Contemplata, a Free Platform for Constituency Treebank Annotation

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 7222–7229 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 7222 Contemplata, a Free Platform for Constituency Treebank Annotation Jakub Waszczuk1, Ilaine Wang2,3, Jean-Yves Antoine2, Ana¨ıs Halftermeyer3 1Heinrich-Heine-Universit¨at, D¨usseldorf, Germany 2LIFAT, Universit´e de Tours, France 3LIFO, Universit´e d’Orl´eans, France waszczuk@hhu.de, jean-yves.antoine@univ-tours.fr, {ilaine.wang, anais.halftermeyer}@univ-orleans.fr Abstract This paper describes Contemplata, an annotation platform that offers a generic solution for treebank building as well as treebank enrichment with relations between syntactic nodes. Contemplata is dedicated to the annotation of constituency trees. The framework includes support for syntactic parsers, which provide automatic annotations to be manually revised. The balanced strategy of annotation between automatic parsing and manual revision allows to reduce the annotator workload, which favours data reliability. The paper presents the software architecture of Contemplata, describes its practical use and eventually gives two examples of annotation projects that were conducted on the platform. Keywords: treebank, syntactic annotation, spontaneous speech, French language, constituent trees 1. Introduction This paper presents Contemplata, an annotation platform that was originally created to fulfil the needs of temporal annotation while providing a generic tool for the annotation of constituency treebanks. The representation and the processing of temporality is im- portant for many understanding NLP tasks. Temporal an- notation has benefitted from the normalisation efforts of the ISO TC37/SC4 committee which has led to the definition of the ISO-TimeML standard (ISO, 2012). While originally developed for English, ISO-TimeML has been applied on a large variety of languages (Italian, Korean, Romanian, Chi- nese, French...) with only slight idiomatic adaptations. The Temporal@ODIL project aims at building a new French corpus annotated with both temporal mentions and tempo- ral relations. The primary motivation behind the project is to focus specifically on spontaneous speech, unlike the French Time Bank (Bittar et al., 2011) which targets writ- ten text. This leads Temporal@ODIL to review the ISO- TimeML standard and propose enhancements, while pre- serving coherence and upward compatibility with the in- ternational standard (Lefeuvre-Halftermeyer et al., 2016). In that respect, the project meets a larger variety of needs in NLP and in corpus linguistics, in addition to offering a coverage of spoken language for temporality annotation. What makes the originality of this proposal is that temporal mentions are delimited not by their minimal chunk (lex- ical head) but by the range of the syntactic subtree that covers the temporal mention. This large-span annotation challenges the annotator’s ability to delimit the temporal mentions with a satisfactory reliability. To ease this de- limitation, we adopt a solution that was investigated for multi-word expressions in the Prague Dependency Tree- bank (Bejˇcek and Straˇn´ak, 2010): temporal mentions are directly defined on the syntactic structures of the utterances. Thus, the delimitation task boils down to the selection of one specific node. Data reliability is therefore favoured by the reduction of the annotator’s cognitive load. In addition, prior knowledge of the syntactic trees also eases the annota- tion of the semantic relations. For instance, …

Content truncated to 500 words