LFG FOR CHINESE: ISSUES OF REPRESENTATION AND COMPUTATION / LFG用于汉语：表示与计算中的若干问题

LFG has been widely used to analyze the English language as well as other languages from a purely linguistic point of view. A relatively new direction in the LFG research field is applying it to language computation, ranging from parsing to machine translation. The LFG-based work in Chinese computing is still rather rare. This paper addresses two aspects of the current framework of LFG in terms of its application to the processing of Chinese: it is quite powerful for linguistic representation, even very sophisticated sentential constructions in Chinese can be explained by LFG easily, but it seems not strong enough for Chinese computation the main weakness is on the computation of semantics. However, it is evident that it is semantic analysis, rather than syntactic analysis, which plays a dominant role in analyzing Chinese sentences. Consequently, the paper suggests some possible augmentation on the framework of LFG: (1) representing semantic constraints in a systematic way; (2) building up the semantic knowledge base to make inference on the augmented framework possible; (3) incorporating statistics into the LFG formalism; and (4) augmenting the operation of unification to fit the statistical computation. LFG 己被广泛应用于从语言学的视角分析英语和其它语言。关于 LFG 研究的 一个新的趋势是将之应用于语言计算，包括语法自动分析、机器翻译等。基于 LFG 的汉语计算相关工作迄今为止尚为数不多。本文掲示了 LFG 现有框架处 理汉语时所呈现出的两重性：一方面，它具有強大的语言表示能力，即使结构 非常复杂的汉语句子也可以很容易地用 LFG 予以描迷；另ー方面，它的计算能 カ仍然略显不足，主要反映在语义计算上。有证据表明，语义分析（而不是句 法分析）在汉语句子分析中占据着支配地位。在上述讨论的基础上，本文进一步提出了对LFG现有框架几点可能的扩展之处：(1) 系统地表示语义约束；(2) 建立相应的语义知识库，使得构造于扩展框架上的推理成为可能；（3) 将统计 机制嵌入 LFG 中；（4) 扩展合一运算以满足统计计算的需要。