Free Word Order in a Constraint-based Implementation of Dependency Grammar

FREE WORD ORDER IN A CONSTRAINT-BASED IMPLEMENTATION OF DEPENDENCY GRAMMAR Tom B. Y Lai Department of Chinese, Translation and Linguistics, City University of Hong Kong Email: cttomlai@cityu.edu.hk Huang Changning Department of Computer Science and Technology, Tsinghua University, Beijing Email: licn@mail.tsinghua.edu.cn ABSTRACT Parsing of sentences based on Dependency Grammar is emulated with a constraint- and unification-based mechanism that preserves single-headedness and projectivity in syntactic dependency. Working on Chinese, and referring to English at times, the authors have treated subcategorization properties of verbs on the assumptions that relative positions of obligatory complements of a verb are fixed, and that optional verbal adjuncts, while relatively free in their choice of positions in the sentence, are constrained to be placed farther away from the governing verb than any complements. A pair of subcategorization lists residing in the governing verb have been used to capture these properties. However when one tries to extend the model to languages with relatively free word order like Japanese, for which the assumptions mentioned above are not valid, straight-forward list-manipulations on subcategorization lists will not be adequate. This paper discusses how additional subcategorization handling mechanisms can be introduced to deal with verbal complements scrambling and intervening adjuncts, while retaining the mechanisms to capture default word order. 1. INTRODUCTION Computational linguists in China [9][26][27] have been studying and experimenting with Chinese sentence parsing based on the tradition of Gaifman, Hays and Robinson [4][7][21] in the development of Dependency Grammar,[25] which is different from other traditions of the grammar formalism [10][24][19][6] in that single-headedness and projectivity are required for dependency relations holding between governor-dependent pairs of words in a sentence. Using different algorithms for the parsing process, they differ in whether syntactic categories are considered in lieu of the actual words and in whether and how word order is taken into consideration, but they all observe the requirements of single- headedness and projectivity. They also label dependency relations with functional labels like subject and object as is done in practically all schools of Dependency Grammar. Unification [22] has been used in Dependency Grammar implementation and theory (e.g. [8][2][18]). The authors have distinguished between a single-headed and projective dependency (constituent) structure and a much less constrained functional structure.[14][15][16][17] Dependency rules [7] are annotated with functional annotation,[1] and a unification-based parser [22][5] has been adapted to produce dependency and functional structures for Chinese (and English) sentences. They have discussed how syntactic phenomena like control are dealt with. They make a distinction between obligatory complements prescribed for by the syntactic properties of the governing verb and optional adjuncts that are taken care of by non-lexical rules in the syntax of the language. A pair of subcategorization lists residing in the lexical entry of the governing verb are used to take care of obligatory verbal complements, making use of constraints in Chinese (and English) that the subject is to the left of the verb while the other verbal complements are on the other side, and that optional adjuncts must not come between the verb and the …

Content truncated to 500 words