Unsupervised Acquisition of Comprehensive Multiword Lexicons using Competition in an n-gram Lattice

Unsupervised Acquisition of Comprehensive Multiword Lexicons using Competition in an n-gram Lattice Julian Brooke∗ Jan ˇSnajder† Timothy Baldwin∗ ∗ School of Computing and Information Systems, The University of Melbourne julian.brooke@gmail.com, tb@ldwin.net † Faculty of Electrical Engineering and Computing, University of Zagreb jan.snajder@fer.hr Abstract We present a new model for acquiring com- prehensive multiword lexicons from large cor- pora based on competition among n-gram candidates. In contrast to the standard ap- proach of simple ranking by association mea- sure, in our model n-grams are arranged in a lattice structure based on subsumption and overlap relationships, with nodes inhibiting other nodes in their vicinity when they are se- lected as a lexical item. We show how the con- figuration of such a lattice can be optimized tractably, and demonstrate using annotations of sampled n-grams that our method consis- tently outperforms alternatives by at least 0.05 F-score across several corpora and languages. 1 Introduction Despite over 25 years of research in computational linguistics aimed at acquiring multiword lexicons using corpora statistics, and growing evidence that speakers process language primarily in terms of memorized sequences (Wray, 2008), the individual word nonetheless stubbornly remains the de facto standard processing unit for most research in mod- ern NLP. The potential of multiword knowledge to improve both the automatic processing of language as well as offer new understanding of human acqui- sition and usage of language is the primary moti- vator of this work. Here, we present an effective, expandable, and tractable new approach to compre- hensive multiword lexicon acquisition. Our aim is to find a middle ground between standard MWE ac- quisition approaches based on association measures (Ramisch, 2014) and more sophisticated statistical models (Newman et al., 2012) that do not scale to large corpora, the main source of the distributional information in modern NLP systems. A central challenge in building comprehensive multiword lexicons is paring down the huge space of possibilities without imposing restrictions which disregard a major portion of the multiword vocab- ulary of a language: allowing for diversity creates significant redundancy among statistically promis- ing candidates. The lattice model proposed here ad- dresses this primarily by having the candidates— contiguous and non-contiguous n-gram types— compete with each other based on subsumption and overlap relations to be selected as the best (i.e., most parsimonious) explanation for statistical irregulari- ties. We test this approach across four large corpora in three languages, including two relatively free- word-order languages (Croatian and Japanese), and find that this approach consistency outperforms al- ternatives, offering scalability and many avenues for future enhancement. 2 Background and Related Work In this paper we will refer to the targets of our lexi- con creation efforts as formulaic sequences, follow- ing the terminology of Wray (2002; 2008), wherein a formulaic sequence (FS) is defined as “a sequence, continuous or discontinuous, of words or other ele- ments, which is, or appears to be, prefabricated: that is, stored and retrieved whole from memory at the time of use, rather than being subject to generation or analysis …

Content truncated to 500 words