Impact of Prosody on the Perception and Interpretation of Discourse Relations: Studies on “Et” and “Alors” in Spoken French

ABSTRACT

We report on three experiments that aim at measuring the role of prosody in the acceptability and interpretation of discourse relations between utterances connected by two French discourse markers, viz. et “and” and alors “then/well.” These two discourse markers are highly polyfunctional: et “and” can express relations of addition, specification, temporality, or consequence, whereas alors “then/well” can express relations of consequence, specification, concession, or topic-shift, among others. The first experiment compares the naturalness of the stimuli in their spoken and written versions. The second experiment uses a forced-choice task where two prosodic versions of the same stimulus are confronted. The third experiment tests whether the prosody of the first segment and of the marker projects the discourse relation continuation. Our results show that the impact of prosody varies across markers and relations, with alors displaying stronger prosodic profiles associated to the discourse relations it introduces.

Notes

1. In the French subcorpus of DisFrEn, 4,494 DM tokens were annotated. The five most frequent DM types are et “and,” mais “but,” donc “so,” alors “then/well,” and hein “right.”

2. Topic change is not considered as a discourse relation in the Cognitive approach to Coherence Relations framework and cannot be described in terms of these binary features.

3. The production study resulted in 1,280 sequences (S1 + DM + S2) produced by 20 naïve speakers. We opted for an acoustic analysis instead of choosing a standardized transcription system like French ToBI (Delais-Roussarie et al., 2015) or Mertens’ transcription system for French intonation and accentuation (Mertens, 2014). The reason is that we wanted to keep the option of producing synthetized stimuli for the perception studies, based on acoustic parameters. Another reason is that we learned from our own experience (Mertens & Simon, 2013) that it is difficult to reach a high inter-rater agreement when prosodic transcription is concerned.

4. Values for pitch measurements are in STs instead of Hz, since this measure is considered as being better adapted to the human auditory system (see e.g., Nolan, 2003) and to the comparison between male and female voices. These values were automatically extracted from Praat (Boersma & Weenink, 2017) using the “ST relatives to 1 Hz” scale and then verified manually.

5. A few stimuli for which it was not the case were re-recorded.

6. The French DM et [e] is monosyllabic and contains no consonants. It was unnatural to pronounce it with an exaggerated lengthening, so that we preferred to modify the f0 range and the duration of the pause before the DM to keep the stimuli as natural-sounding as possible.

7. This ambiguity is measured as the proportion of inter-annotator agreement computed for each utterance during the disambiguation task, carried out with the same stimuli but in their written form (Methods: stimuli preparation, see Didirková et al., 2017).

8. See Model 1, Table 7 in Appendix 1 for further details.

9. All (mean) durations are given in milliseconds.

Additional information

Funding

This work was supported by the Belgian F.R.S./FNRS under grant no. F 4520.16 (MIS Project “PhraDiCo: Prosodic phrasing for discourse comprehension. Perception of prosodic boundaries and their contribution to the construal of discourse coherence,” PI: Anne Catherine Simon). Technical support for cut-and-paste of paired sound files and silent pause duration adjustment was provided by George Christodoulides. We thank Thomas François and Mickaël De Backer for their advice on the statistical analysis of the data and the anonymous reviewers for their insightful comments. Any remaining errors are ours. There are no potential conflict of interest reported by the authors.