A Framework for Cross-lingual/Node-wise Alignment of Lexical-Semantic Resources

A Framework for Cross-lingual/Node-wise Alignment of Lexical-Semantic Resources Yoshihiko Hayashi Faculty of Science and Engineering, Waseda University 2-4-12 Ohukubo, Shinjuku, Tokyo, 169-0072 Japan yshk.hayashi@aoni.waseda.jp Abstract Given lexical-semantic resources in different languages, it is useful to establish cross-lingual correspondences, preferably with semantic relation labels, between the concept nodes in these resources. This paper presents a framework for enabling a cross-lingual/node- wise alignment of lexical-semantic resources, where cross-lingual correspondence candidates are first discovered and ranked, and then classified by a succeeding module. Indeed, we propose that a two-tier classifier configuration is feasible for the second module: the first classifier filters out possibly irrelevant correspondence candidates and the second classifier assigns a relatively fine-grained semantic relation label to each of the surviving candidates. The results of Japanese-to-English alignment experiments using EDR Electronic Dictionary and Princeton WordNet are described to exemplify the validity of the proposal. Keywords: lexical-semantic resource, cross-lingual alignment, relation classification 1. Introduction Given the current situation where a variety of language re- sources have been actively developed and disseminated, in- tegrating these independently developed resources, prefer- ably in a standardized format, has become an important is- sue (Cimiano et al., 2013). In particular, aligning lexical- semantic resources (henceforth, LSR) across languages in the word-sense/lexical-concept level is highly demanded for realizing cross-language semantic applications. There are a few research projects in line with this re- search direction: BabelNet (Navigli and Ponzetto, 2012) maintains a wide-coverage multilingual semantic network, whose construction relies on mapping of Wikipedia (a mul- tilingual encyclopedia system, in a sense) entries to the cor- responding WordNet (Miller and Fellbaum, 2007) (an En- glish semantic lexicon) senses; Uby (Gurevych et al., 2012) combines a variety of lexical resources for English and Ger- man also in a word-sense level by applying a machine- learned classifier. Like these well-known projects, our primary research goal is to establish a computational method to discover cross- lingual correspondences in a word-sense/lexical-concept level, given LSRs in different languages. However we further explore a methodology for classifying discovered cross-lingual correspondences with a broader range of se- mantic relation types, not limited to synonymy. To pursue this research direction, we apply a machine-learned classi- fier as in Uby (Gurevych et al., 2012), but with a variety of semantic relation types, including near-synonymy, hyper- nymy, verb-argument, and so on. This paper presents a whole system architecture for en- abling node-wise alignment of LSRs in different lan- guages, where potential cross-lingual correspondences are first discovered and ranked by applying an existing method (Hayashi, 2013), and then classified by a succeed- ing module. We particularly propose that a two-tiered clas- sifiers configuration could be feasible for the second mod- ule: the first classifier filters out possibly irrelevant corre- spondences, and then the second classifier assigns a rel- atively fine-grained semantic relation label to each of the surviving correspondences. To exemplify the proposal, we conducted a series of Japanese-to-English alignment experiments that employ EDR Electronic Dictionary (EDR; for Japanese)1 (Yokoi, 1995) and Princeton WordNet (PWN; for English). EDR actually is a dictionary …

Content truncated to 500 words