Negative Polarity in Turkish: from negation to nonveridicality

This paper is concerned with the Turkish negative polarity items (NPIs henceforth) and the nature of various contexts in which they are licensed. Based on new data, it is shown that NPIs in Turkish are not only licensed by sentential negation and yes/no questions, they also appear in protasis of conditionals and along with the adjectival suffix –sIz ‘without’. It is argued that the traditional views treating NPIs as appearing only with negation or in downward entailing environments (Fauconnier, 1978; Ladusaw, 1980, 1996), or proposing a hierarchy of NPIs purely based on negation (van der Wouden, 1997; Zwarts, 1996) do not fully capture the distribution of the NPIs in the language. Instead, I propose that a more exhaustive analysis which subsumes negation and questions under its paradigm is provided by adopting the semantic notion of nonveridicality (Giannakidou, 1998, 2002, 2011) where the truth conditions of the proposition in which NPIs appear is not entailed.


Introduction
Negative polarity items (NPIs) are generally defined as words that generally require the presence of another element in order to be licensed in a sentence. The entity that licenses NPIs across languages is usually sentential negation or a negative quantifier (e.g. not or no one in English) even though NPIs can also be licensed by other elements. Consider (1) and (2).
(1) a. John did not have any friends. b. *John had any friends.
(2) a. John did not ever go there. typically co-occur with negation in the sentence. Their occurrence in affirmative sentences, on the other hand, will lead to ungrammaticality, as in (1b) and (2b).
Note that negation is not always sufficient for an NPI to be allowed in the sentence as NPIs are subject to certain structural restrictions. This is shown below.
(3) a. John did not see anyone.
b. *Anyone did not see John.
(4) a. John did not by anything.
b. *Anything was not bought by John.
The NPIs anyone and anything in (3a) and (4a) are licensed by negation and the sentences are grammatical. However, the sentences in (3b) and (4b) are ungrammatical even though negation is present in the structure. This is due to the structural requirement that the negative element must appear higher (i.e. precede) than NPIs.
As was already noted above, negative contexts are not the only environments in which NPIs are allowed to appear. Consider the following sentences. (7) She is taller than he ever imagined.
The fact that the NPIs appear in the structures above without leading to ungrammaticality indicates that they are also allowed to appear in conditionals, yes/no questions and comparatives in English.
In Turkish, too, there are a number of words whose distribution is rather limited in the sense that they appear only in certain environments, excluding others. The elements hiç 'never' in (8a) and kimse / hiçkimse 'no one' in (9a) are typical NPIs in Turkish since they require the presence of negation in the structure. The absence of negation in these sentences leads to ungrammaticality, as in (8b) and (9b). However, as in the case of English, it is not only negation that can license NPIs in the language. NPIs can also occur in yes/no questions, conditional sentences and along with the adjectival suffix -sIz which means without. This is illustrated below.
there-LOC no one-ACC see-AOR-COND we-DAT news give 'Let us know, if you see anyone there.' (12) a. Ben kimse-siz-im.
I no one-without-1SG 'I am without anyone / I don't have anyone around me.' →It is not the case that I brought here something.
'Without being seen by anyone, they entered the bank.' The data above clearly show that NPIs in Turkish can appear in a variety of environments and their distributions are not restricted to sentential negation. The question that arises then is why do NPIs occur only in structures with sentential negation, yes/no questions, conditionals and structures with the adjectival suffix -sIz? To put it differently, what are the characteristics of Turkish NPIs and how would their distribution accounted for in a unified manner? In this paper, I will address these questions in an attempt to provide a full account of the nature of NPIs in the language.
The structure of the paper is as follows: Section 2 provides an exhaustive list of NPIs and contexts in which NPIs occur in Turkish. This section also shows why previous accounts fall short to provide a uniform analysis. Section 3 is an overview of prominent approaches to the NPI phenomena across languages and an illustration of why they do not fully capture the facts. Section 4 introduces the framework namely nonveridicality, that is adopted to account for the distribution and nature of NPIs in Turkish. Section 5 briefly concludes the paper and makes certain suggestions for future work.

NPIs in Turkish
As noted above, there are certain words in Turkish whose behavior is quite similar to NPIs in other languages. That is to say, these words need to appear with certain elements like negation or in certain contexts such as questions. There are only a few studies such as Zidani-Eroğlu (1997), Besler (2000), Kelepir (2000Kelepir ( , 2001 and Yanılmaz (2009) that investigate NPIs but they are generally descriptive in nature and do not offer a formal account of the phenomenon. A discussion of certain characteristics and distribution of NPIs is found in Kelepir (2001:122) where she classifies Turkish NPIs into three different categories based on their morphological characteristics. This is illustrated in (13).
(13) (i) the adverb hiç 'ever', 'at all' (ii) the words that begin with the morpheme hiç such as hiçkimse 'no one ', hiçbirşey 'nothing', hiçbir N 'no N' (iii) the words that do not contain the morpheme hiç such as kimse 'no one', asla 'never', katiyyen 'in any way', sakın 'ever' The list in (13) is somewhat comprehensive, illustrating a variety of NPIs in Turkish. Note, however, that the taxonomy is only about whether or not the NPIs co-occur with a certain morpheme (i.e. hiç). Therefore, the classification does not illustrate the distribution or the function of the NPIs in question. In that sense, the NPIs in each group do not form a unified class with respect to where they can appear or what their function would be in the sentence.
As far as the distribution of certain NPIs and their licensing environments are concerned, Kelepir provides negated sentences and yes/no questions, stating that the word hiç means 'at all' in the former and ever in the latter. Consider the sentences below. As for the distribution of certain NPIs in yes/no questions, Kelepir notes that there is some disagreement among native speakers concerning which NPI sounds grammatical. This is illustrated in (15).
you-DAT no one five-LOC meet-FUT-1PL-ACC say-PAST Q 'Did anyone tell you that we would meet at 5?' According to Kelepir (2001:163), the NPI kimse is generally acceptable in yes/no questions. On the other hand, the NPI hiçkimse is regarded as less acceptable among some speakers whereas some others find it acceptable. Kelepir concludes that the disagreement in the grammaticality judgments may stem from the strength of the NPIs in question. That is to say, hiçkimse is a strong NPI and kimse is not in the sense of Zwarts (1996) and van der Wouden (1997) and the former can only be licensed by negation.
Kelepir also maintains that similar to Slavic (Progovac, 1994;Brown, 1999), Korean and Japanese (Sohn, 1995), most of the Turkish NPIs are licensed by a negative operator or in yes/no questions. She goes on to say that in contrast to NPIs in English, Turkish NPIs cannot be licensed in conditional contexts. She provides the following pair to illustrate that. Based on the ungrammaticality of the structure in (17), Kelepir argues that hiçkimse in Turkish is not allowed in the antecedent of the conditional sentences whereas its counterpart in English can appear in the same environment. This is also noted by Yanılmaz (2009Yanılmaz ( , 2013 who argues that NPIs in Turkish are only licensed by overt negation or the yes/no question marker. Note, however, that this claim does not imply that no other NPI in the language can appear in the antecedent of conditionals. This is due to the fact that the NPI kimse 'anyone' that is often considered to be in syntactic and semantic free variation with the NPI hiçkimse 'anyone'. Consider (18). The grammaticality of the sentences in (18a-b), in contrast to the ungrammaticality of the one in (17), clearly indicates that there is at least one type of NPI that can occur in conditional sentences in Turkish. It also shows that the licensing environments of the two NPIs, despite apparent similarities, do not overlap since kimse has a wider distribution and can be licensed without negation and the yes/no question marker.
Note also that the NPIs functioning as adverbs like asla 'never', katiyyen 'in any way' and sakın 'ever', they are only licensed by negation. The examples are provided below. ① Note that these sentences are not constructed sentences. They and all the new data were taken from the online sources where one can find naturally occurring speech. A simple search would produce more results.
it money in any way touch-NEG 'Don't touch that money in any way.' c. Sakın bura-ya gel-me. ever here-DAT come-NEG 'Don't you ever come here.' The affirmative counterparts of the above sentences would not be grammatical. Moreover, Kelepir (2001:123) maintains that the NPI sakın is used only in imperative constructions where a direct command or a prohibition is usually expressed. The fact that the sentences in (19a-c) are all imperatives supports this claim. However, a closer look will reveal that it is not the case. Consider (20) 'What if they ever lied to you?' As the sentences in (20) illustrate, the adverb sakın can also occur in what is called optative constructions in which a wish or a hope is being expressed in question form. The conclusion then is that the use of the NPI is not restricted to only imperative constructions.
Another NPI functioning as an adverb in verbal clauses would be the word zinhar that means never ever. It is a borrowed word from Persian and interestingly it was not classified as an NPI in previous work. It is licensed by negation just like other adverbs.
this-ACC never ever forget-NEG 'Don't you ever forget that.' Note that the affirmative counterparts of these sentences would be ungrammatical. In that sense, it is not unreasonable to treat it as an NPI in the language. ① There is one last environment where certain NPIs are licensed. This is the one in which the NPI co-occurs with the adjectival suffix -sIz without in the language. What is interesting here is that the NPIs can appear in the absence of verbal negation. This is shown in (22). ① Note that there is another use of 'zinhar' where it means absolutely or definitely. In this case it is not an NPI but is used emphatically in non-verbal sentences. An example would be: Bu zinhar suç / günah / saçma (this absolutely crime / sin / nonsense, 'This is absolutely a crime / sin / nonsense').
never no transfer make-INF-without-INST Esenler-DAT arrive-PASS-PAST 'One can arrive at Esenler without making any transfer.' b. Biz bu ülke-ye hiçbirşey-siz gel-di-k. (Kelepir, 2000) we this country-DAT nothing-without come-PAST-1PL 'We came to this country without anything.' The sentences above illustrate that certain NPIs can appear without verbal negation or the yes/no question marker so long as they appear with the adjectival suffix -sIz.
The data above show that Turkish NPIs are licensed in a number of contexts, similar to their counterparts in English. Table 1 illustrates a list of NPIs and their distribution In Turkish. As can be observed from Table 1, all Turkish NPIs are licensed by verbal negation. However, negation is not the only licensor since NPIs can also appear in yes/no questions even though there are differences in terms of grammaticality judgments in terms of the acceptability of 'Hiç N's between speakers. Based on new data, it was shown that some NPIs are licensed by the adjectival suffix -sIz and some can appear in conditional sentences. In short, Turkish NPIs are allowed to appear in a number of different contexts. The conclusion is that the distribution of NPIs in Turkish is in fact wider than what was proposed in previous work.
The question that arises at this point is why do different NPIs appear only certain contexts and are not allowed in others? To put it differently, what property or properties do their licensers have in common given that different NPIs are licensed in different contexts? That there needs to be an analysis in which the distribution of Turkish NPIs is uniformly captured seems warranted. In the next section, I will review a number of competing analyses to account for the NPI phenomena cross-linguistically.

Cross-linguistic approaches to negative polarity
The investigation of negative polarity items across languages is not a new phenomenon. One of the earliest studies goes back as far as Klima (1964). In his work, Klima investigates negation in English and coins the term "affective contexts" in order to refer to NPI licensing environments. Specifically, he argues that NPIs such as any in English need to be found along with negation or with an expression carrying the syntactic feature [+affective]. In that sense he is one of the earliest who noticed that NPIs need the presence of some element in the structure. However, he does not provide a formal analysis to flesh out the exact nature of this feature. Since then there has been extensive cross-linguistic research NPIs proposing accounts from different points of view. There is a number of important syntactic analyses found in Linebarger (1980Linebarger ( , 1987Linebarger ( , 1991, Laka (1990), Progovac (1992Progovac ( , 1994 while semantic accounts were offered by Ladusaw (1980Ladusaw ( , 1996, Hoeksema (1983), Zwarts (1996Zwarts ( , 1998 van der Wouden (1997), von Fintel (1999 and Giannakidou (1998Giannakidou ( , 2002Giannakidou ( , 2011Giannakidou ( , 2014 among others. On the other hand, there are several analyses that propose a pragmatic account such as Krifka (1992Krifka ( , 1994Krifka ( , 1995, Israel (1997Israel ( , 2004, and Chierchia (2004and Chierchia ( , 2013. All of these investigations explore the relationship between different NPIs and the contexts in which they are allowed to occur in various languages. The analyses have demonstrated that each of them has certain advantages as well as some drawbacks in terms of proposing an exhaustive account of the phenomenon. In the next section, I will review two important approaches to NPIs and argue that they do not fully account for the Turkish data.

Downward entailment and a hierarchy of NPIs
One of the most influential analyses of NPIs found in the literature is found in Ladusaw (1980Ladusaw ( , 1996 where he argues that NPIs are allowed what he calls downward entailing contexts. Adopting Fauconnier's (1978) notion of implication reversal, Ladusaw argues that there is a close relationship between polarity sensitivity and downward entailment. That is to say, NPIs are licensed when they are in the scope of a downward entailing operator. Basically, a downward entailing operator licenses inferences from general properties (i.e. sets) to specific instances (i.e. subsets). In other words, downward entailing functions are order reversing and are closed under subsets. Ladusaw (1980) proposes the following licensing condition for all NPIs: (23) ɑ is the trigger for negative polarity items in its scope iff ɑ is downward entailing. ① Note that the term "trigger" in (23) is required in order to make the licensing of the NPI possible. One of the primary downward entailing operators across languages is negation as it can license the inference from a general property to a specific instance, as in (24). (24a) is a downward entailing context in that an expression denoting a set (i.e. Linguistics) can be substituted for a subset (i.e. Phonology), as in (24b). This is shown in (25) To sum up, expressions denoting sets in downward entailing contexts can be replaced with expressions denoting subsets. The downward entailing contexts, therefore, correctly capture the occurrence of NPIs with negation and certain elements, as shown in (27). The fact that John does not have any books (i.e. set) in (27) entails that he does not have any linguistics books (i.e. subset). The entailment goes downward, from a larger set to a small set. The same is true for the sentence in (27b). In short, Ladusaw's analysis showed that NPIs generally occur in downward entailing contexts, negation being one of them.
Note that the idea of downward entailment and its relationship with NPIs across languages initiated a significant line of research, especially in the area of semantics. Among others, the analyses by Zwarts (1996Zwarts ( , 1998 and van der Wouden (1997) are influential since they helped the further development of this semantic framework. For instance, in a series of papers, Zwarts introduced a hierarchy of NPIs that he called weak, strong and superstrong in English, German and Dutch. According to that, weak NPIs are licensed in any downward entailing contexts across languages. On the other hand, strong NPIs (medium NPIs in van der Wouden's classification) are licenced in anti-additive contexts (i.e. functions satisfying the first de Morgan biconditional law). Last but not least, superstrong NPIs are licensed by classical negation (i.e. anti-morphic contexts). This is illustrated in Table 2. As shown above, a weak NPI like any is licensed in downward entailing contexts. A strong NPI like any book at all or lift a finger is different in the sense that it is licensed in anti-additive contexts as in 'No student brought any book at all'. On the other hand, a superstrong NPI such as one bit requires an anti-morphic context as it is only licensed by classical negation as in 'This book is not one bit interesting'. In that sense, this account provides an appealing classification of NPIs based on different logical properties of negation.
The question that arises at this point is whether we can account for the distribution Turkish NPIs by adopting these semantic analyses. As already shown in Table 1, all NPIs in Turkish are licensed by negation that is a well-attested downward entailing context. Moreover, it is also true that different NPIs are licensed by different negative elements such as classical negation and the adjectival suffix. Therefore, it does not sound unreasonable to argue that Turkish NPIs are all licensed by downward entailing contexts. However, the occurrence of NPIs is not limited to negative contexts as they are also licensed in other contexts. For instance, it was shown that the NPIs hiç and kimse are licensed by yes/no questions that are by definition non-downward entailing contexts. As noted by Giannakidou (2002Giannakidou ( , 2011, yes/no questions are not downward entailing as they denote the propositions corresponding to their true answer (cf. Groenendijk and Stokhof, 1997). In fact, Ladusaw (1980) also notices this issue and proposes the possibility of a biased negative answer in yes/no questions in an attempt to circumvent the problem. However, it is a well-known fact that not every yes/no question in a language, whether they contain an NPI or not, carries a bias towards a negative answer. Note also that other researchers have attempted to make a connection between negation and yes/no questions to account for the occurrence of NPIs in questions. For instance, based on a suggestion by Enç (lecture notes), Progovac (1994:151) maintains that the yes/no question marker -mI in Turkish may be related to the sentential negative marker -mA and thus may license or bind NPIs. Consider the sentences below. If the two elements were somehow related, we would expect them to be in some kind of (e.g. syntactic and semantic) complementary distribution. However, their presence does not lead to ungrammaticality or unacceptability in any way.
Furthermore, -mA is not the only negative marker in Turkish. In existential sentences, the negative predicate that is used is yok 'does not exist' and it would be very hard to look for a surface similarity between the two. Consider (31). If we accepted that certain NPIs appear in yes/no questions because the question marker is somehow related to classical negation, we would expect other NPI licensers to look like verbal negation, phonologically or otherwise. However, this is not the case as it is clear from the sentence in (31b). Therefore, it would make sense to look for the occurrence of NPIs in yes/no questions somewhere else. ① It was already shown that it is not only negation or yes/no questions that can license NPIs in the language. Another important context in which NPIs appear is conditionals. As discussed by Heim (1984) and Giannakidou (2002), protasis of conditionals are not strictly downward entailing contexts. Yet there is at least one type of NPI that is licensed in conditional contexts, as noted below.
① Note that in terms of phonology, the negative morpheme in Turkish has two allomorphs (i.e. -me/-ma) conforming to the vowel harmony in terms of backness whereas the question marker has four allomorphs (i.e. -mı, -mi--mu, -mü), obeying the vowel harmony in terms of both backness and height. This again shows further distinction rather than a potential similarity between the two elements. In short, the fact that there is at least one type of NPI that can appear in protasis of conditionals in Turkish and some other languages clearly indicates that their occurrence is not restricted to downward entailing contexts.
One last context in which the occurrence of an NPI cannot be accounted for by referring to classical negation or downward entailment is one where there is no sentential negation in the structure. Consider the sentences below.
(33) a. Biz bu ülke-ye hiçbirşey-siz gel-di-k. (Kelepir, 2000) we this country-DAT anything-without come-PAST-1PL 'We came to this country without anything. The negative marker in Turkish nonverbal sentences is değil 'not' and there is no negation in (33a). Yet the NPI hiçbirşey can appear in the sentence without leading to ungrammaticality. As noted in Kelepir (2000), this is because the NPI appears with the adjectival suffix -sIz that licenses the NPI. What has not been ever discussed in previous work is the fact that the adjectival suffix licenses NPIs in verbal sentences as well, as shown in (33b) and (33c) above. The grammaticality of these sentences shows that NPIs in Turkish are licensed in -sIz clauses without verbal negation. Note that the fact that -sIz 'without' can license NPIs in Turkish should not come as a surprise. It appears that similar without-clauses can also license NPIs in other languages. Consider the examples from Greek (Xherija, 2015:186 As the grammaticality of the sentences above indicates, NPIs can be licensed by horís 'without' in Greek. The NPIs kanénan and kanéna distagmo can occur without the presence of sentential negation in horís clauses.
The discussion above showed that negation itself, its possible correlation with other elements or downward entailment do not correctly predict the distribution of NPIs in Turkish. This is because of the fact that the occurrence of NPIs is not limited to negation or downward entailing contexts since they can also appear in yes/no questions, conditionals and along with the adjectival suffix -sIz. Therefore, it seems necessary to propose a more inclusive analysis in order to account for the occurrence of NPIs in various contexts in Turkish. In the next section, I will introduce the theory of nonveridicality, originally proposed and further developed by Zwarts (1995Zwarts ( , 1998 and Giannakidou (1998Giannakidou ( , 2002Giannakidou ( , 2011Giannakidou ( , 2014. I will then argue that the facts about Turkish NPIs could be better accounted for by adopting the notion of nonveridicality.

The nonveridicality theory
Nonveridicality as a semantic concept was originated in Zwarts (1995Zwarts ( , 1996 and further developed by Giannakidou in a series of papers (1998,2002,2011,2014). In her work on the NPI phenomena, Giannakidou (2002:6) notes that there are in fact quite a number of non-downward entailing contexts such as non-monotone (i.e. non-downward entailing) quantifiers, conditionals, yes/no questions, habituals, directive intentional verbs, generic sentences and imperatives in which NPIs can occur in such languages as Greek, English and Dutch. Consider the sentences below. The point here is that quantifiers like 'exactly three ' and 'neither' in (35a-b) are both non-monotone quantifiers. Yet they can license the NPI 'any' in English. Similarly, it is well-attested that questions are not downward entailing contexts but the NPI ook maar iets  are both allowed in them. Finally, NPIs can also appear in imperatives, as in (37), that are predominantly not downward entailing contexts.
Based on these facts, Giannakidou makes a generalization where she argues that what those non-downward entailing contexts have in common along with negation is the property of nonveridicality. Basically, nonveridical contexts are those in which the truth of the proposition is not entailed. To better understand what is nonveridicality, let us consider the following pair, taken from Hoeksema (2013:48).
b. John hopes it is raining.
The truth of the proposition in (38a), the fact that it is raining, is not entailed in the context of (38b). Therefore, it is typically a nonveridical context. Note that the same is also true for sentential negation which is equivalent to the sentence in (39).
(39) It is not the case that it is raining.
Thus one can say that if a statement of the form XYZ such that Y denotes a proposition that does not entail Y, then the context X__Z is said to be nonveridical. More technically, both veridicality and nonveridicality are defined in terms of truth, as formulated by Giannakidou (2002:5).
(Non)veridicality for propositional operators (40) (i). A propositional operator F is veridical iff Fp entails p: Fp  p; otherwise F is nonveridical.
(ii). A nonveridical operator F is antiveridical iff Fp entails not p: Fp  p.
Basically, a propositional operator is a proposition embedding function. It is either a sentence modifier (e.g. a sentence-level adverb or a modal operator), or the question operator. (40i) states that F is veridical iff whenever Fp is true, p is also true. If that does not hold, F is nonveridical. Note also that a nonveridical F is antiveridical iff whenever Fp is true p is not true. In that sense, antiveridical operators are a proper subset of nonveridicals. Positive operators like past tense adverbials are veridical and do not license NPIs. In contrast, modal verbs, intensional operators, and questions are all nonveridical and allow NPIs. Antiveridical operators, on the other hand, are negation and without that are the prototypical licensers of NPIs (cf. Giannakidou, 1993;Zwarts, 1995). In short, the idea of nonveridicality subsumes antiveridicality, modality, intensionality, as well as downward entailment. In that sense, it is more exhaustive and covers a wide range of NPI licensing contexts.
Recall that NPIs in Turkish are allowed to occur in a number of different contexts. Specifically, they are licensed by (i) classical (sentential) negation, (ii) yes/no questions, (iii) if-conditionals and (iv) the adjectival suffix -sIz. Consider (41). The sentences in (41a-d) illustrate that the contexts for Turkish NPIs may be somehow limited, especially when one considers the range of environments in which NPIs in Greek and English are licensed. Nevertheless, the sentences in (41a) and (41b) are an instance of sentential negation and yes/no question, licensing the NPIs. In (41c), on the other hand, the NPI appears in the protasis of the conditional sentence and in (41d) the NPI is in the environment of the adjectival suffix. In that sense, these contexts are not all negative or downward entailing contexts but are nonveridical in which the truth of the proposition is not entailed. Specifically, both classical negation and the adjectival suffix constitute the antiveridical contexts. On the other hand, yes/no questions and conditional structures are clearly nonveridical contexts in that the truth of the proposition in these environments are not entailed or guaranteed. Therefore, applying the idea of nonveridicality analysis of Turkish NPIs, we can correctly capture the nature of their licensing environments in a unified manner. Now that it was established that Turkish NPIs are licensed in nonveridical contexts where the truth of the proposition is not entailed, the question that should be addressed is why the NPIs themselves show variation with regard to by which context they are licensed. For instance, the contexts in which kimse appears is wider than hiçkimse and sentence adverbs like asla appear along with negation only. This issue will be discussed in the next section.

Variation in the distribution of NPIs
It is well-attested that different NPIs within and across languages show certain differences with respect to in which context they are allowed. More specifically, not all NPIs are licensed in the same nonveridical context. For instance, the NPIs kimse and hiçkimse in Turkish, even though they are morphologically similar and have been considered to be in free variation, are different in terms of where they can appear. On the other hand, the adverbs such as sakın and asla are licensed by negation and are excluded from other nonveridical contexts. The question that one would raise at this point is why this would be the case. One reason why kimse is allowed in more contexts than hiçkimse would be that historically it did not always function as an NPI in the language. As noted by Kelepir (2001:122-123), kimse used to occur in [+human] indefinite noun phrases similar to the suffix -body in English, as shown below.  Kelepir (2001:123) argues that the NPI kimse in contemporary Turkish has more or less lost this indefinite usage and became a negative polarity item. As for the NPI hiçkimse, Kelepir notes that it is possible that before kimse has lost its indefinite usage, hiçkimse was part of the paradigm in (42) in the sense that hiç combined with the indefinite kimse and yielded a negative polarity item. In time, as kimse lost its indefinite usage, hiçkimse and kimse have become interchangeable.
It was already shown that the two NPIs are not always interchangeable. However, Kelepir's arguments do shed light on the issue at hand. If we assume that the NPI hiçkimse has always been an NPI, as Kelepir suggests, we would expect it to be more sensitive to negation the way superstrong NPIs are sensitive to negation in other languages. In other words, one would argue that as a superstrong NPI, hiçkimse is licensed in rather limited contexts. On the other hand, kimse is allowed in more contexts so long as those contexts are nonveridical. Thus one could claim that historically kimse was not always an NPI but gradually evolved into one in the language. Therefore, it can appear in context that are not strictly negative. In that sense, we could regard kimse as a weak NPI that does not strictly require the presence of negation but can appear in other nonveridical contexts. Therefore, it is not unreasonable to propose a hierarchy of NPIs in Turkish so long as they are licensed in nonveridical contexts. Note also that the idea that an element losing its primary usage and adopting another one or an element whose distribution getting more restricted over time should not come as unexpected. For instance, Hoeksema (2010:837) notes that the Dutch enig 'some/any' displays a diachronic change from nonveridical environments in general to a strict subset thereof, namely negative, conditional and interrogative contexts. Apparently, enig had split into two different uses: emphatic and nonemphatic. The emphatic enig is allowed within a set of environments similar to 'ever' and 'any' in English while the use of nonemphatic enig is currently disappearing.
Last but not least, the data above showed that there is a set of adverbs such as sakın and katiyyen that are licensed only in negative contexts. This in fact is not unexpected since they are all negative adverbs and their behavior is similar to that of superstrong NPIs in other languages.

Conclusion
In this paper, I worked on the negative polarity phenomenon in Turkish. Specifically, I investigated NPIs and their various licensing contexts. Firstly, I showed that the distribution of the NPI sakın is in fact very similar to that of other NPIs functioning as adverbs in that its distribution is not limited to imperative contexts. I also showed that the adverb zinhar shows behavior similar to true NPIs in the language. As opposed to previous analyses, I demonstrated that there is at least one type of NPI namely kimse, which is licensed in non-negative or interrogative contexts, namely in conditional sentences. I also argued that the two NPIs kimse and hiçkimse that were regarded as in free variation in fact display certain distinctions in terms of their distribution. The analysis of various types of Turkish NPIs led to the conclusion that they are licensed in a number of different contexts such as negative sentences, yes/no questions, conditional sentences and the adjectival suffix -sIz. Based on these facts, I proposed that NPIs are in fact allowed in nonveridical contexts where the truth of the proposition is not entailed. That is, the idea of nonveridicality uniformly captures the distribution of all NPIs in the language. Finally, I argued for a hierarchy of NPIs to account for the distinction in their distribution. Further research on the finer distinctions between these NPIs and their licensing contexts will surely shed more light on the phenomenon.