SBIR Phase II: Incorporation of Knowledge Base into Statistical Machine Translation

这个小企业创新研究（SBIR）第二阶段项目体现了一种创新的机器翻译方法。该模型旨在克服开发高质量统计机器翻译（SMT）系统的两个重要瓶颈：(1)无法处理结构问题；(2)对大量平行文本的依赖。当两种语言在结构和词法上存在很大差异时，比如英语和韩语，统计学在处理词序等语法问题上的无能就变得更加明显。对大量平行文本的依赖是一个巨大的挑战，尤其是对语音翻译。基于第一阶段项目的成功测试，本项目提出了一种从输入中自动学习对处理词序和非局部依赖关系至关重要的语言知识的方法，并将其与简单的转换一起合并到SMT中，最大限度地提高了基于知识的方法和统计方法的强度，并最大限度地减少了对不断增加的双语数据量的需求。该方法旨在构建一个基于句法短语的统计机器翻译引擎，该引擎不仅比现有的基于词的统计机器翻译引擎更准确，而且可以减少对大型数据源的需求。拟议项目的主要影响是实现自动翻译质量的潜力，与最好的基于知识的机器翻译引擎的质量一样高；但是用最少的手工制作知识，因此在开发时间和人力资源方面的成本要低得多。虽然该研究特别关注英语和韩语之间的机器翻译，但由此产生的翻译模型可能可用于任何语言对之间的翻译。研究结果将用于开发语音翻译设备，特别是克服与医院患者沟通的语言障碍。它将提供一项关键技术，加速语音翻译应用程序的开发，以降低医疗保健提供者的成本并提高医疗保健质量。此外，所提出的学习语言特征的方法将对许多不同的应用产生影响，包括语音识别、搜索引擎、体裁和主题检测以及文档搜索和查询。最后，通过帮助解决“自动翻译”问题，这一领域对美国和世界其他地区的经济福利和安全至关重要，拟议的研究将对国内和全球产生有益的影响。

This Small Business Innovation Research (SBIR) Phase II project embodies an innovative approach to machine translation. The proposed model aims to overcome two important bottlenecks in the development of a high quality statistical machine translation (SMT) system: (1) inability to handle structural problems and (2) dependence on huge amounts of parallel texts. The inability of statistics to sufficiently handle grammatical problems such as word order becomes more evident when the language pair is very different in structure and morphology, such as with English and Korean. The dependence on a huge amount of parallel texts is a great challenge especially to speech translation. Based on successful tests in the Phase I project, this project proposes a method to learn linguistic knowledge crucial to handling word order and non-local dependencies automatically from input and incorporate it into SMT along with simple transformations, maximizing the strength of both knowledge-based approaches and statistical approaches, and minimizing the need for ever-increasing amounts of bilingual data. The proposed approach aims to build a syntactic-phrase-based statistical machine translation engine that not only is more accurate than the existing word-based ones, but also can decrease the need for large data sources.The primary impact of the proposed project is the potential for achieving automatic translation quality as high as the quality of the best knowledge-based machine translation engines; but with a minimum of handcrafting of knowledge and therefore at a much lower cost in terms of development time and human resources. While the research is specifically concerned with MT between English and Korean, the resulting translation models would potentially be usable for translation between any pair of languages. The result of the research will be used to develop a speech translation device, in particular to overcome language barriers in communication with patients in hospitals. It will provide a key technology that will accelerate development of speech translation applications in order to reduce costs of healthcare providers and to enhance the quality of healthcare. Additionally, the proposed method of learning linguistic features will have an impact on many different applications including speech recognition, search engines, genre and topic detection, and document search and query. Finally, the proposed research will have beneficial impacts nationally and globally by helping to solve the 'automatic translation' problem, an area of paramount importance to the economic welfare and security of the United States and the rest of the world.