If you’ve seen some, you’ve seen them all: Identifying variants of multiword expressions

Proceedings of the 27th International Conference on Computational Linguistics, pages 2582–2594 Santa Fe, New Mexico, USA, August 20-26, 2018. 2582 If you’ve seen some, you’ve seen them all: Identifying variants of multiword expressions Caroline Pasquer University of Tours France Agata Savary University of Tours France Carlos Ramisch Aix Marseille Univ, Universit´e de Toulon, CNRS, LIS, Marseille, France Jean-Yves Antoine University of Tours France first.last@(univ-tours|lis-lab).fr Abstract Multiword expressions, especially verbal ones (VMWEs), show idiosyncratic variability, which is challenging for NLP applications, hence the need for VMWE identification. We focus on the task of variant identification, i.e. identifying variants of previously seen VMWEs, whatever their surface form. We model the problem as a classification task. Syntactic subtrees with previously seen combinations of lemmas are first extracted, and then classified on the basis of features relevant to morpho-syntactic variation of VMWEs. Feature values are both absolute, i.e. hold for a particular VMWE candidate, and relative, i.e. based on comparing a candidate with previously seen VMWEs. This approach outperforms a baseline by 4 percent points of F-measure on a French corpus. Title and Abstract in French Trait pour trait identiques ? Identification de variantes d’expressions polylexicales Les expressions polylexicales (EP), et parmi elles plus particuli`erement les EP verbales (EPV), se caract´erisent par une grande variabilit´e idiosyncrasique de forme. La d´etection et l’identification de ces EPV vari´ees pose ainsi un r´eel d´efi `a la r´ealisation d’applications langagi`eres robustes. Cet article met l’accent sur la tˆache d’identification dans un corpus de variantes d’une EP verbale d´ej`a rencontr´ees. Il propose une strat´egie d’identification bas´ee sur l’extraction de formes candi- dates `a partir de patrons syntaxiques, suivie de leur classification bas´ee sur des caract´eristiques morphologiques et syntaxiques. Ces propri´et´es sont `a la fois absolues (c.-`a-d. concernent l’entit´e consid´er´ee) ou relatives (c.-`a-d. issues de la comparaison avec des EPV d´ej`a rencontr´ees). Les performances du syst`eme r´esultant ont ´et´e ´evalu´ees sur un corpus francophone. Elles montrent une am´elioration de 4 points de F-mesure par rapport `a une baseline bien ´etablie. 1 Introduction Multiword expressions (MWEs) such as red tape, by and large, to make a decision and to break one’s heart are combinations of words exhibiting unexpected lexical, morphological, syntactic, seman- tic, pragmatic and/or statistical behavior (Baldwin and Kim, 2010). Most prominently, they are semanti- cally non-compositional, that is, their meaning cannot be deduced from the meanings of their components and from their syntactic structure in a way deemed regular. For this reason, the presence of MWEs in texts calls for dedicated treatment, whose prerequisites include their automatic identification. The goal of MWE identification is, given some input running text, to identify all MWEs’ lexicalized components present in it (Schneider et al., 2016; Savary et al., 2017).1 Such systems face three main This work is licensed under a Creative Commons Attribution 4.0 International License. License details: http:// creativecommons.org/licenses/by/4.0/ 1A lexicalized component of a MWE, or component for short – marked in bold in examples – is one which is always realized by the same lexeme. For instance, the noun decision …

Content truncated to 500 words