Complement clauses in Russian Sign Language

(1)

Complement clauses in Russian Sign Language

Author: Evgeniia Khristoforova

Student number: 12071528

Supervisor: Roland Pfau

Second reader: Enoch O. Aboh

Qualification: RMA Linguistics

Institution: Amsterdam Center for Language and Communication – UvA

(2)

2

Although complement clauses have been an object of typological research in spoken languages for many years, this syntactic construction has attracted the attention of sign language linguistics only recently. Nevertheless, the existing studies suggest that sign languages employ complement constructions that are very similar to those identified in spoken languages. At the same time, complement clauses in sign languages exhibit modality-specific properties, such as an infrequent use of morphosyntactic and lexical cues for sentential complementation such as overt complementizers or markers of (non-)finiteness. The present study provides the first description of complement clauses in Russian Sign Language. By

combining corpus investigation and grammaticality judgments, I investigate morphosyntactic properties of complement clauses selected by different semantic classes of complement-taking predicates. I found that complement clauses in Russian Sign Language generally come in two types: (i) control constructions with a covert embedded subject and a deranked paradigm of subject agreement on the embedded predicate, and (ii) sentence-like complement clauses. Control complement clauses can be fully

embedded in the main clause or they can occupy a sentence-final position, while only the latter option is available for sentence-like complement clauses. In addition, I observed that sentence-like complement clauses in Russian Sign Language are frequently introduced by a complementizer sign,which is

typologically rather unusual in sign languages. At the same time, the obtained results are compatible with complementation patterns identified in other signed and spoken languages and thus contribute to the typological studies in the field of complex syntactic structures across modalities.

to the memory of Tatyana Davidenko

1 Introduction

Complement clauses, i.e. embedded clauses functioning as an argument of the predicate, are a canonical example of syntactic embedding. An extensive typological research (Givón 1980, Cristofaro 2003, Noonan 2007) reports an astonishing variety of different forms that this construction can take across languages or even within one language across different complement-taking predicates.

Although being relatively well studied in spoken languages, sentential complementation has, until recently, been largely ignored in sign language (SL) research, since sign languages were assumed to be too simplistic to display syntactic complexity. However, recent studies (Pfau et al. 2016) provide convincing counterevidence suggesting that SLs do not only employ complex sentences but the patterns identified in SLs are consistent with syntactic models that have been proposed to account for the complexity in spoken languages. Still, the typological research on complex syntactic structures in SLs is lagging behind, as only a few SLs have been described in terms of the sentential complementation.

This study provides the first description of complement clauses in Russian Sign Language (RSL). By combining a corpus investigation and a grammaticality judgment task, the study aims to identify basic properties of complement clauses across different semantic classes of complement-taking predicates in RSL. The analysis reveals that complement clauses in RSL can be sentence-like (i.e. represent the same structure as a simple clause) or control constructions (i.e. non-finite clauses with a covert embedded subject, whose reference is controlled by one of the arguments in the main clause). Accordingly, different types of complement clauses exhibit different distribution and morphosyntactic properties. In addition, the frequent use of a complementizer sign was observed, which is typologically unusual among SLs. This study also proposes a tentative cross-linguistic and cross-modal comparison. As a result, it reveals that complement clauses in RSL do not only fit into the general picture of complementation strategies across SLs and beyond but also exhibit some language-specific properties, such as the use of a

(6)

6 complementizer and the patterns of verbal deranking in the embedded predicates (i.e. how the

embedded predicative from is different from its main clause counterpart).

The paper is structured as follows. First, the general background on complement clauses is given in section 2. Specifically, it overviews the general typological tendencies in spoken languages and the way they are accounted for from the generative perspective (2.1), and it also introduces the most relevant SL studies and the general information about RSL (2.2). Section 3 presents the various methodological and analytic tools that were implemented in this study. Section 4 is dedicated to the joined results obtained from corpus investigation and the grammaticality judgment task, with each subsection targeting one semantic type of complement-taking predicates. The discussion,

cross-linguistic comparison and tentative theoretical analysis are given in section 4.8.4. The paper is concluded by the general summary of the findings and the perspectives for the future research.

2 Background

In the first part of this section, I introduce some basic notions related to sentential complementation. In so doing, I primarily build upon typological and theoretical studies on spoken languages. I define

complementation as a syntactic and semantic phenomenon in section 2.1.1. Then, I introduce the notion of complement-taking predicates (CTP), which are the ones introducing complement clauses. Further, I move to the different types of complement clauses (section 2.1.2) focusing on their clause-internal morphosyntactic properties (section 2.1.3) and the structure of the complex sentence as a whole (section 2.1.4). I further illustrate the complex interaction between these two dimensions that has been captured by the complementation hierarchy (Cristofaro 2003) (section 2.1.5).

In the second part of the section, I switch to existing SL research on complement clauses. First, I familiarize the reader with some SL background relevant to complementation. Then, building upon the research on four sign languages(SLs) —SL of the Netherlands (Nederlandse Gebarentaal —NGT), French SL (Langue des Signes Française — LSF), Italian SL (Lingua dei Segni Italiana — LIS) and Turkish SL (Türk

İşaret Dili —TiD) — I illustrate that a) SLs do have complement constructions, b) complementation in SLs

can be analyzed using the same set of notions as the ones developed for spoken languages, and c) even though sharing a lot of structural properties, SLs do exhibit modality-specific and language-specific features concerning sentential complementation. I will conclude the section by a brief overview of the relevant research on Russian Sign Language.

2.1 Sentential complementation: definition and basic notions

2.1.1 The struggle of defining sentential complementation

Complement clauses (CCs) have been investigated in numerous typological studies, and, as it often happens with such thoroughly investigated phenomena, it is not easy to propose a definition that will embrace the whole variety of forms that sentential complementation takes across languages.

The first challenge is to disentangle semantic complementation and syntactic complementation. The former is defined as two state-of-affairs, the dependent state-of-affairs providing the specification for the main one (Cristofaro 2003). Although such semantic environment is favorable for syntactic subordination, it is not a necessary condition for complementation to take place. Consider the Muna (Austronesian, Indonesia) example in (1) adopted from (Cristofaro 2003: 96).

(1) a-kona-e ome-gholi ghunteli

1SG:REAL-think-it 2SG:REAL-buy egg ‘I thought you had bought eggs.’

(7)

7 (Van den Berg 1989: 243)

According to Cristofaro (2003), a pronominal form on the verb marked by –e signals that the second clause is not embedded but juxtaposed. In other words, (1) can be interpreted as ‘I think this: you had

bought eggs’. The present study is solely focused on syntactic complementation, which implies syntactic

subordination, and thus does not take examples like (1) into account.

The second challenge concerns the contrasting view of various theoretical frameworks, which defined sentential complementation in a slightly different way. What all these definitions share is that a CC functions as an object of the predicate in the matrix clause. However, the range of constructions defined as complements vary across different frameworks. Thus, the functional approach would define any predication functioning as a complement as an instance of sentential complementation (Noonan 2007). These would include complements to the auxiliary-like elements (e.g. English modals) and non-complement subordinate clauses (e.g. appositive clause and noun non-complements). Such a broad perspective on the sentential complementation is advantageous from a typological point of view, because it takes language variation into account. However, from the structural perspective not all of these entities qualify as CCs. In this study, I adopt a minimalist perspective on sentential

complementation, which defines CCs as Tense Phrases (TPs) or Complementizer Phrases (CP) selected by a matrix predicate. Still scoping over the wide range of construction types, this definition excludes constructions involving auxiliary-like elements, because these select for Verbal Phrases (VPs). It also rules out nominal complement clauses (i.e. appositives and relatives functioning as complements) because these have a Determiner Phrase (DP) as a terminal node.1

By ‘excluding’ all these constructions from further analysis, I do not imply that they will be completely disregarded. Quite the contrary, some of them will be encountered rather frequently in the data as an alternative way to carry out the semantic complementation. The competition between different complementation strategies becomes very relevant when it comes to the emergence of syntactic complementation and its further syntactic integration. However, in order to create reasonable limitations for this study, I will limit the scope to the finite and non-finite TP/CP complements.

2.1.2 Complement-taking predicates

Let us now focus on those predicates which can select for clausal complements, namely complement-taking predicates (CTP). All studies on sentential complementation unanimously agree that semantic properties of CTPs largely govern morphosyntactic properties of their clausal complements. According to the literature (Cristofaro 2003, Noonan 2007), CTPs can be subdivided into the following semantic classes:

 Desiderative predicates express a desire for the complement proposition to be realized. English examples: want, prefer, hope, wish …

 Achievement/implicative predicates express the reason for the success (or failure) or the manner of the achievement (or the lack of thereof) described in the complement clause. English examples: positive achievement predicates — manage, chance, remember; negative achievement predicates — try, forget, fail …

 Manipulative predicates express a relation between an agent, an affectee, and a complement proposition, in which the affectee participates. Contrary to other semantic classes, these CTPs are select for two internal arguments: an affectee and a resulting situation.

(8)

8 English examples: force, make, let …

 Impediment predicates take an agent subject and express the fact and/or the manner of an action that intends to prevent the embedded proposition from happening or functioning in a proper way.

English examples: hinder, prevent, forbid …

 Pretence predicates describe a situation in which the agent subject deliberately behaves as if the embedded proposition or predication was true although he/she knows that it is not. English examples: pretend, imagine …

 Commentative predicates specify the personal attitudes of the experiencer towards the

complement proposition, i.e. provide a comment on the complement proposition in the form of an emotional reaction.

English examples: like, dislike, hate, regret, relieved, sorry …

 Predicates of fearing express fear or concern that the complement proposition will be realized. English examples: be afraid, fear, worry, be anxious …

 Immediate perceptive predicates take an experiencer subject and express the sensory mode and the manner by which the event described in the complement proposition is perceived. English examples: see, hear, feel …

 Knowledge or acquisition of knowledge predicates take an experiencer subject and specify the fact and/or the manner of acquiring a knowledge described in the complement proposition. English examples: know, learn, discover ….

 Propositional attitude predicates take an experiencer as a subject and express the degree of certainty towards the truth value of the embedded proposition while not claiming it to be true. English examples: positive attitude — believe, think, assume; negative attitude — doubt, do not

think, do not know …

In addition to the semantic classes above, yet another three semantic classes, i.e. modal (e.g. can, must), phasal/aspectual (begin, continue) and utterance (say, ask) CTPs, are distinguished in the literature but excluded from the present study for several theoretical and practical reasons. The former two classes are cross-linguistically prone to grammaticalization as auxiliary-like elements (e.g. most English modal predicates). Even though it need not always be the case in a particular language, from typological perspective it is likely for these predicative classes to fall out of the scope of the definition proposed for CTPs in this study. In parallel, the utterance predicates often introduce a direct speech, which lacks syntactic subordination. Consequently, these constructions oftentimes do not qualify for a structural definition of CCs. Besides, some phasal and utterance predicates were already tentatively overviewed in the previous research on the related phenomena in RSL (see section 2.2.4 for details). Thus, in the present study, I am focusing only on the subset of semantically defined CTPs that have the greatest chances to take TP or CP as complements according to the typological studies. The future research, however, should complement this investigation by taking modal, phasal and utterance predicates in account.

(9)

9 Noonan (2007) claims that the semantic classes of CTP described above are attested in most languages of the world. However, languages vary on what properties of CC particular semantic class of CTPs can license. Let us zoom into the attested variation in these properties so that we can further map them on the semantic properties of CTPs.

2.1.3 The typology and syntax of complement clauses: the view from inside

This section is dedicated to the morphosyntactic properties of ССs, which are internal to the subordinate clause. As this study primarily concerns SLs, I focus on those morphosyntactic properties that are

relevant for SLs. Thus, I will skip the grammatical mood and only briefly tackle case-assignment, because no SLs has been reported to have a morphological coding for these categories.

2.1.3.1 Morphosyntactic features of embedded predicates in complement clauses

In a very first approximation, CCs can be divided into sentence-like and non-sentence-like, the former being undistinguishable from the independent sentence if observed in isolation (Noonan 2007). Thus, the embedded predicate preserves all its inflections that it would express in the main clause in a particular language.

In non-sentence-like CCs, the embedded status of CC-predicate is marked either by subjunctive mood or by a defective (or deranked in Stassen's (1985) terms) paradigm of verbal inflection. Languages vary dramatically in terms of what verbal categories are preserved in the deranked forms of embedded predicates and which levels of deranking are attested. However, according to Noonan (2007), the hallmark of the deranking process is the loss of the structural relations with the embedded subject, which surfaces in the non-nominal case-assignment and defective subject agreement, if any of these categories are morphologically encoded in a language. The way the verbal deranking affects the embedded subject will be discussed in section 2.1.3.2. Here, I focus on its circumstances for the inflectional paradigm of the embedded predicate.

In some language, the loss of the subject-predicate relation is the only deranking that the infinitives undergo. Thus, embedded predicates in Classical Greek lose the subject agreement but keep the full range of Tense and Aspect inflections (2).

(2) Fēsì grápsai/ gegráfénai/ gráfein/ grápsein

say-3SG write.AORIST:INF/write.PERF:INF/write.PRES:INF/write.FUT:INF

‘He says that he wrote/ has written/ is writing/ will write.’

(adopted from Noonan 2007: 68) Although finiteness is often associated with the marking of Tense-Aspect-Mood (TAM) categories alone, the Classical Greek case suggests that the subject agreement is a more reliable marker of finiteness. Crucially, shifting the focus of the finiteness diagnostics from TAM markers to subject agreement markers allows us to analyze finiteness in those languages that do not possess systematic morphosyntactic coding for Tense (such as Classical Chinese and all SLs).

In languages such as Russian, verbal deranking leads to the loss of tense marking but keeps aspectual distinctions (Noonan 2007). Ultimately, some languages lose all TAM inflection keeping only transitivity, voice and/or object agreement markers if they are present in a language.

2.1.3.2 Embedded subjects in non-finite embedded clauses

Similarly to predicates, subjects of sentence-like CCs exhibit the same morphosyntactic properties as subjects in the simple clause. However, in non-finite clauses, the embedded subject can be realized in several ways depending on its referential properties and morphosyntactic environment. Thus, if the

(10)

10 subject of the non-finite CC is coreferential to the matrix subject, it can be deleted under identity (i.e. equi-deleted). The subject position is thus occupied by a phonologically null pronoun PRO, whose reference is controlled by the matrix subject, making this construction a control clause (Polinsky 2013). Under the traditional generative account (Chomsky and Lasnik 1993), the absence of phonological forms for PRO is due to the null case checked with a null CP specified for an uninterpretable null case feature, as illustrated in Figure 1.

Figure 1. The syntactic derivation of a control construction

If the embedded subject is not coreferential to the matrix subject, it surfaces as an overt DP. However, contrary to the matrix subject, it cannot check its nominative case due to the non-finite status of the embedded predicate. Nevertheless, following the Extended Projection Principle (EPP), the embedded subject raises to the specifier position of the non-finite TP, where it receives its case from the matrix predicate via head-spec licensing (Figure 2). Such constructions are referred to in the literature as Subject-to-Object raising, which is a particular case of Exceptional Case Marking (ECM) construction.

Figure 2. The syntactic derivation of the Subject-to-Object raising construction

The last scenario to be mentioned is specific to the CTPs that take two objects. In case of manipulative CTPs, for instance, these are the DP corresponding to the Goal of manipulation (i.e. an affectee) and the

(11)

11 CC describing the imposed action. Typically, the affectee is coreferential to the subject of the embedded clause, which gets equi-deleted. The resulting structure is very similar to one proposed for subject control in Figure 2. The crucial difference is that it is the direct object of the main clause, not the subject, that controls the reference of the PRO in the CC, hence the term ‘object control’. As in the case of subject control constructions, the null Co_{specified for an uninterpretable null case feature is presumed} ensuring the phonologically null PRO (Figure 3).

Figure 3. The syntactic derivation of object control construction

The complexity of derivations illustrated in this subsection is necessary to account for the astonishing variety of forms that CCs take across languages. However, if a language lacks overt morphosyntactic coding for such categories as case or TAM (as most SLs do), the underlying structure attributed to sentential complementation is hard to trace.

2.1.3.3 The use of complementizers

In many languages CCs are introduced by a specialized functional element — the complementizer. The type of complementizer is largely determined by the selectional features of the CTP. In its own turn, the complementizer defines many CC-internal morphosyntactic properties (e.g. illocutionary force,

finiteness, subject case, etc.), thus being an essential mediator between the CTP and its clausal complement.

Notwithstanding the contribution that complementizers make to the morphosyntax of the CC, there are languages that lack overt complementizers or do not use them on a regular basis (SLs among others). The UG approach, however, implies that even if the complementizer is not observed, it still can be presumed on the underlying structural level in order to saturate the selectional properties of the CTP as in control constructions above. The null complementizer thus serves the purpose of unifying the structure not only for different types of CCs but also for different languages, which is in accordance with the general principles of UG framework.

(12)

12

2.1.4 The syntactic properties of complement constructions: the view from outside

This section is concerned with two topics related to the structure of the complement construction as a whole, i.e. on the level higher than embedded CP/TP: the syntactic tests for subordination and the distribution of CCs in a sentence.

2.1.4.1 Extraction patterns as a diagnostics for syntactic complementation

Apart from reflecting the underlying structure, the CC-internal phenomena discussed above can serve as a diagnostics for a syntactic complementation. However, if a language lacks a regular morphosyntactic coding for nominative and verbal categories or does not employ complementizers, the complement construction can become indistinguishable from two adjacent independent or coordinated clauses. In this case, one can rely on a number of syntactic tests. In what follows, I will briefly discuss two most widely-used syntactic tests: extraction of wh-element from the embedded clause and topicalization test. There can be numerous other syntactic tests, most of which, however, would be language- or modality-specific. I will return to language-specific tests in sections 3.2.4.

2.1.4.1.1 Wh-extraction out of the embedded clause

The idea of wh-extraction test comes from the observation that in the process of question formation interrogative pronouns (i.e. wh-elements) move from their base-generated position to the CP region, which for most languages which show this pattern is on the left-periphery of the sentences. This movement can be local, but it can also cross the boundary of the embedded CP (unless they are islands for extraction (Ross 1965)). However, extracting the wh-element from one clause only is not available in coordinate clauses as illustrated in (3). As for independent adjacent clauses, the wh-element cannot cross the sentence boundary and hop ahead of the sentence, which it has no syntactic relation with.

(3) a) Whati did John say Mary love eating ti for breakfast?

b) *Whati does John love eating ti and Mary loves eating pancakes for breakfast? c) *Whati Mary is cooking a breakfast. … does John love eating ti, she wonders.

The leftward wh-movement as in (3) is a strong tendency across wh-movement languages. However, a broad typological overview reveals that, there are notable exceptions (Cheng 1991). For instance, the research on SLs reveals that many SLs (e.g. American SL in (4)) tend to exhibit rightward wh-movement (4a), doubling strategies (i.e. two copies of a wh-element on both sides of the sentence)(4b) or in-situ wh-elements (4c) (Cecchetto 2012).

(4) a) JOHN BUY ti YESTERDAY WHATi2

b) WHATi JOHN BUY ti YESTERDAY WHATi c) JOHN BUY WHAT YESTERDAY

‘What did John buy yesterday’

adopted from Cecchetto (2012: 297) Consequently, the application of wh-extraction test in a particular language must be preceded by a fine-grained language-specific analysis of wh-question formation.

2_{Glossing conventions: Signs are glossed in small caps.}_IX_{-3 is a pointing sign, with numbers referring to the person.}

Square brackets refer to the clause boundaries. Underscores or t mark the base-generated position of the moved constituent. Non-manual markers are placed above the glosses, with underscore showing their scope.

(13)

13

2.1.4.1.2 Topicalization out of the embedded clause

The topicalization test follows very similar principles. It involves the fronting of the topicalized constituent up to the Topic Phrase (TopP) in order to establish its pragmatic role. The syntactic

operation is only one of the topicalization strategies (others involve prosodic or morphological marking), but if a language does apply topicalization, this most likely will involve leftward movement. Prominently, the leftward tendency holds for SLs as well (Wilbur 2012). Similar to wh-extraction, topicalization is possible from the embedded clause, but not from one of the coordinated clauses or ahead of independent sentence (5).

(5) a) That sandwich, I want you to try ti

b) *That sandwichi, I ordered ti but John ordered a salad.

c) *That sandwichi Do you remember a cozy restaurant on the corner? I ordered ti there.

Finally, although topicalization is typically associated with the left periphery of the sentence, topicalized constituents can undergo more local leftward movement, i.e. to the position between the subject and the predicate. Such local fronting is referred to as middle-field topicalization and is used to explain SOV/SVO alternations in languages like Russian allowing for a flexible word order (Bailyn 2012). The middle-field topicalization has interesting consequences for raising-to-object constructions discussed previously, because it can trigger discontinuity by separating the raised constituent from the CC with the CTP intervening between them, i.e. S-CC(X)-CTP-CC order. Middle-field topicalization in complement constructions can serve as a diagnostics for a non-finite subordinate CC involve raising-to-object.

2.1.4.2 The sentence position of the CC

As soon as internal morphosyntactic cues alongside with syntactic tests made it evident that we are dealing with a CC, we can investigate how the semantic class of CTP and morphosyntactic properties of its complement influence the structure of the sentence. The relevant parameters of variation are the relative position of the CC within the complex sentence (sentence-initial vs. sentence-final) and the possibility for central-embedding of CCs. The general question that can be asked in this regard is to what extent the distribution of the CC and its mobility differ from the ones associated with a simple NP object.

Typological research reveals that the distribution of the CC can be conditioned by the type of the CC. More specifically, it frequently relies on the sentence-like vs. non-sentence-like opposition. Thus, across various SOV languages, such as Persian, sentence-like CCs are obligatorily extraposed to the sentence-final position notwithstanding the basic word order. Non-finite CCs, however, can be extraposed or can stay in-situ (Noonan 2007:95). As for sentence-initial position of the CC, it is not typically associated with the canonical object position. However, if topicalization via fronting is attested in a language, certain types of CCs can typically be fronted as well — e.g. ‘[That John is married]i I didn’t

know ti’.

2.1.5 The complementation hierarchy

The typological research on sentential complementation revealed systematic relationship between the semantic type of the CTP and the morphosyntactic properties of the CCs. Moreover, several scholars (Givón 1980, Noonan 1985, Cristofaro 2003) noticed that although languages differ in terms of which semantic classes of CTP license a particular construction, there are some universal tendencies. In this

(14)

14 research I am going to focus on the universal implicational hierarchy of complementation proposed by Cristofaro (2003):

Modals (‘must’, ‘can’, ‘be able’, may’), Phasals (‘finish’, ‘start’, ‘stop’, ‘continue’) > Desideratives (‘want’), Manipulatives (‘make’, ‘order’, ‘tell to’) >

Perception (‘see’, ‘watch’, ‘hear’) > Knowledge (‘know’, ‘find out’, ‘realize’), Propositional attitude (‘believe’, ‘think’), Utterance (‘say’, ‘tell’)

(Cristofaro 2003: 125) The rationale behind the hierarchy is that at any point of it if the CTP takes CCs with a deranked predicate then all semantic classes of CTP on the left also take CCs involving predicative deranking.

According to Givón (1980, 2001) and Cristofaro (2003), the universality of the complementation hierarchy can be explained in terms of semantic integration, which is iconically represented on the level of syntax. In addition, the principle of information recoverability is involved: as long as the main

predicate predetermines certain semantic aspects of the embedded state-of-affairs, there is no need to duplicate them in the embedded clause. Givon (2001) summarizes the relationship between semantics and syntax in the sentential complementation:

“The stronger is the semantic bond between the two events, the more extensive will be the syntactic integration of the two clauses into a single though complex clause”

(Givón 2001: 39-40) Building upon the substantial typological evidence, Cristofaro (2003) reveals that the hierarchy works not only for general deranking, but also for any given verbal category. For instance, if the embedded predicate of a particular CTP lacks Tense specification, the embedded predicate of the CCs selected by the CTP on the left is likely not to mark Tense as well.

Despite the overwhelming scope of the typological research on the sentential complementation, to date it never took SLs into account. Nevertheless, over the past ten years sign linguistics yielded several comprehensive accounts of sentential complementation in different SLs, which are overviewed in the following section.

2.2 Complement clauses in SL

Syntactic complexity in general, and sentential complementation in particular, have started to attract the attention of sign linguists only recently, and thus remain understudied. This may be due to the fact that CCs in SL are extremely hard to investigate. As mentioned previously, a considerable amount of nominal and verbal inflection, which can mark the CC, is absent in most SLs. More specifically, SLs investigated so far do not employ systematic morphosyntactic coding for case, Tense and Mood. Verbal agreement is expressed on the certain class of predicates (i.e. agreeing predicates), but absent in other predicates (i.e. plain predicates) (Padden 1983).3_{In addition, most SLs rarely use overt complementizers.}

3_{Verbal agreement in SLs is realized via the extensive use of the signing space (Mathur & Rathmann 2012). The}

discourse referent is associated with a particular point in the signing space — the referential locus. If the referent is present in the moment of speech act, its referential locus corresponds to the actual location in space. Thus, in case of first- and second-person agreement, these are the bodies of the of the signer and the addressee, respectively. If the referent is not present, its locus is established in the signing space by virtue of indexical pointing. In order to realize the verbal agreement, the movement (and sometimes orientation) component of the

(15)

15 Nevertheless, the existing research suggests that SLs employ syntactic complementation as a dominant strategy to encode the corresponding semantic relationship. In what follows, I briefly

overview four studies targeting complementation in LSF (Hauser 2020), NGT (van Gijn 2004), LIS (Geraci & Aristodemo 2016), and TiD (Göksel & Kelepir 2016).

2.2.1 Evidence for subordination: complementation in LSF and NGT

The first two studies, — Hauser (2020) on LSF and van Gijn (2004) on NGT — focused on syntactic subordination as the core component of sentential complementation. For reasons of sparse

morphosyntactic marking of subordination, both authors opted to use syntactic tests to identify the subordination across sentences involving different CTPs.

Building upon the list of CTPs adopted from Noonan (1984) (see section 2.1.2 for a similar list based on Noonan (2007) and Cristofaro (2003)), van Gijn (2004) investigated the syntactic subordination of CCs in NGT by virtue of a grammaticality judgment task. In addition to wh-extraction and

topicalization, which were already discussed, van Gijn applied tests based on subject pronoun copy position4_{and the scope of the negative non-manual markers (NegNMM)}5_{. Both tests were initially} elaborated for American SL (ASL) by Padden (1983).

The experiments on grammaticality judgments showed that CCs in NGT involve syntactic subordination for most CTPs. However, van Gijn found that a) subject pronoun copy and the scope of NegNMMs do not work for NGT the same way they do in ASL and b) the extraction tests showed contradicting results for propositional attitude and utterance predicates. More specifically, the

topicalization test suggested syntactic subordination for these CTPs (6), while wh-extraction appeared to be ungrammatical (6).

(6) a) GOBLINSi MAN IX-3A BELIEVE ti EXIST

‘As for goblins, the man believes that {they} exist.’

(van Gijn 2004: 164) b) *WHOi INGE BELIEVE ti 1-VISIT-3B

‘Who does Inge believe visits him.’

(van Gijn 2004: 145) Additionally, van Gijn did not find any use of complementizers (however, see Bos (2016) on the

grammaticalization of the verb ROEPEN “to call” into a direct speech complementizer in NGT). I will return

to van Gijn’s results concerning particular CTPs in section 5.3.1 for the purposes of cross-linguistic comparison.

predicate is modified such that the starting and end points of the movement correspond to the referential loci of the arguments. In most frequent cases, the starting point is associated with the locus of the subject, and the end point with the locus of the object.

4_{Subject pronoun copy is a right dislocation of a pronominal element referring to the subject of the main clause.}

Based on ASL data, Padden (1988) suggested that only a subordinate clause can intervene between the subject pronoun copy and a clause, the subject of which it represents.

5_{Non-manual markers (NMMs) are facial expressions (e.g. eye brow raise or frown, mouth patterns and eye gaze}

direction) and head and body movements that are linguistically relevant; they may fulfil various phonological, morphological, syntactic and discourse functions (Pfau & Quer, 2010). NMMs of negation attributed to the matrix predicate are claimed by Padden (1988) to scope over the embedded clause but not over conjuncts or adjacent independent sentences.

(16)

16 In her study on CCs in LSF, Hauser (2020) followed a similar approach. By adopting a wider range of CTPs based on Noonan (2007) and Cristofaro (2003), Hauser (2020) investigated the distribution of different types of CCs (control/raising, infinitival, sentence-like) across CTPs and subordination as such in LSF. With a few exceptions, most semantic classes of CTPs in LSF appeared to take finite or non-finite CCs, which supports the claim that semantic complementation reliably entails syntactic subordination. In addition, Hauser (2020) found that several CTPs (such as LIKE and DESPISE) license the obligatory use of the complementizer WHAT asin (7).

(7) MARIA LIKE WHAT AGATHE GO CINEMA

‘Marie likes that Agathe goes to the cinema.’

adopted from Hauser (2020: 185) Further details on the distribution of CC types across CTPs in LSF is provided in section 5.3.1 as a part of cross-linguistic comparison.

2.2.2 The structure of complement constructions: complementation in LIS and TiD

Geraci & Aristodemo (2016) and Göksel & Kelepir (2016) focused on the syntax of complement constructions in LIS and TiD, respectively. Both teams of scholars observed systematic correlations between the type of CCs, the semantics of CTP, and the resulting constituent order in the complement construction. Importantly, both LIS and TiD are SOV languages, which makes it possible to investigate the central-embedding of CCs.

Geraci & Aristodemo (2016), following Geraci, Cecchetto & Zucchi (2008), observed that control and Subject-to-Subject raising CCs in LIS can be found in central (8), sentence-initial and sentence-final positions. However, sentence-like complements are constrained to sentence-initial and sentence-final positions only (8). Moreover, CCs of some CTPs (such as THINK)are limited to sentence-final position.6

(8) a) GIANNI [COW MILK] TRY

‘Gianni tried to milk the cow.’ b) *GIANNI [PIERO BIKE FALL] TELL

‘Gianni said that Piero fell off the bike.’

(Geraci & Aristodemo 2016: 117 – 119) In addition, Geraci & Aristodemo (2016) report that wh-extraction (which is rightward in LIS) is possible out of control constructions, but is heavily constrained for sentence-like CCs — a wh-element can be extracted CCs only if accompanied by a pronominal pointing sign, coreferential to the extracted constituent.

The study by Göksel & Kelepir (2016) focused on syntactic complementation in TiD based on elicitation and grammaticality judgments. Göksel & Kelepir (2016) found that, similar to LIS, CTPs differ in terms of the syntactic position of their CCs. Thus, Göksel & Kelepir (2016) identified two groups of CTPs, WANT-type and KNOW-type. The WANT-type imposes the embedded position of the CC, which is in accordance with the canonical object position in TiD. The KNOW-type, on the contrary, selects sentence-final CCs. Prominently, the distribution of CTPs among two types in TiD overlaps with a similar division in

6_{Geraci & Aristodemo (2016) also found that under certain conditions it is possible for sentence-like CCs to be}

centrally embedded. The conditions involve role-shift, null matrix subject or the specific use of space while signing the matrix predicate. See the original paper for details.

(17)

17 LIS concerning the selection of sentence-like and non-finite CCs. However, Göksel & Kelepir (2016) emphasize that they have no evidence to assume that the position of CC in the sentence is conditioned by the finiteness of the CCs as it is claimed for LIS. Without excluding such a possibility, Göksel & Kelepir (2016) nevertheless stick to the semantics of CTP as a predictor for a particular position of the CCs.

The typological overview given in this section leads to the following three points. First, SLs do have syntactic complementation, and this is realized via subordination mechanisms very similar to those identified in spoken languages. Second, SLs nevertheless constitute a separate group of languages with their own peculiarities. These can be modality-specific (e.g. the use of NMM and signing space) or they may wide-spread in SLs but much less frequent in spoken languages (e.g. the lack of overt

complementizers). Finally, although sharing a number of properties, SLs can display individual language-specific properties as illustrated throughout this chapter. All this being said, let us turn to the basic description of the SL in focus, which is Russian Sign Language.

2.2.3 RSL: the background

Russian Sign Language (RSL) is a SL used by deaf and hard-of-hearing people in Russia (Zajtseva 1991). RSL is a relatively young language as its emergence dates back only to the 19th century when the first specialized schools for deaf children were founded. The latest census dated to 2010 revealed more than 120,000 native RSL signers.

The syntax of RSL attracted scholars’ attention only recently. The point of departure was the study on basic word order in RSL by Kimmelman (2012). This author found that while the subject is generally preferred in the sentence-initial position, the position of the object can be central or clause-final depending on the class of the predicate, its aspectual marking and the complexity of the object. The study by Kimmelman (2012) targets simple clauses, but very little research has been devoted to the study of complex sentences.

So far among complex structures that have been investigated in RSL are conditionals (Burkova 2012), quotations (i.e. direct and indirect speech) (Kimmelman & Khristoforova 2018) and relative clauses (Khristoforova & Kimmelman 2020). The research on quotation by Kimmelman & Khristoforova (2018) is probably the most relevant to the present investigation as utterance predicates introducing quotations are frequently attributed to CTPs. By integrating corpus study and grammaticality judgments, we observed that quotations may exhibit a mixed behavior of indexical elements (i.e. pronominal signs are interpreted within the context of quote, while locative signs — within the context of narration), which complicates the distinction between direct and indirect speech in RSL (see Quer (2011) and Herrmann & Steinbach (2012) for similar phenomenon in other SLs). In addition, we found an

inconsistent use of role-shift NMM which do not straightforwardly align with the indexical shift as it was initially expected.7_{Finally, apart from very few entries of complementizer}_WHAT_{, current syntactic tests} (i.e. wh-extraction and center embedding) failed to reveal any evidence of syntactic subordination in RSL quotations.

Another relevant study targets a phasal predicate NOT.STOP,which conveys a continuative aspect

(Khristoforova 2019). The grammaticality judgments tasks revealed that a) NON.STOP can equally precede or follow the lexical predicate, and b) the position of the phasal predicate has no effect on the

7_{Role-shift is defined as a range of specific NMM (e.g. body and head sideward leans and turns, a change of eye}

gaze direction, facial expressions, etc.) used in order to “shift into the role” of the character, whose speech, thought or actions are reported (Lillo-Martin 2012). Commonly accompanied by the shifted interpretation of indexical elements (e.g. pronouns, deictic elements, time and locative adverbials, etc). For instance, in quotation constructions involving role-shift a first person pronoun does not refer to the signer, but to the subject of the main clause as in direct speech in spoken language, e.g. “Johni said: ‘Ii am tired’.

(18)

18 agreement values on the lexical predicate. As a possible analysis, I proposed verbal serialization8_{, which} implies partially shared argument structure for two adjacent predicates but allows them to occur in different orders and inflect independently from each other (Haspelmath 2016).

These two studies motivated the exclusion of phrasal and utterance predicates from the current study: a) they introduce constructions, which seem to fall outside of the scope of the structural

definition of CCs adopted in this study, and b) there is already at least some research on these two predicative classes, while other semantic classes remain uninvestigated.

Sentential complementation has never been systematically investigated in RSL. The present study aims at filling this gap by investigating a large portion of CTPs in relation to the morphosyntactic properties of their CCs and the syntactic structure of complement construction as a whole. It will contribute to the typology of the sentential complementation in spoken and sign languages and enrich our knowledge on the complex syntax in RSL.

3 Methodology

3.1 Corpus investigation

The first step was to investigate complement clauses (CCs) produced in an (almost) naturalistic setting, i.e. in the on-line RSL corpus9_{(Burkova 2012), which contains around 200 video recordings of different} text types (picture-based storytelling, interviews, spontaneous narrations, etc.). Most recordings are annotated in ELAN software (https://tla.mpi.nl/tools/tla-tools/elan/), which is one of the most widely used software for a multi-layered annotations. The corpus annotations contain four layers, or tiers: separate tiers for sign glosses for the right and left hand, sentence-by-sentence translation and comments.

The following section introduces two different query strategies, which were applied to identify CCs introduced by different CTPs in the corpus.

3.1.1 Query strategies 3.1.1.1 Bigram search

As mentioned, the corpus contains two tiers with glosses for right- and left-hand configurations. Using a structured search engine available in ELAN, I searched for the combinations of CTP + predicate (V) on the distance from zero up to two glosses from each other. For instance, the query is satisfied by WANT

APPLE BUY; WANT RED APPLE BUY; WANT BUY with distances between CTP and V estimated as 1, 2 and 0

respectively. The reverse sequence, i.e. V+CTP, with up to two intervening glosses, was also retrieved. Such method allows to capture all potentially possible syntactic structures involved in

complement constructions of different CTPs. However, two main problems with this method arise: (a) the gloss tier does not contain any syntactic parsing. Thus, the query will find sequences such as

WANT APPLE. BUY IMP ONE (“I want and apple. Buy me one!”), which does not instantiate CC but perfectly satisfies the query;

(b) the query won’t find complex cases with more than two intervening signs although they are likely to be possible in a language (e.g. WANT RED SWEET APPLE QUICKLY BUY );

8_{Complementation was not considered as an option as the grammaticality judgment task involved primarily}

intransitive predicates. However, in the light of the outcomes of the present study, this analysis should be tested. It is thus expected that if NOT.STOP is a phasal CTP, the embedded transitive predicate will move to the central position alongside with its objects resulting in V-O-NOT.STOP or O-V-NOT.STOP.

(19)

19 Increasing the distance between glosses would solve problem (b), but would also enhance the number of errors of type (a) in the output. Given that the final diagnostics is made manually for each individual case, two-gloss distance seems to be a reasonable trade-off between not missing CCs in the corpus and decreasing the time and effort for the annotator.

3.1.1.2 Search among translations

The previous method was supplemented by a search on the translation tier. This strategy relies on the stipulation that CCs in RSL are likely to be translated by CC constructions in Russian. Thus we can search for Russian CCs in translations and expect them to align with CCs in RSL. Alike the bigram approach, this method is not biased to any particular CC structure. In addition, it circumvents the limiting factor of the distance between target glosses described above, because the translations are made sentence-by-sentence, and thus the sentence boundaries coincide with the boundaries of a respective annotation.

However, this method is far from being ideal on its own, as the stipulation CCs in RSL will translate into CC constructions in Russian is not linguistically valid (if one takes a broad typological perspective). Although, indeed, a functional definition of CCs implies that they are used in the same semantic environment across languages, it does not entail that we will find a syntactically embedded CC (see section 2.1.1). Nevertheless, I decided to use this method as a first window into CCs in RSL.

In combination, these two strategies yielded a significant amount of CC in the corpus: 224 annotated CCs introduced by 238 tokens of CTPs representing 36 types (Table 1).10_{CTPs with less than} five entries were collapsed into one group.

Table 1. Corpus entries for complement-taking predicates

CTP entries WANT 54 SEE 35 THINK 32 TRY 24 ASK.FOR 20 KNOW 18 WANT.NOT 7 LIKE 5 LOVE 5 others (< 5 entries per CTP) 38 Total 238

The obtained data covers most traditionally identified semantic classes of CTPs (Table 2). However, predicates of fearing, impediment predicates and pretence predicates were represented by less than five cases and thus excluded from the analysis.

10_{The number of CTPs tokens is higher due to multiple entries of two-level complementation, i.e. a CTP}

(20)

20

Table 2. Semantic classes of CTPs represented in the corpus

CTP semantic class entries

Desiderative 65 Perceptive 34 Achievement 23 Knowledge 18 Manipulative 18 Commentative 16

Positive propositional attitude 15 Negative propositional attitude 15

Utterance (interrogative) 8 Modal 7 Utterance (declarative) 6 Fear 4 Impediment 2 Pretence 2

The attentive reader may notice that although I excluded modals and utterance predicates from the analysis (see section 2.1.2 for motivation), these classes of predicates still appear in Table 2. The reason is that certain predicates appeared to be able to display a range of semantic meanings outside of their designated semantic class. For instance, the predicate THINK frequently introduced utterance-like clauses

as in “I thought: “Okay, let’s go to the cinema””. In these cases, the semantic class “utterance(declarative)” was assigned.

3.1.2 Corpus data analysis

Obtained entries of CCs were additionally described for the following parameters:  Properties of CTP:

o semantic class (cf. “desiderative”, “commentative”, etc.)

o verbal characteristics: argument structure, agreement (if present), transitivity  NMMs and their scope

 Properties of the CC

o the properties of the embedded subject(overt or covert; coreferential to the matrix subject or not)

o the properties of the embedded predicate (predicative class (i.e. agreeing, classifier or plain); agreement values if relevant; aspectual marking if present)

o complementizer (specified if present) o CC-external constituent order

o CC-internal constituent order

3.2 Grammaticality judgment tests

Even though the corpus investigation yielded a considerable number of CCs, the obtained data is incomplete because it cannot provide negative evidence. In other words, while the corpus shows us what is possible in a language, there is still a chance that certain grammatical constructions are

infrequent (and thus did not make it to the corpus). For instance, WANT being the most frequent among

the CTPs is nevertheless found mostly in control contractions. However, it is well-attested that want can form Subject-to-Object raising constructions in some languages such as English. Is the lack of CCs other

(21)

21 than control due to a language- or modality-specific constraint, or it is just an artifact of statistics? In order to answer this and other similar questions, grammaticality judgment tests were designed.

3.2.1 Participants

Four RSL native signers participated in the experiment. All participants were born and raised in Moscow, and thus dialectal variation is unexpected (Table 3). In addition, all signers come from deaf families.11

Table 3. Sociolinguistic profiles of the participants

Participant Gender Year of birth Place of birth Deaf in family

1 Female 39 Moscow parents;

grandparents

2 Female 37 Moscow parents

3 Female 31 Moscow parents;

grandparents

4 Male 49 Moscow parents

3.2.2 Materials

A detailed description of constructions (i.e. syntactic tests) for which the grammaticality judgments were collected is given in section 3.2.4. Here, I focus on how these constructions were represented in the experiment.

Each syntactic test for a particular CTPs was targeted by at least two sentences in RSL. Each utterance was accompanied by a picture illustrating animate characters involved in an every-day activity/situation. The visual materials were implemented to provide a necessary context for an

utterance. Otherwise, transformations resulting from discourse factors (such as topicalization) might be judged as infelicitous. The materials contained 8 pictures in total, providing a context for 6-10

utterances.

Each stimulus was signed and recorded by the researcher (i.e. the author), who is a hearing signer with an intermediate level of RSL proficiency. The questionnaire containing 58 stimuli recordings was created. For each stimulus, the participants were asked to evaluate the utterance using a Likert scale (from 0 —“absolutely ungrammatical” to “4” — “absolutely grammatical”).

3.2.3 Procedure

Due to the lockdown restrictions resulting from COVID-19 prevention measurements implemented in Russia and the Netherlands, the research was conducted online via Zoom application (https://zoom.us). Video calls involving the participant and the researcher were initiated and recorded by means of the “Record the screen” function.

During the experiment, the participant was asked, first, to repeat the sentence in the video and then to evaluate it grammaticality. Specifically, the question addressed to the participant was whether they would find this sentence correct in a natural conversation in RSL. If the participant gave the stimulus a low score (“2” or lower), they were asked to propose the correct variant with a similar meaning to ensure that the low score was not due to some unrelated reason, such as incorrect use of a lexical sign.

11_{Deaf children born in hearing families (which is 90% of the cases) are typically exposed to SL later. Research}

revealed that the age of SL acquisition affect the performance of such signers on the level of syntactic

comprehension and production (Boudreault & Mayberry 2006). Consequently, SL research typically focuses on deaf signers who have been exposed to the SL of their parents since birth.

(22)

22

3.2.4 Syntactic tests

As mentioned in section 2.1.4.1, numerous syntactic tests have been elaborated to determine the basic properties of complement constructions across languages. Specifically, these tests involve various types of morphosyntactic transformations (e.g. topicalization, change in constituent order, etc.), which are judged by the native speakers of the target language as acceptable for one type of constructions but not for other. Given the absence of overt markers of subordination (i.e. specific functional signs,

morphemes or NMMs), grammaticality judgment task involving syntactic tests is the most convenient way to investigate the underlying structure of CCs in RSL.

In this study, I focused only on those tests that a) identify structural properties of CCs specific to RSL and b) make the results comparable to those obtained for other SLs. Thus, while some tests are traditional for this field (e.g. topicalization), others have never been applied to other SLs (e.g. middle-field topicalization, non-first subject agreement on the predicate).

It is important to highlight that the grammaticality judgment task was implemented as a complementary method to fill the gaps left by the corpus investigation. Consequently, if for a particular CTP one of the constructions described below was sufficiently represented in the corpus, it was not additionally tested in the grammaticality judgment task. In what follows, I describe each test in detail and explain my motivation to include it in the questionnaire.

3.2.4.1 Test 1. Overt non-coreferential subjects

The test aims to answer the question whether CTPs taking control constructions in the corpus are capable of taking other types of CCs. For instance, in the example below the CC contains an overt subject MAN,which is referentially different from the matrix subject, CAT.

Example of stimulus:

CAT IX-3a WANT IX-3b MAN 3b-FEED-3a

SjMC _SiCC

‘The cat wants the man to feed him.’

The corpus investigation revealed that some CTPs in RSL (e.g. WANT, TRY, LIKE, etc.) can take control CCs, but whether they are limited to this type of CC or not remained unclear from the corpus study. If the example stimulus above is assessed as grammatical (i.e. it receives a mean score significantly higher than “2”), then we can conclude that the CTP in focus is capable of taking constructions other than control.

Note, however, that this test does not tell us exactly which construction the CTP takes apart from the control — the overt coreferential subject in CC can be attributed both to finite CCs or ECM constructions. The two can be distinguished by looking at the finiteness properties of the embedded predicate, which is the target of the Test 2.

3.2.4.2 Test 2. Non-first subject agreement on the embedded predicate

This test reveals if the target CTP can take a finite clause as a complement. As mentioned in section 2.1.3.1, one of the defining properties of a finite predicate is its ability to inflect for subject agreement. Recall that not all predicates in SL can inflect for subject agreement. However, those, which can, are expected to exhibit agreement with the subject of the CC if this CC is finite. Following this logic, if the agreeing predicate TEASE in the stimulus below can be inflected for subject agreement while embedded, then we can safely postulate that the CTP takes finite CCs.

(23)

23 Example of stimulus:

MOUSE3a LOVE 3a-TEASE-3b CAT3b AGR.S-V-AGR.O

‘The mouse loves to tease the cat.’

All the stimuli for this test are limited to non-first subject agreement for the reasons that will become apparent further on.

3.2.4.3 Test 3. The use of a complementizer in relation to the referential properties of the

CC-subject

The present test reveals two things: (i) if a complementizer can be used with a particular CTP and (ii) if its use is conditioned by the referential properties of the embedded subject. Thus, both stimuli (a) and (b) have a complementizer WHAT introducing the CC, but (a) has non-coreferential subjects in the CC and matrix clause, while (b) has coreferential subjects.

Example of stimuli:

a) MAN KNOW WHAT CAT HUNGRY

SiMC

COMPL SjCC

‘The man knows that the cat is hungry.’

b) GIRL IX-3A KNOW WHAT IX-3A BEAUTIFUL

SiMC _COMPL_SiCC ‘_{The girl knows that she is beautiful.’}

The corpus robustly reveals that RSL signers make use of a complementizer WHAT in CCs. However, it is not clear from the corpus if all CTPs permit complementizers equally and what constrains the use of the complementizer (if anything). One working hypothesis is that some CTPs require complementizer use if their non-finite CCs is strictly controlled. In other words, the complementizer may be the only way for these CTPs to license a CC with an overt subject referentially different from the matrix subject. Under this hypothesis, complementizer use with coreferential subjects is expected to be at least less

acceptable due to its redundancy.

3.2.4.4 Test 4. Sentence-initial topicalization

The topicalization test is traditionally applied to identify the syntactic dependency between two clauses (see section 2.1.4.1.2 for details).12_{Although so far we have been taking for granted that all our target} predicates take an embedded clause as a complement, this need not always be the case. Complex constructions in RSL often lack of an overt complementizer and can often be interpreted as coordinated clauses or adjacent independent sentences (i.e. paratactic construction). In order to disambiguate such sentences, we can extract a constituent from the second clause to the left periphery of the sentence. Thus, in the following example stimulus, we extract the embedded object BALL IX-3 to the sentence-initial position of the sentence.

12_{The wh-extraction test was disregarded due to the fact that there is no comprehensive study on wh-question}

formation in RSL. As long as we do not know what strategies of question formation are adopted in RSL (i.e. leftward movement, rightward movement, doubling or in-situ) (however see Esipova (2014) for a tentative account), we cannot reliably use this test.

(24)

24 Example of stimuli:

BALL IX-3 MOUSE SEE CAT SEARCH _

‘This ball, the mouse see that the cat is searching.’

As mentioned, topicalization is only possible if the clause from which we extract the constituent is structurally embedded. Thus, if the example stimulus receives a high grammaticality score, we can claim

CAT SEARCH BALL to be syntactically subordinate.

3.2.4.5 Test 5. Middle-field object dislocation

This test investigates the construction involving middle-field dislocation of the embedded object. This type of syntactic transformation was found in the corpus for some CTPs (eg. WANT, TRY) but not for others. For instance, in the example stimulus the embedded object BALL appears between the matrix subject CAT and the CTP LOVE. In order to avoid confusion with a sentence-initial topicalization, the

matrix subject was overt in all of the stimuli. Example of stimulus:

CAT BALL LOVE PLAY _

S CC(O) CTP [V _]CC

Lit., ‘Cat ball loves playing.’ ‘Cat loves playing with a ball.’

Complement constructions with such constituent order have not been previously investigated for any SL. Thus, unlike previous tests, this test is data-driven in the sense that we cannot say what

grammaticality or lack thereof would mean until we collect the judgments across all CTPs. Nevertheless, several a priori hypotheses based on spoken languages can be proposed to explain the position of the embedded object.

Hypothesis 1: the CTP functions as light verb or auxiliary-like element.

As mentioned in section 2.1.1, auxiliary-like elements (such as English modals) are claimed to head the respective functional projections (ModP or AspP) and thus select for VP, not a clause. One of the possible consequences of this, is that in SOV languages or in languages with a flexible word order (as RSL), the object can appear between the matrix subject and the verbal cluster resulting in S O Aux+V order as in the example stimulus above.

Hypothesis 2: the middle-field object dislocation is related to topicalization.

Alternatively, sentence-medial position of an embedded object can be analyzed as an instance of middle-field topicalization discussed in section 2.1.4.1.2. If this is true, then middle-field position of an embedded object should obey the same constraints as a sentence-initial topicalization.

The results for this test are expected to provide evidence in favor of one of the hypotheses above. If an object can be fronted to middle-field position across all CTPs that permit for sentence-initial

topicalization, middle-field topicalization is the most likely analysis for the middle-field object

dislocation. If only a limited number of CTPs with non-finite CCs allows for such operation, then these predicates are good candidates for a light verb analysis.

(25)

25

3.2.4.6 Test 6. Pre-verbal complement clauses

The corpus revealed that some CTPs allow the CC to appear in the position between the matrix subject and the CTP, while for other CTPs, such constituent order was not attested. The test aims to check whether preverbal CCs are indeed CTP-specific, or they are possible across all CTPs.

Example of the stimuli:

MOUSE [CAT SEARCH BALL] SEE S [ ]CC CTP

Lit. ‘The mouse that the cat search for a ball sees’ ‘The mouse sees that the cat is searching for a ball’

3.2.4.7 The selection of target CTPs for a grammaticality judgment task

As mentioned, if a construction involved in any of the Tests (1-6) appeared in the corpus with a particular CTP, it was not additionally tested for this CTP in the grammaticality judgment task. Thus, in Table 4 “x” indicates that the corresponding type of construction is found in the corpus for the

respective CTP in a sufficient amount (more than five entries) and need not be additionally tested.

Table 4. Tested constructions represented in the corpus

Test 1 Overt non-coreferential subjects Test 2 Non-1st subject agreement on the embedded predicate Test 3.1 Complementizer (coreferential subjects) Test 3.2 Complementizer (non-coreferential subjects) WANT SEE x x TRY x x x KNOW x x ASK.FOR x x x x LIKE x LOVE THINK X x x x Test 4 Sentence-initial topicalization Test 5 Middle-field object dislocation Test 6 Central-embedding of CC WANT x SEE x TRY X x KNOW ASK.FOR LIKE X LOVE x x THINK x

3.2.5 Statistical analysis of the grammaticality judgment results

As mentioned previously, the grammaticality of the stimuli was assessed on a 5-point Likert scale. Each of the Tests (1-6) was represented by at least two sentences and thus received two scores from each

Complement clauses in Russian Sign Language