Establishing a New State-of-the-Art for French Named Entity Recognition

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 4631–4638 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 4631 Establishing a New State-of-the-Art for French Named Entity Recognition Pedro Javier Ortiz Suárez1,2, Yoann Dupont1,2, Benjamin Muller1,2, Laurent Romary1, Benoît Sagot1 Inria1, Sorbonne Université2. 2 rue Simone Iff, 75012 Paris1, 21 rue de l’École de médecine, 75006 Paris2. {pedro.ortiz, benjamin.muller, laurent.romary, benoit.sagot}@inria.fr yoa.dupont@gmail.com Abstract The French TreeBank developed at the University Paris 7 is the main source of morphosyntactic and syntactic annotations for French. However, it does not include explicit information related to named entities, which are among the most useful information for several natural language processing tasks and applications. Moreover, no large-scale French corpus with named entity annotations contain referential information, which complement the type and the span of each mention with an indication of the entity it refers to. We have manually annotated the French TreeBank with such information, after an automatic pre-annotation step. We sketch the underlying annotation guidelines and we provide a few figures about the resulting annotations. Keywords: Named Entity Recognition, French, Language Modeling 1. Introduction Named entity recognition (NER) is the widely studied task consisting in identifying text spans that denote named en- tities such as person, location and organization names, to name the most important types. Such text spans are called named entity mentions. In NER, mentions are generally not only identified, but also classified according to a more or less fine-grained ontology, thereby allowing for instance to distinguish between the telecommunication company Or- ange and the town Orange in southern France (amongst others). Importantly, it has long been recognised that the type of named entities can be defined in two ways, which underlies the notion of metonymy: the intrinsic type (France is always a location) and the contextual type (in la France a signé un traité ‘France signed a treaty’, France denotes an organization). NER has been an important task in natural language pro- cessing for quite some time. It was already the focus of the MUC conferences and associated shared tasks (Marsh and Perzanowski, 1998), and later that of the CoNLL 2003 and ACE shared tasks (Tjong Kim Sang and De Meulder, 2003; Doddington et al., 2004). Traditionally, as for in- stance was the case for the MUC shared tasks, only per- son names, location names, organization names, and some- times “other proper names” are considered. However, the notion of named entity mention is sometimes extended to cover any text span that does not follow the general gram- mar of the language at hand, but a type- and often culture- specific grammar, thereby including entities ranging from product and brand names to dates and from URLs to mon- etary amounts and other types of numbers. As for many other tasks, NER was first addressed using rule-based approaches, followed by statistical and now neu- ral machine learning techniques (see Section 3.1. for a brief discussion on NER approaches). Of course, evalu- ating NER systems …

Content truncated to 500 words