Principle-based Parsing for Chinese

Language, Information and Computation(PACLIC 11),1996, 363-371 Principle-based Parsing for Chinese Charles D. Yang and Robert C. Berwick Artificial Intelligence Laboratory Massachusetts Institute of Technology 545 Technology Square, NE43-769 {charles, berwick}@ai.mit.edu Abstract This paper describes the implementation of Mandarin Chinese in the Pappi system, a principle-based multi-lingual parser. We show that sub- stantive linguistic coverage for new and linguistically diverse languages such as Chinese can be achieved, conveniently and efficiently, through pa- rameterization and minimal modifications to a core system. In particular, we focus on two problems that have posed hurdles for Chinese linguistic theories. A a novel analysis is proposed for the so-called BA-construction, along with a principled computer implementation. For scoping ambigu- ity, we developed a simple algorithm based on Jim Huang's Isomorphic Principle. The implementation can parse fairly sophisticated sentences in a couple of seconds, with minimal addition (less than 100 lines of Prolog code) to the core parser. This study suggests that principle-based pars- ing systems are useful tools for theoretical and computational analysis of linguistic problems. 1 Introduction Natural languages are complex and their syntactic properties seem to differ from one another quite dramatically. Traditional parsing technologies utilize language-particular, rule-based formalisms, which usually result in large and inflexible systems (Marcus 1980). For a recent example of the rule-based ap- proach to Chinese parsing, see (Lee et al., 1991). In recent years, computational linguistics has seen the development of the Principle-based Parsing (Berwick et al 1991). A principle-based parser transpar- ently reflects the structure of the contemporary linguistic theory, the Principles and Parameters framework (Chomsky 1981). It is believed that languages are constrained by a small number of universal principles, with linguistic variations largely specified by parametric settings. The merit of principle-based parsing is two-fold. As a tool for linguists, it is directly rooted in grammatical theories. Therefore, linguistic problems, par- ticularly those that involve complex interactions among linguistic principles, can be cast in a computational framework and extensively studied by drawing directly on an already-substantiated linguistic platform. It is designed from the start to accommodate a wide range of languages — not just 'Eurocenteric' 363 Romance or Germanic languages. Japanese, Korean, Hindi and Bangla have all been relatively easily modeled in PAPPI (Berwick and Fong 1991, Berwick forthcoming). As a tool for engineering, it inherits the useful design principle of modularity among at least some of the principles. Differences among languages reduce to distinct dictionaries, required in any case, plus parametric variation in the principles. To show how this project may be concretely executed for Chinese, we first review the basic phrase structures and parameters for Chinese. Our principle source is Jim Huang's seminal dissertation (1982). Then we turn to two par- ticular problems that have generated much interest and productive work in the literature, the BA-construction and the scoping non-ambiguity in Chinese. Both theoretical and computational analyses are presented. The paper con- dudes with some general observations on principle-based parsing, in relation to linguistics and computation. 2 Structures and Parameters in Chinese "Chinese exhibits a …

Content truncated to 500 words