ODIL_Syntax: a Free Spontaneous Spoken French Treebank Annotated with Constituent Trees

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 5301–5307 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 5301 ODIL Syntax : a Free Spontaneous Spoken French Treebank Annotated with Constituent Trees Ilaine Wang1,2, Aurore Pelletier2, Jean-Yves Antoine1, Ana¨ıs Halftermeyer2 1LIFAT, Universit´e de Tours, France 2LIFO, Universit´e d’Orl´eans, France {ilaine.wang, anais.halftermeyer}@univ-orleans.fr, jean-yves.antoine@univ-tours.fr Abstract This paper describes ODIL Syntax, a French treebank built on spontaneous speech transcripts. The syntactic structure of every speech turn is represented by constituent trees, through a procedure which combines an automatic annotation provided by a parser (here, the Stanford Parser) and a manual revision. ODIL Syntax respects the annotation scheme designed for the French TreeBank (FTB), with the addition of some annotation guidelines that aims at representing specific features of the spoken language such as speech disfluencies. The corpus will be freely distributed by January 2020 under a Creative Commons licence. It will ground a further semantic enrichment dedicated to the representation of temporal entities and temporal relations, as a second phase of the ODIL@Temporal project. The paper details the annotation scheme we followed with a emphasis on the representation of speech disfluencies. We then present the annotation procedure that was carried out on the Contemplata annotation platform. In the last section, we provide some distributional characteristics of the annotated corpus (POS distribution, multiword expressions). Keywords: treebank, syntactic annotation, spontaneous speech, French language, constituent trees 1. Introduction The development of large treebanks has provided use- ful training data for the development of robust and large coverage parsers. While the first treebanks relied on a constituency-based representation of syntactic structures, dependency treebanks have met for more than one decade an irrepressible surge of interest, due to a conjunction of practical and theoretical reasons: • efficient dependency parsing algorithms (as well as the related machine learning techniques) have been pro- posed, leading to state-of-the-art performances; • dependency trees are closer to predicate-argument structures than constituent trees, which prepares in an easier way further semantic-oriented tasks such as in- formation extraction, question answering, etc.; • dependency structures are assumed to fit better the rep- resentation needs of free word order languages, while constituency-based structures must struggle on such languages with serious discontinuity problems. This explains that current treebanks are based on a depen- dency representation, as shown for instance by the Univer- sal Dependencies initiative1 or the CoNLL shared tasks2 from 2006 to 2009. Following a different path, this pa- per presents a constituency treebank: ODIL Syntax. This seemingly odd choice is motivated by specific annotation needs. Indeed, ODIL Syntax constitutes the first annotation layer of a larger ressource, ODIL Temp, which describes 1UD: https://universaldependencies.org/ 2CoNLL shared tasks: https://www.conll.org/ previous-tasks all the temporal entities and the temporal relations that can be found in the ODIL corpus. The temporal annotation of ODIL Temp relies on an extension of the TimeML standard (ISO, 2012). The main originality of this extended anno- tation scheme is to delimit temporal mentions not by their minimal …

Content truncated to 500 words