Human Language Technology and Pattern Recognition.pdfVIP

下载本文档

5
0
约2万字
约 28页
2017-04-11 发布于江苏
举报
版权申诉

Human Language Technology and Pattern Recognition.pdf

1、原创力文档（book118）网站文档一经付费（服务费），不意味着购买了该文档的版权，仅供个人/单位学习、研究之用，不得用于商业用途，未经授权，严禁复制、发行、汇编、翻译或者网络传播等，侵权必究。。
2、本站所有内容均由合作方或网友上传，本站不对文档的完整性、权威性及其观点立场正确性做任何保证或承诺！文档内容仅供研究参考，付费前请自行鉴别。如您付费，意味着您自己接受本站规则且自行承担风险，本站不退款、不进行额外附加服务；查看《如何避免下载的几个坑》。如果您已付费下载过本站文档，您可以点击这里二次下载。
3、如文档侵犯商业秘密、侵犯著作权、侵犯人身权等，请点击“版权申诉”（推荐），也可以打举报电话：400-050-0827(电话支持时间：9:00-18:30)。
4、该文档为VIP文档，如果想要下载，成为VIP会员后，下载免费。
5、成为VIP后，下载本文档将扣除1次下载权益。下载后，不支持退款、换文档。如有疑问请联系我们。
6、成为VIP后，您将拥有八大权益，权益包括：VIP文档下载权益、阅读免打扰、文档格式转换、高级专利检索、专属身份标志、高级客服、多端互通、版权登记。
7、VIP文档为合作方或网友上传，每下载1次，网站将根据用户上传文档的质量评分、类型等，对文档贡献者给予高额补贴、流量扶持。如果你也想贡献VIP文档。上传文档

Human Language Technology and Pattern Recognition

Improving SMT by Using Multiple Translation Hypotheses Sa?a Hasan, Evgeny Matusov, Arne Mauser, David Vilar, Richard Zens, Hermann Ney TC-STAR OpenLab on Speech Translation Trento, Italy – March 30 - April 1, 2006 Human Language Technology and Pattern Recognition Lehrstuhl für Informatik VI Computer Science Department RWTH Aachen University, Germany S. Hasan: Multiple Translation Hypotheses 1 / 23 TC-STAR OpenLab: March 30 - April 1, 2006 Outline 1. Introduction 2. Related work 3. Rescoring I Reranking N -best lists I Different methods for rescoring I Experiments 4. System combination I Computing alignments I “Voting” on confusion networks I Experiments 5. Conclusions S. Hasan: Multiple Translation Hypotheses 2 / 23 TC-STAR OpenLab: March 30 - April 1, 2006 Introduction I SMT systems (e.g. phrase-based decoders) . use a combination of various models during generation . are capable of producing single-best output . generate word graphs / N -best lists with multiple translation hypotheses I Observation: all MT systems make errors I Assumption: different MT systems make different errors (due to utilizing different models / generation strategies / tweaks) I Two possibilities for improvement: . rerank multiple translation candidates from a single MT system → Rescoring . generate consensus translations from various MT systems → System Combination S. Hasan: Multiple Translation Hypotheses 3 / 23 TC-STAR OpenLab: March 30 - April 1, 2006 Related work I Rescoring . discriminative and minimum error rate training [Och Ney 02, Och 03] . different discriminative reranking techniques [Shen Sarkar+ 04] . syntactical features for rescoring [Och Gildea+ 04, Hasan Bender+ 06] . clustered language models [Hasan Ney 05] I System combination . successful approaches to system combination in automatic speech recognition (ASR) like ROVER [Fiscus 97] . sentence selection algorithms [Nomoto 04, Paul Doi+ 05] ? selection of hypotheses based on scores of statistical and other models ? ap