| 摘要: |
| 当前,系统融合是在机器翻译的后处理上进行.提出了在解码过程中来融合翻译模型,融合了主流两个翻译系统的翻译模型(层次化的基于短语的文法Hiero 和括号转录文法BTG).并从理论和实践的角度探索了现在主流的两种解码方法.同时,所提出的解码方法解决了伪歧义或一致性问题.在实验结果上得出:多文法模型融合的标志性要好于成员翻译模型;新的解码方法标志性好于传统解码方法(Viterbi 解码). |
| 关键词: 超图 推导 规则 翻译模型融合 伪歧义 一致性翻译 |
| DOI：10.3724/SP.J.1001.2012.04165 |
| 分类号: |
| 基金项目:国家自然科学基金(60736014); 国家高技术研究发展计划(863)(2006AA010108); 黑龙江省教育厅科学技术研究项目(12521073) |
|
| Research on the Translation Model Combination Based on Hypergraph |
|
LIU Yu-Peng, LI Sheng, ZHAO Tie-Jun
|
|
Department of Computer Science and Technology, Harbin Institute of Technology, Harbin 150001, China
|
| Abstract: |
| The system combination performs under post-processing, but the paper introduces a translation model combination, which combines two mainstream translation models (Hiero and MaxEnt-based BTG) during decoding. To the spurious ambiguity and consensus problem, the paper introduces new decoding method to solve two problems. In experiment, translation model combination is significantly better than member model, and the new decoding method is better. |
| Key words: hypergraph derivation rule translation model combination spurious ambiguity consensus translation |
