Werdy: Recognition and Disambiguation of Verbs and Verb Phrases with Syntactic and Semantic Pruning

Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 374–385, October 25-29, 2014, Doha, Qatar. c⃝2014 Association for Computational Linguistics Werdy: Recognition and Disambiguation of Verbs and Verb Phrases with Syntactic and Semantic Pruning Luciano Del Corro Rainer Gemulla Max-Planck-Institut f¨ur Informatik Saarbr¨ucken, Germany { delcorro, rgemulla, weikum }@mpi-inf.mpg.de Gerhard Weikum Abstract Word-sense recognition and disambigua- tion (WERD) is the task of identifying word phrases and their senses in natural language text. Though it is well under- stood how to disambiguate noun phrases, this task is much less studied for verbs and verbal phrases. We present Werdy, a framework for WERD with particular focus on verbs and verbal phrases. Our framework first identifies multi-word ex- pressions based on the syntactic structure of the sentence; this allows us to recog- nize both contiguous and non-contiguous phrases. We then generate a list of can- didate senses for each word or phrase, us- ing novel syntactic and semantic pruning techniques. We also construct and lever- age a new resource of pairs of senses for verbs and their object arguments. Finally, we feed the so-obtained candidate senses into standard word-sense disambiguation (WSD) methods, and boost their precision and recall. Our experiments indicate that Werdy significantly increases the perfor- mance of existing WSD methods. 1 Introduction Understanding the semantics of words and multi- word expressions in natural language text is an important task for automatic knowledge acquisi- tion. It serves as a fundamental building block in a wide area of applications, including semantic parsing, question answering, paraphrasing, knowl- edge base construction, etc. In this paper, we study the task of word-sense recognition and dis- ambiguation (WERD) with a focus on verbs and verbal phrases. Verbs are the central element in a sentence, and the key to understand the relations between sets of entities expressed in a sentence. We propose Werdy, a method to (i) automati- cally recognize in natural language text both sin- gle words and multi-word phrases that match en- tries in a lexical knowledge base (KB) like Word- Net (Fellbaum, 1998), and (ii) disambiguate these words or phrases by identifying their senses in the KB. WordNet is a comprehensive lexical resource for word-sense disambiguation (WSD), covering nouns, verbs, adjectives, adverbs, and many multi- word expressions. In the following, the notion of an entry refers to a word or phrase in the KB, whereas a sense denotes the lexical synset of the entry’s meaning in the given sentence. A key challenge for recognizing KB entries in natural language text is that entries often consist of multiple words. In WordNet-3.0 more than 40% of the entries are multi-word. Such entries are challenging to recognize accurately for two main reasons: First, the components of multi-word en- tries in the KB (such as fiscal year) often consist of components that are themselves KB entries (fis- cal and year). Second, multi-word entries (such as take a breath) may not appear consecutively in a sentence (“He takes a deep breath.”). Werdy addresses the latter …

Content truncated to 500 words