FrSemCor: Annotating a French Corpus with Supersenses

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 5912–5918 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 5912 FrSemCor: Annotating a French corpus with supersenses L. Barque1, P. Haas2, R. Huyghe3, D. Tribout4, M. Candito5, B. Crabb´e5, V. Segonne5 1 Universit´e Paris 13 & LLF, 2 Universit´e Paris 13 & Lattice, 3 Universit´e de Fribourg 4 Universit´e de Lille & STL, 5 Universit´e de Paris & LLF {lucie.barque, pauline.haas}@univ-paris13.fr, richard.huyghe@unifr.ch delphine.tribout@univ-lille.fr, {marie.candito, vincent.segonne}@linguist.univ-paris-diderot.fr Abstract French, as many languages, lacks semantically annotated corpus data. Our aim is to provide the linguistic and NLP research communities with a gold standard sense-annotated corpus of French, using WordNet Unique Beginners as semantic tags, thus allowing for interoperability. In this paper, we report on the first phase of the project, which focused on the annotation of common nouns. The resulting dataset consists of more than 12,000 French noun tokens which were annotated in double blind and adjudicated according to a carefully redefined set of supersenses. The resource is released online under a Creative Commons Licence. Keywords: semantic annotation, supersenses, corpus 1. Introduction This paper presents the construction of a French corpus whose nouns are manually annotated with semantic classes, hereafter “supersenses”. The project aims at compensating for the lack of semantically annotated data that can be used in both NLP and linguistics research. Our goal is to provide a gold standard resource, based on a careful semantic an- notation designed and supervised by lexical semantics ex- perts. In this paper, we focus on the annotation of more than 12,000 occurrences of common nouns in a French treebank corpus. Labels used in semantically annotated corpora are of two types: they can either be linked to sets of predefined word senses, as in a dictionary, or correspond to bare seman- tic classes that are not pre-associated to a given lexicon. In Semcor (Landes et al., 1998), a well-known sense an- notated corpus of English, occurrences of a word are as- sociated with a predefined list of senses (represented by synsets) from the Princeton WordNet (Miller et al., 1990). WordNet being known to include very fine-grained lexical distinctions, its hierarchical organization has been seen as different levels of semantic granularity. The top level of WordNet ontology, namely the “Unique Beginners” (UBs) have been taken as coarse-grained categories for both man- ual and automatic annotation, which has been proved to limit the difficulty of making fine-grained sense distinctions (Palmer et al., 2007; Navigli, 2006; Navigli, 2009). First used by Ciaramita and Johnson (2003) and Ciaramita and Altun (2006) as ”supersenses”, WordNet UBs are adopted in a growing number of studies devoted to coarse seman- tic tagging (Johannsen et al., 2014; Flekova and Gurevych, 2016). For semantic annotation projects based on WordNet UBs1, the corpus may or may not be pre-tagged with UBs, de- pending on whether or not a WordNet lexicon is available with sufficient coverage for the target language. Pretagging was used for instance for Danish in the …

Content truncated to 500 words