【作者】 刘东飞；

【导师】 钟珞；

【作者基本信息】 武汉理工大学 ， 计算机应用技术， 2009， 博士

【摘要】 随着我国经济与对外交流的不断发展,翻译市场规模迅速扩大,利用计算机软件等翻译工具越来越受到用户的青睐,各类机器翻译的技术和方法层出不穷。机器翻译的方法主要分为两大类,分别是基于规则的方法和基于语料库的方法。基于规则的机器翻译最大的难题就是语言歧义的消解;基于语料库的方法是采取翻译记忆的模式,用户利用已有的原文和译文,建立起一个或多个语料库,在翻译过程中,系统自动搜索库中相同或相似的翻译资源,并给出参考译文。但是,目前大量基于翻译记忆模式的翻译软件所依赖的语料库都是人工构建的,语料库容量有限,且更新缓慢。随着因特网的发展,基于Web的全文搜索引擎越来越成为信息检索的主要手段,若能将翻译技术与Web信息搜索技术结合起来,就可以为用户提供完善的、实时的和动态的翻译服务。从研究对象的角度来看,Internet所包含的数据呈现出海量性、半结构化、多样性、动态性、分布式和异构性的特点。特别是经过多年的积累,Internet网站上存储了海量的中英文双语网页资源。双语语料库对机器翻译、机器辅助翻译、双语词典编纂、双语术语自动提取、双语对比研究以及双语教学的巨大价值已经得到越来越多的认可。对于机器翻译或机器辅助翻译而言,双语语料库主要在两方面发挥作用。一方面,双语语料库可以持续不断地为基于记忆的翻译引擎提供翻译实例,另一方面,双语语料库又是一个宝藏,可以从中挖掘学习各种细粒度翻译知识,供机器翻译、机器辅助翻译使用,这些翻译知识对翻译任务的各个环节都会起到积极作用本课题的研究目的是通过对网络搜索引擎、自动积累双语语料库、机器翻译的研究,利用网络机器人技术、网页噪音过滤技术、语句匹配入库算法、数据挖掘、分词技术、双语对齐技术、智能用户接口技术、个性化搜索技术、元搜索及结果融合与排序算法、全文检索技术以及JAVA应用等多项技术,研究基于网络大规模语料库和双语网页搜索的智能双语搜索方法和搜索引擎。这既可以为专业人员提供自动的双语网页资源,又可以为用户提供高质量的辅助翻译,避免翻译的重复劳动,提高翻译质量,具有很好的市场价值和社会效益。本文在研究过程中,主要的创新工作如下:(1)在互联网海量信息中,专门针对含有中英双语信息的单页文本和双页对照文本进行研究。根据两者不同的特点,对双页双语文本首次提出了双页双语语料挖掘算法DBWCM(Double Bilingual Wepage Corpus Mining);对单页双语文本设计了分步处理算法,提出了识别与净化单页双语文本的算法IPSBW(Identification and Purification of the Single Bilingual Webpage)和双语句对匹配与语料挖掘算法BSMCM(Bilingual Sentences Matching and Corpus Mining)。通过执行它们提取网页中大量存在的双语平行翻译语料,形成大规模的语料库,为辅助翻译搜索打下坚实基础。(2)针对当前搜索引擎的用户界面现状,利用概念检索和潜在语义分析的思想,通过建立双语同义词词典对查询关键词进行双语扩展,并为用户输入提供扩展提示,实现良好的用户界面,提高了双语辅助搜索的智能性和查全率。同时为提高查询的准确率,满足用户的个性化需求,研究了显式和隐式两种获取与更新用户个性化信息的方法,提出了双语关键词个性化扩充算法PEBK(Personalized Expansion of the Bilingual Keyword)和双语查询结果的个性化排序算法PSBR(Personalized Sort of the Bilingual Results),以便使查询结果达到因时因人因地的不同,得到用户真正想看到的结果。(3)为了扩大双语搜索的范围,将元搜索技术应用在双语翻译搜索中,通过分析元搜索的结果融合方法,针对使用PageRank算法的问题,提出了结合相关度的改进算法RSBS(Results Sort of the Bilingual Search),并对算法的效果进行验证。(4)最后,在形成双语语料库的基础上,基于Java和Lucene工具,设计网络机器人模块、网页识别与净化模块、语料匹配入库模块、索引模块、检索模块、个性化查询模块,用户接口等七个主要模块,实现双语搜索,为用户提供辅助翻译服务。智能双语辅助翻译搜索涉及的研究领域非常广泛,它涵盖了人工智能、语言学、机器翻译、搜索引擎、Web数据挖掘、数据库等多个领域知识。本文所做的工作尽管为利用网络实现辅助翻译提出了一些可行方法,但是,对于构建一个高效且智能化程度高的基于网络的辅助翻译系统还有许多问题有待进一步研究。更多还原

【Abstract】 With the development of economic and international communication, translation market is quickly expanding. Translation using computer software, called machine translation (MT) becomes popular and MT based tools attract users’ interests. There are various machine translation technologies which can be mainly classified into two categories: (1) rule based approaches and (2) corpus based approaches . The former has difficulty in solving language disambiguation , while the advantage of the later is the utilization of translation memory. In corpus based approaches, users can utilize originals and their corresponding translations to build one or several corpora. Then when a translation task is arrived, system will automatically search the corpora for the same or similar originals and finally return translation results.However, the corpora of many sorts of translation software using translation memory are manually built. The capacity of these corpora is limited and their update is slow. With the development of Internet, Web search engines are widely used as an important information retrieval means. We argue that the combination of translation technology and Web information retrieval can provide satisfactory, realtime and dynamic translation service for users.Judging from the object of study, data contained by Internet appears out great capacity , partly structurization, diversity, dynamic , distributed and the isomerism characteristic. Chinese-English pair of Web page resource having been stored on Internet website great capacity especially after having accumulated for many years.The great value of bilingual corpus for machine translation , machine aided translation,bilingual dictionary to compile , the bilingual terms drawing automatically, bilingual contrast studies as well as bilingual education already gets more and more approved. for machine translation or the machine aided translation ,bilingual corpus can produce effect in the field of two mainly. The bilingual corpus can provide translation example unceasingly for the translation engine based on translation memory,on the other hand, bilingual corpus are a buried treasure from which we can excavate the various fine-grained degree translator knowledge for machine translation and machine aided translation, these translate knowledge may arrive at a positive role to each links of translation .This thesis studies how to build Web-based large scale bilingual corpus, retrieval mechanism and search system implementation of searching intelligent bilingual Web pages. With the help of general Web search engines, bilingual corpus can be built automatically and MT can be realized. Furthermore, we make use of internet robot technology, Web page filtering technology, sentence matching algorithm, data mining, word segmentation technology, bilingual matching technology, intelligent user interface, personalized search, meta-search, rank aggregation algorithms, text information retrieval, Java programming, and so on. Our research can provide not only bilingual Web pages for professionals, but also high-quality translation service for market users. Our system can improve translation quality and efficiency by avoiding manual translation, thus it has great market and social profits.Our thesis is summarized as follows.(1) Within the large-scale Web information sources, we focus on finding Chinese-English bilingual Web pages including single-page bilingual ones and counter double-page bilingual ones .. For the latter, we propose a novel algorithm, called DBWCM (Double Bilingual Webpage Corpus Mining); for the former, we design a step-by-step approach, propose two novel algorithms, i.e., IPSBW(Identification and Purification of the Single Bilingual Webpage) and BSMCM(Bilingual Sentences Matching and Corpus Mining) . By using those algorithms, alarge number of original texts and their corresponding translation versions areextracted from Web pages. Thus we can build a large-scale bilingual corpus whichlays a solid foundation for aided translation..(2) We investigate the current user interfaces of search engines. Based on concept search and latent semantic analysis (SLA) , we build a bilingual synonymicon to expand search keywords, which suggests suggest related keywords given the input keywords of users. In addition, our bilingual query expansion improves traditional user interfaces, the intelligence and recall of bilingual aided search. Also, in order to improve search precision, we study how to learn and update users’ preferences by explicit and implicit relevance analyses, and then personalized search results. PEBK (Personalized Expansion of the Bilingual Keyword) and(Personalized Sort of the Bilingual Results) are two novel proposed algorithms for personalization by considering the contexts of users such as time, interests, location, and so on.. (3) Meta-search are applied in bilingual translation to broaden search coverage. We analyze rank aggregation algorithms of meta-search and raise the problems of PageRank algorithm. Moreover, we propose a novel algorithm to enhance rank aggregation, called RSBS (Results Sort of the Bilingual Search) . Experimental results show that our algorithm is effective.(4) Last, we set up a bilingual search system based on our bilingual corpus by using Lucene and Java programming language. Seven function models are devised in our retrieval system. They are internet robot model, Web pages identification and purification model, bilingual Web pages matching model, index model, retrieval model, personalized search model, and user interface model. Our bilingual search system can provide users with aided translation service.Intelligent bilingual aided translation search covers several research fields. It has contained much field knowledge such as artificial intelligence , linguistics , machine translation completely , search engine , the Web data mining , the data base and so on. This thesis proposes some effective aided translation approaches by using Internet. In the future, there are still some interesting topics in building a efficient and highly intelligent aided translation system which are needed further exploration.更多还原