4Couv: A New Treebank for French

4-Couv: A New Treebank for French Philippe Blache, Gr´egoire Montcheuil, Laurent Pr´evot, St´ephane Rauzy Laboratoire Parole et Langage - UMR 7309 5 avenue Pasteur, 13100 Aix-en-Provence, France firstname.lastname@lpl-aix.fr Abstract The question of the type of text used as primary data in treebanks is of certain importance. First, it has an influence at the discourse level: an article is not organized in the same way as a novel or a technical document. Moreover, it also has consequences in terms of semantic interpretation: some types of texts can be easier to interpret than others. We present in this paper a new type of treebank which presents the particularity to answer to specific needs of experimental linguistic. It is made of short texts (book backcovers) that presents a strong coherence in their organization and can be rapidly interpreted. This type of text is adapted to short reading sessions, making it easy to acquire physiological data (e.g. eye movement, electroencepholagraphy). Such a resource offers reliable data when looking for correlations between computational models and human language processing. Keywords: Treebank, French, Constituency, Treebanking Tools, Experimental Linguistic 1. Introduction Several treebanks already exist for French. The most pop- ular one is the French Treebank (Abeill´e et al., 2003, FTB) and its different evolutions or enrichments (Schluter and van Genabith, 2007; Candito et al., 2010). Other French de- pendency treebanks are also available, for example through the Sequoia project (Candito and Seddah, 2012; Candito et al., 2014) or the Universal Dependencies project (Nivre et al., 2015). However, and this constitutes the first motivation for developing a new treebank, it still remains necessary to develop other resources in order to increase the size and the variety of available material. A second and even more important reason to build specific treebanks is the type of application it can be used for. More precisely, classical tree- banks are usually made of newspaper articles, rather long, and which interest when reading can be poor. This can con- stitute a drawback (or even a bias) when using such texts in human experimentation (in particular reading). In such cases, short, semantically coherent and self-contained texts are preferable for the acquisition of physiological responses from reading subjects. The solution we propose in this perspective consists in building a corpus of book backcovers1 , that are short texts, easily available and frequently used. We present in this pa- per the corpus2, its annotation and first results. 2. Corpus description Backcovers are short texts of different genres: extract, syn- opsis, genesis of the book, comment about the work, or a combination of them. Each text is semantically self- contained, and interesting (minimizing attention and com- prehension drops). A text contains between 4-10 sentences and 80-200 tokens (80% having at most 30 tokens). We have build such a corpus, which is still under evolution. At this stage, it consists in 8,000 texts from different pub- lishers (from Pocket and Gallimard publishers, with their agreement). These texts have been selected manually by 1The French appellation of a backcover …

Content truncated to 500 words