Automatic Period Segmentation of Oral French

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 6389–6394 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 6389 Automatic Period Segmentation of Oral French Natalia Kalashnikova¹,4, Loïc Grobol², Iris Eshkol-Taravella³, François Delafontaine¹ LLL UMR 7270¹, Lattice², MoDyCo UMR 7114³, ZAION4 10 Rue de Tours, 45065 Orléans , 1 Rue Maurice Arnoux, 92120 Montrouge, 200 Avenue de la République 401B, 92001 Nanterre, 18bis Rue de Villiers, 92300 Levallois-Perret natalia.kalashnikova@sorbonne-nouvelle.fr, loic.grobol@ens.fr, ieshkolt@parisnanterre.fr, francois.delafon- taine@etu.univ-orleans.fr Abstract Natural Language Processing in oral speech segmentation is still looking for a minimal unit to analyze. In this work, we present a comparison of two automatic segmentation methods of macro-syntactic periods which allows to take into account syntactic and prosodic components of speech. We compare the performances of an existing tool Analor (Avanzi, Lacheret-Dujour, Victorri, 2008) developed for automatic segmentation of prosodic periods and of CRF models relying on syntactic and / or prosodic features. We find that Analor tends to divide speech into smaller segments and that CRF models detect larger segments rather than macro-syntactic periods. However, in general CRF models perform better results than Analor in terms of F-measure. Keywords: spoken language, automatic segmentation, period, oral french, CRF, macro-syntactic units 1. Introduction Automatic segmentation of oral speech is one of the main components of spoken language processing applications such as speech recognition, data extraction, etc. However, sentences, the relevant units for written language are ill- suited for these tasks. Lacheret and Victorri (2002) suggest to replace it by the notion of prosodic period, grounded in observations and analyses of spontaneous spoken language. A semi-automatic tool for speech segmentation in periods, Analor, was developed (Avanzi, Lacheret-Dujour, Victorri, 2008) within this theory. However, this approach only takes into account prosodic characteristics of speech, which – since segmentation depends not only on prosody, but also on syntax and semantics – seems overly restrictive. We propose to go beyond these limitations by working within the Fribourg model of macro-syntax, which defines periods in both syntactic and prosodic terms. In order to develop an automatic tool capable of detecting macro- syntactic periods, we cast this segmentation task as a sequence labelling problem and study the adequacy of machine-learning models using Conditional Random Fields (Lafferty et al. 2001) relying on lexical, syntactic and prosodic features. This work is a part of the SegCor1 project whose goal is to develop several tools for automatic segmentation of linguistic units, including periods which will be our focus. The result of one of these tools for the segmentation of chunks (Eshkol-Taravella et al., 2019) is used in this study. The article is presented as follows. Section 2 represents the related work of periods ; section 3 describes the corpus and goals of this study ; section 4 illustrates the main points of the manual annotation that serves as a reference to test the automatic methods; section 5 presents experiments of automatic annotations ; the results are analyzed in section 6 ; and section 7 proposes …

Content truncated to 500 words