WASP-Bench: an MT lexicographers’ workstation supporting state-of-the-art lexical disambiguation

WASP-Bench: an MT Lexicographers' Workstation Supporting State-of-the-art Lexical Disambiguation Adam Kilgarriff, David Tugwell Information Technology Research Institute (ITRI), University of Brighton Lewes Road, Brighton, UK {adam.kilgarriff, david.tugwell}@itri.brighton.ac.uk Abstract Most MT lexicography is devoted to developing rules of the kind, “in context C, translate source-language word S as target-language word T”. Very many such rules are required, producing them is laborious, and MT companies standardly spend large sums on it. We present the WASP-Bench, a lexicographer's workstation for the rapid and semi-automatic development of such rule-sets. The WASP- Bench makes use of a large source-language corpus and state-of-the-art techniques for Word Sense Disambiguation. We show that the WSD accuracy is on a par with the best results published to date, with the advantage that the WASP-Bench, unlike other high- performance systems, does not require a sense-disambiguated training corpus as input. The WASP-Bench is designed to fit readily with MT companies' working practices, as it may be used for as many or as few source language words as present disambiguation problems for a given target. Keywords MT-lexicography, Lexical Disambiguation, workbench. Introduction Choosing the right target language word is the central problem of MT. The difficulty usually lies in the fact that the source language word has different meanings, which translate differently, or that it translates differently in different contexts. Most MT lexicons have extensive sets of rules which say “for source-word s, if the context is c, translate as target-word t”. The problem with these rules is that they are laborious to write, and you need a lot of them. In this paper we present a corpus-based lexicographers' workstation, in which this process is automatically supported. The software finds, from a large source-language corpus, a candidate set of contexts C which appear salient for a source word, so are good candidates for the left hand side of context is c ⇒ translate as t rules. The lexicographer provides the appropriate value of t for a few of these, in a process that takes just a few minutes per source word. The software then applies the highest-performing Word Sense Disambiguation (WSD) algorithm currently available (Yarowsky (1995)) to bootstrap a long list of rules from the ‘seeds’ the lexicographer has provided. The list is ordered by confidence: the rules at the top of the list are the ones we can be surest of. Thus to disambiguate, and thereby select the appropriate target word t, for an instance of s in a text to be translated, all an MT system need do is work through the list until it finds a rule where the left hand side matches. It can then select the target- language word that the rule specifies. The WASP-Bench is designed to fit readily with MT companies' working practices. It can be used for as many or as few source language words as present disambiguation problems for a given target, making it suitable for lexicon maintenance as well as development. The rule-sets, or “word sense profiles”, comprise simple, surface-context-based rules and should …

Content truncated to 500 words