French MultiWord Expressions representation and parsing

old_uid17254
titleFrench MultiWord Expressions representation and parsing
start_date2019/01/25
schedule11h
onlineno
location_infosalle de Reunion C334
summaryMany NLP tasks, such as natural language understanding, require a representation of syntax and semantics in texts. MultiWord Expressions (MWEs), which can be described as a set of (not necessarily contiguous) tokens that exhibit some idiosyncratic properties (Baldwin and Kim, 2010), to quote Sag et al. 2001 are "a pain in the neck for NLP" .  MWEs are difficult to predict as their syntactic behavior tends to be unpredictable: they can have an irregular internal syntax and a non-compositional meaning. MWEs-aware NLP systems are also hard to evaluate, because until recently and the PARSEME COST initiative (Savary et al, 2017) there were only few annotated corpora annotated with MWEs (Laporte et al. 2008). I will first present my previous works on named entity recognition (Dupont et al, 2017), showing how they are related to MWEs, before delving deeper into MWEs. I will present in more details how they are a challenge, and how we can represent them using metagrammars (Savary et al., 2018), more precisely within the FRMG framework of de la Clergerie, (2010).
responsiblesSeddah