1. INTRODUCTION
This paper explores the uses of a database of corpora of naturally occurring interactions – CLAPI (Corpus de LAngue Parlée en Interaction) – for the systematic study of linguistic phenomena within social interaction.Footnote 1 It focuses on the specific complexities of interactional data, on the possibilities for archiving them and making them accessible to the scientific community and on the possibilities for browsing them in order to sustain qualitative analysis and/or to enhance quantitative approaches.
The paper is divided into two parts: the first deals with the presentation of the CLAPI workbench, the objectives, principles and solutions which have been adopted for the conception of its architecture. The second part focuses on the analysis of the discourse particle voilà: based on a previous multimodal qualitative analysis of this particle by Bruxelles and Traverso (Reference Bruxelles, Traverso and Drescher2006), the study exploits CLAPI larger corpora and search engines in order to verify, generalise and finally reorient previous findings. We demonstrate therefore how it is possible to highlight the specific contributions of particular analyses by applying them to larger quantities of data and by employing specific search tools.
2. CLAPI, A WORKBENCH FOR THE STUDY OF FRENCH LANGUAGE IN INTERACTION
2.1. Which corpora for interactional linguistics?
The need for readily available, naturally occurring audio and video recorded interactional data is now generally accepted by various disciplines within linguistics and the social sciences. On the one hand, social interaction is the ‘natural habitat’ of language – the site where grammar emerges (Hopper Reference Hopper1987) and changes, where children acquire their first language, where non-natives are socialised in a second language, and where social identities are displayed, negotiated and achieved. On the other hand, only naturalistically gathered data can make available for analysis complex details on which the organisation of talk is based – details which are neither imaginable, for instance by introspection, nor reproducible, for example in experimental tasks. On the basis of such considerations, interactional linguistics has explored both traditional grammatical forms and previously neglected verbal and vocal resources, as well as multimodal resources, as they are used for the organisation of talk in interaction: this has opened up new insights on syntax (Auer Reference Auer2005; Ford, Fox and Thompson Reference Ford, Fox, Thompson, Ford and Thompson2002), phonetics (Local, Kelly and Wells Reference Local, Kelly and Wells1986), prosody (Couper-Kuhlen and Selting Reference Couper-Kuhlen and Selting1996), the lexicon (Clift Reference Clift2001), and multimodality (Goodwin Reference Goodwin2000), in ordinary speakers’ talk, as well as in that of children (Wootton Reference Wootton1997), or aphasic participants (Goodwin Reference Goodwin and Goodwin2003). This has also fuelled new reflections and higher demands on transcription practices and standards (Jefferson Reference Jefferson and Lerner2004), as well as on audio and video recording practices and technologies. The field of conversation analysis has played an historical role in underlining the necessity for naturalistic data, of repeated listening and visioning of these data by the analyst, and of detailed transcriptions for the study of the way in which participants orient themselves towards the relevant resources mobilised for achieving the interactional order.
These insights have emerged very recently, since the 1960s, in the constitution of large collections of data, as a basis for a systematic exploration of the details of talk in interaction – for example, for the study of sequencing in openings (Schegloff Reference Schegloff1968, Reference Schegloff and Givon1979), or for the study of repair (Schegloff, Jefferson and Sacks Reference Schegloff, Jefferson and Sacks1977; Schegloff Reference Schegloff and Psathas1979). Nowadays, the existence of browsable data banks allows us to enlarge these systematic analyses; conversely, past systematic explorations of data collections fuel the development of search tools and request engines which are not reduced to the identification of single forms but try to access aggregates of forms, characterised by their distribution within various sequential positions. Interactional linguistics (Ochs, Schegloff and Thompson Reference Ochs, Schegloff and Thompson1996; Selting and Couper-Kuhlen Reference Selting and Couper-Kuhlen2001) challenges the way in which searches are done in large corpora, by stressing the importance of time and therefore of a series of successive moments within the turn as well as in the sequence (Mondada Reference Mondada and Condamine2005; Fox Reference Fox2007). Turns develop incrementally, step by step, in an emergent way, within the temporal unfolding of their production, sensitive to the contingencies of context and adjusting to the coordinated conduct of other participants: this view invites the development of a ‘positionally sensitive grammar’ (Schegloff Reference Schegloff, Hovy and Scott1996), where grammatical forms – as well as other resources – are described relative to their contribution to the sequential organisation of interaction. This description focuses on actions implemented in systematic practices mobilising specific (linguistic but also multimodal) resources arranged within specific configurations or constructions, positioned in specific sequential environments. This grounds the hypothesis ‘that there may, in fact, be grammatical formats for sequentially-specific actions’ (Fox Reference Fox2007: 304).
This conception of grammar in interaction (Ochs, Schegloff and Thompson Reference Ochs, Schegloff and Thompson1996) articulates two kinds of intertwined demands for larger corpora: on the one hand, this kind of inquiry is grounded in the details observable in naturally occurring social interactions, as they are recorded, transcribed, and stored in corpora constituted by primary data (recordings) and secondary data (their annotations and representations). On the other hand, this kind of inquiry relies on search engines which are ‘positionally sensitive’, which allow us to search for aggregates of concomitant forms in specific sequential environments.
Both dimensions inspire the development of specific data banks for interactional corpora. These are still in fieri: in most of the national corpora – such as the British National Corpus (BNC) or the Real Academia Española corpus (Corpus de Referencia del Español Actual, CREA) – the spoken part is quantitatively marginal and accessible only through very simple transcripts; data banks focusing on spoken language recorded in social interactions in natural settings (and not for instance through sociolinguistic interviews) are still very few. A notable exception is the Talk Bank (http://talkbank.talkbank.org/), started in 1999 by Brian MacWhinney, and which is devoted to the study of human and animal communication, but which does not offer any equivalent search mechanism capable of browsing the corpora of the data bank.
2.2. The development of the CLAPI talkbank
CLAPI (Corpus de LAngue Parlée en Interaction – http://clapi.univ-lyon2.fr) has been developed in order both to make accessible to a larger scientific community corpora of naturally occurring data and to make possible new kinds of (semi)automatic requests on these data.
The project was launched in the 1990s at the GRIC research lab in Lyon (Groupe de Recherche sur les Interactions Communicatives), which then became the ICAR research lab (Interactions Corpus, Acquisition, Représentations). This research unit had a long history of interactional studies based on corpora collected through fieldwork in various social settings. The first database was aimed at storing and conserving the corpora constituted by the members of the centre, associated with a set of metadata describing them. In this early version, each corpus was illustrated through a downloadable excerpt of sound and transcript, but was not associated with the entire corpus (for a presentation of this stage of the project in 2001, see Bruxelles and Traverso Reference Bruxelles, Traverso, Pusch and Raible2002).
The second database, CLAPI, was created in 2001 by the Groupe ICOR; it consists not only of a storage space for corpora but of an articulated workbench, where primary and secondary data are archived, described by metadata, and made browsable thanks to search engines (for a presentation of the project in 2004, see Balthasar and Bert Reference Balthasar and Bert2005; for an up-to-date presentation see Groupe ICOR, to appear b).
The project has thus evolved in time, taking into consideration various dimensions:
• It evolved from a data bank for the internal purposes of a closed group into a data bank open to the larger scientific community, thus varying the users’ profiles (both of data producers and data consumers) implemented through a codification of rights and accesses;
• It evolved from a data bank considered first of all as a storage space into a data bank considered as a collaborative workbench, offering various tools in order to make corpora and phenomena more accessible, thanks to XML tagging;
• It gradually merged older corpora, interesting as regards their legacy and for a diachronic approach to interactional data, and newer corpora, produced using state-of-the-art recording practices and transcribed, aligned, and tagged according to contemporary standards.
These changes show that a data bank is a dynamic object that evolves in time, implying a constant reflexive discussion about its constitution, its principles, and its choices.Footnote 2
2.3. Current state of the CLAPI talk bank
Currently, the data bank comprises 120 hours of audio and video, constituting 296 recordings and 477 transcripts – of which 40 hours are aligned with their transcripts in an XML format (the equivalent of 350,000 words). The collated corpora document various social interactions, from dinner conversations to medical interactions, from legal discussions to service encounters – recorded over the last 25 years: the data bank is dedicated to corpora of social interactions in French, recorded within a diversity of settings in a naturalistic way.
In order to make these data available, the workbench has been developed according to several principles and choices, governing the metadata, the transcript and formats, and the access. At these three levels, the issues are to enhance data accessibility, to offer the possibility of efficiently selecting a subset of data for specific analytical purposes and to provide interoperability through browsable and searchable corpus structures and resource descriptions. In this development, the relevant legal and ethical constraints have been scrupulously taken into account.
2.3.1. Metadata
Metadata represent a key resource that characterises, thanks to a standard set of descriptors, a corpus in sufficient detail and accuracy for users to determine whether or not it is of relevance to their particular enquiry. Metadata of the CLAPI talk bank are both oriented towards a wide scientific community of users and applications as well as towards the specific features of interactional corpora and needs of interactional research. In order to respond to these rather complex demands, 75 descriptors have been elaborated, taking into account current work on social interaction as well as standards introduced by various international initiatives, such as the Dublin Core Metadata Initiative (which proposes a rather general vocabulary of fifteen properties for use in resource description), the OLAC initiative (which proposes an XML metadata format to interchange language resources within the framework of the Open Archives Initiative), and the ISLE metadata initiative (IMDI, which is oriented to multimedia/multimodal language resources paying special attention to disciplines such as corpus linguistics, anthropology and language and speech engineering).
The descriptors cover a set of generic or specific fields organised as follows:
• General information on the corpus (name, date and place of collection. . .);
• Authors (owner(s), collectors, transcriptors. . .);
• Recordings (name, date and place of recording, length, type: audio/video/computer mediated traces . . .);
• Transcripts (transcript conventions, software, alignment and tagging, information on the quality of the transcript);
• Participants (number, sociolinguistic characteristics of each participant);
• Interactional genre (medical consultation, conversation, etc.);
• Access conditions of the transcript and the recordings;
• Bibliography.
In addition to the standard information that is required in a database, special attention has been devoted to:
1) The different agents who are involved in the creation of a corpus throughout the successive steps of its production (i.e. fieldwork, data recording, digitalisation and anonymisation, transcription, alignment, annotation etc.). It is the ‘corpus owner’ who decides the conditions under which her/his corpus is made accessible.
2) The scientific documentation of the corpus: previous studies on the corpus are listed (and if possible, they are full-text accessible). When recording a corpus in the database, these descriptors can be filled either by selecting predefined categories or by proposing a descriptive text – which provides for more specific details. When consulting the database, they facilitate the selection of a set of corpora for automatic queries, and they can be used as component of the query itself (see 3.4).
2.3.2. Transcript and formats
As a talk bank specifically devoted to the documentation of naturalistic interactional data, CLAPI is oriented towards the preservation and accessibility of the relevant detailed features of social interaction. This has consequences for the way in which secondary data (transcription) are made out of the primary data (recordings).
Transcription for interactional research implies that the specificity of spoken production is preserved: a transcript convention has been established to this end (ICOR Convention), inspired by current standards in interactional studies, such as the Jefferson Conventions (Jefferson 2004) and the GAT conventions (Selting et al. Reference Selting, Auer, Barden, Bergmann, Couper-Kuhlen, Günthner, Quasthoff, Meier, Schlobinski and Uhmann1998). It encompasses a list of relevant phenomena, related to both the verbal production and the multimodal resources (gestures, gazes, body postures, facial expressions, etc.) used by participants in their exchanges, as well as the temporal organisation of these resources within interaction (coordination with other co-participants’ conduct).
In this excerpt, the main features of interactional transcription are available:
• each piece of talk is related to a participant (initials);
• the characteristics of talk production are preserved: repetition, false start (1038, 1041 for instance), pronunciation (made visible by the use of an ‘adapted orthographic representation’, as in 1038 ‘t'as’, instead of the standard ‘tu as’);
• precise temporal indication on turn organisation is given: overlap (indicated by the brackets, in 1044–1045 for instance); gap (1040, 1042);
• multimodal information is provided on a separate line and located in relation to talk (for instance line 1040, + indicates the place where the gesture described in the following line stops or begins).
In addition to the attention devoted to this sort of transcription, the number and complexity of relevant phenomena make it necessary to align the recordings with the transcripts, with one of the currently available pieces of software such as Praat, CLAN, or ELAN.
On the basis of the ICOR transcript convention, tools have been developed for providing an XML tagging of the main interactional phenomena (participants, tokens, verbal productions, overlaps and pauses) making automatic search possible. The issue of adapted orthography has been solved by the development of an automatic orthographic variants recognition tool.
In the current state of the database, each corpus contains both the original transcripts and conventions chosen by the original owner and the ICOR-XML transcript. The result is a preservation of the original features, the use of a standard, and the browsability of a large number of non-standard resources.
2.3.3. Access
CLAPI provides various tools for interactional research, making possible the browsing of the corpora in a selective way.
A first set of tools automatically generates a synoptic view of the main characteristics of each corpus, such as its lexicon and lexical frequencies, the distribution of interactional phenomena (participants, tokens, verbal productions, overlaps and pauses) throughout the entire transcript, and the more frequent correlations between these phenomena.
A second set of tools offers various searching possibilities. The simplest search tools (‘co-occurrences of a token’, ‘contextual analysis of a token’) allow for the search of a linguistic form or token in order to make concordances, with the results displayed in an aligned form. More complex search tools allow for the search of various tokens, in various positions within the turn, and taking into consideration their co-occurring within specific interactional structures (‘complex search’, ‘requêtes multicritères’). For example, it is possible to search for two tokens which are separated by three others, the first being at the end of a turn and the second at the beginning of an overlap or just after a pause). Thus, searches are able to deal with linguistic and interactional items, as well as with standard and non-standard original items, thanks to the automatic tagging of the latter. Searches within the transcripts can be combined with searches within the metadata.
2.3.4. Legal and ethical constraints
These possibilities offered by the CLAPI database – accessing metadata, consulting primary as well as secondary data and browsing corpora – are crucially dependent on the legal and ethical constraints which are particularly complex and delicate in the case of naturalistic data.Footnote 3 Taking these issues into consideration, in order to respect both the privacy of the recorded participants and provide access to the data for a large community, the CLAPI workbench has implemented a series of solutions. These solutions concern, on the one hand the categorisation of the items to be accessible and, on the other the categorisation of the users.
Corpora items can be accessible in various ways: the whole of a corpus can be consulted (recordings and transcription), or only a part of it can be available (i.e. a subpart of the corpus and/or the transcript only, or the transcript cum audio but not video), or the corpus can be accessible by the means of the search mechanisms only (as the result of a query). Various users’ profiles have also been implemented, allowing different kinds of accessibility, depending, for restricted corpora, on previous agreements made with the corpus’ owner.
In general terms, a good accessibility of the corpora has been made possible, while respecting the lega and ethical demands of the corpus owner(s): a generic user of the talk bank, without special authorisation, can gain access to all the metadata, to the transcript conventions, to the freely accessible corpora, to about 80 exemplary subcorpora (transcripts and signal), and, through search engines, to 70% of the XMLised data set (30h, 265,000 words). The remaining portion of the data bank is accessible by agreement with the corpus owner(s).
3. FROM QUALITATIVE TO QUANTITATIVE ANALYSIS: THE CASE OF VOILÀ
The second part of this article is devoted to the study of the discourse particle voilà in the large range of corpora accessible in the CLAPI database.Footnote 4 It is based on a previous multimodal interactional study of this particle (Bruxelles and Traverso Reference Bruxelles, Traverso and Drescher2006) that has been carried out on a video recording of a working meeting of architects, the Mosaic corpora (Détienne and Traverso (eds.), to appear). Before presenting these findings, we recall the main outcomes of previous studies of voilà.
3.1. Previous studies of voilà
3.1.1. Deictic and Conversation Structuration Marker
The particle voilà has a core deictic value, which is still visible in its morphology: voilà is composed of the verb voir (‘to see’) in the imperative mood and of the adverb là (‘here’). Voilà was opposed to voici, as indicating distance (vs. proximity), but this opposition is no longer significant in spoken French. In their 1999 and 2001 papers, Plauché and Bergen explain in detail the diachronic evolution that has led, along with a metaphoric extension, from this core usage of voilà as a deictic – in which the particle draws attention to an object in the setting – to its temporal usage (‘voilà l'instant que nous attendions’, lit. ‘here is the moment we were waiting for’) and to its usage as a discourse deictic (‘voilà deux exemples’, lit. ‘here are two examples’).
Voilà has also been studied by the Geneva researchers (Roulet et al., Reference Roulet, Auchlin, Moeschler, Rubattel and Schelling1985) in their work on Conversation Structuration Markers (‘marqueurs de la structuration de la conversation’, MSC). Following Gülich's pioneer work on Gliederungssignale (1970), they define these items as forms (morphemes or syntagms) that are recurrent in discourse production and characterised by a loss of their original lexical meaning. The category of Conversation Structuration Markers is based on the role these items play in interaction: they display the way in which participants structure the co-construction of discourse. Roulet et al. develop an approach based on a hierarchical model (Reference Roulet, Auchlin, Moeschler, Rubattel and Schelling1985: 99–100) that distinguishes between a ‘monological level’ (the level of interventions) and a ‘dialogical level’ (that of exchanges). Within this framework, they identify the following features for voilà. At the level of intervention, voilà is one of the rare markers that are not constrained in their syntagmatic position (i.e. to the beginning or end of a constituent): voilà can precede as well as follow the constituent it marks. As for structure, voilà marks the constituent as the main act or intervention in the sequence. In this analysis, voilà, as well as quoi and pi(s), does not articulate constituents at the exchange level. It mainly marks monological relations and it scarcely appears in dialogical contexts (Roulet et al. Reference Roulet, Auchlin, Moeschler, Rubattel and Schelling1985: 108). For Auchlin (Reference Auchlin1981: 95, footnote 11), the main value of voilà on this monological level is to indicate that the constituent is conclusive. Working within the same framework, Filliettaz (Reference Filliettaz, Cavé, Guaïtella and Santi2001) points out that this description does not account for all the usages of voilà. Specifically, it does not account for the cases in which voilà is used along with a ‘non-verbal action’ (‘action non langagière’), as it is frequently the case in his data (service encounters). He thus proposes that in such usages, voilà is not a Conversation Structuration Marker, but is used in order to clarify the communicative feature of a non-verbal action (2001: 194–199).
Although we are not working within the same framework, the main outcomes of these studies needs to be kept in mind in order to be verified as part of our quantitative study: voilà mainly acts at a monological level (i.e. it is used within a turn and not to articulate turns by different speakers); it has a prominent conclusive value and, in some of its uses, a co-gestural value.
3.1.2. A multimodal study of voilà
Bruxelles and Traverso's study (Reference Bruxelles, Traverso and Drescher2006) is based on the video recording of a meeting between three architects. Their analysis integrates the gestural dimension of interaction: gaze, gestures to documents, position of the body, etc. They have identified three values for the particle voilà in their data: a) a use of voilà as a deictic particle (excerpt 2); b) a use of the particle as a structuration marker, specifically in transitions between two successive topics or activities (excerpt 3); c) a use of voilà as an agreement marker (excerpt 4). This usage is particularly found in highly collaborative sequences, and voilà frequently has a conclusive value.
C's pointing gesture (line 160: C points at the concerned area on the plan and shows it with a repetitive gesture from left to right) is done immediately after he has uttered ‘voilà’. This gesture lasts several seconds, and is accompanied by several other deictic verbalisations: ‘(voilà) c'est là’ (‘here it is’), ‘c'est celle-ci’ (‘it's this one’) (160, 163).
The other elements that accompany ‘voilà’ in this transition are found in nearly all the transitions between activities throughout the entire meeting:Footnote 5
• Other particles ‘alors’ (‘then’), ‘bon (‘well’) and ‘c'est bon’ (‘okay’);
• Numerous pauses;
• Gestural markers: the speaker retracts his current gesture and handles the plan that will be used in the next activity.
• Furthermore, in this excerpt, a meta-pragmatic formulationFootnote 6 is used (line 3).
In this excerpt, two occurrences of ‘voilà’ are used to answer requests for confirmation marked by the particle ‘hein’ (translatable as ‘you see’).
The authors have identified three main features that distinguish the different usages:
• voilà isolated or combined. This feature distinguishes cases when voilà constitutes the complete turn. It is the case for the deictic marker and for the agreement marker, but not for the structuration marker. A related feature seems to be that isolated voilà often has an interjectional feature.
• Monological (within one participant's turn) or dialogical (second pair part) use. Only the structuration marker is clearly used to articulate components within a participant's turn; the two other categories relate turns of different participants.
• Backward or forward orientation. This issue only concerns the agreement marker and the structuration marker, for the deictic marker is related to the local item (be it an object or an idea) that is designated. As to the agreement marker, it mainly has a closing feature (thus a backward orientation), which is displayed by the positive assessments, and the summaries that are often uttered along with it. The issue is less clear for the structuration marker. The authors put forward two alternative views: it could be understood as having an opening feature (thus a forward orientation), for it is followed by topical or actional changes; on the other hand, it can be seen as a constitutive element within a configuration of the form ‘marker + pause + marker’ that projects a transition. Through such a configuration, participants display that they are achieving a transition and they devote a certain amount of time to this achievement. This permits them to prepare the introduction of the next activity but also gives them the opportunity to go back to the previous one for an expansion or an addition.Footnote 7 This issue of the marker's orientation will be one of those we will try to clarify through the current study.
The authors have identified additional features that can be summarised in the following table:
3.2. CLAPI studies of voilà
Before examining how the CLAPI database tools make it possible to check these findings on a larger range of corpora, we can provide general quantitative information about voilà on the basis of its frequency in the database.
3.2.1. An overall view of voilà in the CLAPI database
There is no orthographic variant of voilà in the database (no shortened version).Footnote 8 Seven ‘related’ forms are found: one misprint (‘voila’ without the accent on ‘a’), and six forms ‘voilÀ’, in which the capital letter marks that the final vowel is pronounced louder.
There are 871 occurrences of voilà in the CLAPI database. A quantitative approach (in addition to a qualitative one) is thus possible and relevant. The particle is one of the most frequent, along with mais, ben, alors (see ICOR, forthcoming a and b).
The distribution of voilà in the different corpora accessible in the database is heterogeneous: for a similar length (about 60 minutes), the number of occurrences ranges from 14 (in the corpus ‘Français des années 80’, which is composed of recordings of sociolinguistic interviews) to 90 (in the corpus ‘Vente à domicile’, recordings of a door-to-door salesman selling encyclopaedias). The type of activity developed in the interaction seems to affect the use of the particle: for instance, the manipulation of objects seems to be related to a high rank of use, whereas the situation of interview seems to reduce its use. A link also appears with the situation of mediated talk: the use of voilà seems to increase in phone-calls. This result is somewhat contradictory to the ‘multimodal’ use of voilà. We can hypothesise that it is linked to the necessity for the participants to display in a very visible manner both the fact that they are on line (receipt tokens) and the way in which they structure their interaction.
3.2.2. Constitution of a subcorpus on the basis of the metadata
In order to carry out this extended study of voilà, we have selected in the database a range of corpora that are, to a certain extent, homogeneous with the Mosaic corpus: i.e. XML tagged corpora, recorded in natural settings, in multiparty situations.
The subcorpus is constituted of 23 transcripts. It represents various interaction genres: conversations between friends, working or institutional meetings, medical consultations, service encounters, and classroom interactions. This variety of situations prevents bias that could be linked to an over-representation of a certain type of situation. The number of participants in these situations ranges from two to twelve.
The subcorpus represents 114,457 tokens, and encompasses 312 occurrences of voilà.
3.3. Analysis of voilà with CLAPI database tools
We aim to show how the CLAPI database tools can be used to confirm or modify the outcomes emanating from the qualitative study. We have used the tools working on the linguistic and interactional context of the particle: co-occurrence of voilà with tokens and interactional phenomena that are tagged in the base (i.e. overlap and pauses), and location of the particle within the turn-at-talk.
Various means to access these results are usable in the database: either automatic search of co-occurrences (‘co-occurrences of a token’, ‘contextual analysis of a token’), or ‘complex search’ (‘requêtes multicritères’) in which the researcher can choose the number and the type of criteria he/she wants to use.
3.3.1. Co-occurrences of tokens
The automatic search of co-occurrences shows that the left- and right-hand co-occurrent tokens of voilà are mainly other discourse particles. What emerged from the analysis of a single corpus (Bruxelles and Traverso Reference Bruxelles, Traverso and Drescher2006) is thus not linked to a specific corpus or to an interaction genre, but holds for a variety of corpora. The automatic search of co-occurrences indicates that, in more than half of the cases (56%), voilà appears in a series of particles.
A second result concerns the particles that are used together with voilà. The most frequent is ben (which could be translated in many of its usages into ‘well’): 16 occurrences in the first left-hand position (‘ben voilà’); 20 occurrences in the second left-hand position (‘ben x voilà’). This frequent co-occurrence had not been identified in the qualitative study, although it stands for the three usages that have been described: ah ben voilà is a typical deictic marker; ben voilà is used for introducing reported speech (and marks a transition); at the end of a turn, sometimes following ‘et’ (‘and’), ‘ben voilà’ has a conclusive value (for instance, it is uttered by a dentist at the end of a dental treatment that has lasted for about two minutes – corpus Dentiste).Footnote 9 A further examination of this series ben voilà would be of interest, specifically when compared with the reverse series voilà ben, which only appears four times in the data.
The next most frequent token in the left-hand context of voilà is puis/pis (‘puis/pis voilà’, ‘then voilà’): 16 occurrences. Interestingly, it also seems to function with a conclusive value, although puis (‘then’) semantically denotes that something will be added (an item in a list, the next episode in a story, etc.).
In the right-hand context, the most frequent co-occurrent token is not a particle, but the pronoun c’ (19 occurrences, out of which 18 are ‘c'est’, ‘it is’). The second most frequent token is a second occurrence of voilà (‘voilà voilà’): 12 occurrences (out of which seven take place at the beginning of the turn, and two consist of a series of three ‘voilà’). The next most frequent token in the right-hand context of voilà is the affirmative morpheme oui, and its variant ouais (9 ‘oui’; 7 ‘ouais’). Out of these 16 occurrences, 9 are composed of the two tokens alone ‘voilà oui/ouais’. Two additional observations can be made concerning this result. First, it confirms the use of voilà as an agreement marker (Bruxelles and Traverso's results, table 1, column 3). Significantly, non (‘no’) rarely occurs along with voilà (2 occurrences), and there is no occurrence of si (with its negative meaning) in this position. Second, the reverse position of the two tokens (‘oui’ in the left-hand context of ‘voilà’, ‘oui voilà’) only occurs four times in the data. Moreover, this last usage does not correspond to a collocation, but to the successive production of the two tokens separated by a pause or uttered with opposing prosodic formats. This clarification is made possible by the immediate access to the signal that the database permits. This access constitutes an important asset of the database, for it facilitates the checking of the transcript, thus avoiding errors due to an inaccurate transcription.
An important outcome of the automatic search of co-occurrences of voilà is to have shown the importance of the clusters of particles. Furthermore, it suggests that an order of particles within the clusters can be significant. This idea of a sequential order of particles and interactional phenomena has already been verified in previous qualitative studies, specifically in transitions (Mondada and Traverso Reference Mondada and Traverso2005; Bruxelles, Greco and Mondada forthcoming). The possibility of automatic search with the CLAPI database tools is a valuable source of help for this type of investigation.
3.3.2. Location of voilà within the turn
With the aid of the tool ‘contextual analysis of a token’, we have access to the different positions of a token within the turn. Voilà presents a very specific functioning in this respect: it is used as frequently at the beginning of the turn as at the end (in 50% of its occurrences, it is the first token in the turn; in 39%, it is the last). By comparison, the location of the other particles within the turn is clearly different: ben is in 40% of its occurrences at the beginning of the turn and in 3% at the end; donc is in 25% of its occurrences at the beginning of the turn and in 5% at the end; mais is in 25% of its occurrences at the beginning of the turn and in 6% at the end. This specificity of voilà is to be understood in relation to the turn length:Footnote 10voilà is as frequent at the beginning as at the end of the turn, because it frequently occurs in short turns, i.e. fewer than four tokens. This usage represents 38% of its occurrences (four times more often than the other particles).
This result is consistent with Bruxelles and Traverso's (Reference Bruxelles, Traverso and Drescher2006) findings on the use of voilà as a deictic or an agreement marker, for which they have identified that the marker constitutes the complete turn or is used in short turns.
3.3.3. Co-occurrence of voilà with other interactional phenomena
The two interactional phenomena that are tagged in the CLAPI database (in addition to the turns, tokens and participants) are pauses and overlaps.
• Pauses: voilà is used twice as frequently after a pause (12%) than before one (6%). In the latter case (and at a distance of three tokens before the pause), it is noteworthy that the most frequent usages are not voilà as deictic, but those of voilà in transitions. This also confirms what numerous previous studies on transitions had shown: on achieving transitions, participants make use of long and numerous pauses. This emphasises the temporal unfolding of interactional activities.
• Overlap: voilà frequently occurs in overlap, and specifically in the overlapping segment, 15%, (versus 2% in the overlapped segment). This specificity is related to the use of voilà as a deictic or an agreement marker (cf. ICOR, forthcoming b).
On the basis of this overall view of voilà in the database, we turn now to a detailed examination of the particle in what has emerged as an intriguing and recurrent context: voilà preceded by an uncompleted syntactic construction. This configuration has been discovered by observing the possible materials occurring before and after voilà and by taking into account their retrospective or prospective orientations, as well as their sequential functions (as closing a particular unit). However, these features have not been identified directly by an automatic search, but thanks rather to the possibility of accessing the transcript and the signal of the stretches of interaction that have been found as hits in response to an automatic query. This configuration had not been noticed by Bruxelles and Traverso (Reference Bruxelles, Traverso and Drescher2006). It is nevertheless interesting in comparison with their results for it may challenge their distinction between the two categories of structuration (which has been related to transitions) and agreement marker (to which a closing value has been attributed), as well as between prospective and retrospective orientations of particular markers (such as et puis). More specifically, it raises the issue of the possibility and the relevance of sorting out closing and opening values for voilà as a structuration marker, since it can co-occur with resources that have either a forward orientation or a backward orientation.
3.4. Back to qualitative analysis: Voilà as an interactional hinge
Automatic searches have confirmed but also revealed different characteristic positions of voilà, as well as various other linguistic materials immediately preceding or following it. These findings invite us to return to the corpus for a qualitative in-depth description of some of these environments, in order both to refine the analysis and to give hints for further developments of more sophisticated search tools.
The study of co-occurrences of voilà, coupled with tokens, pauses and overlaps by using the ‘complex search’ tool has enabled us to isolate a recurrent environment that seems interesting for highlighting some of the jobs achieved by voilà as an interactional resource. This environment is constituted of a syntactically uncompleted Gestalt followed by voilà in turn-final position.
Here is a first set of excerpts showing this phenomenon in a particular configuration:Footnote 11
In both cases, voilà in final position closes a turn, which is produced by a second speaker overlapping the current speaker. In both cases, this turn is syntactically incomplete: in excerpt 5, line 6, the article (‘le’) projects a noun which is not uttered; in excerpt 6, line 6, the verb ‘avoir’ projects an argument which is not uttered. In both cases, the overlapping speakers withdraw in favour of the overlapped speakers. Moreover, both produce a contribution that acts out the same kind of relevant next action at that sequential moment: in excerpt 5, after a slight pause (4), FAB and SOP self-select almost at the same time, for proposing a kind of conclusion to what precedes; in excerpt 6, L answers to C with a negative utterance (5) which is very similar to M's answer (4).
So, in both cases, two speakers are not only overlapping but also producing a similar kind of turn, responsive to a previous one. The second speaker seems to withdraw before completing the turn s/he is engaged in. S/he thus displays that her/his own turn needs not to be completed, since it is achieved by the first, overlapped turn. In this sense, the syntactical incompleteness of the second turn is a product of the orientation towards the fact that action completion is achieved by the first. In this environment, voilà introduces a right boundary of an otherwise syntactically uncompleted turn: it closes the Gestalt.
We can notice that, in excerpt 5, a first collaborative conclusion has already been produced just before (1–2), closed by the one-unit turn ‘voilà’, co-occurring with a closing gesture (image 1).
The environment described in these two excerpts displays one of the interactional jobs achieved by voilà, consisting in closing a Gestalt which otherwise could appear as (syntactically) unachieved.
This invites us to search for other cases in which voilà occupies the final position of an uncompleted turn:
In these three excerpts, unlike the case in the previous ones (5, 6), we don't have two concurrent turns achieving the same action, but we still have a syntactically uncompleted turn, closed by a final ‘voilà’.
In excerpt 7, M begins with ‘je’ (3) projecting a verb but not producing it; in excerpt 8, SOP engages in the production of a comparative expression with ‘le moins’ (4) but self-repairs it and does not complete it; in excerpt 9, SOP, after a pause, adds an extension to her previous turn with ‘et puis euh’ (7) but does not continue. In all the cases, the speaker starts a syntactical structure but stops before completing it: (s)he closes the turn by ‘voilà’, thus displaying that, although syntactically uncompleted, it has reached its interactional completion.
Thus, voilà is a resource by which speakers mark the turn as complete, in spite of its syntactic incompleteness. This usage shows that participants in interaction rely on various kinds of turn completion and that completion is locally defined within the ongoing course of interaction, in a flexible and changeable way.
This feature of voilà is also attested in a similar environment, constituted by ‘et puis voilà’ (as in excerpt 9 above) or ‘et voilà’, in which the particle immediately closes a construction, even if something more has been projected.
In the following two cases, it is possible to contrast, within the same excerpt, two uses of ‘et puis’: the first projects and implements more to come; the second is immediately followed by ‘voilà’:
In excerpt 10, SOP is proposing various initiatives to take within a business strategy: the first one is formulated with ‘et pis’ + an action description (‘on IRA VOIR la direction’, 2); the second one, added after a pause (4), is not completed and followed by ‘voilà’ (5).
Similarly, in excerpt 11, the participants are collaboratively elaborating the description of the way in which objects are spatially distributed in supermarkets: FAB introduces a first explanation (1–3), which is reformulated by JEB, line 4. JEB pursues his description, in line 8 (‘et puis euh’), and is collaboratively joined by FAB, who produces the next part of the turn (9 and then 11, 13). In line 15, SOP closes this explanation by ‘et voilà’, accompanied by a closing gesture (images 2, 3 and 4).
Thus, in both cases, we have first a use of et puis followed by a syntactical turn completion and a second use of et (puis) followed by ‘voilà’. In the latter case, ‘voilà’ closes down a Gestalt that is still open.
These analyses focus on environments where voilà closes an uncompleted Gestalt. This accounts for the participants using voilà as an interactional resource for doing the closings in an environment where completion is not achieved at the formal level. Moreover, this draws attention not only to the position of voilà within the turn, but also to the resources that co-occur before and after it. These resources can be characterised by:
• their completion or incompletion;
• their stronger or weaker projective potential;
• their possible orientation not only onwards – prospectively – but also backwards – retrospectively.
Voilà deals with these features, which can coexist within a multi-unit turn: a multi-unit turn can be incrementally formatted in such a way that it is oriented either to the previous or to the next turn, this direction being possibly reoriented during the temporal unfolding of the turn. Voilà works as a closing device, distinguishing between these Gestalts and providing for their boundaries. This can be seen in the following two excerpts:
In excerpt 12, JEB tells an anecdote about an asparagus-flavoured biscuit and then goes on with a more general topic concerning innovative products. Line 8, after the punch line of the story (3), and SOP's (4) and FAB's (5) comments, JEB initiates the closing of this sequence in overlap (6). After a pause (7), he nevertheless initiates a turn projecting more to come (‘c'est euh’ 8), then immediately closes it by ‘voilà’, although it is uncompleted (see the previous excerpts for similar cases). After it, ‘donc’ opens another prospective development.
In excerpt 13, we have another kind of material preceding ‘voilà’: ‘ouais’ orients retrospectively in a responsive and aligning way to what precedes; after ‘voilà’, ‘après’ is clearly oriented forwards.
Thus, voilà can work as a hinge between different kinds of materials, oriented to different directions. Further inquiries will have to explore both sequential positions (before and after voilà) and formal resources, which can be implemented within them, having different orientations (prospective or retrospective). Together, these configurations of forms and positions define a positional grammar invoked by interactional linguistics.
4. CONCLUSIONS
In this paper, we have given an overall view of how quantitative and qualitative approaches can be used jointly in interactional linguistic research. Specifically, we have underlined how automatic research in a large range of data can support detailed analysis of interactional phenomena.
The first part of the paper was devoted to the CLAPI database, presenting its functionalities and the various opportunities it offers for studies in different linguistic fields concerned with the spoken language. The CLAPI database is both an archive of naturally-occurring interactional data, easily browsable thanks to a rich set of metadata, and a workbench encompassing a series of automatic search tools. Although the base has been designed and developed from an interactional point of view, it is also a resource for other types of approaches to spoken interaction.
The analysis presented in the second part of the paper has shown how a careful consideration of similar sequential environments leads to identification of recurring interactional configurations, made of various interactional resources used by participants. It has illustrated how the constitution of a collection of cases for exploring these sequential environments is made easier and even made possible in larger corpora by search tools able to identify various items (such as voilà combined in an ordered way with other items, et puis and/or après, donc, etc.), to specify the sequential position of certain items (voilà in a turn-final position, voilà within or just after an overlap), to recognise concomitant gestures (‘closing gesture’ simultaneous with voilà) and even to detect uncompleted turn-constructional units. As the study has shown, currently available CLAPI tools do not automatically produce the collection of relevant cases: this has to be done manually. Therefore, search tools represent a more heuristic way of browsing large corpora than a tool that automatically produces final results; moreover, this manual dimension of the search also involves the essential check and refinement of the transcripts, allowing for the discovery of new relevant features (to give only one example, the resolution onset of the overlap has been transcribed during the analysis, since it was not annotated in the original transcripts; all transcripts have been refined for the particular purposes of the analysis). The use of the database involves both automatic and manual analyses: in turn, the latter allow the development of new search tools or functions, which originate in analytical practices and respond to analytical needs.
Transcription conventions
Data were transcribed according to the ICOR convention (see http://icar.univ-lyon2.fr/projets/ICOR/ICAR_Conventions_ICOR.pdf).
Multimodal details have been transcribed according to the following conventions:
* * delimitate participant's gestures and actions descriptions.
+ + (one symbol per participant).
*–> gesture or action described continues across subsequent lines.
*–>> gesture or action described continues until and after excerpt's end.
*>> gesture or action described begins before end of excerpt.
—>* gesture or action described continues until the same symbol is reached.
. . .. gesture's preparation.
— gesture's apex is reached and maintained.
,,,,, gesture's retraction.
im the exact point where screen shot (image) has been taken is indicated,
# with a specific sign showing its position within turn-at-talk.
