Using Extra-Linguistic Material for Mandarin-French Verbal Constructions Comparison

Using Extra-Linguistic Material for Mandarin-French Verbal Constructions Comparison ⋆ Pierre Magistry a,b, Laurent Pr´evot c, Hintat Cheung a, Chien-yun Shiao a, Yann Desalle d, and Bruno Gaume d aGraduate Institute of Linguistics, National Taiwan University, Taiwan {d97142005, hintat, r96142005}@ntu.edu.tw bCLCLP - Institute of Linguistics, Academia Sinica, Taiwan cLaboratoire Parole et Langage, CNRS � Universit´e de Provence, France laurent.prevot@lpl-aix.fr dCLLE-ERSS, CNRS � Universit´e de Toulouse, France {desalle, gaume}@univ-tlse2.fr Abstract. Systematic cross-linguistic studies of verbs syntactic-semantic behaviors for ty- pologically distant languages such as Mandarin Chinese and French are difficult to conduct. Such studies are nevertheless necessary due to the crucial role that verbal constructions play in the mental lexicon. This paper addresses the problem by combining psycho-linguistics and computational methods. Psycho-linguistics provides us with a bilingual corpus that fea- tures verbal construction associated with carefully built extra-linguistic material (short video clips). Computational approaches bring us distributional semantic models (DSM) to measure the distance between linguistic elements in the extra-linguistic space. These models allows for cross-linguistic measures that we evaluate against manually annotated data. In this pa- per, we discuss the results, potential shortcomings involving cultural variability and how to measure such bias. Keywords: Cross-linguistic study, Distributional Semantic Models, Psycholinguistics, Ex- tralinguistic context 1 Introduction The study reported in this paper is part of a broader project (M3: Model and Measurement of Meaning) which investigates lexical organization of French and Mandarin. The project combine psycho-linguistic and computational approaches for investigating the verb lexical organization for both languages. We therefore deal with two kinds of data: On the one hand the productions ob- tained through psycho-linguistic experiments realized both with French and Mandarin Chinese speakers; On the other hand we systematically exploit existing electronic resources. However, this paper does not address on the later aspect of this work. We focus here on crossing the results of both methodologies: psycho-linguistics experiments and analyses with computational and sta- tistical methods. As suggested by (Biber, 1988), particular attention is given to extra-linguistic material that is used here as a context for applying a distributional approach. Distributional ap- proaches are usually applied to linguistic material. However, in this study we attempt to use the ⋆ We want to thank Karine Duvignau and Marion Fossard for collecting the French data as well as Jialin Wu, Yufeng Yao, Miaoru Huang, Tehsin Wu and Sumen Chen for collecting the Mandarin data. The research pre- sented in this paper is supported by the ANR-NSC (France-Taiwan) bilateral project M3 (Modeling and Mea- surement of Meaning). More information about M3 project on these websites: http://erss.irit.fr/flexsem (France) http://140.112.147.149:81/m3/ (Taiwan). We are also grateful to the anonymous reviewers. Copyright 2009 by Pierre Magistry, Laurent Pr´evot, Hintat Cheung, Chien-yun Shiao, Yann Desalle, and Bruno Gaume 325 23rd Pacific Asia Conference on Language, Information and Computation, pages 325–334 videos themselves for unraveling some aspects of the mental lexicon. The experiments proposed in the following sections demonstrate the validity of such an approach. Our work can be situated in a growing body of semantic studies that approximate semantic …

Content truncated to 500 words