Types and Records for Predication

摘要

This paper studies the use of records and dependent types in GF (Grammatical Framework) to build a grammar for predication with an unlimited number of sub-categories, also covering extraction and coordination. The grammar is implemented for Chinese, English, Finnish, and Swedish, sharing the maximum of code to identify similarities and differences between the languages. Equipped with a probabilistic model and a large lexicon, the grammar has also been tested in wide-coverage machine translation. The first evaluations show improvements in parsing speed, coverage, and robustness in comparison to earlier GF grammars. The study confirms that dependent types, records, and functors are useful in both engineering and theoretical perspectives.