Picking them up and Figuring them out: Verb-Particle Constructions, Noise and Idiomaticity

CoNLL 2008: Proceedings of the 12th Conference on Computational Natural Language Learning, pages 49–56 Manchester, August 2008 Picking them up and Figuring them out: Verb-Particle Constructions, Noise and Idiomaticity Carlos Ramisch♣♦, Aline Villavicencio♣♠, Leonardo Moura♣ and Marco Idiart♥ ♣Institute of Informatics, Federal University of Rio Grande do Sul (Brazil) ♦GETALP Laboratory, Joseph Fourier University - Grenoble INP (France) ♠Department of Computer Sciences, Bath University (UK) ♥Institute of Physics, Federal University of Rio Grande do Sul (Brazil) {ceramisch,avillavicencio,lfsmoura}@inf.ufrgs.br, idiart@if.ufrgs.br Abstract This paper investigates, in a first stage, some methods for the automatic acquisi- tion of verb-particle constructions (VPCs) taking into account their statistical prop- erties and some regular patterns found in productive combinations of verbs and par- ticles. Given the limited coverage pro- vided by lexical resources, such as dictio- naries, and the constantly growing number of VPCs, possible ways of automatically identifying them are crucial for any NLP task that requires some degree of semantic interpretation. In a second stage we also study whether the combination of statis- tical and linguistic properties can provide some indication of the degree of idiomatic- ity of a given VPC. The results obtained show that such combination can success- fully be used to detect VPCs and distin- guish idiomatic from compositional cases. 1 Introduction Considerable investigative effort has focused on the automatic identification of Multiword Expres- sions (MWEs), like compound nouns (science fic- tion) and phrasal verbs (carry out) (e.g. Pearce (2002), Evert and Krenn (2005) and Zhang et al. (2006)). Some of them employ language and/or type dependent linguistic knowledge for the task, while others employ independent statis- tical methods, such as Mutual Information and Log-likelihood (e.g. Pearce (2002) and, Zhang et al. (2006)), or even a combination of them (e.g. c⃝ 2008. Licensed under the Creative Commons Attribution-Noncommercial-Share Alike 3.0 Unported li- cense (http://creativecommons.org/licenses/by-nc-sa/3.0/). Some rights reserved. Baldwin (2005) and Sharoff (2004)), as basis for helping to determine whether a given sequence of words is in fact an MWE. Although some re- search aims at developing methods for dealing with MWEs in general (e.g. Zhang et al. (2006), Ramisch et al. (2008)), there is also some work that deals with specific types of MWEs (e.g. Pearce (2002) on collocations and Villavicencio (2005) on verb-particle constructions (VPCs)) as each of these MWE types has distinct distributional and linguistic characteristics. VPCs are combinations of verbs and particles, such as take off in Our plane took off late, that due to their complex characteristics and flexible na- ture, provide a real challenge for NLP. In particu- lar, there is a lack of adequate resources to identify and treat them, and those that are available provide only limited coverage, in face of the huge number of combinations in use. For tasks like parsing and generation, it is essential to know whether a given VPC is possible or not, to avoid for example us- ing combinations that sound unnatural or ungram- matical to native speakers (e.g. give/lend/?grant out for the conveying of something to someone or some …

Content truncated to 500 words