Linguistics in Amsterdam 5-1 (January 2012)Vadim Kimmelman: Word order in Russian Sign Language

To refer to this article use this url: http://www.linguisticsinamsterdam.nl/05/nr01/a01

3 Methodology

3.1 Different approaches to data collection

next section

Word order in SLs has been investigated using many different methodologies, which in some cases yielded different results, as I have already discussed with reference to ASL (section 2). In several papers, grammaticality judgements and assessment of constructed examples was the main means of data collection (Fischer 1975; de Quadros 1999). This means is very convenient to test complex theories, but the obvious drawback is that the intuition of native signers is not a very objective measure, and assessment of the grammaticality of different word orders is notoriously difficult. As a variant of this methodological strategy, a researcher who is a native signer may use her/his own intuition; see, for example, de Quadros (1999) (the author is a bilingual child of Deaf parents) and Kegl et al. (1996) (one of the authors is a Deaf native signer). One should notice, however, that the intuition of a researcher can be biased by theoretical considerations.

A much more reliable methodology is the use of naturalistic corpus data (e.g. Friedman 1976; Deuchar 1983; Bouchard & Dubuisson 1995; Nadeau & Desouvrey 1994; Quinto 2000; Wilbur 2002; Sze 2008). However, this method has several serious drawbacks, too. First, in a naturalistic set of data, it is not always possible to find the full variety of constructions and test all factors that can influence word order. Second, in naturalistic narratives, sentences in which more than one argument is overtly expressed are very rare. Third, for SLs, it is particularly difficult to create a balanced and sufficiently large corpus that might include different genres.

Quite often researchers use an experimental approach to elicit the data necessary for determining word order, namely, a picture description task (e.g. Volterra et al. 1984; Boyes-Braem et al. 1990; Coerts 1994; Vermeerbergen 1996, 2004; Saeed et al. 2000; Leeson 2001; Sze 2003; Vermeerbergen et al. 2007; Johnston et al. 2007; Milković et al. 2006). This method makes it possible to avoid some of the drawbacks of the other approaches. First, it is possible to specifically test different factors that can influence word order by carefully creating the relevant stimuli. Second, when a situation in the picture is described in one or two sentences, the signers are forced to use sentences with several overt arguments. Third, this method is relatively simple and less time-consuming than collecting a large corpus of natural discourse. Of course, the problem is that data obtained under experimental conditions are less natural than spontaneous data; this approach also favours narrative genre, while for some purposes other genres are necessary.

Some researchers combined different methodologies to study word order in SLs. For example, Massone & Curiel (2004), in their research on word order in LSA, did not only use a sentence interpretation task and an experimental task (describing pictures and films), but also analyzed naturalistic narratives and dialogues. The results of their research were later discussed with a large group of native signers who provided their intuitions on the word orders used. This is an example of a very thorough and reliable methodology.

In order to analyse word order in Russian Sign Language, I decided to combine two methods: an analysis of (semi-)naturalistic corpus data and an experiment (picture-description task). The aim of using the corpus data was to assess some general principles of word order in RSL based on a naturalistic data and to create hypotheses concerning the factors that may have an influence on word order. The aim of the experiment was to test these hypotheses. This approach still has the drawback of being biased toward the narrative genre. This might be a problem since word order in, for instance, conversations might be different; this question remains for the future research. In the following, I describe the corpus (Section 3.2) and the experiment (Section 3.3). In both sections, I provide information about the stimuli, the procedure, the subjects, and the transcription.

3.2 Corpus

Stimuli: The corpus of narratives that I analyzed was collected and annotated by Prozorova (2009). It consists of 13 stories told by nine signers. Two stories were based on the The Pear Film (Chafe 1980), the other 11 stories were based on several comic strips by H. Bidstrup.

Procedure: In the case of The Pear Film, the signers were asked to watch the movie twice and then retell the story for recording. In this case only the hearing researcher was present in the room. In the case of comic strips, two signers participated. One of them was given time to look at one of the comic strips, the strip was then removed and the signer was asked to tell the story. The procedure was then repeated with another comic strip. The first story was used to familiarize the signer with the procedure, and only the second story was used for later analysis. Subsequently, the signers switched roles, and the addressee (the second signer) told two different stories to the first signer. Occasionally one or both of the signers told one more story based on a different strip.

Informants: The corpus was not collected specifically to analyze word order, or even the grammar of RSL; the aim of creating it was to analyze the prosodic structure of RSL discourse. The requirement that only native signers with similar background should contribute to the corpus was not strictly followed. Nine Deaf signers participated: four men and five women. The average age of the informants at the time of the recording was 31 years. Seven were born and raised in Moscow, and also studied there. Two other participants were born and raised in Magadan, but at the time of data collection had already lived and studied in Moscow for several years. Five came from Deaf families, but the remaining four acquired RSL only at school (approx. at the age of 6); they also used spoken Russian at home. Therefore, the signers can be divided into two groups: five native signers, with RSL as their first language acquired in early childhood (all from Moscow), and four competent signers (with different regional background). Keeping this in mind, I compared the word order data from these two groups but did not observe any significant differences. The data will therefore be pooled for further analysis.

Transcription: The corpus was annotated by Prozorova (2009) for the purposes of prosodic analysis. She transcribed it in ELAN with several transcription tiers: right hand, left hand (rough translations of the signs), PHases of the signs (for the definition of phases see Kita, van Gijn, & van der Hulst 1998), boundary movements (movements marking prosodic boundaries in RSL, see below), eye blinks, and DISCOURSIVE UNITS. Translation of the signs was done with the help of a native signer. In the tier DISCOURSIVE UNITS, discoursive units designated by boundary movements were translated (see Section 3.4.2 for the prosodic analysis of RSL). I added three tiers to her transcription, namely ORDER (with labels such as S, V and O assigned to signs/constituents), CLAUSE (where these labels were grouped into clauses), and TRANSLATION (where I translated the sentences). The procedure of determining S, V and O and the definition of clauses used is given below.

3.3 Experiment

Stimuli: For the experiment the procedure proposed by Volterra et al. (1984), including their stimuli was used. The original set of stimuli contained 18 pairs of pictures, consisting of three groups. Six pairs of pictures represented reversible situations (for example “The boy embraces an old lady”), six pairs of pictures represented non-reversible situations (for example, “The boy opens the door”), and six pairs of pictures represented locative situations (for example, “The ball is under the table”). Some of the original pictures seemed somewhat unclear, so I asked the artist A. Rysaeva to create other pictures instead, preserving the type of the situation. Furthermore, I knew from the corpus data that the verb class can influence the word order. I therefore decided to include four more pictures which could be described with plain or agreeing verbs (and not with classifier constructions). Two of the new pictures represented reversible situations and two non-reversible situations. The list of all pictures is given in Appendix I.

Procedure: In the original experiment by Volterra et al. (1984), the signer was given a set of pairs of pictures. The situations on the pictures in each pair differed in one aspect (for example, on one picture a boy was closing the door, and on the other opening it). One of the pictures in a pair was marked by a cross. The addressee (another native signer) was given the same set of pairs, but without marking. The signer was asked to describe the marked pictures in each pair such that the addressee could identify it.

I decided not to use this procedure because I was concerned about the possibility that the contrast between the two pictures could produce the grammatical category of contrast, which in turn could influence word order. Thus in my experiment one signer was given a set of pictures (instead of pairs of pictures) which s/he was asked to describe to another signer. The pictures were given in a randomized order, so that different types of situations were mixed. The results of this procedure turned out to be satisfactory for the purposes of my research.

The experiment was conducted in two stages. First four native signers were given the 18 pictures (excluding the four additional pictures). In the data obtained, however, the number of plain and agreeing verbs was too small for an analysis, so in addition, two other native signers were given all pictures (including the additional ones) to obtain more sentences with plain and agreeing verbs.

Signers: Six Deaf native signers participated in the experiment, three men and three women. The average age of the participants at the time of the experiment was 33 years. All signers but one came from Deaf families. Two signers claimed to have learned RSL only at school, but their RSL competence is assessed as very high by the Deaf community; they use RSL in their daily life and at work. Five of the signers were born and raised in Moscow, while one woman was born in Kirov, but had studied and lived in Moscow for several years. This group of signers is more homogeneous than the corpus data group, but is still not totally homogenous.

Transcription: As discussed below, I assume that there is no one-to-one association between prosodic units (as defined by Prozorova 2009) and syntactic units (clauses). Because of this assumption and the complexity of prosodic transcription, I did not analyze the experimental data prosodically although I do consider the dependency between word order and prosodic marking, as explained in Section 3.4.2. The transcription was made in ELAN and contained the following tiers: PICTURE (the number of the stimulus picture), RIGHT HAND, LEFT HAND, ORDER, SENTENCE, and TRANSLATION. The translation of the signs was done by me with the help of two native signers.

3.4 Analysis

3.4.1 Syntactic labels and clause boundaries

In order to describe word order, the discourse data had to be divided into sentences/clauses. In order to do this, several methodological decisions were taken. In general, my approach to analysing the data was deliberately pre-theoretical, in other words, I tried to assume as little as possible without empirical proof.

First, I want to comment on simultaneity. As already mentioned in section 1, because of the availability of several articulators (the right and the left hand, different parts of the face, the torso), SLs can simultaneously express more than one meaning. For example, in the sign depicted in Figure 1, the left hand signs the lexical sign chair while the right hand simultaneously articulates a classifier construction meaning ‘small animal’; the combination of the two hands expresses the meaning ‘a small animal [a cat] sits on the chair’.

FIG2

Figure 1: Simultaneous construction meaning (in the context) ‘the cat sits on the chair’

At first sight, it seems that in such cases one cannot talk about word order, as there is apparently no order between the sign chair and the classifier construction. Closer inspection, however, reveals that in reality this is not the case. In fact, in the sentence that this picture was taken from (12), the sign chair was uttered first, and then the classifier construction on the right hand was placed in relation to the left hand, which was held stationary in the signing space.

In most of the simultaneous construction in the data, the onset of one of the signs started earlier than the other. In such cases it is possible to establish a word order, as in (12), which was analyzed as displaying the Object Verb order. If a construction turned out to be fully simultaneous, it was not analyzed, as the main goal was to observe how RSL uses word order.

Second, I decided not to use prosodic boundaries as a criterion to determine sentence/clause boundaries (still, in section 3.4.2, I will look at the correlation between them). I had two reasons for this decision. Firstly, I agree with researchers who consider prosody a separate level of grammar which is not reflected directly and unambiguously in syntactic structure (Nespor & Vogel 1986). Clauses and sentences are syntactic objects, and there is no guarantee that there will be a one-to-one mapping between the entities of syntax and the entities of the prosodic level. Secondly, it has been shown that in many SLs, sentence boundaries are not marked consistently by any prosodic clues (Hansen & Heßmann 2007). On the difficulty of defining sentences in SLs, see also the comprehensive discussion in Sze (2008).

Third, I decided to divide the discourse into clauses, not sentences. I remained agnostic with respect to the question how these clauses are combined into sentences, even in cases in which semantically one clause was clearly subordinate to another. The motivation behind this decision was that one has to look at simple cases (=clauses) first in order to understand the basic principles of word order in RSL, and only then approach more complex structures (=sentences).

In previous studies on word order in SLs, it was not always clear how the researchers defined clauses and sentences (Crasborn 2007). Some researchers used prosodic patterns as one of the criteria (for instance, Coerts 1994). In some recent studies (Johnston et al. 2007; Vermeerbergen et al. 2007), clauses were defined semantically. I adopted a similar procedure.

To determine clauses the following definition was used:

This definition is semantic: basically it defines a predication, not a clause. However, I think that semantic predication is a much closer approximation to the syntactic notion of a clause than a prosodic unit. Note that according to this definition, a clause always contains one verbal predicate. There were two regular exceptions:

1. Nominal and adjectival clauses (no verb). In some clauses the predicate was semantically a noun (referring to an object (14) or a person) or an adjective (expressing a stative non-verbal property, like ‘beautiful’). I decided to analyze those cases separately. This decision is open to criticism, because one needs to prove that these clauses do indeed form a separate category[8], but I will demonstrate that this decision did not influence the results.

2. Verbal echoes and sandwiches (more than one verb). I decided to consider cases in which a verb was doubled to be one clause. The motivation behind this was that verb doubling is a prominent phenomenon in the grammar of many SLs, and it was important to include it in the analysis of RSL, too. I analyzed a sequence containing two (or more) verbs as one clause (a) if all copies of the verb referred to one situation (that is, if two occurrences of the verb go referred to two going activities, they were analyzed as separate clauses), (b) if between the two occurrences of the verb only the arguments and adjuncts of this verb appeared, and (c) if the occurrences of the verb were identical or different only in aspect, agreement, or non-manual marking.

There are several constructions which, in the present study, were not considered a single clause. First, there are the so-called split sentences, which are characterized by the fact that the subject of the first clause is the object of the second clause within a sentence (15).

Second, some researchers consider verbs like sit, stand etc. as semi-auxiliaries used to localize referents in cases like (16)[9].

In all cases like split sentences and sentences with verbs of location, I defined the clauses with the help of the usual criteria. Hence, I would divide an example like (15) into two clauses, and an example like (16) into three. Again, I am not excluding the possibility that split sentences are special constructions or that verbs of location have special properties, but this demands additional fine-grained analysis which I leave for future research. For the purpose of a first approximation to RSL word order, it was not necessary.

Having defined clauses for RSL, it was then necessary to define subjects, objects and verbs. The procedure was again semantic and similar to the one used in Vermeerbergen et al. (2007) and Johnston et al. (2007). I used the label V to mark a verbal predicate, the label S to mark the most Agent-like argument in a clause, the label O to mark other arguments in the clause[10], Adv to mark adverbs (semantically modifying the verb), N to mark nouns used as predicates in nominal clauses, and A to mark adjectives (modifying arguments or being predicates in adjectival clauses). In the case of verbs of movement or location, I considered locations to be objects because semantically they are (obligatory) arguments of these verbs (a different decision was made, for example, in Sze 2008). For instance, if a cat is sitting on a chair, the chair is clearly an argument, not an adjunct.

The corpus contained 773 clauses (see the next section for the procedure of clause identification), 457 of which contained only one verbal sign without any arguments. The experimental data contained 229 clauses, 111 of which contained only one verbal sign.

3.4.2 Prosody

In this paper, I also look at the dependency between word order and prosodic marking (see the end of this section for motivation). In order to do so I use theoretical assumptions introduced by Prozorova (2009) that I want to very briefly outline in this section.

Prozorova (2009) analyzed her corpus of RSL to study prosody; the aim of her research was to determine how prosodic units are formed and how prosodic boundaries are marked. Working in the framework of information flow (Chafe 1994), she claimed that RSL discourse can be divided into units that she called elementary discoursive units (EDU) (a term offered by Kibrik & Podlesskaja 2009). EDUs are comparable to the more common notion of prosodic phrase.

Prozorova claimed that EDUs are consistently marked in the RSL discourse by head (and/or body) movements. According to her, there are two types of head movements: shifts and returns. Shifts are short movements from the default position in any direction, while during returns, the head returns to the default position and the shoulders are usually relaxed. Shifts mark the boundaries of EDUs, while returns mark the boundaries of bigger units which she called super-discoursive units. Additionally, boundaries of both types of units can be marked by eye blinks, pauses, and other prosodic markers, but all of these additional markers are optional.

Therefore, EDUs are formally defined: if there is a boundary shift, then there is a boundary, and the interval between two boundaries is an EDU. However, Prozorova (2009) showed that EDUs are also semantically and syntactically prominent in that they usually represent one event with one main participant, and syntactically constitute a clause. This observation confirms that these prosodic units are linguistically relevant objects.

However, the mapping between EDUs and clauses is not necessarily one-to-one. Clauses are syntactic units, while EDUs are prosodic units. For example, in the corpus I encountered the following sequence of two clauses (17).

In the first of these clauses, the locative object room constitutes its own EDU, while in the second clause the whole clause constitutes one EDU. This is parallel to the way an English sentence can de divided into intonation phrases in several ways:

Do the sentences in (18a-c) consist of different numbers of clauses, and does (17b) contain one clause, while the (manually) identical (17a) consists of two clauses? The answer can be “yes”, but it has to be proved, and in the context of my pre-theoretical approach, I do not have any reason to argue that the syntactic structures of (17a) and (17b) are different (and definitely not that (17a) contains more than one clause).

Prosody is also relevant when speaking about locative sentences. Padden (1988) argued for constructions like (11) that they are multi-clausal. Whether this is true (for RSL) is an open question, but if it is true I would expect to see more prosodic breaks in this type of constructions (given that there is a correlation between prosodic boundaries and clause boundaries).

Although I did not use prosodic boundaries to identify clause boundaries, I decided to analyze the possible dependency between division into prosodic units and word order. Before approaching the data, I formulated the hypothesis in (19):

This hypothesis will be discussed in Section 5.