Automatic Domain Assignment for Word Sense Alignment

Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 414–418, October 25-29, 2014, Doha, Qatar. c⃝2014 Association for Computational Linguistics Automatic Domain Assignment for Word Sense Alignment Tommaso Caselli TrentoRISE / Via Sommarive, 18 38123 Povo, Italy t.caselli@gmail.com Carlo Strapparava FBK / Via Sommarive, 18 38123 Povo, Italy strappa@fbk.eu Abstract This paper reports on the development of a hy- brid and simple method based on a machine learning classifier (Naive Bayes), Word Sense Disambiguation and rules, for the automatic assignment of WordNet Domains to nominal entries of a lexicographic dictionary, the Senso Comune De Mauro Lexicon. The system ob- tained an F1 score of 0.58, with a Precision of 0.70. We further used the automatically as- signed domains to filter out word sense align- ments between MultiWordNet and Senso Co- mune. This has led to an improvement in the quality of the sense alignments showing the validity of the approach for domain assign- ment and the importance of domain informa- tion for achieving good sense alignments. 1 Introduction and Problem Statement Lexical knowledge, i.e. how words are used and ex- press meaning, plays a key role in Natural Language Processing. Lexical knowledge is available in many different forms, ranging from unstructured terminolo- gies (i.e. word list), to full fledged computational lexica and ontologies (e.g. WordNet (Fellbaum, 1998)). The process of creation of lexical resources is costly both in terms of money and time. To overcome these lim- its, semi-automatic approaches have been developed (e.g. MultiWordNet (Pianta et al., 2002)) with differ- ent levels of success. Furthermore, important informa- tion is scattered in different resources and difficult to use. Semantic interoperability between resources could represent a viable solution to allow reusability and de- velop more robust and powerful resources. Word sense alignment (WSA) qualifies as the preliminary require- ment for achieving this goal (Matuschek and Gurevych, 2013). WSA aims at creating lists of pairs of senses from two, or more, (lexical-semantic) resources which de- note the same meaning. Different approaches to WSA have been proposed and they all share some common elements, namely: i.) the extensive use of sense de- scriptions of the words (e.g. WordNet glosses); and ii.) the extension of the basic sense descriptions with addi- tional information such as hypernyms, synonyms and domain or category labels. The purpose of this work is two folded: first, we exper- iment on the automatic assignment of domain labels to sense descriptions, and then, evaluate the impact of this information for improving an existing sense aligned dataset for nouns. Previous works has demonstrated that domain labels are a good feature for obtaining high quality alignments of entries (Navigli, 2006; Toral et al., 2009; Navigli and Ponzetto, 2012). The Word- Net (WN) Domains (Magnini and Cavaglia, 2000; Ben- tivogli et al., 2004) have been selected as reference do- main labels. We will use as candidate lexico-semantic resources to be aligned two Italian lexica, namely, Mul- tiWordNet (MWN) and the Senso Comune De Mauro Lexicon (SCDM) (Vetere et al., …

Content truncated to 500 words