Experiments in German Treebank Parsing

LECTURE NOTES IN ARTIFICIAL INTELLIGENCE(2003)

引用 5|浏览15
暂无评分
摘要
State-of-the-art parsers for English are based on probabilistic grammars induced from treebanks. German, a language with a free-er word order and a richer morphology, is likely to pose new problems to techniques developed especially for English. This paper describes one of the first parsing experiments using probabilistic context-free grammars extracted from the German treebank NEGRA. Particularly, we introduce "partial-parent encoding" for NEGRA, a variant of the parent-encoding technique. Evaluation shows that grammatical functions, and partial-parent encoding improve the performance of a parser for German. Moreover, we find evidence that extending the size of the treebank as well as adding more linguistic information will improve the results.
更多
查看译文
关键词
word order
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要