LETRAS: A Learning-based Framework for Machine Translation of Low Resource Languages

LETRAS项目研究机器翻译（MT）技术发展的新方法，目标是建立一个通用框架，支持为只有有限数量的数据和电子形式的资源可用的语言构建机器翻译原型系统。该项目的研究重点是从有限数量的抽取双语数据中自动学习翻译迁移规则。一种新的运行时翻译“引擎”通过建立一个可能部分翻译的大结构，然后应用有效的搜索技术来恢复最佳翻译，将源语言句子映射到它们的目标语言对等体。在最后一个阶段，自动规则细化模块帮助系统学习如何根据用户提供的翻译错误反馈来纠正和改进其不完善的翻译规则。几个语言对的机器翻译原型系统正在构建中，作为项目的一个组成部分，并与外部研究小组合作。原型指导我们的研究并测试我们的新想法。与此同时，我们与当地研究人员和当地社区的合作促进了母语信息技术的发展，并以我们最先进的机器翻译研究教育当地研究人员。原型包括一个希伯来语到英语的机器翻译系统（与以色列海法大学合作）；inupiak -to- english MT系统（与阿拉斯加大学费尔班克斯分校和阿拉斯加巴罗的inupiat遗产中心合作）；以及与巴西圣保罗大学合作建立的卡里亚纳-葡萄牙语MTsystem。阿拉斯加合作项目由美国国家科学基金会极地项目办公室（OPP）提供支持，与以色列和巴西的合作由美国国家科学基金会国际科学与工程办公室（OISE）提供支持。OISE还为玻利维亚的计划之旅提供资金，以探索一个可能的艾马拉到西班牙的项目。这些项目潜在的长期影响是深远的——使世界上许多语言的机器翻译得以发展，这反过来又为土著和少数民族社区积极参与21世纪信息丰富的活动打开了大门。

The LETRAS project investigates novel approaches to development of MachineTranslation (MT) technology, with the goal of establishing a general frameworkthat supports building MT prototype systems for languages for which onlylimited amounts of data and resources in electronic form are available. Theresearch focus of the project is on automatic learning of translationtransfer-rules from limited amounts of elicited bilingual data. A new run-timetranslation "engine" maps source language sentences to their target languageequivalents, by building a large structure of possible partial translationsand then applying effective search techniques for recovering the besttranslation. In the last stage, an automatic rule refinement module helps thesystem learn how to correct and improve its imperfect translation rules, basedon feedback on translation errors provided by users. MT prototype systems forseveral language pairs are being constructed as an integral part of theproject and in collaboration with external research groups. The prototypesguide our research and test out our new ideas. At the same time, ourcollaborations with local researchers and native communities promote thedevelopment of information technology for native languages and educate localresearchers with our state-of-the-art MT research. The prototypes include aHebrew-to-English MT system (with University of Haifa, Israel); anInupiaq-to-English MT system (with University of Alaska, Fairbanks, and theInupiat Heritage Center in Barrow, Alaska); and a Karitiana-to-Portuguese MTsystem (with University of Sao Paulo, Brazil). Support for the Alaskacollaboration is being provided by NSF's Office of Polar Programs (OPP), andsupport for the collaborations with Israel and Brazil is being provided byNSF's Office for International Science and Engineering (OISE). OISE is alsoproviding funding for a planning trip to Bolivia to explore a possible Aymara-to-Spanish project. The potential long-term impact of the projectis profound - enabling the development of Machine Translation for manylanguages of the world, which in turn opens the door for active participationof native and minority communities in the information-rich activities of the21st century.