From co-occurrence to constructions: patterns of discourse markers and disfluencies across registers in English and French

Spoken language is characterized by online processes of production and comprehension happening over time. A natural consequence of this temporal nature is the presence of discourse markers (henceforth DMs, e.g. Schiffrin 1987) that can generally be defined as syntactically optional pragmatic expressions “fulfilling structuring functions with respect to local and global content and structure of discourse” (Fischer 2000: 20) such as well, but or I mean. These procedural devices express a wide range of discourse functions, from connective meanings such as cause or contrast, to interactional meanings such as turn-taking or monitoring. This heterogeneity within the DM category is reflected in preferences of syntactic category and position, as well as register and crosslinguistic variation (e.g. Aijmer 2013). DMs tend to co-occur with so-called disfluencies (e.g. Shriberg 1994) such as filled pauses, identical repetitions or false-starts, especially in impromptu speech though their presence is not excluded from more prepared and monologic situations. Through high frequency of use, some clusters of DMs and disfluencies tend to acquire constructional status insofar as they meet the criteria of high frequency, form-meaning pairing, sensitivity to context and flexibility in degrees of abstraction: for instance, the turn-initial combination "but um" expressing a disagreeing response in spoken English can be schematized as "DM + um", "turn-initial DM + um" or "conjunction + filled pause", etc., forming a paradigm potentially instantiated by other clusters. I argue that the high frequency and meaningful variation of co-occurrence patterns vouch for their treatment as constructions or schemas in usage-based terms, taking these notions slightly away from their lexico-grammatical core to a more discursive or interactional level, in line with recent works (e.g. Fischer & Alm 2013). In this paper, I will present the corpus-based method and major findings regarding the variation of these patterns across languages and registers. In doing so, the aim is to build a case for a constructional approach to patterns of DMs and disfluencies which would allow for their fine-grained investigation in varying degrees of abstraction, zooming in and out from broad categories to specific functions of DM lexemes (e.g. but um expressing disagreement). This proposal not only extends the scope of constructions from lexis-grammar to discourse but also takes into account the interactional and multimodal nature of authentic (spoken) communication, which has mostly been neglected so far. Following valid tertia comparationis and operational definitions, DMs and disfluencies have been annotated in a comparable corpus of English and French encompassing eight different interaction settings from private conversation to news broadcast. After combining information on the function and position of DMs and their co-occurring disfluencies, using statistical models such as conditional inference trees or multiple correspondence analysis, a number of interesting patterns emerged from the data: for instance, additive DMs (e.g. and) are typically utterance-initial and cluster with filled or silent pauses, while monitoring DMs (e.g. you know) are more strongly associated with the final position and disruptive disfluencies such as false-starts or truncated words. This paper paves the way for further study of the acquisitional and processing profile of these candidate constructions.