Abstract: Automatic acquisition of translation templates is very important for MT system to improve its translation quality and its ability of adapting to new domain.In this paper，tree-to-string method is applied to extract translation equivalences.Error-driven learning method is used to acquire translation templates.A knowledge optimization tool is used to filter translation templates.Then these templates are applied to a transfer-based MT system，and “863” dialog corpus is used as open test corpus.The experiment shows that when new acquired and optimized templates are used，evaluation score for translation of open test corpus is improved.

Key words: translation template, translation equivalence, error-driven

摘要： 翻译模板自动获取是提高MT译文输出质量和领域适应能力的关键性因素。利用Tree-to-String方法抽取等价对，使用错误驱动的学习方法从中获取翻译模板并进行优化。将优化后的翻译模板用于一个基于转换的机器翻译系统中，同时使用“863”对话语料对其进行评测。实验结果表明：当使用自动获取并经优化的模板进行翻译时，开放测试语料的译文评测分数有一定程度的提高。

关键词: 翻译模板, 等价对, 错误驱动