Generation for Dialogue Translation Using Typed Feature Structure Unification

Generation for Dialogue Translation Using Typed Feature Structure Unification Yoshihiro UEDA ATR Interpreting Telephony Research Labs. Sampeidani, Inuidani, Seika-cho Kyoto 619-02, Japan ryu%atr-la.atr.co.jp@u unet. u u. net Kiyoshi KOGURE NTT Basic Research Laboratories 9-11, Midori-cho 3-chome, Musashino-shi Tokyo 180, Japan kogure%atom.ntt.jp@ uunet.uu.net Abstract This article introduces a bidirectional grammar generation system called feature structure-directed generation, developed for a dialogue translation system. The system utilizes typed feature structures to control the top-down derivation in a declarative way. This generation system also uses disjunctive feature structures to reduce the number of copies of the derivation tree. The grammar for this generator is designed to properly generate the speaker's intention in a telephone dialogue. 1. Introduction It is important for the generation part of the dialogue translation system to reflect the speaker's intention (illocutionary force). To properly translate the illocutionary forces, the Intention Translation Method has been developed at ATR (Kogure et al., 1989). This generator was developed as a part of the dialogue translation system. Bidirectional grammar is helpful in maintain- ing tile grammar/lexicon (Appelt, 1987). A feature structure representation has been adopted for analysis result and generation input because it can keep various information including illocutionary forces and pragmatics in a consistent way. The infinite application of grammar rules is a common problem of the existing top-down unification-based generators (Shieber et al., 1989). The solution adopted here is to control the generation process by selecting appro- priate rules to apply. Typed feature struc- tures (A~'t-Kaci, 1986) are utilized to describe the control of the generation process in a declarative way. This description can also be used to avoid the derivation of unnecessary trees and to increase the generation efficiency. Another problem of the top-down generators is making multiple copies of the phrase structure when the generation process encounters multiple rule candidates. This problem can be treated by introducing disjunctive feature structures. This article first describes the advantages of typed feature structures and disjunctive feature structures in sections 2 and 3. The grammar for the generation system and the generation results are shown in section 4. The current status of this project and future tasks are described in section 5. 2. Introducing Typed Feature Structures 2.1 Selecting Appropriate Rules The basic mechanism of this generator is the top-down application of the grammar rules and construction of the feature structures of the daughter nodes. It is important to avoid the derivation of unnecessary phrase structures by selecting appropriate rules to apply in order to increase the efficiency. Consider the following rules taken from D-PATR (Karttunen, 1986). VP =HC*=> (VP XP) (<!m sem cont> == <!head-dtr sem cont>) (<lhead-dtr !subcat first> == <!comp-dtr-l>) (<!head-dtr Jsubcat rest> == <Jm !subcat>)) VP =CH=> (VP PP) (<!m sem cont> == <!head-dtr sem cont>) (<!head-dtr !subcat first> == <!comp-dtr-l>) (<Phead-dtr !subcat rest> == <!m isubcat>)) (1)I (2) The construction of the semantic repre- sentations given to the mother nodes of these ~ In this rule, =HC*=> link shows that the first element of the right hand syml×~ls becomes …

Content truncated to 500 words