Another Stride Towards Knowledge-Based Machine Translation

Anothe r St ride Towa rds Knowledge-Based Machine Translation Masaru Tomita Jaime G. Carbonell Computer Science Department CarnegieMellon University Pittsburgh, PA 15213 12 April1986 Abstract Building on the well-established premise that reliable machine translation requires a significant degree of. text comprehension, this paper presents a recent advance in multi-lingual knowledge- based machine translation (KBMT). Unlike previous approaches, the current method provides for separate syntactic and semantic knowledge sources that are integrated dynamically for parsing and generation. Such a separation enables the system to have syntactic grammars, language specific but domain general, and semantic knowledge bases, domain specific but language general. Subsequently, grammars and domain knowledge are precompiled automatically in any desired combination to produce very efficient and very thorough real-time parsers. A pilot implementation of our KBMT architecture using functional grammars and entity-oriented semantics demonstrates the feasibility of the new approach? 1. Introduction This paper introduces a new approach to knowledge-based machine translation for well-defined domains, integrating two recent advances in computational linguistics: entity-oriented parsing [16] and functional grammars [4, 19]. The entity-oriented formalism has several strengths in representing semantic knowledge for circumscribed domains, but has limitations in representing general syntactic knowledge. Functional grammar formalisms, such as lexical functional grammar (LFG) and functional unification grammar (UG), on the other hand, can represent general syntactic knowledge, but are severely limited in their ability to represent general semantic information. In our approach, the semantic and syntactic knowledge bases are developed separately in the entity-oriented and functional grammar formalisms, and a multi-stage grammar preoompfler compiles tfiem into a single knowledge base which contains both syntactic and semantic information in a form suitable for efficient real-time parsing. Our integrated approach copes with limitations of both entity-oriented and functional grammar formalisms, retaining the advantages of each. The approach is particularly well suited for machine translation, where knowledge of multiple languages must be represented in a uniform manner. Knowledge-based machine translation (KBMT) [8] is the process of applying syntactic knowledge of the source language and semantic knowledge pertinent to the source text in order to produce a canonical language-free meaning representation, which may then be rendered in many different languages. The analysis process of producing a meaning representation is far more complex than that of using target-language knowledge to express the meaning, representation in the target language, because the former is a many-to-one mapping, whereas the latter may be coerced into a one-to-one mapping. 2 Whereas KBMT is in principle far superior to conventional transfer grammar techniques requiring a human translator (the "posteditor") to clean Lip syntactic and semantic errors [5, 8], in practice semantic analysis requires fairly thorough coverage of the domain. This ravenous hunger for domain knowledge makes KBMT more' practical for domains in which the development of the knowledge base can be amortizecl over very large numbers of texts to translate .. domains such as stocks and other security negoti~:Ltions, doctor-patient communication, weather forecasts, banking transactions, financial reports, economic analyses, invoices and purchase orders, etc. Thus, KBM] is particularly well-suited for multi-lingual Iranslation …

Content truncated to 500 words