The functional definition of relative clauses that I adopt is based on insights from Keenan and Comrie (1977), Downing (1978), Cristofaro (2003), Givón (2009) and Hendery (2012). Functional equivalents of relative clauses – abbreviated as RCs – are characterized by the following features and exemplified in Example (1):

The main function of the relative clause I read is singling out the referent, book, for further elaboration in the main clause, was very interesting. It thus fulfills the function of specification subserving the referential function. On the other hand, the property of being interesting is outside of this grouping, since it is expressed by the main-clause predicate at a higher level of grammatical organization.

The functional definition of the RC constructions rests on a more general functional definition of subordination, which involves a cognitive asymmetry established between two or more states of affairs (Croft 1995; Cristofaro 2003). In terms of the information structure, the subordinated event in RCs tends to present non-asserted, given information, whereas the non-subordinated event asserts, that is, provides new information (Cristofaro 2003; see also Lambrecht 1996). The fact that relative clauses subserve the referential function of the nominal might explain their major discourse functions. It has been observed that relative clauses often act as topics in discourse by presenting referents as anchors (Prince 1981: 236) for subsequent main predication.

In sum, somewhat simplified, (restrictive) relative clauses are used universally as modifiers of head nouns, to identify or further restrict the domain of reference. The relative clause performs this function by depicting an event – known to the hearer and thus ‘presupposed’ – in which the co-referent head noun was a participant as subject, object, indirect object. (Givón 2015). The functional scheme of the relative clause construction is summarized in Figure 1 below. This scheme is going to be essential for the subsequent discussion (Sections 6 and 7) of the changes undergone by the structural cues of the RC construction in ISL, as well as for the motivation of those changes.

Whereas the conceptual definition of the relative clause construction is relatively well-established in the literature, its form varies widely both cross-linguistically and diachronically (e.g., Keenan & Comrie 1977; Hendery 2012). Three major formational parameters are usually distinguished (see Hendery 2012 for a comprehensive overview): the degree of relative clause integration in terms of adjunction/ embedding; internal/external position of the nominal head in relation to the specifying predication; and prenominal/postnominal position of relative clauses in relation to the nominal head.

One of the most crucial RC dimensions is the degree of the syntactic integration between the specifying and major predication: the syntactic, formational relation between a relative clause and the matrix clause may vary from loosely connected (adjoined in parataxis) to closely integrated (embedded in hypotaxis) (e.g. Lehmann 1986, 1988; Hopper & Traugott 2003). The adjoined/ embedded parameter can be further broken down into formational features such as the presence/absence of subordination markers (e.g., relative pronouns, relativizers, etc.), balanced/deranked verb forms, special word order/main clause word order, etc.

Parataxis involves the juxtaposition of two autonomous clauses, neither of which is syntactically dependent on the other, under a unifying intonation contour (Hopper & Traugott 2003), as in Example (2) below. Even though the semantic relation between the two clauses is expressed by prosodic means only, the two autonomous clauses together constitute a complex sentence (cf. Givón 2015).

A construction is more integrated (hypotactic) when a clause is dependent on the main clause. At the right-hand end of the clause linkage continuum are complex sentences consisting of a main clause and an embedded clause. In English, relative clauses can be preceded by a head noun phrase followed by a relative pronoun (Ex. 3). In some contexts, a relative pronoun can be substituted with the invariant relativizer that, which corelates with the high level of embedding, as well as the use of non-finite verb forms (Ex. 4) (Bruyn 1995; Hopper & Traugott 2003; but see Romaine 1984).

Apart from the presence and the nature of the grammatical connecting element, there is another type of structural components whose importance for the RC construction cannot be overestimated. The recent dynamic models of functional subordination (e.g., Langacker 2014) argue that prosody (rhythm, pauses and intonation) not only reflects functional aspects of subordination, but, at the same time, enhances or weakens the degree of clause integration, since intonation units are treated as basic units of thought packaging. Since prosodic constituents have special salience as the level we primarily attend to, they represent windows of attention (Du Bois 1985; Chafe 1994; Langacker 2014 inter alia). Yet, the contribution of prosody to the clause combining in subordination is often overlooked.

Moreover, prosodic characteristics do not constitute all-or-nothing phenomena: the degree of clausal integration might interact with the number (e.g., the presence of pauses, pitch reset, a type of a boundary tone or a pitch accent, etc.) and the intensity of prosodic characteristics (length of pauses, strength of pitch reset, etc.) (Cohen & t Hart 1965; Couper-Kuhlen 1996). Applying this idea to a wide range of relative clauses, Birkner (2008; 2012), checked the widely assumed correlation between semantic characteristics of RCs (e.g., restrictive/ non-restrictive scopes) with particular prosodic designs. The author demonstrated that the link between the semantics of the RC type and prosody is indeed very complex – although non-restrictive relative clauses occur more frequently as separate prosodic constituents, restrictive relative clauses also exhibit this format quite often. The reason for the mismatch might be due to factors beyond the sentence itself, such as information structure and interaction.

Relative clauses can be classified further on the basis of the grammatical function of the head noun and the nominal gap into SS (subject-subject), SO (subject-object), OS (object-subject) and SO (subject-object) relative e clauses (e.g., Sheldon 1974). The present study focuses on subject-subject relative clauses, whereas the variation in other formational parameters is left completely open. In fact, it is the major goal of the study to determine which formational features vary in a young sign language diachronically, and how they interact with each other. In order to tackle this question, Section 2.3 will summarize which dimensions of the RC construction are usually modified in language change.

In spite of the diversity of RC structural types outlined above, the following universal constraints have been hypothesized to shape the diachronic rise of all REL-clause types (e.g., Givón 2015): a) the communicative goal of using REL-clauses and b) the general developmental trend from juxtaposed clause linkage (parataxis) to syntactically integrated construction (hypotaxis). It has been a common assumption that complex constructions in modern languages have their roots in juxtaposition of independent clauses in an earlier stage of the language (Lehmann 1985; Hewitt 1987; Ramat 2000; Hopper & Traugott 2003; Givón 2009 inter alia), or, to put it in Hopper and Traugott’s words (2003: 168), they are a “unidirectional cline from relatively free juxtaposition to syntactic and morphological bondedness”.

Cross-linguistically there are two major ways in which clause subordination might arise: either via the integration of two independent sentences within one sentence or via expansion, that is, the reinterpretation of a thing-like (nominal) participant as a propositional (clausal) participant (Heine & Kuteva 2007, chapter 5; Givón 2009; 2015).1 The integration pathway, or, in Givón’s (2009) terminology, the clause-chaining pathway, involves a starting point where two chained clauses with separate intonation contours become subsumed under a unifying intonation contour, and is illustrated in (5), an example from Bambara, which is ambiguous between a chained construction and a relative clause.

The expansion pathway attempts to account for the process by which devices that first served to structure independent sentences come to assume functions of subordination. One specific instantiation of this pathway is the restructuring of a non-restrictive (parenthetical) relative clause (6a) into the restrictive relative clause (6b). Crucially, according to this pathway, the word order is reorganized and the demonstrative DIE is grammaticalized into the relative pronoun. In the process of grammaticalization, the demonstrative loses its strong contrastive focus semantics and becomes phonologically reduced.

Unfortunately, historical information on this sort of grammatical change in the languages of the world is scanty (Givón 2009: 119), and many of the reconstructions proposed are based on applying the methodology of grammaticalization theory to synchronic linguistic data, even if a small number of the reconstructions are also supported by attested historical evidence (McConvell 2006; see Hendery 2012 for an overview).

One reason why the parataxis-to-hypotaxis hypothesis is so difficult to support empirically might come from the claim that the onset of either subordination pathway – integration and expansion – seems to involve one crucial and obligatory step – adjustment and merger of intonation contours (Sankoff & Brown 1976; Traugott & Hopper 2003; Givón 2009; Mithun 2009). The investigation of the RC genesis in young contact languages, such as pidgins and creoles (e.g., Sankoff & Brown 1976; Bickerton 2016) might provide a solution for this problem. Bruyn (1995) demonstrated that RCs in Sranan, a creole language spoken in Suriname, developed as a result of a complex interplay between several interrelated processes: (i) the development of relative markers from demonstrative pronouns, (ii) the grammaticalization of anaphoric relations in adjacent clauses, and (iii) gradual integration of the clauses from parataxis to full subordination.

Yet, due to the limitations of the writing systems and the scarcity of language documentation, diachronic changes of prosodic integration are not systematically reflected even in (relatively) young spoken languages. The lack of solid empirical evidence led a number of researchers to conclude that the parataxis-to-hypotaxis hypothesis should be abandoned altogether (Bednarczuk 1980; Harris & Campbell 1995; Roberts 2007; Weiß 2020 inter alia). The discussion in the current section highlights what is missing to complete the argument: diachronic evidence supporting or refuting proposed ‘pathways of change’, and the role of intonation in this process. The present study aims to provide exactly this sort of evidence.

The field of emerging sign languages has laid important foundations for understanding how grammatical complexity in general and subordination in particular arise. Young sign languages provide a natural laboratory for investigating this question, both because they can emerge de novo at any time, and because the cues of linguistic organization are overtly visible. Importantly, findings imply that conceptual organization of communicative messages precedes linguistic organization, an implication that will be relevant for the present study. This section will first consider what we know about the emergence of manual relative clause markers, and then will proceed to a more global restructuring of subordinate constructions, including their non-manual marking.

The objective of the present study is to trace the changes that affect prosodic and morpho-syntactic cues in the RC equivalents produced in the three age groups of ISL signers. The following hypotheses will serve as reference points for the overall discussion in the present study:

ISL is the established language of the deaf community in Israel (Meir & Sandler 2008). It is a young sign language, roughly 90 years old, which arose with the formation of the deaf community around the 1930s, together with the establishment of the first Israeli School for the Deaf in 1932 in Jerusalem. Immigrants from all over the world further contributed to the emerging signing system. A conventional local sign language evolved, and today, ISL is used by roughly 10,000 people in a wide range of settings, including the educational system, deaf organizations, interpreting programs, the media and even the parliament. The linguistic structure of ISL has been investigated in earlier work (e.g., Meir & Sandler 2008) and its emergence has more recently become the object of study (e.g., Meir 2016).

In this study, we recruited Deaf participants representing different generations of ISL users. The equivalents of relative clauses were videotaped from three age groups of ISL signers: 20–35, 36–50, and 51–72 (Table 1). The methodology relies on Labov’s Apparent Time hypothesis, which assumes that a person’s language does not change significantly after puberty and infers diachronic changes from synchronic data collected from different age groups (Labov 1963).

The study used a communicative task created to fit all three age groups of signers. The task was built in such a way so that it would necessitate the use of Subject-Subject (S-S) relative clause equivalents on the one hand, and, on the other hand, would be free from spoken language interference.

The study employed the technique of referent identification, which builds on the insight of Zukowski’s (2009) original design targeting the identification of referents with expanded noun phrases. In the task, two identical characters, distinguishable only by their participation in different events (e.g., a girl swinging vs. a girl rocking on a rocking horse), are introduced by a native deaf experimenter. Pictures of both characters engaging in an additional activity are then shown (e.g., in Figure 4, one girl is shown eating ice cream in addition to swinging, and the other one is shown drinking in addition to riding a horse).

The participant appointed as ‘the director’ examined the pictures on the cards, observed the changes that had occurred and told a deaf ‘partner’ from the same age group which character was doing which action. A native deaf experimenter asked a double question about the observed situation (e.g., Which girl is eating ice-cream and which girl is drinking?). The director’s answer to this double question enabled the partner to choose the appropriate picture from a set of fours cards containing mixed combinations of the referents and their actions. Due to the double question, the expected answer is the equivalent of a restrictive relative clause embedded into a full matrix clause “The girl who is swinging is eating an ice cream, and the girl who is riding the rocking horse is drinking.” Signers received four sets of picture pairs. This means that each signer came up with a maximum of eight responses (eight equivalents of S-S restrictive relative clauses).

The data were glossed and translated with the assistance of a native Deaf consultant. All coding was implemented using ELAN, a video annotation software (Crasborn & Sloetjes 2008). The relevant categories of coding included glosses produced by both hands, rhythmic cues of prosodic constituency (e.g., holds, reduplications, blinks, etc.), gaze direction, facial expressions (e.g., brow raise, lid tightening, cheek raise, etc.), head and torso movements (e.g., head movements forward and backward, nods), and others. Special importance was allotted to the coding of manual signals of prosodic constituency, because the analysis is dependent on the parsing of an utterance into prosodic units. The parsing of the signing stream into prosodic constituents is based on the presence or absence of such signals for prosodic boundaries as pauses and holds, reiteration of the final sign, as well as slowing down or increasing the size of a sign, all at the end of prosodic constituents (Sandler 1999). Since the aim of the study was to provide empirical evidence for the emergence of the relative clause construction, the following criteria were chosen for the coding and analysis of the structural cues4:

With no exception, all the communicative partners in the referent identification task were able to correctly identify the referents described by the relative clause equivalents.

With no exception, in all the RC clause equivalents produced by the ISL signers, the clause which was intended to be presupposed in the task (specifying predication) preceded the clause with the main predication (the one which was intended to be asserted in the task). The nominal head appeared in the majority of the RC equivalents in the mid and younger ISL group (67% and 75% respectively). In contrast, older signers produced nominal heads less than in half of all the responses – 41% (Table 2). In addition, in all the age groups, all the responses where the nominal head was explicitly signed, it preceded the specifying predication. In other words, the RC equivalent appeared post-nominally for all age groups.

Forward head movement (FHM) in relative clauses is quite frequent with older signers and becomes even more frequent in the mid and younger age groups, which indicates its increasing systematicity as ISL develops. Points appear in only 15 per cent of the older signers’ responses, whereas they appear in the majority of the mid age group and younger group responses (60 per cent and 68 per cent respectively). The scarcity of demonstratives in the older group production indicates a lower degree of their conventionalization. Table 3 also demonstrates that in the raw data squint appears three times as often in the relative clause equivalents of the middle-aged and younger signers than in those produced by the older group. The multiple logistic regression analysis run for these data demonstrates that the frequencies of FHM, squint and pointing strongly correlate with signer’s age group, with a p-value < 0.01.

In sum, the results of the analysis indicate that ISL signers of all the three age groups were able to identify the referents regardless of the linguistic structure of RC equivalents. Yet, the age groups differed in the degrees of the structural integration of the RC equivalents. Older signers’ equivalents included fewer structural cues and they were less likely to align with the RC constituents than those produced by the younger signers. As a result, the distributional scope of squint diachronically stays behind the distributional scope of FHM, whereas the distributional scope of the pointing is, in turn, one generation behind the scope of squint. In other words, the signals of the RC in ISL gradually change towards a tight and unambiguous mapping with the RC structure.

The feature that strikes at the first sight at the RC equivalents produced by the older age group is that many of them look choppy and disintegrated. This overall impression can be accounted for by several objective measures presented in the section 5 above. First of all, almost half of them (45%) are free – produced without the nominal head, as in Figure 7 below.7 Importantly, pointing signs tend to substitute the nominals in many older signers’ relative clause equivalents, and appear utterance-initially in a separate intonational phrase, as in the example below. As a result, the RC equivalent constitutes at least three disintegrated intonational phrases.

Forward head movement (FHM) is very frequent in the RC equivalents produced by the older ISL signers – more than 70 per cent of older signers’ responses display this intonational component. This fact already indicates a high degree of FHM conventionalization in this context in the older age group, which comes into a sharp contrast with a low frequency of squint and demonstratives in this age group – 29% and 16% of the responses respectively.

The distributional patterns of the structural cues – FHM, squint and pointing – in relation to linguistic prosodic constituency in RC equivalents might explain varying degrees of their conventionalization. According to the first distributional parameter – ‘linguisticness’, FHM is highly integrated into the linguistic system of ISL: in 96% of all the FHM tokens, the scope of FHM is aligned with prosodic constituents as determined by the manual cues, such hold, reduplication, or sign size. In contrast, squint produced by older signers in a quarter of all the cases does not obey prosodic constituency and appears to be not aligned with boundaries of corresponding intonational phrases. This happens, when squint crosses prosodic boundaries or occurs in the middle of an intonational phrase.

A similar conclusion can be drawn about pointing signs. As demonstrated in Figure 7 above, in those rare cases when points do appear in the older signers’ responses, they usually constitute a separate prosodic unit. In such cases, the demonstrative sign is not integrated prosodically and is used in a gesture-like way, by pointing to a real-life object in the signer’s environment (e.g., a card). In this sense, its linguistic status is borderline.

The FHM distribution in headed RC equivalents and the patterns of FHM alignment for the older signers presented in Section 5, Results, demonstrate that this intonational marker most often co-occurred with the nominal head of RC equivalents at the early stages of ISL development, as in Figure 8 below. In the responses produced by the older signers, the FHM toward the addressee co-occurring with the nominal signals that the lexical material marked by this nonmanual cue serves as an anchor for subsequent material. Furthermore, as pointed out in the previous research, FHM is the most frequent marker of ISL topics (Dachkovsky & Sandler 2009; Dachkovsky et al. 2013). This function of FHM is also corroborated by the basic attention anchoring function of FHM in human communication in general (Lynn 2013). In sum, in the older age group, FHM mostly serves the information structuring function of the marker of topic/comment asymmetry. Thus, in Figure 8, the topic of the utterance, the referent ‘man’, is first introduced, and then the predication of watching is anchored to this topic.

The link suggested here between clausal dependency and information structure categories, such as topic-comment, is not novel. As explained above, under the functional approach the notion of subordination is related to two basic discourse distinctions, foreground/ background and figure/ ground,8 and referred to as the Asymmetry Assumption by Cristofaro (2003) and Langacker (2014). It rests on the assumption which assumes an asymmetrical cognitive relation between two events, such that one event imposes its own profile over the other.

Figure 8 also demonstrates that, apart from FHM, squint co-occurs only with the nominal constituent of the relative clause equivalent, which appears to be the topic of the utterance. Squint co-occurring with the topic constituent indeed represents a default scope of squint in the older age group, in those rare cases when squint appears in the oldest age group’s relative clause equivalents at all. Since topics serve as anchors that help to relate new information to what is already known (accessible) to the addressee, it makes sense that topics themselves may be known to different degrees, that is, that they can have different accessibility rankings (Krifka 2008; Dachkovsky et al. 2013). The Low Accessibility topic (Ariel 1991) in the present example stands in explicit competition with the other referent on the second card of the task. Therefore, the squint here “points” to the less automatically accessible item in the discourse, and instructs the addressee to focus on it.

The most common distributional pattern of the cues in the older age group is schematized in Figure 9 below. It illustrates the fact that older signers’ RC equivalents are very fragmented and disintegrated, since the non-manuals cues usually span only over the nominal referent, so that the specifying predication is prosodically not related to the nominal. Moreover, the pointing sign often appears in a separate intonational unit, resulting in a disjoint sequence of 3–4 intonational phrases.

The mid age group’s RC equivalents present a drastic change in some parameters of the RC construction and the entrenchment of the previous changes for other parameters. The most striking difference between the older age group and the mid age signers’ RC equivalents is the increase in the RC complexity. The overwhelming majority of the RC equivalents in this age group are preceded by the nominal head, which contrasts sharply with the half of the free RC responses in the older age group. The frequency and the distribution of the RC cues can, at least partially, account for integration of the nominal head.

FHM becomes even more frequent in this age group – approximately 90 per cent of RCs produced by the middle-aged signers had FHM. This indicates that FHM has become completely conventionalized and tightly associated with the RC construction. Unlike FHM, squint undergoes an abrupt change in its frequency – it increases from 29 to 89 per cent of the responses. The usage of the pointing demonstratives also increases sharply – they appear in the majority (60%) of the total number of RC equivalents, but they are still not as widespread as squint or FHM.

In terms of linguisticness, the mid age group signers align all the three signals with linguistic prosodic constituents, which differs from the older group’s behavior of squint and pointing signs. The most common (70 %) distribution of FHM in the mid age group is the co-occurrence of FHM with the whole RC together with the nominal under one unifying contour. This means that at this stage the signers are able to unify the nominal and the specifying predication into one constituent by the virtue of FHM spread over both of them, as in Figure 10, where FHM spreads both over the nominal GIRL IX and over the specifying predication SWING IX. At this stage the specifying predication loses the status of an independent predication and becomes reanalyzed as the property of the nominal. This change is encoded by the alignment of FHM both with the nominal topic and the specifying predication.

Starting with the middle-aged group, FHM ceases to be a marker of merely the (nominal) anchoring point and becomes a marker of the dependence and asymmetry between the whole RC equivalent (together with its nominal) and the main predication, thereby moving toward a closer correlation with the functional schema of the RC construction. In other words, FHM becomes the signal of dependency at a higher, constructional level, rather than a mere marker of topic anchoring at the level of information structure.

Changes in the squint distribution do not exactly match the changes in its frequency. As demonstrated in Figure 10 above, squint in the middle-aged signers tends to exhibit the following peculiar pattern: first, it is aligned with the nominal referent of the relative clause; then it relaxes; and afterwards it re-activates and aligns with the specifying predication. In other words, the distribution of squint in this age group is transitional between a Low Accessibility reading attributed to the topic constituent and a marker of the nominal together with the specifying predication. I propose that a reset and a further spread of squint over the specifying predication contributes to the reading under which the action performed by the referent is re-analyzed and interpreted as the property belonging to the referent, providing its specification and assisting its unique identification. Nonetheless, for the middle-aged signers this process of the reanalysis is still not completed, and the partial termination of squint after the nominal and its further reset is the evidence of the fact that the meaning change is still on its way. Thus, the distribution of squints in the middle-aged group demonstrates a gradual drift from the marker of a Low Accessibility topic to the specification marker associated with the function of the RC as the property specifying the referent.

Unlike a dramatic increase in the frequency, the distribution of the pointing signs in relation to the structural RC constituents has not changed significantly in comparison with the older group. Although a quarter of the pointing sings in the mid age group occur both immediately after the nominal referent and at the prosodic boundary between the clauses, as in Figure 10 above,9 still the majority of the pointing tokens (68%) appear after the nominal head only (boy, girl, woman, man), as in Figure 11 below.

In this example, the pointing sign modifies the nominal boy and immediately follows it within the boundaries of the same intonational phrase delineated both by rhythmic cues (reduplication) and by the spread of the non-manual signals (head movement forward and squinted eyes). This kind of distribution suggests that the function of the pointing is still adnominal – it modifies the noun and identifies the referent expressed by the noun.

In sum, in the mid age group, the FHM function approximates the functional profile the restrictive relative clause, while the usage of the pointing sign lags behind and still modifies the nominal referent itself. The distribution of the squint indicates that occupies the intermediate position between those two functions. The scheme in Figure 12 summarizes the most common distribution of the RC cues in the mid age group.

In the 3rd, youngest age group, the intonational non-manual cues continue their drift toward a close matching with the functional profile of the RC construction. First of all, the overwhelming majority of the RC equivalents in this age group appear with the nominal referent – the nominal head precedes the specifying predication in 80% of the responses, indicating that structurally complex referents can be integrated as part of the construction.

FHM appears in the overwhelming majority of the RC equivalents. Its scope systematically spreads between the nominal referent and specifying predication. The tendency of FHM to group together the nominal with the specifying predication, as a unifying contour, observed with the middle-aged signers, becomes even stronger with younger signers (92% of all the responses with the nominal referents) and points at the conventionalization of this type of distribution as the language matures. This type of distribution encodes that the specifying predication corresponding to the RC has been fully integrated into the nominal constituent, has lost its independent predication status and has acquired the status of the property of the nominal. As a result, the complex nominal constituent containing the RC predication is in the asymmetric relations with the main predication. In sum, the FHM has become a subordination signal in the newly emerging RC construction in ISL. The example in Figure 13 demonstrates a typical pattern of FHM movement observed with the younger signers.

The scope of the second intonational component, squint, also adjusts itself to the functional profile of the RC construction in the youngest age group. Figure 13 also demonstrates that, in contrast to the middle-aged signers’ distributional pattern, the behavior of squint in the youngest group parallels that of FHM – it tends to span the nominal and the specifying predication together, without a reset, taking the whole construction into its scope. In such cases, squint does not mark the referent only, but spans over the whole nominal constituent, similarly to the function of the definite article in English, which acts as a determiner for the whole noun phrase and not only the referent (noun). That is, the function of squint in the youngest group has been reanalyzed from the signal cuing the information structure status of the referent to the one subserving the constructional function of referent specification.10

Although the span of the intonational cues – FHM and squint – in the youngest group integrates the nominal referent with the specifying predication into one intonational phrase, the distribution of more than a third of pointing signs has still a double valence – they can occur after the nominal head and at the boundary between the clauses, as in Figure 13 above. In this sense, the distribution of the morpho-syntactic marker of the RCs lags behind the functional changes of the intonational cues. This interaction of the manual and non-manual cues is summarized in Figure 14 below:

Yet, a different, more integrative pattern of the demonstrative emerges in a third of all the cases of pointing in the youngest group. In this type of distribution, the pointing sign appears exclusively at the right boundary of the relative clause, thereby finalizing the integration of the nominal referent with the specifying predication into one complex nominal constituent, as in Figure 15 below.

Thus, we are observing the reorganization of the whole relative clause construction where various structural devices – facial expressions, head movements and manual signs – obtain more and more constructional functions, thereby establishing and highlighting the constructional boundaries (Figure 16).

The discussion of the individual stages in the changes of all the RC structural components in ISL demonstrates that we are observing a gradual and incremental genesis of the relative clause construction where various structural devices – facial expressions, head movements and manual signs – are reorganized and gradually obtain more construction-related functions. The complex of the non-manual signals act as a skeleton framing the whole construction and uniting the referent and the specifying predication into one nominal component. The reorganization of the structural cues and the crystallization of the RC construction can be summarized by the following scheme in Figure 17. The hypotheses entertained in Section 3.4 will be considered below in the light of this scheme.

The present study has a number of limitations. One important caveat is that the paper focuses on only one type of relative clauses – subject-subject relative clauses, leaving other types of relative clauses for future research. This fact leads to an obvious difficulty in determining certain aspects of syntactic structure of relative clauses.

Another caveat concerns the choice of participants in the three age groups: the participant’s age of acquisition and the generation order might be confounding factors. Specifically, the older signers in the study are not native signers, most of them having acquired ISL later in life. On the one hand, this situation truly represents the first generation of ISL users. On the other hand, the contribution of each factor in language emergence cannot be clearly determined. In order to overcome this limitation, I am planning a follow-up study that would investigate non-native deaf signers’ production of relative clauses across the three age groups matched for the age of ISL exposure.