Pos-Patterns or Syntax? Comparing Methods for Extracting Word Combinations

This paper reports on work carried out in the framework of an ongoing project aimed at building an online, corpus-based lexicographic resource for Italian Word Combinations. Our aim is to compare two of the most commonly used methods for the automatic extraction of word combinations from corpora, with a view to evaluate their performance – and ultimately their efficacy – with respect to the task of acquiring word combinations for inclusion in the lexicographic combinatory resource. 1. WORD COMBINATIONS: LEXICOGRAPHY AND NLP It is widely acknowledged that lexicographers‘ introspection alone cannot provide comprehensive information about word meaning and usage, and that investigation of language in use is fundamental for any reliable lexicographic work (Atkins and Rundell 2008). This is even more true for dictionaries that record the combinatorial behaviour of words, where the lexicographic task is to detect the typical combinations a word participates in. In fact, it was much harder to study lexical combinatorics empirically before the advent of large corpora and the definition of statistical techniques for the analysis of word associations (Hanks 2012). This paper reports on work carried out in the framework of an ongoing project called CombiNet aimed at building an online, corpus-based lexicographic resource for Italian Word Combinations. We use the term Word Combinations (WoCs) to encompass both Multiword Expressions (MWEs) – namely WoCs characterised by different degrees of fixedness and idiomaticity that act as a single unit at some level of linguistic analysis, such as idioms, phrasal lexemes, collocations, preferred combinations (Calzolari et al. 2002, Sag et al. 31 PRIN Project 2010-2011 Word Combinations in Italian (n. 20105B3HE8) funded by the Italian Ministry of Education, University and Research (MIUR). URL: http://combinet.humnet.unipi.it.