2.1. Negatively Marked Elements in Turkish
Turkish has a set of negatively marked elements requiring licensing by negation. A class of these are formed with the morpheme
hiç ‘n-at all’ appearing on its own or as part of a paradigm of morphologically complex expressions, such as
hiçbir şey ‘n-one thing’ (7). In examples like the latter,
hiç is written together with
bir ‘one’ by orthographic convention. Some expressions, like
asla ‘never’ and
katiyyen ‘under no circumstances’ do not include the morpheme
hiç but nevertheless show similar licensing restrictions (also see
Kelepir 2001).
(7) | a. | hiç | ‘n-at all, n-ever, n-amount’ |
| b. | (hiç) kimse | ‘n-person’6 |
| c. | hiçbir şey | ‘n-thing’ |
| d. | hiçbir yer | ‘n-place’ |
| e. | hiçbir zaman | ‘n-time’ |
There are no (near) homonyms to
hiç-expressions; the items in (7) represent the entire set of items corresponding to both typical English NPIs like
ever and
any x and English negative quantifiers like
never and
no x. This is irrespective of factors such as stress, unlike, for example, in Greek (
Giannakidou 2000). If the
hiç morpheme is missing, these elements are plain indefinites:
(8) | a. | birisi/birileri | ‘someone’ (lit: one of that/them) |
| b. | bir şey | ‘something’ |
| c. | bir yer | ‘some place’ |
| d. | bir ara | ‘some time’ (lit: one interval) |
Turkish has only a limited set of expressions that require licensing by negation in NPI-like fashion, other than
hiç-expressions. One such case is minimizers, which appear to be NPIs with a concealed
even connotation, as in
tek kuruş ‘single dime’ and
bir Allah’ın kulu ‘a soul (lit: one servant of god)’. (9) shows that each of these three kinds of negative dependents requires sentential negation. (9a) has a
hiç expression, (9b) includes
katiyyen ‘under no circumstances’, and (9c) contains a minimizer NPI. Sentential negation is expressed by the suffix
–mA in the verbal complex.
(9) | a. | Recep hiçbir şey oku*(-ma)-z. |
| | Recep n-one thing read-neg-aor |
| | ‘Recep doesn’t read anything.’ |
| b. | Kedi katiyyen dışarı çık*(-ma)-yacak. |
| | cat under.n.circumstances out go-neg-fut |
| | ‘The cat will under no circumstances go outside.’ |
| c. | Bütün gün dükkan-da dur-du-m ama bir Allah’ın kul-u gel*(-me)-di. |
| | all day shop-loc stay-past-1sg but one god-gen servant-poss3sg come-neg-past |
| | ‘I was at the shop all day but not a soul stopped by.’ |
We focus in this section on the first class of these negation-requiring items with hiç, and argue that they are so-called neg-words or NCIs, not standard NPIs. To support this, we will draw upon comparisons between hiç-expressions and minimizer NPIs, among other evidence.
2.2. NPI Licensing and Negative Concord in European Languages
The main difference between NPIs and neg-words is that the latter, but not the former, can induce negation on their own in particular contexts, for instance, in fragment answers (see
Giannakidou 2006;
Giannakidou and Zeijlstra 2017). To set up the comparison, we will first look at the behavior of neg-words and plain NPIs in other European languages.
Typical NPIs, such as English
anybody and
ever, are indefinites. They do not encode negativity in themselves but generally require licensing by DE or NV environments, including negation (10) (see
Ladusaw 1992;
Giannakidou 2000). The fact that they do not encode negation is manifested by the inability to answer a question negatively on their own (11).
(10) | I will *(not) meet anybody this afternoon. |
(11) | A: Who did you say you will meet? |
| B: #Anybody. (intended: ‘Nobody.’) |
By contrast, negative quantifiers such as English
nobody do not require (in fact, reject in standard varieties unless double negation is intended) licensing by a negative marker and always encode negativity by themselves.
(12) | I will (*not) meet nobody this afternoon. |
| (intended: ‘I will not meet anybody this afternoon.’) |
(13) | A: Who did you say you will meet? |
| B: Nobody. |
Neg-words are elements that, in certain contexts, require licensing by negation, just like standard NPIs, while in other contexts they behave like negative quantifiers, giving rise to NC effects. NC is the phenomenon where multiple elements that in certain configurations yield semantic negation of their own, jointly yield only one. English does not employ neg-words; it is not an NC language (at least in the standard variety), but many languages do. One such language is Italian. Italian
nessuno is best translated as ‘anybody’ in (14a), but as ‘nobody’ in (14b). (14c) shows that
nessuno is fine in a fragment answer, inducing negation on its own.
(14) | a. | Non ha telefonato a nessuno. | |
| | Not has called to n-body | |
| | ‘She hasn’t called anybody.’ | |
| b. | Nessuno ha telefonato. | |
| | n-body called | |
| | ‘Nobody called.’ | |
| c. | A: A chi ha telefonato? | |
| | to whom has called | |
| | ‘Whom has she called?’ | |
| | B: A nessuno. | |
| | to n-body | |
| | ‘Nobody.’ | (Zeijlstra 2004) |
In languages like Italian, postverbal neg-words must be licensed by a preverbal negative element (either the negative marker or a preverbal neg-word). Conversely, preverbal neg-words may not be licensed by the negative marker, as shown in (15). NC languages with such an asymmetry are called Non-strict NC languages.
(15) | a. | *(Non) ha telefonato a nessuno. | |
| | Not has called to n-body | |
| | ‘She hasn’t called anybody.’ | |
| b. | Nessuno (*non) ha telefonato. | |
| | N-body called | |
| | ‘Nobody called.’ | (Zeijlstra 2004) |
Not all languages with neg-words have such restrictions with respect to the position of the neg-word. In so-called Strict NC languages, illustrated by Czech in (16), the negative marker must always accompany a neg-word, irrespective of the position of the neg-word in the clause, unlike in Italian. Only in fragment answers can neg-words induce semantic negation on their own (c).
(16) | a. | Dnes nikdo *(ne-)volá | |
| | today n-body neg-calls | |
| | ‘Today nobody calls.’ | |
| b. | Dnes *(ne-)vola nikdo. | |
| | Today neg-calls n-body | |
| | ‘Today nobody calls.’ | |
| c. | A: Co jsi viděla | |
| | What aux.2sg saw. sg.f | |
| | ‘What did you see?’ | |
| | B: Nic. | |
| | n-thing | |
| | ‘Nothing.’ | (Radek Šimík, p.c.) |
Next to the fact that plain NPIs can never induce semantic negation, but neg-words can (in any case in fragment answers), there is another characteristic difference between standard NPIs and neg-words. Most NPIs can be licensed, in addition to negation, in contexts such as polar and content questions and antecedents of conditionals (
Ladusaw 1979;
Giannakidou 2000;
Giannakidou and Zeijlstra 2017 a.o.).
7 Neg-words, however, are only licensed by the negative marker (
Zeijlstra 2004) and, in some cases, in polar questions. Hence, an English NPI like
any is licensed in content questions (17a), whereas neg-words like
nic ‘n-thing’ in Czech are not (17b).
8(17) | a. | Who noticed anything suspicious? | |
| b. | *Kdo viděl nic? | |
| | who saw.sg.m n-thing | |
| | (‘Who saw anything (at all)?’) | (Radek Šimík, p.c.) |
Another notable difference between standard NPIs and neg-words is that the licensing of neg-words is subject to syntactic locality, but NPI licensing is not. For this reason, the English NPI
any is licensed in an adjunct island (18a), but Italian
niente is not (18b):
(18) | a. | Gianni doesn’t work in order to earn any money. | |
| b. | *Gianni non labora per guadagnare niente argente. | |
| | Gianni not works in.order.to earn no money | |
| | (‘Gianni doesn’t work in order to earn any money.’) | (Zeijlstra 2022) |
2.3. NPI Licensing and Negative Concord in Turkish
The first of our claims regarding Turkish negative dependencies is that Turkish expressions with
hiç are neg-words and not plain NPIs (
contra Kelepir 2001); hence, Turkish is an NC language. It is, in fact, a Strict NC language, as the subject position is also licensed by sentential negation (19b).
(19) | a. | Bugün hiçkimse-yle toplantı yap*(-ma)-yacağ-ım. |
| | today n-body-com meeting do-neg-fut-1sg |
| | ‘I will meet nobody today.’ |
| b. | Hiçbir öğrenci toplantı-ya gel*(-me)-di. |
| | n-one student meeting-dat come-neg-past |
| | ‘No student came to the meeting.’ |
Evidence for this comes from the fact that neg-words require a negative marker as a licenser (19), similar to NPIs, but can appear on their own in a fragment answer with the meaning of a negative quantifier (also see
Kelepir 2001) (20).
(20) | A: | Kim-inle toplantı yap-acak-tı-n? |
| | Who-com meeting do-fut-past-2sg |
| | ‘Who did you say you were meeting?’ |
| B: | Hiçkimse-yle. |
| | n-body-com |
| | ‘With nobody.’ |
Turkish neg-words further behave on par with NC items in that they require licensing by the negative marker and not just any DE/NV context, as shown in (21) below (also see
Kelepir 2001). Contexts such as content questions and antecedents of conditionals do not license
hiç-expressions; the negative marker is required.
(21) | a. | *Kim şüpheli hiçbir şey fark et-ti? |
| | who suspicious n-one thing notice do-past |
| | (‘Who noticed anything suspicious?’) |
| b. | *şüpheli hiçbir şey fark ed-er-se-niz biz-i ara-yın. |
| | suspicious n-one thing notice do-aor-cond-2pl we-acc call-imp.2pl |
| | (‘Give us a call if you notice anything suspicious.’) |
Kelepir (
2001) notes these facts concerning the ability of
hiç-expressions to give rise to semantic negation in fragment answers and their inability to occur in NPI-licensing contexts other than negation, and discusses them in the context of whether they are NPIs (like
anybody) or negative quantifiers (like
nobody). She argues that the fact that these expressions do not induce double negation readings under negation (nor under multiple hiç-expressions;
Kelepir 2001, p. 160) and the fact that they are licensed by
–sIz ‘without’ are evidence that they are not negative quantifiers. She briefly considers and dismisses the possibility that they are neg-words, based on the fact that
hiç-expressions require licensing also in the subject position, unlike Romance-type neg-words (as in 19b). However, if Turkish is a Strict NC language, as we claim, this argument would be moot. We give below two further arguments based on a contrast between Turkish minimizers and
hiç-expressions to substantiate our claim that the latter are not licensed on par with NPIs in the same language.
First, consider the syntactic contexts in which minimizer NPIs are licensed in Turkish. Unlike
hiç-expressions, these expressions cannot yield negation and hence cannot appear in fragment answers (22). Also unlike
hiç-expressions, they are fine in content questions and conditionals (23).
(22) | A: | Kim-inle toplantı yap-acak-tı-n? |
| | Who-com meeting do-fut-past-2sg |
| | ‘Who did you say you were meeting?’ |
| B: | *Bir Allah’ın kulu-yla. |
| | one god-gen servant-com |
| | ‘*With a soul.’ |
(23) | a. | Kim tatil-de bir Allah’ın kul-u-yla görüş-tü? |
| | who holiday-loc one god-gen servant-poss3sg-com meet.up-past |
| | ‘Who met a soul during the holidays?’ |
| b. | Bun-u bir Allah’ın kul-un-a söyle-r-se-n sen-in-le |
| | this-acc one god-gen servant-dat say-aor-cond-2sg you-gen-com arkadaşlığ-ı kes-er-im. |
| | friendship-acc stop-aor-1sg |
| | ‘If you tell a soul about this, I will stop being friends with you.’ |
Another difference between Turkish minimizer NPIs and neg-words is their ability to appear in non-negated
ne…ne… ‘neither nor’ constructions, as exemplified in (24b). This is one of the very few documented constructions in Turkish that express negation without requiring the negative marker on the verb.
9 In failing to mark negation on the verb, the construction is similar to the English
neither…nor… construction, both in meaning and structure (see
Jeretič 2022;
Gračanin-Yüksek 2023).
First, contrast the negated (24a) and non-negated (24b)
ne…ne… constructions. The non-negated construction that we are interested in needs to carry focus (cf.
Şener and İşsever 2003). We indicate this by underlining the rightmost word of the
ne…ne… sequence where the resulting nuclear pitch accent falls (see 24b). In the negated construction in (24a), the nuclear pitch accent falls on the verbal complex, where the negative morpheme is found. In the non-negated construction in (24b), there is no negation on the verb, yet the interpretation is negative, as in (24a).
(24) | a. | Ne Emre ne Merve okul-a git-me-di. |
| | NE Emre NE Merve school-dat go-neg-past |
| | ‘Neither Emre nor Merve went to school.’ |
| b. | Ne Emre ne Merve okul-a git-ti. |
| | NE Emre NE Merve school-dat go-past |
| | ‘Neither Emre nor Merve went to school.’ |
Negated
ne…ne… constructions like in (24a) should straightforwardly provide a licensing context for negative dependencies due to the presence of the negative marker. This is indeed true for all negative dependents, including
hiç-expressions (25).
(25) | Ne Emre ne Merve hiçbir yer-e git-me-di. |
| NE Emre NE Merve n-one place-dat go-neg-past |
| ‘Neither Emre nor Merve went anywhere.’ |
The non-negated construction, however, has limited licensing powers. It can license minimizer NPIs, as in (26a).
10 Hiç-expressions, on the other hand, are not licensed by the non-negated
ne…ne… construction (26b). This suggests, again, that
hiç-expressions are genuine neg-words requiring the negative marker in the clausal spine.
(26) | a. | (?)Ne Emre ne Merve tek kuruş harca-dı. |
| | NE Emre NE Merve single dime spend-past |
| | ‘Neither Emre nor Merve spent a dime.’ |
| b. | *Ne Emre ne Merve hiç para harca-dı. |
| | NE Emre NE Merve n-amount money spend-past |
| | (‘Neither Emre nor Merve spent any money.’) |
All of this indicates that Turkish expressions with
hiç are neg-words. Unlike other kinds of NPIs, they may appear without negation in fragment answers, yet they need to be licensed otherwise. And, again unlike other NPIs, they need to be licensed strictly by clausal negation. Other DE/NV contexts or the likes of the non-negated
ne…ne… construction cannot license them.
11 This indicates that this dependency does not involve regular NPI licensing but rather NC (see
Ladusaw 1992;
Giannakidou 2000;
Zeijlstra 2004, and others).
One point of discussion, though, concerns polar questions. We have noted that in Turkish and elsewhere, neg-words cannot appear in content questions. In many NC languages, they are also banned from polar questions. Japanese is such an example (see
Giannakidou 2000 for more examples). Negation is required to license the neg-word
nani-mo ‘n-thing’ (lit.
what-also) in (27a). Content questions are not licensing environments (27b). Neither are polar questions (27c).
(27) | a. | Taroo-wa nani-mo mie*(-naka)-tta. | |
| | Taroo-top n-thing see-neg-past | |
| | ‘Taroo didn’t see anything.’ | |
| b. | *Dare-ga nani-mo mi-mashi-ta-ka? | |
| | who-nom n-thing see-hon-past-q | |
| | (‘Who saw anything?’) | |
| c. | *Taroo-wa nani-mo mi-mashi-ta-ka? | |
| | Taroo-top n-thing see-hon-past-q | |
| | (‘Did Taroo see anything?’) | (Daiki Matsumoto, p.c.) |
While we have seen that Turkish presents the same pattern as Japanese with respect to declaratives and content questions, the situation with polar questions is different. Polar questions constitute an NC context in Turkish (28a; also see
Kelepir 2001). However, this is only true for a subset of polar question configurations. The focus-sensitive polar question clitic must be on the verbal complex to license NC. In other placements (indicated in triangular brackets), licensing is not possible (28b) (
Kamali 2011).
(28) | a. | Hiçkimse yemek yap(-ma)-dı mı? |
| | n-body dinner make-neg-past pq |
| | ‘Did(n’t) anyone make dinner?’ |
| b. | Hiçkimse <mi> yemek <mi> yap*(-ma)-dı? |
| | n-body pq dinner pq make-neg-past |
| | (‘Did NOONE make dinner?’) |
| | (‘Did noone make DINNER?’) |
This begs the question as to why neg-words may appear in this particular polar question configuration. It is known from Non-strict NC languages (e.g., Spanish/Italian;
Zeijlstra 2004) that neg-words may appear in polar questions, but this has not been reported for Strict NC languages. We leave this question open for further research (but see
Kamali and Matsumoto, forthcoming, for a recent proposal).
2.4. The Domain of Negative Concord in Turkish
Before drawing a final conclusion, we would like to briefly investigate the locality of NC in Turkish. Long-distance licensing of NC is generally considered impossible (
Giannakidou 2000;
Zeijlstra 2004), but certain matrix predicates and their morphosyntactic associates have been shown to provide transparent domains (see, for example,
Giannakidou and Quer 1997). Turkish
hiç-expressions, on the other hand, appear to be licensed across clause boundaries with a few exceptions.
Turkish uses a few strategies to embed clauses. The standard method is nominalization, where the embedded subject carries the genitive case instead of the nominative, and the predicate carries a nominalizing morpheme. The internal structure of the nominalization is not fully morphologically transparent, but possessive agreement inside and case marking outside the embedded clause clearly indicate that it is nominalized (29a). To see this, compare (29a) with root embedding in (29b), which is also grammatical under certain matrix predicates, potentially with an accusative-marked embedded subject.
(29) | a. | Biz kedi-nin su iç-tiğ-in-i san-ıyor-uz. |
| | we cat-gen water drink-nomin-poss3sg-acc think-pres-1pl |
| | ‘We think that the cat is drinking water.’ |
| b. | Biz kedi(-yi) su iç-iyor san-ıyor-uz. |
| | we cat-acc water drink-pres think-pres-1pl |
| | ‘We think the cat is drinking water.’ |
It has been known since
Kornfilt (
1984) that some embedded clauses in Turkish are transparent to licensing by matrix negation, while others are opaque.
Kelepir (
2001) shows that the complement clauses of a set of verbs, including neg-raising verbs (30), perception verbs (31), and attitude verbs (32), are transparent to the long-distance licensing of
hiç-expressions. However, factive verbs introduce opaque domains (33). The same holds for root embeddings (34).
12(30) | Biz kedi-nin hiçbir şey iç-tiğ-in-i/-e |
| We cat-gen n-one thing drink-nomin-poss3sg-acc/-dat |
| san-m-ıyor-uz/inan-m-ıyor-uz. |
| think-neg-pres-1pl/believe-neg-pres-1pl |
| ‘We don’t think/believe that the cat is drinking anything.’ |
(31) | Biz kedi-nin hiçbir şey iç-tiğ-in-i |
| we cat-gen n-one thing drink-nomin-poss3sg-acc |
| gör-me-di-k/duy-ma-dı-k. |
| see-neg-past-1pl/hear-neg-past-1pl |
| ‘We didn’t see/hear the cat drink anything.’ |
(32) | Biz kedi-nin hiçbir şey iç-me-sin-i |
| We cat-gen n-one thing drink-nomin-poss3sg-acc |
| iste-m-iyor-uz/onayla-m-ıyor-uz. |
| want-neg-pres-1pl/approve-neg-pres-1pl |
| ‘We don’t want the cat to drink anything.’ |
| ‘We don’t approve of the cat’s drinking anything.’ |
(33) | *Biz kedi-nin hiçbir şey iç-tiğ-in-i |
| we cat-gen n-one thing drink-nomin-poss3sg-acc |
| öğren-me-di-k/bil-m-iyor-uz. |
| learn-neg-past-1pl/know-neg-pres-1pl |
| ‘*We didn’t find out that the cat was drinking anything.’ |
| ‘*We don’t know that the cat is drinking anything.’ |
(34) | ?*Biz kedi(-yi) hiçbir şey iç-ti san-m-ıyor-uz. |
| We cat-acc n-one thing drink-past think-neg-pres-1pl |
| ‘We don’t think the cat drank anything.’ |
We may add that complement clauses of reportative verbs, which are neither neg-raising nor factive, are also transparent to NC.
(35) | Biz kedi-nin hiçbir şey iç-tiğ-in-i |
| we cat-gen n-one thing drink-nomin-poss3sg-acc |
| söyle-m-iyor-uz/iddia et-m-iyor-uz. |
| say-neg-pres-1pl/claim do-neg-pres-1pl |
| ‘We are not saying/claiming that the cat is drinking anything.’ |
Locality domains such as adjunct islands still create opaque domains for NC, though, for reasons not properly understood by us, in certain adjuncts
hiç-items may appear. (36) provides two examples under the same adverbial clause with differing acceptability.
(36) | a. | *Ahmet hiçbir yer-e git-sin diye para kazan-m-ıyor-uz. |
| | Ahmet n-one place-dat go-opt so.that money earn-neg-pres-1pl |
| | (‘We do not earn Money so that Ahmet travels anywhere.’) |
| b. | Hiçkimse onayla-sın diye çalış-m-ıyor-uz. |
| | n-body approve-opt so.that work-neg-pres-1pl |
| | ‘We aren’t working so that anybody approves.’ |
This short review suggests that long-distance licensing of NC is the rule rather than the exception in Turkish. We leave it to future research to explore why this may be the case and what may connect factive domains, root embedding, and certain adverbial clauses in their inability to license long-distance NC.
2.5. Are Turkish Neg-Words Universals or Existentials?
One of the central questions in the domain of neg-words and NPIs concerns whether neg-words/NPIs are universals or indefinites/existentials. NPIs are widely considered to be existentials licensed under a licensing operator such as negation (starting with
Ladusaw 1979). However, it is also possible that they are universals obligatorily outscoping negation. This possibility arises because it is normally impossible to tease apart the narrow scope existential constellation ¬ > ∃ from the wide scope universal constellation ∀ > ¬, as the two are truth-conditionally equivalent (37) (see
Giannakidou 2000;
Shimoyama 2011).
(37) | Sue didn’t meet anybody. |
| ¬∃x.[Human(x) & Meet(s,x)] |
| ∀x.[Human(x) →¬Meet(s,x)] |
In the domain of neg-words,
Ladusaw (
1992) and
Zeijlstra (
2004) assume the existential analysis, whereas
Szabolcsi (
1984) and
Giannakidou (
2000) make a case for the universal analysis.
Shimoyama (
2011) has argued that Japanese NPIs are universals, but as Shimoyama’s primary data are argued to be neg-words by others (
Watanabe 2004 a.o.), we assume her analysis to be applicable to neg-words.
Giannakidou (
2000,
2006) proposes a set of diagnostics, such as modification by
almost/absolutely, the inability to bind donkey-pronouns, and the absence of neg-words in predicate nominals, to distinguish universals from existentials. For instance,
I saw almost somebody is strongly degraded, while
I saw almost everybody is fine. Also, universals, unlike existentials, do not bind donkey anaphora:
The students that have {somethingi/anythingi/*nothingi/*everythingi} to say should say iti now. Similarly, universals cannot appear in predicate nominals, but existentials can:
Mary is {a/*every} doctor. Then, she shows that neg-words in Greek pattern with universals like
every x and not with existentials like
some x in English in these respects, indicating that they must be universals.
Shimoyama (
2011) introduces fixed scope quantificational elements into the picture to sidestep the equivalence illustrated in (37). Suppose Q
+ is a scope-taking element that outscopes negation. With such an element, only universal NPIs can give rise to the scope construal ∀ > Q
+ > ¬, because this construal is not equivalent to any construal involving Q
+, ¬ and ∃. By contrast, with Q
−, a scope-taking element that takes scope below negation, only existential/indefinite NPIs can give rise to the scope construal ¬ > Q
− > ∃, as it is not equivalent to any construal involving ¬, Q
− and ∀.
Shimoyama then tests the quantificational status of Japanese NPIs based on examples like (38). The adverb
hudan ‘usually’ is a Q
+ adverb: it has to outscope negation. With the neg-word
dare-mo in the subject position, we observe the decisive reading ∀ > Q
+ > ¬, which indicates that the neg-word is a universal.
13 The reading Q
+ > ¬ > ∃ (or the equivalent Q
+ > ∀ > ¬) may be present as well, but this reading is uninformative exactly because it has existential and universal equivalents.
(38) | Nihonzin gakusei-no dare-mo hudan(-wa)/taitei sankasi-nakat-ta. |
| Japanese student-gen n-body usually-top/mostly participate-neg-past |
| ‘For every Japanese student, it was usually the case that they did not participate.’ |
On the flipside, the construal ¬ > Q− > ∃ would constitute evidence for the existential analysis. To demonstrate the quantificational status of an NPI unequivocally, one must test this diagnostic reading too, this time with a Q− adverb.
Now, we first check the quantificational status of Turkish
her-expressions based on Giannakidou’s diagnostics (
Giannakidou 2000,
2006). Here, we apply two tests: modification with
almost and the occurrence of predicate nominals. The test involving donkey anaphora cannot be applied to Turkish as these require relative clauses where neg-words are independently banned in Turkish. Other diagnostics by Giannakidou are more language-specific and only hold for Greek or are no longer theoretically up to date (in the case of diagnostics based on syntactic locality
14 or existential commitment
15). The two diagnostics we apply below point in the direction of neg-words being universals:
(39) | a. | Hemen hemen hiçbir öğrenci ders-e gel-me-di. |
| | almost n-one student class-dat come-neg-past |
| | ‘Almost no student came to class.’ |
| b. | *Leyla hiçbir doktor değil. |
| | Leyla n-one doctor not |
| | (‘Leyla is no doctor.’) |
As seen, in Turkish, the neg-word may be modified by
almost and may not appear in a predicate nominal. These are universal-like characteristics:
every can also be modified by
almost and cannot appear in a predicate nominal, whereas the facts are reversed for the existential
some. Hence, these data would suggest that Turkish neg-words are universals.
16Next, let us look at Shimoyama’s diagnostics. Turkish and Japanese share important morphosyntactic features such as SOV constituent order and wh-in-situ, in addition to being Strict NC languages. In the context of this discussion, a particularly relevant parallelism is the relative scope between the subject position and negation. In Korean and partly also Japanese, the subject position is typically interpreted outside of the scope of negation (
Shimoyama 2011). Turkish has this feature as well. Consider (40). If we have a class of ten students and three of them skipped class, this sentence can be truthfully uttered, indicating that the subject outscopes negation. In the reading where the subject takes scope under negation, seven to ten students are said to have skipped class, where (40) would not be a true statement.
17(40) | Üç veya daha fazla öğrenci ders-e gel-me-di. | |
| three or more many student class-dat come-neg-past | |
| ‘For three or more students, they did not come to class.’ | 3 or more > ¬ |
| ??’It is not the case that three or more students came to class.’ | ??¬ > 3 or more |
According to Shimoyama, if the subject position typically outscopes negation, there is reason to suspect that a subject neg-word may be outscoping negation as well. Structures with a neg-word subject should then potentially yield a reading with a universal outscoping negation (∀ > ¬), as Shimoyama claims to be the case in Japanese, and not a reading with an existential under the scope of negation (¬ > ∃).
Let us then see what Shimoyama’s tests for neg-words reveal for Turkish
hiç-expressions. Turkish also has quantificational adverbs with a preferred Q
+ (Q > ¬) reading such as
çoğu zaman ‘mostly’, just like the Japanese
hudan. The reading where the adverb scopes under negation requires untypical prosody if at all possible (see Note 17).
(41) | Ahmet çoğu zaman bisiklet-e bin-m-iyor. | |
| Ahmet most times bicycle-dat ride-neg-pres | |
| ‘Ahmet doesn’t ride a bike most of the time.’ | mostly > ¬ |
| ??‘It is not most of the time that Ahmet rides a bike.’ | ??¬ > mostly |
In the Japanese examples considered by Shimoyama, the negative element in the subject position scopes over the quantificational adverb, hence automatically over the negation that the adverb scopes over, leading to the ∀ > Q+ > ¬ reading that provides evidence for the high scope universal analysis.
In Turkish, we do not find this to be the case. (42), which now has a neg-word subject, is marginally acceptable to begin with, and in the allowed prosody (to the extent that it could be controlled), the critical ∀ > Q > ¬ reading is absent.
18(42) | ??Hiçbir komşu-m çoğu zaman bisiklet-e bin-m-iyor. | |
| n-one neighbor-poss1sg most times bicycle-dat ride-neg-pres | |
| ??‘No neighbor of mine rides a bike most of the time.’ | ??¬ > ∃ > Q |
| Not: ‘For all my neighbors, most of the time they do not ride a bike.’ | *∀ > Q > ¬ |
On the flipside, an adverb that supports a Q
− (¬ > Q) reading can provide evidence for the narrow scope existential analysis by providing (or failing to provide) the decisive construal ¬ > Q > ∃. One such adverb in Turkish is
çok ‘much’.
(43) | Ahmet çok ye-me-di. | |
| Ahmet much eat-neg-past | |
| ‘Ahmet didn’t eat much.’ | ¬ > Q |
| ??‘There is much Ahmet didn’t eat.’ | ??Q > ¬ |
Now, with a neg-word subject, if we get a ¬ > Q
− > ∃ construal, then we can safely conclude that the neg-word has an existential interpretation. As shown in (44), this reading is indeed available (in addition to the non-decisive ¬ > ∃ > Q
−).(44) | Hiçkimse çok ye-me-di. | |
| N-body much eat-neg-past | |
| ?‘It’s not much that was eaten.’ | ¬ > Q > ∃ |
So, based on Shimoyama’s tests, scope-sensitive quantificational adverbs in Turkish provide evidence for the existential analysis and not for the universal analysis: they do not allow the ∀ > Q > ¬ reading (with Q+ elements) that would be evidence for the universal analysis. In fact, they allow the ¬ > Q > ∃ reading (with Q− elements), which is evidence for the existential analysis.
Let us also test this with modals. Just like adverbials, modals can be employed as the disambiguating Q element in these tests. As we will see in detail in
Section 4, this is a particularly clear test because Turkish modals always take fixed scope with respect to negation, leading to unambiguous scope construals. Below, what we gloss as
abil1 is a Q
− element, as it has to scope below negation (45a).
Abil2, on the other hand, has to scope above negation (45b).
19(45) | a. | Dans ed-e-me-z-sin. | |
| | dance do-abil1-neg-aor-2sg | |
| | ‘You may not dance.’ | ¬ > ◊/*◊ > ¬ |
| b. | Dans et-me-yebil-ir-sin. | |
| | dance do-neg-abil2-aor-2sg | |
| | ‘You are allowed not to dance.’ | ◊ > ¬/*¬ > ◊ |
Now, let us take the two instances of existential modal in (45a–b) for our test. If the Q
− modal leads to a ¬ > ◊ > ∃ reading with a neg-word subject, this is evidence for a narrow scope existential analysis of the neg-word. Indeed, we do find this evidence. This sentence is ambiguous between a total prohibition and an individual prohibition reading, the first of which exemplifies the diagnostic configuration.
(46) | Hiçkimse dans ed-e-me-z. | |
| n-body dance do-abil1-neg-aor-2sg | |
| ‘Nobody may dance.’ | ¬ > ◊ > ∃ |
| ?‘There is no particular individual who may dance. (Mary because her knee is injured, John because he is grounded, etc.)’ | ?¬ > ∃ > ◊ = ∀ > ¬ > ◊ |
If, on the other hand, the Q
+ modal
abil2 leads to a ∀ > ◊ > ¬ reading with a neg-word subject, this would be evidence for a wide scope universal analysis of the neg-word. However, this reading is marginal if at all acceptable. The sentence has the total permission reading, and not a reading where for each individual there is separate permission granted not to dance.
(47) | Hiçkimse dans et-me-yebil-ir. | |
| n-body dance do-neg-abil2-aor-2sg | |
| ‘It is permitted that nobody dances.’ | ◊ > ∀ > ¬ = ◊ > ¬ > ∃ |
| ??‘Each is individually permitted not to dance.’ | ??∀ > ◊ > ¬ |
Hence, the data on modals and quantificational adverbs converge. Shimoyama’s wide scope universal analysis of Japanese negative elements is not tenable for Turkish neg-words. According to these tests, Turkish neg-words behave like existentials.
At this stage, the results are conflicting. Giannakidou’s diagnostics point in the direction of Turkish
hiç-expressions being universals; Shimoyama’s, in the direction of existentials. We can therefore not draw a safe conclusion here, though
Penka (
2011) and
Gajić (
2016) have argued that some of Giannakidou’s diagnostics for a universal analysis, for instance,
almost-modification, are also possible in languages where neg-words are clearly existential (e.g., Italian or Bosnian-Serbian-Croatian), casting doubt on the validity of this test.