more concerning given the added complexity of the neither..nor configuration. However, the contrast between the -mA negation and değil is real: in my empirical work on the data presented in Jeretič (2022), the judgments on double negation readings in the scope of ne..ne were robust. If the difficulty of double negation readings were the reason for NC readings with değil, then why should we not also observe NC readings with -mA? As we see in the following section, this paper confirms that there is indeed a contrast in judgments between -mA and değil, as well as reveals a contrast between -mA and yok, further dispelling the possibility of the NC readings being an illusion.

The second assumption is one that is in accordance with the syntactic literature, and questioning it results in proposing a rather unconventional syntactic mechanism. Indeed, değil is overtly in the scope of the ne..ne phrase, and it is unlikely that it can undergo LF movement above the ne operator, even less covert, which here would be ATB movement out of a coordination. There is no other known mechanism that would allow for a negation pronounced inside the coordination to be interpreted above the ne..ne operator. However, it may be worth challenging current knowledge, and considering the possibility for a structure of the type [[ne A ne B] X] to be able to be linearized as ne A X ne B (X), in certain cases. This would allow for an operator taking scope above the coordination to appear inside the first junct. We discuss a special case of this possibility in light of certain results in Section 3.5.2, but dismiss it for the general case.

Finally, the third assumption is taken to be relatively uncontroversial, that there should be a directionality to an agreement mechanism. If NC did not have directionality, many basic facts about NC systems would lose their explanation. For example, in Turkish specifically, we should expect -mA to always be licensed in the scope of ne..ne, which is not what we observe.

3.4. Interaction of the Three Negation Markers with ne..ne Phrases

In this section, I present the design and results of a questionnaire that was sent to Turkish native speakers to test for the behavior of each sentential negation marker in the scope of ne..ne, namely to check whether or not NC readings were available.

3.4.1. The Questionnaire

I selected a series of questions that tested for the availability of NC readings for each of the three Turkish negation markers in the scope of a ne..ne coordination. To this end, the participants were asked to judge whether such sentences were acceptable in a context that targeted the NC reading, i.e., where the NC reading of the sentence was acceptable but not the double negation one.

I tested two types of sentences that forced clausal coordination, i.e., forward gapping structures and full overt clausal coordination. The target sentences with forward gapping are found in (26-a), (27-a), (28-a), and those with full overt clausal coordination in (26-b), (27-b), (27-d), (28-b). For the plain sentential negation marker -mA, the only reading expected in the scope of ne..ne (according to Jeretič, 2022) is a double negation reading; thus, these sentences were not expected to be felicitous in the contexts given, which targeted the NC reading. The değil sentences were expected to be acceptable for some speakers in the scope of ne..ne, again given the observation in Jeretič, 2022; and for yok, there was no expectation, given that there were no previous reports of the readings of it in the scope of ne..ne.

Finally, I added control sentences in which the negated predicate appeared after the coordination structure ((26-c), (27-c), (28-c)). These were fully expected to have a NC reading, according to the literature on optional NC with ne..ne phrases (Gračanin-Yüksek, 2023; Jeretič, 2018, 2022; Şener & İşsever, 2003), but could be expected to be judged ungrammatical due to the prescriptive rule against using a negation marker in ne..ne sentences. Such sentences have been analyzed in Jeretič, 2018, 2022 and Gračanin-Yüksek, 2023 as cases in which the negation is outside of the ne..ne coordination (see Section 3.2), thus allowing it to scope above the [uNeg] of the ne..ne coordination and license it (see more details of this analysis in Section 3.2). Thus, these sentences acted as controls for the validity of this questionnaire’s methodology as a way to retrieve NC readings in the minimally different target sentences.

(26)	Context: I’m feeling bad. I realize that it’s because I haven’t eaten fruits or vegetables.
	a.	Ne meyve ye-me-di-m  ne (de) sebze.
		ne  fruit     eat-neg-pst-1s   ne (also)  vegetable
		Target reading: I ate neither fruits nor (also) vegetables.	not expected
	b.	Ne meyve ye-me-di-m   ne (de) sebze  ye-me-di-m.
		ne  fruit  eat-neg-pst-1s  ne (also) vegetable eat-neg-pst-1sg
		Target reading: I ate neither fruits nor (also) vegetables.	not expected
	c.	Ne meyve ne (de) sebze  ye-me-di-m.
		ne  fruit  ne (also) vegetable eat-neg-pst-1s
		Target reading: I ate neither fruits nor (also) vegetables.	expected

(27)	Context: Both Devrim and Umut are healthy.
	a.	Ne Devrim hasta değil  ne (de) Umut.
		ne  devrim  sick   neg.cop   ne (also)  umut
		Target reading: Neither Devrim nor Umut is sick.	expected sometimes
	b.	Ne Devrim hasta değil  ne (de) Umut hasta değil.
		ne  devrim  sick   neg.cop   ne (also)  umut sick   neg.cop
		Target reading: Neither Devrim nor Umut is sick.	expected sometimes
	c.	Ne Devrim ne (de) Umut hasta değil.
		ne  devrim ne (also)  umut sick neg.cop
		Target reading: Neither Devrim nor Umut is sick.	expected
	d.	Yesterday, Devrim was tired and Umut sick. Today they are both feeling better.
		Ne Devrim yorgun değil  ne (de) Umut hasta değil.
		ne  devrim  sick   neg.cop    ne (also) umut  sick   neg.cop
		Target reading: Neither is Devrim tired nor Umut sick.	expected sometimes

(28)	Context: I would like to eat something fresh, unfortunately …
	a.	Ne meyve yok  ne (de) sebze.
		ne  fruit  neg.ex   ne (also)  vegetables
		Target reading: There are no fruits or vegetables.	no expectation
	b.	Ne meyve yok  ne (de) sebze     yok.
		ne  fruit  neg.ex   ne (also)  vegetables neg.ex
		Target reading: There are no fruits or vegetables.	no expectation
	c.	Ne meyve ne (de) sebze    yok.
		ne  fruit  ne (also) vegetables neg.ex
		Target reading: There are no fruits or vegetables.	expected

Participants were asked to judge whether these sentences were grammatical, and if yes, whether they were felicitous in the given context.

A few other questions were asked. In particular, one factor that seemed to matter for some speakers in preliminary investigations was whether only the predicate was negated or the full clause. For this reason, the following sentences were also added (with corresponding controls):

(29)	Context: I’m feeling great, rested and healthy.
	Ne	hasta	değil-im	ne	(de)	yorgun	(değil-im).
	ne	sick	neg.cop-1s	ne	(also)	tired	(neg.cop-1s)
	Target reading: I’m neither sick nor tired.

(30)	Context: I’m fasting tomorrow, and won’t be able to consume anything.
	Ne	yi-yecek	değil-im	ne	(de)	iç-ecek	(değil-im).
	ne	eat-fut	neg.cop-1s	ne	(also)	drink-fut	(neg.cop-1s)
	Target reading: I will neither eat nor drink.

(31)	Context: I’m hungry and thirsty, that’s because today …8
	Ne	ye-me-di-m	ne	(de)	iç-me-di-m.
	ne	eat-neg-pst-1s	ne	(also)	drink-neg-pst-1s
	Target reading: I neither ate nor drank.

3.4.2. Results

Twenty-three Turkish speakers answered the questionnaire. Two speakers reported no NC readings in coordinations when the negation came after (or in) the second junct, i.e., sentences (26-c), (27-c), (28-c), which were meant to be controls for the validity of the task. The rejection of NC readings in these sentences corresponds to the above-mentioned prescriptive view on ne..ne sentences, which are not supposed to co-occur with sentential negation. Furthermore, these speakers did not report any NC readings in the target sentences. Since I couldn’t differentiate between these responses being purely a result of a prescriptivist mindset and the speakers’ grammar, I excluded them from further consideration.

The individual results for the remaining 21 speakers on the target sentences are recorded in

Table 1. Questionnaire responses for each negative marker in the scope of ne..ne (* = ungrammatical; no NC = target NC reading unacceptable; NC = target NC reading acceptable).

	-mA		değil			yok
	clause	verb	clause	adjective	future pred.	clause
	(26)	(31)	(27)	(29)	(30)	(28)
no NC detected: 12/21
3	*	*	*	*	*	*
9	no NC	no NC	no NC	no NC	no NC	no NC
NC detected: 9/21
A	no NC	NC	no NC	no NC	no NC	no NC
B	no NC	no NC	no NC	no NC	NC	no NC
C	no NC	no NC	no NC	no NC	NC	NC
D	no NC	no NC	no NC	NC	NC	no NC
E	no NC	NC	no NC	NC	NC	NC
F	no NC	NC	NC	NC	no NC	NC
G	no NC	NC	NC	NC	NC	no NC
H	no NC	NC	NC	NC	NC	NC
I	NC	NC	NC	NC	NC	NC

. I coded for whether the sentence was deemed ungrammatical (*), unacceptable in the context given (no NC), or acceptable in the context given (NC). For speakers in which no non-negativity was detected (first part of the table), I recorded the cumulative results of those who judged the sentences plain ungrammatical and those who judged them as simply unacceptable in the context given; for speakers who gave any response that corresponded to a NC reading of a sentential negation marker in the scope of ne..ne (second part of the table), I listed the individual responses.

Twelve speakers did not report any NC readings in the sentences. Three of those reported ungrammaticality for any negation marker in the coordination. The other nine speakers reported NC readings for some of the negation markers, and results were quite varied, with no systematic pattern emerging.

Six speakers accepted NC readings with -mA, but five of those accepted them only when the negated verb alone was inside the coordination; further consultation with two of those speakers revealed that such NC readings were only possible under a very specific prosody.

Eight speakers accepted NC readings with değil. Three (G,H,I) accepted them across all three sentence types tested; two (D,E) did so except when full clauses were coordinated; two (B,C) accepted NC only in the future predicate sentence; and one (F) accepted NC except in the future predicate sentence.

Five speakers allowed NC readings with yok (and speaker G said it was borderline, i.e., worse than değil but better than -mA in a full clause).

Many speakers reported markedness of NC readings of negation markers inside the scope of ne..ne, and some reported that it was only possible with a marked prosody. Furthermore, several speakers who did accept NC readings of değil and/or yok did so readily in the cases of forward gapping structures ((27-a), (28-a)), but either dispreferred or rejected cases of overt clausal coordination ((27-b), (27-d), (28-b)). However, one speaker (H) reported the opposite preference, accepting the full clausal coordination structures but not the forward gapping ones.

The questionnaire did not test for double negation readings, as they are not the focus of this paper. However, based on certain participants’ comments and expectations, the double negation reading was the reading in the “no NC” cases, but also available as an alternative reading in the “NC” cases. Note also that for speakers B and C, and one of the speakers in the second line of the “no NC detected” group, their responses were too vague to determine whether they were judging the sentences ungrammatical or simply infelicitous.

Finally, I add that while these results are reported directly from the questionnaire with no further consultation with speakers, some speakers were closer consultants who provided judgments at various times (E, G, and 2 from the no-NC group). Their responses revealed the subtlety of these judgments overall, but also some robustness in their judgments across each time they were asked.

3.5. Discussion of the Results

Given the assumptions spelled out in Section 3.4.1, the single negation readings of a given negation marker in the scope of ne..ne are a diagnostic of its non-negativity. The main result from Section 3.4.2 is that non-negativity was sometimes detected for all three negation markers in Turkish. But the overall picture is complex: there is no evident systematicity of the negativity status of each negation marker, and negation overall, within and between speakers. In what follows, I attempt to provide explanations for this complex picture.

3.5.1. On the Differences Between the Negation Markers

A striking result is that the picture is not homogeneous across speakers, negation markers, and sentence types. A little over half of the speakers asked rejected NC readings altogether. For these speakers, we cannot conclude anything about the negativity of their negation markers: it is both consistent with a grammar in which sentential negation markers are all negative, or one in which a negation is non-negative but locality constraints (or something else) disallow its licensing by the negative operator introduced by ne..ne.

All but one speaker rejected NC readings of -mA when the ne..ne phrase coordinated a full clause, confirming the observation from (Jeretič, 2022). However, what was unexpected was the acceptance of NC readings by five speakers (besides the one that also accepted the full clause) when the negated predicate alone was in the coordination. Because of the systematicity of this contrast, we discuss an analysis in Section 3.5.2 that maintains the status of -mA as negative for the speakers who accept the ne..ne coordination with the -mA-negated verb. This analysis is based on the suspension of one of the core assumptions laid out in Section 3.4.1, namely, the one in which an element spelled out inside the scope of a coordination is not, under certain conditions, interpreted inside the coordination. Such a systematic rejection of NC readings with full clauses was not observed with değil and yok. This analysis of NC readings with -mA does not explain the whole picture of NC readings observed in this study, which is important to account for the asymmetry between -mA on the one hand, which is always negative but for one speaker, and değil/yok on the other, which can be non-negative.

For six speakers, results were inconsistent for the different sentence types for değil, and there were not enough syntactic frames for yok to determine differences between sentence types. There was also no correlation between responses for değil and responses for yok. In general, there was not enough data to determine whether there are systematicities in these results. Instead, one could argue that for those speakers with inconsistent değil results, değil is non-negative, and the lack of systematicity across sentence types is due to the unavoidable markedness of using a negation marker whose contribution is both optional and trivial, pushing speakers to sometimes judge those sentences unacceptable. With this assumption, değil was non-negative for nine, and yok for five (given the noisiness of these results, and the higher number of sentences tested for değil than yok, it is likely that the number would go up for NC readings of yok if more sentences were tested).

There is thus a contrast between -mA, which was non-negative for 1 speaker out of 21, versus değil or yok that were for 9 speakers out of 21. This difference can be attributed to morphosyntactic and semantic differences found between -mA on the one hand and değil/yok on the other. -mA is a bound morpheme that occupies the head of the negative projection and simply contributes semantic negation. Değil and yok instead are free morphemes, and it is possible that they do not occupy the same syntactic position. They have a verbal function, as a copula and an existential, respectively, and thus may be occupying the head of a verbal projection instead. As verbs, they are not simply negation, but probably manipulate an event argument as well. How could these differences affect the negativity status of each negation marker? There might, for instance, be a preference to assign negative semantics to the head of a negative projection, or to a marker whose function is negation only. I remain agnostic as to the answer to this question, since there are no available empirical or theoretical reasons to choose between them. Another possible explanation for the divide between -mA and değil/yok has to do with frequency. If we assume negativity to be an arbitrary lexical property, speakers may have enough data with -mA in ne..ne phrases to understand that it is negative, and thus assign negative semantics to it. In contrast, değil and yok are less frequent. Thus, we can hypothesize that as speakers acquire these markers, they do not have enough information in their input to determine their negativity status, and therefore choose one of the two options for each marker. The lack of information in the input is not surprising given the lengths we need to go as linguists to uncover the negativity status of each of these markers.

There are not enough data to say whether for a given speaker, the negativity status of değil and yok is the same. But it is plausible that it does not have to be the same, and the negativity status of these markers is an arbitrary lexical property. Furthermore, if the differences in results between speakers reflect differences in negativity status across these speakers, it suggests that this property is not one that is acquired from the input, but at least sometimes arbitrarily assigned. This claim is also supported by the fact that there are no obvious syntactic differences between değil and yok, making it again dubious that there is a particular property about these markers that should affect its negativity status.

If the negativity status of negation is underspecified in the input, it is nevertheless of interest that speakers do not converge on one meaning for negation.9 This could be due to the lack of learning heuristics to select one or the other option, making meaning assignment arbitrary. A more interesting scenario is that there are opposing pressures in the assignment of meaning to exponents. On the one hand, we could expect a preference for a grammar that is one-to-one between semantic operators and their exponents. The existence of this pressure finds support in the fact that NC languages tend to assign non-negative semantics to negative elements that are also existential quantifiers (e.g., hiçkimse, ‘nobody’). The fact that değil and yok are candidates for being non-negative might come from the fact mentioned above that they perform an additional function than simply negation: one is a copula and the other an existential. On the other hand, we might expect a preference for the semantics of an operator to be associated with an overt exponent, which would make learners be pushed to assign negative semantics to değil and yok, in combination with their copular and existential meaning. Due to two competing pressures, one that prefers the semantics of an operator to be associated with an overt exponent and the other to have only one grammatical function attached to an exponent, we end up with variation in the result of the learning process.

While the unsystematic results for değil and yok could make us question the validity of the questionnaire’s methodology, the relative systematicity observed with -mA provides support for it. Indeed, it is less likely that speakers will have illusions of grammaticality with some specific markers rather than others.

More generally, any attempt to explain these results differently than making it dependent on a lexical property of these markers runs into the difficulty of explaining why değil/yok but not -mA allow for NC readings.

3.5.2. On the NC Readings of ne..ne Coordinations of -mA-Negated Verbs

We now discuss the curious contrast found with five speakers, where NC can sometimes be observed with -mA, but only if the bare negated predicate is coordinated, as in (31), repeated below in (32), and any addition of a contentful object or subject seems to remove that possibility, as in (26), repeated below in (33).

(32)	Context: I’m hungry and thirsty, that’s because today …
	Ne	ye-me-di-m	ne	(de)	iç-me-di-m.
	ne	eat-neg-pst-1sg	ne	(also)	drink-neg-pst-1sg
	Target reading: I neither ate nor drank.

(33)	Context: I’m feeling bad. I realize that it’s because I haven’t eaten fruits or vegetables.
	Ne	meyve	ye-me-di-m	ne	(de)	sebze	ye-me-di-m.
	ne	fruit	eat-neg-pst-1sg	ne	(also)	vegetable	eat-neg-pst-1sg
	Target reading: I ate neither fruits nor (also) vegetables.

A Sketch of an Analysis

I propose that the contrast between (33) and (32) is due to the combination of two factors: bare verb coordination in the syntax and the inability in Turkish to suspend a verbal suffix from its verbal complex (with the exception of suffixes following participial suffixes, namely progressive -Iyor, future -AcAK, reported past -mIş, and aorist -r, as discussed in Kornfilt, 1996).10

Instead of yielding an ungrammatical structure, I propose that some speakers allow the exponents of the verbal suffixes to lower and distribute over both juncts, despite scoping above the coordination in the syntax.

(34)	Structure:	Spellout:
		→ ne yemedim ne içmedim

I assume that this type of lowering is possible only when the coordination ‘looks like’ a verb. In other words, if arguments of the verbs are inside the coordination, the lowering operation can no longer ‘see’ the verb inside the coordination, and therefore cannot distribute over this, accounting for the lack of NC readings in sentences like (33).

Evidence from the Interaction of ne..ne and -sIz (‘Without’)

Another piece of evidence for this analysis comes from the interaction of -sIz (‘without’), and a ne..ne phrase. The following sentence was also included in the questionnaire.

(35)	Context: I want to eat vegetable alone, with nothing on top.
	Ne	et-siz	ne	(de)	balık-sız	sebze	ist-iyor-um.
	ne	meat-without	ne	(also)	fish-without	vegetables	want-prog-1s.
	a.	Target reading: I want vegetables without meat or fish.
	b.	Other available reading: I don’t want vegetables without meat or fish.

Four speakers out of twenty-three accepted an NC reading when -sIz was inside the scope of the ne..ne phrase (a judgment which was reported in Jeretič (2022) to be unavailable). These were speakers A, F, G, H from Table 1. Thus, all speakers who reported this reading were also speakers who NC reading of -mA-negated verbs inside ne..ne (and only speaker E accepted NC with -mA inside ne..ne but not with -sIz). This appears to be a high degree of correlation, encouraging a common proposal for both phenomena.

The proposal for the spellout rule on top, if extended for affixation of -sIz to NPs, can predict this reading. Indeed: if -sIz is spelled out inside the coordination, but is merged above in the syntax, it can license the [uNeg] of the ne..ne phrase (just as it can license NCIs more generally).

The NC reading in (35) arguably constitutes a more robust piece of evidence for a spellout rule of the type proposed in this section than the evidence with -mA. In fact, it is difficult to explain it in any other way.

I asked one speaker who provided this judgment so as to check some predictions of this theory in the context of an analysis of ne..ne phrases as presented in Section 2.1 (i.e., as proposed in Jeretič, 2018, 2022, followed up by Gračanin-Yüksek, 2023). The next few sentences are this speaker’s judgments.

One crucial prediction is that if -sIz scopes above the ne..ne phrase to license the [uNeg] of the ne operator, then this whole ne..ne -sIz string is a constituent. In other words, the ne..ne phrase is not underlyingly a full clause, in contrast to Jeretič’s (2018, 2022) analysis with all other ne..ne sentences without a sentential negation marker. (35) also has a non-NC reading, which is most likely a reflection of a structure in which -sIz is simply inside the ne..ne coordination in the syntax, i.e., what could be called the surface reading.

In sum, I propose that the two readings of (35) are due to a structural ambiguity between the following two options: a clausal ne..ne coordination with -sIz inside it, resulting in a double negation reading, and a small ne..ne coordination with -sIz outside of the ne..ne phrase in the syntax, licensing it and thus obligatorily yielding NC.

I test this theory employing two tests (from Jeretič, 2022) that force clausal or non-clausal coordination. First, the clausal coordination test is obtained by including the clause in the first junct (forcing ellipsis in the second). I show this below.

(36)	Context: I want to eat vegetables alone, with nothing on top.
	#Ne et-siz    sebze    ist-iyor-um  ne (de) balık-sız.
	ne  meat-without vegetables want-prog-1s ne (also) fish-without
	a.	unavailable: I want vegetables without meat or fish.	(*NC)
	b.	I don’t want vegetables without meat or without fish. (i.e., I want vegetables
		with both meat and fish.)	(DN)

The only possible reading in this case is a double negation reading, which supports the idea that (a) the DN reading is due to clausal coordination and (b) the NC reading is only possible if the coordination is local.

Second, I apply the constituency coordination test of right-dislocation on the ne..ne phrase, which is only possible if the ne..ne coordination is non-clausal.

(37)	Sebze    ist-iyor-um,   ne et-siz       ne (de)   balık-sız.
	vegetables want-prog-1s ne meat-without ne (also) fish-without
	a. I want vegetables without meat or fish.	(NC)
	b. not available: I want vegetables neither without meat nor without fish. (i.e., I
	want vegetables with both meat and fish.)	(*DN)

In this case, only the NC reading is available, bearing out the prediction that a NC reading must be obtained in a case in which the ne..ne -sIz complex is a constituent, and furthermore, the double negation reading must be obtained if the coordination is clausal.

This is a particularly striking result, given that the right-dislocation of a ne..ne phrase is known to always be accompanied with a main predicate that has sentential negation, since the constituent ne..ne phrase must be licensed by the closest negation operator available, i.e., sentential negation.11 This is, of course, not necessary if it has been licensed by -sIz instead.

One might be tempted to offer an alternative explanation in which -sIz carries [uNeg] and is licensed by a local negation, thus allowing that negation in (37) to scope over the ne..ne coordination, and thus license at once the ne..ne phrase and the two -sIz morphemes. However, this proposal is unlikely to work, as -sIz in no other configuration engages in NC. In (38-a) we see -sIz surviving without sentential negation, and in (38-b) we see it unable to engage in NC when sentential negation is present.

(38)	a.	Et-siz      sebze       ist-iyor-um.
		meat-without vegetables want-prog-1s.
		I want vegetables without meat.
	b.	Et-siz      sebze     iste-mi-yor-um.
		meat-without vegetable want-neg-prog-1s.
		(i) I don’t want vegetables without meat.
		(ii) unavailable: I don’t want vegetables with meat.

And we can check that this is not because of the unavailability of NC between sentential negation and an adverb of this type, as shown below for a typical NCI in the scope of ‘with’ (-lI), which can be licensed by negation.

(39)	Hiçbir	et-li	sebze	iste-mi-yor-um.
	no	meat-with	vegetables	want-neg-prog-1s.
	I don’t want any vegetables with meat.

Further Prediction

This proposal makes further predictions about the scope of -mA and -sIz above other coordinations. I asked one speaker who accepted NC readings in such a situation, and they tentatively accepted the following sentence, which is only felicitous if negation scopes above the conjunction:

(40)	Hem	ye-me-di-m	hem	iç-me-di-m,	sadece	ye-di-m.
	and	eat-neg-pst-1s	and	drink-neg-pst-1s	only	eat-pst-1s
	I didn’t both eat and drink, I only ate.						$\neg >\wedge$

This was not a steady judgment, and more investigation into it is necessary across speakers and configurations in order to see if it can be accessed in all cases.

3.5.3. The Contrast Between Forward Gapping and Overt Full Clausal Coordination

A preference was sometimes reported for forward gapping structures over full clausal coordination and forward gapping structures (except for one speaker who reported the opposite preference). I will not provide a full analysis of this observation, but it still calls for some explanation, as it is suggestive of an alternative solution to the results, namely, one in which negative markers are negative, but sentences like (27) and (28) are formed from small ne..ne phrases without ellipsis, combined with a negative verb, and one ne junct that can be extraposed to a final position, as shown below.

(41)	Ne Devrim [ne de Umut] hasta değil [ne de Umut].

First, this potential solution faces a direct violation of the Coordinate Structure Constraint, which prevents the extraction of a single junct. If we somehow allow this violation to occur in this case, such a solution would still be problematic because the identical operation would have to be blocked in the presence of a -mA marked predicate, since the forward gapping sentence in (26-a) (and corresponding structure with a subject coordination) does not have an NC reading. Thus, the following structure is not generated.

(42)	Unavailable: Ne Devrim [ne de Umut] gel-me-di [ne de Umut].

This suggests that the reason why the cases of the type [ne X değil ne Y] are not due to extraposition of the second ne junct [ne Y] (which would allow an explanation where [ne X ne Y] is a constituent that can be licensed by değil). This is thus another example of an alternative explanation to the data that cannot deal with differences between negation markers.

Instead, forward gapping structures with değil should be treated in the same way as with -mA: as the result of ellipsis of değil in the second junct, as is predicted from the syntax proposed in Jeretič (2022).

Why the NC interpretation with an overt değil in the second junct is more difficult than when it is covert remains to be understood. This effect could simply be due to a general dispreference for coordinations to have repeated elements in both juncts. In the case of değil and yok, these repeated elements are furthermore semantically vacuous, and do not seem to contribute to the grammaticality or meaning of the sentence in any way. Perhaps this is why (at least for some NC accepters), the sentences with değil or yok in the first junct are degraded in the first place. Adding another superfluous değil/yok thus makes it even more degraded.

3.6. Accounting for the NC Readings

To end this section, I provide the formal details of an analysis in which a given negation marker is non-negative. I illustrate it for değil, but it can be readily exported to yok and -mA.

When NC is available (except for cases with coordination of the -mA-negated verb), the sentential negation marker is non-negative, and has the same status in the NC system as other NC items like the negative indefinite hiçkimse (‘nobody’). When there is no NC available, the negation marker is negative and has to ability to license NC items: it is therefore as proposed for -mA in Jeretič (2022).

For the speakers that have double negation readings when değil and yok are in the scope of ne..ne, the analysis is the same as with -mA: the markers carry [iNeg] and thus, because they are in the scope of the ne operator, they cannot license it, and a covert negative operator merges above the coordination.

In cases in which değil and yok (or -mA, for one speaker) in the scope of ne..ne yield a NC reading, we simply assume that they are NCIs for those speakers, that is, they carry [uNeg], and are licensed by the negative operator c-commanding the ne..ne phrase. So, here is a derivation for a simple sentence with değil:

(43)	Hiçkimse	hasta	değil-di.
	nobody	sick	cop.neg-pst
	Nobody was sick.

The NegOp is licensed here thanks to the [uNeg] introduced by değil, present on the clausal spine (as opposed to those introduced by hiçkimse).

In clausal ne..ne phrases, the reason why NC is possible is that the NegOp need not appear in between the NeP and the NegP, as shown below.

(44)	Ne	Ali	hasta	değil	ne	Beste	<hasta değil>.
	ne	Ali	sick	cop.neg	ne	Beste	sick cop.neg
	Neither Ali nor Beste is sick.

Nothing prevents another NegOp to appear below the NeP, since it is licensed by değil’s [uNeg] on the clausal spine. This explains the double negation reading always available for this string for speakers who also allow NC readings.

The explanation is identical for the variation in the availability of NC readings for yok. The negativity of değil and yok appears to be independent, which is expected if they are lexicalized separately.

4. Some Cross-Linguistic Support for Our Conclusions

4.1. A Proof of Concept from Kazakh

In this section, I provide some recently discovered data from Kazakh that suggests non-negativity of its own negation marker while avoiding some of the concerns previously evoked. These are data first observed by Ótott-Kovács (2023), which show the interaction of sentential negation and the iterated disjunction ‘either..or’, also called ne..ne (and thus most certainly historically related to the Turkish ne..ne, though today corresponding in meaning with Turkish ya..ya).

An example is given below:

(45)	Ne	Ajʃa	bi	bile-di	ne	Ajnur	æn	ajt-tɯ.
	either	Aisha	dance	dance-pst.3	or	Ainur	song	say-pst.3
	a.  Only available: Aisha danced or Ainur sang.								$p\vee q$
	b.  Unavailable: Aisha danced and Ainur sang.								* $p\wedge q$
	c.  Unavailable: Aisha didn’t dance, nor did Ainur sing.								* $\neg (p\vee q)$

She shows that when the ‘either..or’ construction coordinates negated tensed clauses, negation can take semantic scope above the disjunction, as shown in (46-b) in addition to the expected lower scope, as shown in (46-a).

(46)	Ne	Ajʃa	bi	bile-me-di	ne	Ajnur	æn	ajt-pa-dɯ.
	either	Aisha	dance	dance-neg-pst.3	or	Ainur	song	say-neg-pst.3
	a.  Available: Aisha didn’t dance or Ainur didn’t sing.								$\neg p\vee \neg q$
	b.  Available: Neither did Aisha dance nor did Ainur sing.								$\neg (p\vee q)$

(46-b) cannot be taken to be the result of an inclusive reading of disjunction, otherwise we should expect (45-b) to be available. Thus, it can indeed only be the result of negation scoping above the disjunction. However, the negation marker in these sentences appears inside the scope of the coordination. Thus, for the semantic contribution of negation to take the scope above, Ótott-Kovács (2023) proposes that the negation marker is non-negative and agrees with a higher negative operator, as shown in the following structure.

(47)	[[ne Ajʃa bi bile-me[uNeg]-di ne Ajnur æn ajt-pa[uNeg]-dɯ] Op¬[iNeg]].

These data are independent of having NC between the negation and the ne..ne phrase and only involve the scope of semantic elements, thus forcing the explanation to not rely on the assumption that NC cannot occur between a semantic negation and a [uNeg] above it.

4.2. Russian ‘Neither..nor’ Coordinations

Russian, recall from Zeijlstra’s (2004) original classification, is a strict NC language (with base SVO word order). This means that all NCIs must co-occur with sentential negation. It also means that preverbal NCIs cannot be licensed by sentential negation, as they are in a higher position and thus cannot be licensed by it. This has led Zeijlstra (2004) to propose that sentential negation is in fact non-negative.

In light of the ‘neither..nor’ test introduced in this paper, this fact makes the prediction that negation can appear inside ‘neither..nor’ coordinations in Russian. This prediction is borne out as shown below.

(48)	Ni	Masha	*(ne)	pela,	ni	Olja	*(ne)	tancevala.
	nor	Masha	neg	sang	nor	Olja	neg	danced
	Neither did Masha sing nor Olja dance.

It is furthermore obligatory, providing support for an alternative source of its necessity in phrases with an NCI present, despite it appearing below the subject.

This fact provides support for both the scope of a ‘neither..nor’ as a test for the negativity of sentential negation, and Zeijlstra’s (2004) claim that negation in strict NC languages (with SVO word order) is non-negative.

5. Conclusions

In this paper, I have discussed the status of the negativity of negation in NC systems in SOV languages, proposing to shift our current understanding limited to SVO languages and inspecting this question for Turkish.

In particular, the surface NC pattern in SOV languages depends not only on whether the negation marker is negative but on the underlying position of subjects, which seems to be variable across SOV languages. Furthermore, if a language is found to have subjects at the vP level (i.e., under NegP) and a strict NC pattern, the negativity status of negation remains underdetermined.

Turkish appears to be such a language, namely with SOV word order, subjects that can remain below NegP, and a strict NC pattern. I introduced an independent diagnostic for the negativity of negation, namely the scope of ne..ne (“neither..nor”) phrases, which introduce a negative operator that is both clausemate and c-commanding the position of sentential negation. If embedding a negative marker in a ne..ne phrase yields NC, there are good reasons to believe that that marker is non-negative.

I showed that there is variation in the negativity status of negation across sentential negation markers and speakers of Turkish. While the plain negation -mA is almost always negative, copular değil and existential yok differ in their negativity across speakers. I suggest that the variability of değil and yok is due to the lack of sufficient information in the input. Furthermore, when learners are faced with the task of assigning negative semantics or not to the negative marker, they do not converge on one possibility, because there are competing pressures: one that prefers for an operator to have an overt exponent, and one that prefers for an exponent to be linked to only one semantic operator. These two opposing preferences can possibly be the reason for the variable typology of negative indefinites: either they expone both negation and an existential quantifier, as in non-NC languages, or only the existential quantifier, as in NC languages.