Detection of Reformulations in Spoken French

Detection of Reformulations in Spoken French Natalia Grabar1, Iris Eshkol-Taravela2 1CNRS, UMR 8163, F-59000 Lille, France; Univ. Lille, UMR 8163 - STL - Savoirs Textes Langage, F-59000 Lille, France 2CNRS UMR 7270 LLL, Universit´e d’Orl´eans, 45100 Orl´eans, France, natalia.grabar@univ-lille3.fr, iris.eshkol@univ-orleans.fr Abstract Our work addresses automatic detection of enunciations and segments with reformulations in French spoken corpora. The proposed approach is syntagmatic. It is based on reformulation markers and specificities of spoken language. The reference data are built manually and have gone through consensus. Automatic methods, based on rules and CRF machine learning, are proposed in order to detect the enunciations and segments that contain reformulations. With the CRF models, different features are exploited within a window of various sizes. Detection of enunciations with reformulations shows up to 0.66 precision. The tests performed for the detection of reformulated segments indicate that the task remains difficult. The best average performance values reach up to 0.65 F-measure, 0.75 precision, and 0.63 recall. We have several perspectives to this work for improving the detection of reformulated segments and for studying the data from other points of view. Keywords: Spoken Corpora, Reformulation, Reformulation Marker, Paraphrase, Supervised Machine Learning 1. Introduction Reformulations may occur in written and spoken lan- guages, in which they show different functions (Flottum, 1995; Rossari, 1992): in spoken language, they mark the elaboration of ideas, and are punctuated by hesita- tions, false starts, and repetitions (Blanche-Benveniste et al., 1991), in written documents, we usually find the fi- nal result of the reformulation process (Hag`ege, 1985). It is considered that reformulation is the activity of speakers built on their own linguistic production or on the one of their interlocutor, with or without specific markers. The objective is then to modify some aspects (lexical, syntactic, semantic, pragmatic) but to keep the semantic content con- stant (G¨ulich and Kotschi, 1987; Kanaan, 2011). Specific reformulation markers may provide the formal mark-up of reformulations. Reformulation is closely related to paraphrases, in that way that reformulated sequences can produce the paraphrases (Neveu, 2004). Reformulation and paraphrase play an im- portant role in languages: • When studying languages, a common exercise con- sists of paraphrasing expressions in order to control their understanding by students; • In the same way, it is possible to control the under- standing of ideas. The first exercises of the kind have appeared with the exegesis of ancient texts: sacred texts (Bible, Koran, Torah) first, and then theological, philosophical and scientific texts; • More naturally, speakers use the reformulation and paraphrase in order to precise and to better transmit their thoughts. It is also common to find reformula- tions in written language: between various versions of the same literary piece of work (Fuchs, 1982), of the Wikipedia articles (Vila et al., 2014), or of scientific articles. The authors can thus rewrite several times their text until they produce the one that suits them at last. Reformulation and paraphrase also play an important role in different NLP applications (Madnani and Dorr, 2010; Androutsopoulos and Malakasiotis, 2010; …

Content truncated to 500 words