The next station: chunking of değİl ‘not’ collocations in Turkish Sign Language

1.1 What this article is about

The present paper focuses on the usage-based constructional change of the basic negator değİl ‘not’ and explores the effects of chunking on the sequential collocations [i.e., sign + değİl] in Turkish Sign Language (TİD). This manual negation marker is produced with a flat hand, palm facing outward, a single down-to-up movement, and a backward head-tilt (henceforth bht)^[1] (see Figure 1) (e.g., Gökgöz 2009, 2011; Makaroğlu 2021; Pfau 2016; Zeshan 2006). While there have been many formal or typological attempts to describe clausal negation phenomena in TİD, a full description of değİl from the perspective of the theory of grammaticalization is still lacking.

Figure 1:

The negation marker değİl ‘not’ in TİD (Dikyuva et al. 2017: 220).

It has been previously suggested that değİl occupies a clause-final position and there is no way to negate a clause without it (Kubuş 2008; Pfau 2016; Zeshan 2006).^[2] Regarding the chunking mechanism of değİl, Zeshan (2003) first observed that when it cliticizes to a preceding predicate, several aspects of its form change. She listed the reduced form of değİl in five categories, namely (i) movement reduction, (ii) duration shortening, (iii) assimilation of orientation, (iv) assimilation of location and (v) assimilation of handedness, without providing an explanation in terms of grammaticalization theory. All these differences have the effect of reducing the ‘phonological content’ of the clitic and/or assimilating the clitic to the host sign by carrying over formational properties from the host sign to the clitic. Based on her initial observations, she concludes that the negative clitics in TİD are particularly common, and these typically involve high-frequency host signs such as ‘know’, ‘like’, ‘understand’, etc. While there appears to be a correlation between the frequency of [sign + değİl] collocations and their chunking properties, to our knowledge, no examination of the frequency effects on the grammaticalization of değİl or other grammatical markers has been performed for TİD so far. In contrast to spoken languages, there have been few studies on the second stage of grammaticalization (i.e., from free clitic to bonded grammatical affix) in sign languages (henceforth SLs). However, with the progress of corpus-based approaches, frequency effects in grammaticalization from a synchronic perspective have become far more visible and acknowledged. To our knowledge, Wilkinson (2016) carries out the first in-depth research on collocations in the signed modality, and explores the effects of frequency on the collocations including a manual negative marker not in American Sign Language (ASL).

By adopting Construction Grammar (henceforth CxG), this paper aims to measure degrees of grammaticalization in collocations using corpus-based variables. As grammaticalization extends beyond specific lexical units and encompasses collocations (i.e., conventionalized word sequences) (Bybee 2006: 713), this study focuses on the chunking effects in language use, examining the sequential collocations of the manual negator değİl ‘not’ with other manual signs and analyzing abstract schemas relating to the occurrence of its collocations in both token and type. This study addresses the earlier observations regarding the formational properties of the manual negator değİl ‘not’ and investigates the following research questions:

1. Which phonetic reductions and/or phonological units^[3] occur as the frequency of [sign + değİl] collocations increases?

2. How can the fusional characteristics of reduced [sign + değİl] collocations be classified into phonological schemas?

3. What factors are responsible for the morphological productivity of a schema in signed collocations?

4. Does the increased schematicity of [sign + değİl] collocations trigger a semantic change?

In a CxG framework, as elaborated by Traugott and Trousdale (2013), grammaticalization is examined from a constructional perspective, where equal attention is paid to changes in form and meaning. Grammatical constructionalization is therefore generally accompanied by changes in the degree of schematicity, productivity, and compositionality which additionally suggests that a grammaticalizing construction may become more productive when it is employed with an increased number of collocations and when its token frequency rises (Traugott and Trousdale 2013: 114). Thus, in line with Traugott and Trousdale’s approach, this paper argues that the chunking process is not language-specific within the spoken modality, and the frequency of [sign + değİl] collocations is strongly correlated with phonetic reduction and duration. By conducting a frequency analysis, the current study also claims that the fusional characteristics of [sign + değİl] collocations can be classified under four reduced constructional schemas: (i) [bilmek + değİl] ‘know not’ type, (ii) [anlamak + değİl] ‘understand not’ type, (iii), [İyİ + değİl] ‘good not’ type, and (iv) [lazım + değİl] ‘should not’ type. It is noteworthy that some highly frequent collocations may be considered potential prefabricated units due to their increased autonomy, and the durations of the collocations decrease depending on their frequency in accordance with Zipf’s law. This study also demonstrates that reduction in movement appears to be a criterion for a scheme to be productive in signed modality, and the constructional change of [sign + değİl] collocations leads to a type of semantic transition that can be referred to as “subjectification”.

Although nonmanual markers play a grammatical role in TİD, as mentioned earlier, this study will focus on the constructional changes of the manual negator değİl ‘not’. I will present the phonological schemas that emerge from the analysis of collocations attested by frequency. At this point, I will not address the question of whether there is a reduction or loss of the negative nonmanuals (e.g., backward head tilt, eyebrow raising, etc.) in the constructionalization of the değİl ‘not’ collocations in TİD. This issue will be left open for following studies.

1.2 Usage-based linguistics approach in the signed modality

In the usage-based linguistics approach, human language is viewed as a dynamic system where all aspects of linguistic structure are shaped by pragmatic and cognitive mechanisms that influence language use (e.g., Bybee 2001; Bybee and Hopper 2001; Diessel 2004). That is to say, usage-based approaches (Bybee 2006, 2010) adopt the idea that the mental grammar of language users is “formed by the recurrent exposure to specific utterances” (Hoffmann 2018: 184). In other words, if a construction (i.e., pairing of FORM and MEANING) is experienced by a language user repeatedly enough, it becomes stored or entrenched (c.f., Croft and Cruse 2004: 276–8).

For more than two decades, a good deal of research has acknowledged the importance of frequency effects in many areas of linguistics, including L1/L2 acquisition studies, language processing, and grammaticalization (e.g., Bybee 2006; Ellis 2002). However, most of our current understanding of frequency effects is based on investigating spoken languages. Moreover, usage-based approaches have only been taken seriously in the last decade in the case of SLs (Hou 2022; Janzen 2018; Lepic 2019; Lepic and Occhino 2018; Wilcox 2014; Wilcox and Occhino 2016; Wilkinson 2016).

Frequency is the mechanism that determines emergence in a language, that is, grammaticalization, and underlines the basis for the usage-based model (Bybee and Hopper 2001; Hopper 1987). Under this view, it is expected that high-frequency collocations undergo phonological reduction in the schematic as an emerging pattern. From a usage-based linguistic point of view, multi-word expressions are essential in facilitating both language acquisition and language processing. They are constructed by the chunking mechanism, which requires two or more linguistic elements to co-occur repeatedly through recurrence and reuse. Subsequently, chunking of words enables language users to create ‘prefabs’ and store them as exemplars in memory (Bybee 2010).

The preliminary observations on frequency effects in SLs clearly show that chunking and automation are general domains of cognitive processing mechanisms for all human languages (Hou 2022; Lepic 2019; Wilkinson 2016). For her study, Wilkinson (2016) specifies three frequent collocations for not in ASL (i.e., [not + have.to], [why + not], and [not + understand]) and argues that ASL is equally sensitive to the frequency effects of chunking as are spoken languages. She further identifies three patterns in the portrayal of the frequent not collocation: (i) displaying loss of analyzability in the internal structure, (ii) exhibiting automatization and (iii) attesting semantic bleaching. Furthermore, she observes the pragmatic strengthening of subjectivity (i.e., the speaker’s participation) in the collocations [why + not] and [not + understand]. Börstell et al. (2016) examines the distribution and duration of compounds named as reduced or non-reduced and identified that the reduced compounds are much shorter in duration than unreduced compounds. Lepic (2019) then specifies two ASL verb-argument constructions: [interpreter + bring.in] and [take.to + hospital] and pointes out that even if one word order is more typical than the other, such constructs are not completely fixed in terms of their order. He also states that they can be conventional multi-word forms, which have a more easily analyzable structure in relation to not collocations. Most recently, Hou (2022) examines recurring sequences of multiple sign forms, i.e., look and the family of ‘look’ signs, and finds that as for its syntactic environment, look has two distinct functions and has been grammaticalized from a visual perception verb to a stance verb. Keeping these observations in mind, similar to previous studies in a usage-based perspective, the main motivation for studying constructional change is to examine frequency effects in the signed modality.

1.3 The structure of the article

The present study aims to systematically investigate the potential effects of frequency on [sign + değİl] collocations. The study is organized as follows: Section 2 provides an overview of the literature on grammaticalization in the signed modality. Section 3 presents the corpus data used in the analysis, along with the methodology employed in the TİD Corpus. Section 4 outlines the frequency profile of [sign + değİl] collocations (see Section 4.1), the schematic characteristics (see Section 4.2), the phonological schemas of [sign + değİl] collocations (see Section 4.3), the frequency effects on grammaticalization (see Section 4.4), and semantic-pragmatic change, particularly in terms of subjectivity (see Section 4.5). Subsequently, Section 5 explores the interaction between the frequency and collocations in the grammatical changes discussed in this paper. Finally, Section 6 concludes and provides theoretical insights for future studies.

3.1 Corpus

The data were taken from the TİD Corpus (see Dikyuva et al. 2017 for detail), a partially annotated corpus that was established in 2015 and consists of recordings of 116 native signers^[5] from 26 distinct cities across Turkey. The participants, who were between the age of 12–60, were either born deaf or became deaf before the age of three. The corpus was collected using semi-structured elicitation tasks that involved a variety of visual and written stimuli such as images, films, and topic lists covering wide range of subjects. The topic lists were used solely as a prompt to initiate signing conversations, during which participants were free to discuss anything related to the topic or introduce their own topics in conversation anytime. In addition to the topic lists, visual stimuli such as silent movies (e.g., Pixar short videos) and images employed in other studies were also used to elicit conversation.

3.2 Participants

For the present study, the sample consists of 500 min of dyadic video conversations that covered both randomly selected and free topics with 50 TİD native signers from 14 distinct cities: Adana, Ankara, İstanbul, İzmir, Erzurum, Konya, Mersin, Eskişehir, Kocaeli, Denizli, Edirne, Diyarbakır, Çanakkale and Kars. The sample also includes 60,594 sign tokens (approximately 7.6 % of the TİD Corpus).

3.3 Procedure

The sample was annotated in terms of the grammaticalization properties of değİl, using the ELAN annotation tool (ELAN 2022). Since the objective of this paper is to document the grammaticalization properties of değİl in negative clauses, other negative units in TİD (e.g., hiç ‘never’, hayır ‘no’) have been excluded in the initial coding. If a grammatical marker involves a fully grammaticalized or fused marker, it becomes difficult for researchers to define and categorize their linguistic units at an initial glance. Due to the extensive changes in their internal forms, the structure of some signs (e.g., verbs) in collocations turns out to be unanalyzable. That’s why, 51 tokens (e.g., [tanımak + değİl] ‘recognize not’) had to be excluded from distributional analysis (but not from descriptive or formal analysis). Finally, this study presents initial results of 1,219 predictable [sign + değİl] collocations collected, and token frequency is discussed in terms of the grammaticalization of değİl based on the usage-based grammar. All tokens extracted from the TİD corpus were assigned a code following the scheme [city code.video file-S:start-E:end] for reference. For instance, the code [06.07-S:01:21-E:01:23] refers to an exemplar was collected in Ankara and comes from video file 07, which starts at time point 01:21 and ends at 01:23.

4.1 Token frequency of [sign + değİl] collocations

Based on a corpus analysis of 207 unique collocations (types) and 1,219 [sign + değİl]^[6] tokens, the distributional findings of this study provide insights into the frequency characteristics of [sign + değİl] collocations. As for the frequency list in Table 1, the top 20 most frequent [sign + değİl] collocation types are listed in order of their occurrences. As can be seen, the statistics indicate that the most frequent type of [sign + değİl] (i.e., [bilmek + değİl] ‘know not’) in the database represents 18.05 % of the current database. Furthermore, a limited number of unique collocations make up a significant percentage of the frequency of occurrence, with the top 20 [sign + değİl] collocation types comprising 69.32 % of the data. These findings suggest that değİl is typically associated with specific co-occurring signs.

Table 2 provides an illustration of the distributional tendencies of sign categories in [sign + değİl] collocations. The results demonstrate that the manual negator değİl ‘not’ frequently emerges with verbs, constituting 820 tokens and 81 unique constructions. In contrast, the instances of değİl combined with classifiers are considerably low, amounting to only 15 tokens. These findings indicate that değİl ‘not’ is not productive with non-conventional forms in TİD, a pattern that is consistent with the observations made for the use of negation sign not in ASL (Wilkinson 2016: 103).

Regarding adjectival predicates, they are the second most common type co-occuring with değİl, accounting for 198 tokens and 45 unique constructions (i.e., 16.24 % of the data). Thus, it is not surprising that the most productive co-occurring signs with değİl ‘not’ are either verbal or adjectival predicates in TİD. In contrast, değİl ‘not’ is rarely observed to combine with other fingerspelled forms.

Fingerspelling is a common phenomenon in SLs, with its frequency ranging from 2 % to 10 % in signing conversations (McKee and Kennedy 2000; Padden and Gunsauls 2003; Schembri and Johnston 2007). However, in TİD literature, while Taşcı (2012) estimates that fingerspelling constitutes 8 % of the language, a study by Dikyuva et al. (2017: 58) finds that it only accounts for 0.95 % of the dyadic conversations in the TİD corpus. Interestingly, the frequency of fingerspelling in [sign + değİl] collocations appears to be consistent with the findings of Dikyuva et al. (2017), which suggests that it may be a language-specific phenomenon.

Despite the fact that ASL data includes more diversified genres (e.g., monologues, narratives, group conversations) (c.f., Wilkinson 2016) compared to TİD, it is worth noting that there is a significant difference in the distribution of certain sign categories (i.e., verbs, adjectives, and fingerspelling) between the two languages, as highlighted in gray in Table 3. Specifically, verbs in [sign + değİl] collocations comprise 67.27 % of the TİD data, but only 36.0 % of the not collocations in ASL. Additionally, adjectives in ASL are two times, fingerspelling is almost 12 times more common than TİD. These differences raise the novel question of whether genre-specific and/or language-specific factors influence the frequency of word categories listed in Table 3.

When examining the interplay between frequency and duration of collocations, as depicted in Figure 2, it is worth noting that there exists a strong correlation between the frequency of the collocations and their duration (c.f., Börstell et al. 2016). In other words, this suggests that the duration of multi-word combinations reduces proportionally to the frequency of [sign + değİl] collocations, following Zipf’s law.

Figure 2:

Duration (x) plotted against frequency (y) on a logarithmic scale over [sign + değİl] collocation types.

As shown in Figure 2, the average duration of the collocations with the lowest frequency is nearly twice as long as that of the most frequent [sign + değİl] collocations (i.e., [bilmek + değİl] ‘know not’ and [anlamak + değİl] ‘understand not’). Consequently, it can be inferred that the frequently occurring collocations have significantly shorter durations than the less frequent ones. This observation supports the usage-based approach to the emergence of grammatical items and aligns with the assumption that sign language (specifically TİD) is analogous to spoken language without any modality-specific characteristics.

To our knowledge, except for Wilkinson (2016)’s three-point Likert scale (i.e., high, medium, and low frequency), no specific cut-off criteria for frequency sets of collocations in SLs have been proposed. To address this issue more comprehensively, I propose an alternative cut-off point, i.e., five-point Likert scale. By examining the frequency profile of 50 top-ranked değİl collocations (as given in Figure 3), I am able to evaluate the test’s ability to distinguish between different sets of collocations, which are labeled with distinct colors).

Figure 3:

Frequency profile of the 50 top-ranked [sign + değİl] collocation types.

Figure 3 illustrates that the very high frequency set is characterized for signs with more than 100 tokens, including only two unique constructions, i.e., [bilmek + değİl] ‘know not’ and [anlamak + değİl] ‘understand not’. This set contains 412 tokens, which represents 33.8 % of the data. The high frequency değİl set, on the other hand, ranges from 20 to 99 tokens and encompasses 257 tokens with six unique collocations, such as [iyi + değİl] ‘good not’, accounting for 21.08 % of the data. The moderate frequency set ranges from 10 to 19, and contains 209 tokens with 15 distinct collocations, such as [unutmak + değİl] ‘forget not’, representing 17.15 % of the data. The low frequency set, which ranges from 5 to 9, contains 94 tokens with 14 distinct collocations, such as [kazanmak + değİl] ‘gain not’, making up only 7.71 % of the data. Finally, the very low frequency range of one to four includes 247 tokens with 170 distinct collocations, such as [önce + değİl] ‘before not’, accounting for 20.26 % of the data.

To sum up, based on the data presented in Table 4, the eight highest [sign + değİl] unique collocations (or 3.87 % of the collocation types) – comprising the very high and high frequency sets – make up 54.88 of the data. It is often argued that describing grammatical collocations can provide crucial insights into how bonded constructions or grammaticalizations emerge through repeated chunking over time. Building on this perspective, the five frequency sets and their fusional properties have significant implications for understanding the relationship between the schematic characteristics of değİl ‘not’ collocations and their distribution. In the following sections, I will propose an analysis of the schematic characteristics of değİl ‘not’ constructions.

4.2 Schematic patterns from high frequency [sign + değİl] collocations

The question of phonological formation is central to understanding grammaticalization in SLs. In the visual-gestural modality, the issue is the phonological fusion of the two-combination. While phonological factors that underlie schematic characteristics in grammaticalization have often been ignored, an examination of the fusional similarities in constructional change provides numerous examples of the systematicity of the schematic patterns. Therefore, it makes sense to identify the fusional characteristics of [sign + değİl] collocations with regard to schematicity. However, determining the specific pattern or criteria to use in grammaticalization classification is challenging due to the complex chunking properties of SLs.

The analysis reveals that the most frequent collocation is [bilmek + değİl] ‘know not’ with 220 tokens, representing 17.99 % of the dataset. As stated in the Introduction, the değİl ‘not’ sign is produced with a flat hand, palm facing outward, and a single down to up movement (see Figure 1). As demonstrated in Figure 5, the reduced [bilmek + değİl] ‘know not’ collocation is realized with a single orientation change at the marked location for bilmek ‘know’ (Figure 4). Due to the movement assimilation, the repeated phonological elements of bilmek ‘know’ are lost in the constructional change. The distributional pattern of the fused [bilmek + değİl] ‘know not’ collocation shows that all the tokens in the dataset are invariably subjected to phonological fusion (Figure 5).

Figure 4:

Lexical form of bilmek ‘know’ (bİlmek ‘know’, Makaroğlu and Dikyuva (2017) https://tidsozluk.aile.gov.tr/vidz_proc/0006/degiske/06-01_cr_0.5.mp4).

Figure 5:

[bİlmek + değİl] ‘know not’ in reduced form.

The second most frequent collocation, [anlamak + değİl] ‘understand not’ is attested with 192 tokens, representing 15.7 % of the dataset. As can be observed, the chunking of [anlamak + değİl] ‘understand not’ includes a reduced phonological form, and the değİl sign assimilates its movement, handshape, and location to the co-occurring sign anlamak ‘understand’. The final movement of the collocation is realized with a handshape (i.e., flat-o to 5-open) and orientation (i.e., left to front) change. Figures 6 and 7 illustrate the lexical form of anlamak ‘understand’ and the reduced [anlamak + değİl] ‘understand not’, respectively.

Figure 6:

Lexical form of anlamak ‘understand’ (anlamak ‘understand’, Makaroğlu and Dikyuva (2017) https://tidsozluk.aile.gov.tr/vidz_proc/0012/degiske/12-01_cr_0.5.mp4).

Figure 7:

[anlamak + değİl] ‘understand not’ in reduced form.

However, the 36th most frequent collocation, [düşünmek + değİl] ‘think not’ which shares similar phonological (i.e., location, side of head) and semantic (i.e., mental-state-verbs) properties, is only attested with 5 tokens (also categorized as low frequency type according to Table 4), representing 0.41 % of the dataset. Note that all instances of the [düşünmek + değİl] ‘think not’ collocations have been observed in non-reduced two distinct lexical units. This clearly supports the hypothesis that TİD users’ language use leads to chunking through repeated usage.

Figure 8:

Lexical form of İyİ ‘good’ (İyİ ‘good’, Makaroğlu and Dikyuva (2017) https://tidsozluk.aile.gov.tr/vidz_proc/0033/degiske/33-01_cr_0.5.mp4).

The third most frequent collocation, [iyi + değİl] ‘good not’ is attested with 74 tokens, representing 6.05 % of the dataset. The analysis reveals that 68 out of 74 instances of the [iyi + değİl] ‘good not’ collocations are reduced to a monosyllabic form and are realized with handshape (i.e., flat-o to 5-open) and an orientation change (i.e., up to down), as seen in Figures 8 and 9.

Figure 9:

[İyİ + değİl] ‘good not’ in reduced form.

The fourth most frequent collocation, [lazım + değİl] ‘should not’ is attested with 54 tokens, representing 4.42 % of the dataset. As illustrated in Figure 11, [lazım + değİl] ‘should not’ is realized with single path movement and handshape change (i.e., L to 5-open) at the canonical location for lazım ‘should’. Movement assimilation is observed, and the repeated phonological elements of lazım ‘should’ are lost in the constructional change of [lazım + değİl] ‘should not’ collocations. The analysis indicates that [lazım + değİl] ‘should not’ has consistently undergone this type of phonological fusion (n = 53) except for a single token (categorized as [iyi + değİl] ‘good not’ type). Figures 10 and 11 show the lexical form of lazım ‘should’ and the reduced [lazım + değİl] ‘should not’ respectively.

Figure 10:

Lexical form of lazım ‘should’ (lazım ‘should’, Makaroğlu and Dikyuva (2017) https://tidsozluk.aile.gov.tr/vidz_proc/0042/degiske/42-01_cr_0.5.mp4).

Figure 11:

[lazım + değİl] ‘should not’ in reduced form.

The seventh most frequent collocation, [bakmak + değİl] ‘look not’ is attested with 27 tokens, representing 2.21 % of the dataset. The analysis reveals that 22 out of 27 instances of [bakmak + değİl] ‘look not’ collocations are attested with two distinct reduced forms (or schemas) contrary to the lexical form of BAKMAK ‘look’ (Figure 12). The first one, as illustrated in Figure 13, is produced with both orientation (i.e., back to forward) and handshape (i.e., 1–5-open) changes, similar to [lazım + değİl] ‘should not’ type (see Figure 11). However, the second reduced form, as seen in Figure 14, is produced with only an orientation change (i.e., from back to forward), similar to the [bilmek + değİl] ‘know not’ type (see Figure 5), and has evolved through bleaching of its original meaning into ‘not visiting or experiencing something’. Therefore, its use has come to be associated with a weakened semantic meaning. The speaker/signer distance analysis thus clearly indicates that the second reduced [bakmak + değİl] ‘look not’ (bilmek + değİl ‘know not’ type^[7]) is more subjective than the first reduced form [lazım + değİl] ‘should not’ type) and has “more restricted interpretation” (Traugott and Dasher 2002).

Figure 12:

Lexical sign bakmak ‘look’ (bakmak ‘look’, Makaroğlu and Dikyuva (2017) https://tidsozluk.aile.gov.tr/vidz_proc/0013/degiske/13-01_cr_0.5.mp4).

Figure 13:

[bakmak + değİl] ‘look not’ in first reduced form ([lazım + değİl] ‘should not’ type).

Figure 14:

[bakmak + değİl] ‘look not’ in second reduced form ([bilmek + değİl] ‘know not’ type).

Actually, I would pose Visconti’s question (inspired by Traugott 2010) of whether it is possible to determine factors of subjectification that are replicable across construction types (Visconti 2013). The fact that the second reduced form of [bakmak + değİl] ‘look not’ is predominantly used in the first person subject, as all instances of it within the dataset have been seen in only the first person, suggests that its usage is subject to restrictions on the first person. This supports Visconti’s view (2013: 13) that subjectification is driven by the inherently subjective nature of speech.

The 8th most frequent collocation [önemli + değİl] ‘important not’ is attested with 26 tokens, representing 2.17 % of the dataset. The analysis reveals that 13 out of the 26 instances of the [önemli + değİl] ‘important not’ collocations are realized with two distinct orientation changes in each syllable (i.e., right to left then left to right) (as seen in Figure 19 below). This means that these 13 tokens of [önemli + değİl] ‘important not’ have not undergone phonological chunking and are produced by a sequence of two distinct lexical items as shown in Example (2).

Example (2) clearly denotes the literal negation of the adjective “crucial” or “essential”, conveying to the addressee that the injury is not significant enough to require an operation. In contrast, in the phonologically reduced form, semantic bleaching occurs, causing it to lose its literal meaning and develop new meanings or discursive functions, as illustrated in (3).

To observe the actual usage in the corpora, the signer first states that he could have succeeded if the necessary conditions were met. He then deems it not a priority or necessity for the audience at that point and changes the topic by using [önemli + değİl] ‘important not’. As shown in (3), the discursive usage of the reduced [önemli + değİl] ‘important not’ collocation is typically used in the stand-alone position and appears outside the syntactic structure. The functional analysis indicates that it can signal a topic change and perform a structural function. Given the evidence, signers use it mostly when they believe that the information is irrelevant or unnecessary for the audience. Therefore, it can be classified as audience-oriented and functions primarily in the monitoring of the speaker’s utterance. Based on the examples above, it is possible to argue for the significant role that frequency of occurrence plays in phonological reduction and present fused schemas of [sign + değİl] collocations that result from constructional change. These results lend support to the usage-based claim that language structure emerges from language use as measured by frequency of occurrence.

4.3 Phonological schemas of [sign + değİl] collocations

4.3.2 [anlamak + değİl] ‘understand not’ type collocations

The schematic structure of [anlamak + değİl] ‘understand not’ type collocations, which have a phonological pattern, indicates that the movement types of signs are aperture (as illustrated in Figure 7) or orientation change, and they do not have body-anchored locations. Although the tokens in [anlamak + değİl] ‘understand not’ show phonological fusion by assimilating the location and, in some cases or sub-schemas, the handshape of değİl to co-occurring signs, they are seen in disyllabic form unlike [lazım + değİl] ‘should not’ type collocations (Figure 11). In Figures 15 and 16, it can be easily seen that [bulmak + değİl] ‘find not’ is realized with two distinct orientation changes in each syllable (i.e., right to left, then left to right), and the phonological character of değİl is represented by a sole orientation change in the final movement.

Figure 15:

Lexical form of bulmak ‘find’ (bulmak ‘find’, Makaroğlu and Dikyuva (2017) https://tidsozluk.aile.gov.tr/vidz_proc/0179/degiske/179-01_cr_0.5.mp4).

Figure 16:

[bulmak + değİl] ‘find not’ in reduced form.

4.3.4 [lazım + değİl] ‘should not’ type collocations

To provide a brief characterization of [lazım + değİl] ‘should not’ type collocations, they contain the minimum condition that the movement of the co-occurring sign has a straight path movement. The tokens in [lazım + değİl] ‘should not’ type collocations show phonological fusion by assimilating the handshape değİl ‘not’ (open-5) and, in some cases or sub-schemas, changing the orientation of the palm with the wrist at the syllable offset. Thus, the collocation is reduced into a complex monosyllabic form, where the handshape of the syllable onset represents the co-occurring sign and the syllable offset is formed with the handshape of değİl ‘not’. For instance, out of 18 [gitmek + değİl] ‘go not’ collocations, 15 were reduced into a monosyllabic form, where the syllable offset is formed by changing the handshape (i.e., B to 5-open) and the orientation with the wrist (i.e., sagittal to coronal plane). The figures below illustrate both the lexical form of gitmek ‘go’ (Figure 17) and the reduced [gitmek + değİl] ‘go not’ (Figure 18).

Figure 17:

Lexical form of gitmek ‘go’ (gitmek ‘go’, Makaroğlu and Dikyuva (2017) https://tidsozluk.aile.gov.tr/vidz_proc/0005/degiske/05-01_cr_0.5.mp4).

Figure 18:

[gitmek + değİl] ‘go not’ in reduced form.

Based on these observations, the current study thus argues that even though the grammaticalization of değİl ‘not’ is directly related to the frequency of its collocations, phonological similarity between two signs is not necessary to motivate constructional change. The phonological shape of the high frequency [sign + değİl] collocations, of course, depends on the phonotactic properties of the co-occurring sign. Here, too, [sign + değİl] collocations have undergone specific phonological changes or fusion in relatively consistent and predictable ways and can be classified in four distinct phonological schemas. Table 5 summarizes the phonological schemas for [sign + değİl] collocations:

Table 5:

Phonological schemas of [sign + değil] collocations.

	[bilmek + değil] ‘know not’ type	[anlamak + değil] ‘understand not’ type	[iyi + değil] ‘good not’ type	[lazım + değil] ‘should not’ type
Phonological requirement	(+) Body-anchored	(−) Body-anchored	(−) Body-anchored	(±) Body-anchored, straight path movement
Syllabicity	Monosyllabic	Disyllabic	Monosyllabic	Monosyllabic
Phonological realization	Orientation change	Aperture or orientation change	Aperture and orientation change	Aperture or orientation change

In this respect, it has been expected that a categorization of [sign + değİl] collocations in terms of schematic properties would be an interesting data point to test the constructionalization in signed modality. Thus, the hypothesis that I will be pursuing in this study is that the grammaticalization of değİl ‘not’ schemas is both morphologically and phonologically defined.

4.4 Frequency effects on grammaticalization of değİl

In this section, I will focus on the relationship between the frequency sets and constructional change. I will present data to support my claim that the high token frequency of collocations becomes schematic and fused structures through repeated chunking over time. As shown in Figure 19, with the exception of [uygun + değİl] ‘suitable not’ (the 10th most frequent) and [önemli + değİl] ‘important not’ (the 8th most frequent), at least 80 % of the tokens of all high-frequency collocations are reduced. The non-reduced forms of the collocations (labeled in gray) represent a very small percentage of the current data. This analysis suggests that these collocations undergo a constructional change parallel to their frequency, indicating a clear tendency.

Figure 19:

Distribution of the top 10 [sign + değİl] collocations and their reduced schemas.

The data reveals that some frequent [sign + değİl] collocations are observed with distinct reduced schemas. To exemplify, the constructional change of the [bakmak + değİl] ‘look not’ is possible with two phonological schemas: the [bilmek + değİl] type (labeled in orange) and the [lazım + değİl] type (labeled in purple).^[8] This suggests that a [sign + değİl] collocation will have distinct schemas and sub-schemas if the phonotactic properties of its co-occurring elements match. A semantic perspective on this will be provided in detail in Section 5.4.

Table 6 presents an overview of the distribution profile of the grammaticalization schema in değİl ‘not’ collocations. The değİl sign frequently undergoes a fusional change in four different schemas (827 tokens), and these fusional schemas make up 67.84 % of the current token database. Thus, it can be concluded that the constructional change accounts for almost two-thirds of the entire değİl ‘not’ collocation dataset. From the viewpoint of lexical word categories, the co-occurring signs in each schema do not possess the same properties, meaning they are not limited to a single word category. Various instances, such as [adjective + değİl], [verb + değİl], [modal + değİl] etc. are observed (see Table 2 above). For instance, the categories of co-occurring signs in [lazım + değİl] type here are verb (e.g., [gitmek + değİl] ‘go not’), modal (e.g., [lazım + değİl] ‘should not’) and adjective (e.g., [iyi + değİl] ‘good not’).

As Table 6 shows, all grammaticalization schemas of değİl ‘not’ are associated with at least six (and at most 26) distinct co-occurring signs in the current data. In other words, each grammaticalization schema is not restricted to just one type of collocation. However, upon examining frequency sets of the collocations, a few değİl constructions with a very low frequency, such as [seçmek + değİl] ‘choose not’ (one, reduced with [lazım + değİl] ‘should not’ type, and two unreduced tokens), [açıklamak + değİl] ‘explain not’ (only one reduced token with [bilmek + değİl] ‘know not’ type) and [alışmak + değİl] ‘get-used-to not’ (only one reduced token with [bilmek + değİl] ‘know not’ type) exhibit phonological reduction. Therefore, as opposed to the token-based claims, I argue that token frequency is not the sole predictor of productivity (which remains schematically open)^[9] in constructional constructions. For further details and discussions, see Bybee (1995), Bauer (2001) and Barðdal (2008).

A closer inspection of the distribution of the schemas reveals that the lazım + değİl ‘should not’ type is ranked the third most frequent of the four schemas, with 184 occurrences in değİl collocations, but it is listed as the most frequent one according to its type frequency (n = 26). Thus, it can easily be argued that the morphological productivity of the unified schemas of değİl is changing and that type frequency and token frequency are not directly correlated.

Furthermore, considering syllable properties and type frequency of schemas, I have counted 56 monosyllabic types of değİl constructions, but only six disyllabic constructions. A combination of these additional findings can yield new insights into the notion of productivity in the signed modality. This means that monosyllabicity is the other triggering factor for the morphological productivity of grammaticalization schemas in SLs. In other words, the monosyllabic constructions include high type frequency and are schematically open.

The same is true for not in ASL, where the most frequent collocations (i.e., [not + have-to], [why + not], and [not + understand]) are reduced to monosyllabic constructions (see Wilkinson 2016). This claim is further supported by the fact that disyllabic değİl constructions (i.e., [anlamak + değİl] ‘understand not’ type), which are morphologically restricted, have the least grammaticalization schema compared to the other three in the low and very low frequency sets (see Table 6). Thus, the disyllabic [anlamak + değİl] ‘understand not’ type should be unproductive in grammaticalization process, as it is rarely attested with the collocations having low frequency.

In this sense, the disyllabic schema may not be extendable to possible reduced değİl collocations, as there is only one instance of [anlamak + değİl] ‘understand not’ type collocation (i.e., [bulmak + değİl] ‘find not’ with 3 tokens) in both frequency sets (i.e., low frequency and very low frequency). This example shows that there is a parallelism between productivity and the monosyllabicity of abstract schemas. This also means that monosyllabicity and a higher type frequency are needed for a schema to be productive in the signed modality.

It should also be pointed out here that 187 out of 246 tokens of [anlamak + değİl] ‘understand not’ type constructions belong to one reduced type (i.e., [anlamak + değİl] ‘understand not’). This specific example shows that the [anlamak + değİl] ‘understand not’ type schema represents lower type frequency and higher degree of specificity, and [anlamak + değİl] ‘understand not’ collocation plays an important role in shaping the schematicity of it. Following Barðdal (2008: 177), I argue that an inverse correlation exists between the importance of the frequency of tokens and the importance of the type frequency. Thus, token frequency is critical for the productivity domain of the lower type frequency of a construction.

Turning now to the token frequency sets, there is a high and consistent correlation between token frequency sets and constructional change in [sign + değİl] collocations. This grammaticalization view entails that grammaticalization is a gradient with high token frequency. As Figure 20 shows, the distribution of constructional change in the three most frequent sets (i.e., very high, high, and moderate) is notably higher than at least 60 % of the tokens in each set. In contrast, the proportion of grammaticalization in the very low frequency set ranges from one to four and constitutes less than 15 % of its tokens. This implies that the constructional change of değİl ‘not’ increases in proportion to frequency, in line with the traditional usage-based model, which highlights the significant impact of collocation frequency on grammaticalization in the signed modality.

Figure 20:

Distribution of the frequency sets and their schemas.

Overall, this analysis is also consistent with the general idea of grammaticalization, which has been studied in spoken languages (e.g., Bybee 2006; Bybee and Hopper 2001) and sign (e.g., Börstell et al. 2016; Hou 2022; Janzen 2012; Lepic 2019; Wilkinson 2016). The hypothesis that frequency has an essential role in constructional change has been advocated. In summary, this section argues that frequent collocations demonstrate a robust interaction between frequency and phonetic reduction.

4.5 Semantic account of [sign + değİl] collocations

As previously stated, one of the hallmarks of constructional change involves semantic-pragmatic transition (e.g., semantic bleaching, semantic weakening and desemanticization). While early CxG accounts have primarily focused on the typical changes of subjectification for the past thirty years (c.f., Traugott 2010), the position of subjectification in the later stages of grammaticalization has yet to be fully explored in both spoken and sign languages (see Narrog 2015). In this section, building on Traugottian hypothesis that “meanings become increasingly based on the speaker’s subjective belief state or attitude toward the proposition” (Traugott 1989: 31), I will primarily investigate the semantic changes in the constructional change of [sign + değİl] collocations.

Firstly, the verb duymak ‘hear’ in TİD (as shown in Figure 21) typically refers to the act of perceiving or becoming aware of a sound, and our data reveals two distinct reduced sub-schemas in its usage. As illustrated in Figure 22, [duymak + değİl] ‘hear not’ is manifested through both orientation change (i.e., left to front) and a handshape alteration (i.e., 1–5-open). However, the other reduced form, as depicted in Figure 23, is formed solely by an orientation change and does not involve a handshape modification.

Figure 21:

Lexical sign duymak ‘hear’ (duymak ‘hear’, Makaroğlu and Dikyuva (2017) https://tidsozluk.aile.gov.tr/vidz_proc/0097/degiske/97-01_cr_0.5.mp4).

Figure 22:

[duymak + değİl] ‘hear not’ in first reduced form with HS change.

Figure 23:

[duymak + değİl] ‘hear not’ in second reduced form without HS change.

It is worth noting that while the first reduced [sign + değİl] collocation, which has the opposite meaning (with no change) of the given word, duymak ‘hear’ by default, the second reduced form has become autonomous in specializing in ‘breaking any law, agreement, principle, or anything that should be treated with respect’.

Furthermore, vermek ‘give’ in TİD (as shown in Figure 24), means ‘to offer something to someone’ and is captured with two distinct reduced sub-schemas in our data. As demonstrated in Figure 25, [vermek + değİl] ‘give not’ is realized through an orientation change (i.e., back to forward) and a handshape alteration (i.e., flat-o to 5-open). However, the second reduced collocation, as illustrated in Figure 26, is solely formed with an orientation change (i.e., back to forward) with a specific handshape (i.e., closed fist). Interestingly, contrary to the first reduced form, the second reduced collocation expresses a volitional attitude, such as a desire or preference, conveying the meaning of ‘not desire to offer or give’ (Traugott 1989). This can be considered a development from non-subjective to subjective interpretation (Traugott and Dasher 2002: 40).

Figure 24:

Lexical sign vermek ‘give’ (vermek ‘give’, Makaroğlu and Dikyuva (2017) https://tidsozluk.aile.gov.tr/vidz_proc/0018/degiske/18-01_cr_0.5.mp4).

Figure 25:

[vermek + değİl] ‘give not’ in first reduced form with HS change.

Figure 26:

[vermek + değİl] ‘give not’ in second reduced form without HS change.

In signed modality, there is a possible development of the fused değİl ‘not’ collocations into a more subjective construction through nonmanual means. As illustrated below, both reduced forms [iyi + değİl] ‘good not’ have the same manual properties or schema (i.e., [iyi değİl] ‘good not’ type), but the nonmanual part of the second collocation (Figure 27) clearly differs from the first one (Figure 9). In particular, it involves a specific mouth gesture (i.e., closed lips are protruded with a single motion, as if kissing someone) and is used to convey the speaker’s/signer’s beliefs or opinions about the degree of goodness of a proposition. Thus, the construct loses its objective content and is instead deployed for subjectivity, indicating a greater degree of the speaker’s/signer’s involvement.

Figure 27:

[İyİ + değİl] ‘good not’ in reduced form with mouth gesture.

As is widely known, gestures vary in conventionality across different speech communities. In addition, many researchers have suggested that gestures become linguistic forms when they fulfill certain lexical and functional roles in sign languages. Thus, the process of gestures becoming lexicalized and grammaticalized in SLs involves deaf individuals progressively classifying and subsequently entrenching and conventionalizing gestural usage events that are already in use by the hearing population (Janzen 2012; Janzen and Shaffer 2002; Wilcox 2004, 2007).

Wilcox (2004, 2007 explains that there are two possible ways of incorporating gestures into SLs: Route I involves a free-standing, non-conventional gesture that develops into a lexical unit in a SL. For instance, the gestural source of öĞretmen2^[10] ‘teacher’ is a gesture produced with a wagging index finger (back to front), which is still in use among hearing people in Turkey to signal reprimand. Route II, on the other hand, begins with a ‘bound’ gesture (e.g., facial expression) that is first incorporated into the intonation contour of sign language expression.

In line with Wilcox’s proposal, I argue that the mouth gesture associated with the reduced form of [iyi + değİl] ‘good not’ derives from a mouth gesture accompanying brown frowning, which is still used by Turkish people to signal disapproval (see Figure 27), and can be classified as a Route II gesture. Over time, this culture-specific gesture takes a developmental route, bypassing any lexical stage, and acquires a subjective reading in these collocations. It is worth noting that the eyebrow part of this disapproval gesture always modulates the nonmanual feature of the reduced form of the [iyi + değİl] ‘good not’ collocation and changes it from eyebrow raising to frowning, contrary to the basic nonmanual pattern of TİD negation (i.e., backward head tilt and/or eyebrow raising) in all instances. However, as seen in Figure 9, the eyebrow raising is attested in the reduced form of the [iyi + değİl] ‘good not’ collocation without a subjective reading. This makes the mouth gesture accompanying brown frowning a suitable candidate for Route II.

Although subjectification in the grammaticalization of SLs still requires further clarification and elaboration, the examples presented here suggest that the late-stage development of [sign + değİl] collocations can lead to a type of semantic transition that we may call “subjectification”. These observations also indicate that the constructional change brings about a semantic shift from non-subjective to subjective. Note that all the subjective reduced collocations mentioned above, except for Figure 28 (which involves nonmanual subjectification) do not entail any handshape change in their linguistic realization. Thus, this study raises new questions about subjectivity in SLs, such as whether this phonetic cue can be considered a way of subjectification. Therefore, it would be worthwhile to explore further any fusional patterns of subjectification in SLs.

Figure 28:

Turkish gesture meaning disapproval – Turkish football manager Fatih Terim.