Anchoring Floating Quantifiers in Japanese-to-English Machine Translation

Anchoring Floating Quantifiers in Japanese-to-English Machine Translation Francis Bond, t Daniela Kurz t and Satoshi Shirai t t NTT Communication Science Laboratories 2-4 Hikari-dai, Seika-cho, Soraku-gun, Kyoto, Japan, 619-0237 {bond, shirai}@cslab, kecl. ntt. co. jp Department of" Computational Linguistics, University of the Saarland Postfach 1150, D-66041 Saarbriicken, Germany kurzOcoli .uni-sb.de Abstract In this paper we present an algorithm to an- chor floating quantifiers in Japanese, a language in which quantificational nouns and numeral- classifier combinations can appear separated from the noun phrase they quantify. The algo- rithm differentiates degree and event modifiers from nouns that quantify noun phrases. It then finds a suitable anchor for such floating quan- tifiers. To do this, the algorithm considers the part of speech of the quantifier and the target, the semantic relation between them, the case marker of the antecedent and the meaning of the verb that governs the two constituents. The al- gorithm has been implemented and tested in a rule-based Japanese-to-English machine trans- lation system, with an accuracy of 76% and a recall of 97%. 1 Introduction One interesting phenomenon in Japanese is the fact that quantifiers can appear in two main po- sitions, as pre-modifier in a noun phrase (1), or 'floating' as adjuncts to the verb phrase, typi- cally in pre-verbal position (2). 1'2 (1) watashi-wa 3-ko-no k~ki-w__p_o tabeta I-TOP 3-CL-ADN cake-Ace ate I ate three cakes (2) watashi-wa kgki-wo 3-ko tabeta I-TOP cake-ACe 3-CL ate I ate three cakes Quantifier 'float' of numeral-classifier combi- nations is widely discussed in the linguistic liter- 1Quantifiers are shown in bold, the noun phrases they quantify are underlined. 2This phenomenon exists in other languages, such as Korean. We will, however, restrict our discussion to Japanese in this paper. ature. 3 Much of the discussion focuses on iden- tifying the conditions under which a quantifier can appear in the adjunct position. The expla- nations range from configurational (Inoue, 1983; Miyagawa, 1989) to discourse based (Downing, 1996; Alam, 1997), we shall discuss these fur- ther below. There has been almost no discus- sion of other floating quantifiers, such as quan- tiflcational nouns. We call the process of identifying the noun phrase being quantified by a floating quanti- fier 'anchoring' the quantifier. The necessity of anchoring floating quantifiers for many nat- ural language processing tasks is widely recog- nized (Asahioka et al., 1990; Bond et al., 1996), and is important not only for machine transla- tion but for the interpretation of Japanese in general. However, although there are several NLP systems that incorporate some solution to the problem of floating quantifiers, to the best of our knowledge, no algorithm for anchoring floating quantifiers has been given. We propose such an algorithm in this paper. The algorithm uses information about case-marking, sentence structure, part-of-speech, noun and verb mean- ing. The algorithm has been implemented and tested within the Japanese-to-English machine translation system ALT-J/E (Ikehara et al., 1991). The next section describes the phenomenon of quantifier float in more detail. We then pro- pose our algorithm to identify …

Content truncated to 500 words