A Quantitative Comparative Study of Prosodic and Discourse Units, the Case of French and Taiwan Mandarin

A Quantitative Comparative Study of Prosodic and Discourse Units, the Case of French and Taiwan Mandarin Laurent Prévot Aix Marseille Université CNRS, LPL Aix-En-Provence, France laurent.prevot@lpl-aix.fr Alvin Cheng-Hsien Chen National Changhua University of Education English Department Changhua City, Taiwan alvinworks@gmail.com Shu-Chuan Tseng Academia Sinica Institute of Linguistics Taipei, Taiwan tsengsc@gate.sinica.edu.tw Klim Peshkov Aix Marseille Université CNRS, LPL Aix-En-Provence, France klim.peshkov@lpl-aix.fr Abstract Studies of spontaneous conversational speech grounded on large and richly annotated cor- pora are still rare due to the scarcity of such resources. Comparative studies based on such resources are even more rarely found because of the extra-need of comparability in terms of content, genre and speaking style. The present paper presents our efforts for estab- lishing such a dataset for two typologically diverse languages: French and Taiwan Man- darin. To the primary data, we added morpho- syntactic, chunking, prosodic and discourse annotation in order to be able to carry out quantitative comparative studies of the syntax- discourse-prosody interfaces. We introduced our work on the data creation itself as well as some preliminary results of the bound- ary alignment between prosodic and discourse units and how POS and chunks are distributed on these boundaries. 1 Introduction Interest for the studies of discourse prosody inter- face has arisen in the last decade as illustrated by the vitality of the events and projects in this do- main. However, while theoretical proposals and descriptive works are numerous, quantitative sys- tematic studies are less widespread due to the cost of creating resources usable for such studies. In- deed, prosodic and discourse analysis are delicate matters requiring lower-level processing such as the alignment with speech signal at syllable level (for prosody) or at least basic syntactic annotation (for discourse). Moreover, many of these studies are dealing with read or monologue speech. The ex- tremely spontaneous nature of conversational speech renders the first levels of processing complicated. Previous works (Liu and Tseng, 2009; Chen, 2011; Bertrand et al., 2008; Blache et al., 2009; Afan- tenos et al., 2012) give us the opportunity to pro- duce conversational resources of this kind. We then took advantage of a bilateral project for working on conversational speech in a quantitative fashion, and this for two typologically diverse languages: French and Taiwan Mandarin. We believe this combination of linguistic resources and skills for these two lan- guages is a rather unique situation and allows for comparative quantitative experiments on high-level linguistic analysis such as discourse and prosody. Our objective is to understand the commonalities and the differences between discourse prosody in- terface in these two languages. More precisely, we look at how prosodic units and discourse units are distributed onto each other. In spirit, our work is closely related to the one of (Simon and Degand, 2009; Lacheret et al., 2010; Gerdes et al., 2012), however our focus here are the insights we can get from a comparative study. More- over our dataset has a more conversational nature than the datasets studied in their work. About the data, (Gerdes et …

Content truncated to 500 words