Contrasting the category of discourse markers in spoken corpora. A usage-based approach to French and English

This presentation reports the method and quantitative results of a corpus-based approach to discourse markers (DMs), a diverse category of pragmatic expressions functioning as metadiscursive devices such as “so”, “you know” or “well” (e.g. Brinton 1996, Cuenca 2013), in spoken French and English. Despite a proliferation of case studies in numerous (especially written) languages, a paradigmatic approach to the full category of DMs is still lacking today, and the present project aims precisely at filling this gap. It is believed that, through manual corpus annotation, full coverage of this complex pragmatic phenomenon is a necessary – albeit challenging – methodological step towards building a reliable resource to further investigate a wide range of research questions. Within the analytical potential that such a resource offers, the present paper will focus on contrastive and register variation of form-function patterns. For this purpose, the bilingual dataset DisFrEn was compiled from existing spoken corpora in English and French, mainly ICE-GB (Nelson et al. 2002) and VALIBEL (Dister et al. 2009). The main principle behind this corpus design is context variation across eight interactional settings: conversation, face-to-face interview, radio interview, private phone call, classroom lesson, political speech, news broadcast and sports commentaries. After some technical treatment, all texts are sound-aligned and balanced across the two languages, amounting to 15 hours of recordings and 163,620 words. This research heavily relies on discourse annotation, by applying to DisFrEn an extensive, crosslinguistic and operational protocol (Crible in press): DMs are first identified following an inclusive definition of the category (hence a “bottom-up” approach with a tertium comparationis and no closed-list selection); these items are then described along a number of i) syntactic (part-of-speech, position in three types of linguistic units), ii) functional (macro-functions and micro-functions) and iii) contextual variables (co-occurrence with other DMs and with markers of (dis)fluency such as pauses or repetitions). The annotation of DisFrEn uncovered a total of 8743 DMs, with a rate of 53 DMs per 1,000 words. Overall, DMs are significantly more frequent in French than in English, and in spontaneous situations such as conversations or phone calls. The distribution of each variable is computed in absolute, relative and normalized frequency, comparing languages and registers. Results show that DMs do not behave similarly in English and French, with clear profiles at most levels of analysis. Major findings will be illustrated in the presentation. By cross-tabulating different analytical levels, recurring patterns of combination between certain types of DMs, their function(s) and their co-text will emerge from the data and inform us of contrastive preferences and their relative “conceptual frequency” (Hoffmann 2004) or “onomasiological salience” (Geeraerts et al. 1994), i.e. the frequency of a structure with respect to all its paradigmatic competitors, as opposed to textual or corpus frequency. Such exhaustivity in the annotation uncovers the inter-relation between variables in a fully connected model of DMs and (dis)fluency, in keeping with Schmid’s “from-corpus-to-cognition-principle” (2000: 39) or Gilquin’s work (2008) on salience and prototypicality