PIRE: Investigation of Meaning Representations in Language Understanding for Machine Translation Systems

-这一国际研究和教育伙伴关系(PERR)将约翰·霍普金斯大学和布朗大学的高级和初级研究人员与捷克共和国查尔斯大学和德国萨尔兰大学的同行联系起来。这个由约翰·霍普金斯大学的弗雷德里克·耶利内克领导的国际团队将研究语音识别/重建和机器翻译(MT)系统中使用的语言意义的正式表示。他们的目标是通过应用各种用于深层句法/语义表示的形式模型来增强当前的语音识别系统，以便其改进的机器翻译系统的输出成为连贯的、语法的文本。项目的补充教育部分包括向参与项目的美国研究生介绍欧洲开发的语言形式，并培训他们将这些形式应用于自然语言处理中的问题。学生将接受捷克语或德语的语言培训，并将在国外至少度过一个学期，在那里他们将在查尔斯大学继续接受技术语法表达方面的语言培训，或在萨尔兰大学继续接受中心驱动的短语结构语法方面的语言培训。在博士项目的最后阶段，每个人都将返回捷克共和国或德国，与欧洲导师合作进行研究，纳入这些最先进的语言处理技术。合作研究、年度研讨会和交叉培训的成果将通过整合自然语言处理的形式意义表示和统计方法来推动计算语言学领域的发展，以便能够利用现代计算机资源更快地将其他语言的言语交流翻译成英语。如果成功，这项工作可能会彻底改变自动语音识别的语言建模，这样即使是自发的语音也可以被翻译成流畅的、重构的文本，有效地捕捉到原文的原意。这个计算语言学的跨学科帝国通过使美国和欧洲的专家能够结合互补的人才并在相互感兴趣和能力很强的领域共享研究资源，实现了促进科学知识的计划目标。更广泛的影响包括将美国研究生早期职业介绍到由领先的语言学家、计算理论家和人类语言技术专家组成的国际专业网络。

----This Partnership for International Research and Education (PIRE) links senior and junior researchers from Johns Hopkins University and Brown University with counterparts from Charles University in the Czech Republic and Saarland University in Germany. The international team, led by Frederick Jelinek at Johns Hopkins, will investigate formal representations of linguistic meaning for use in speech recognition/reconstruction and machine translation (MT) systems. Their goal is to augment current speech recognition systems by applying a variety of formal models for deep syntactic/semantic representation so that output of their refined MT system becomes coherent, grammatical text.The projects complementary education component involves introducing participating U.S. graduate students to European-developed linguistic formalisms and training them to apply those formalisms to problems in natural language processing. Students will have language training in Czech or German and will spend at least one semester abroad where they will further their linguistic training in tectogrammatical representation at Charles University or head-driven phrase structure grammar at Saarland University. In the final stages of their Ph.D. program, each will return to the Czech Republic or Germany to work with European mentors on research that incorporates these state-of-the art language processing techniques. Results from the collaborative research, annual workshops and cross-training should advance the field of computational linguistics by integrating formal meaning representations and statistical methods for natural language processing so that modern computer resources can be exploited to more rapidly translate verbal communications from other languages into English. If successful, this work could revolutionize language modeling for automatic speech recognition so that even spontaneous speech may be translated into fluent, reconstructed text that efficiently captures the intended meaning of the original. This interdisciplinary PIRE in computational linguistics fulfills the program objective of advancing scientific knowledge by enabling experts in the United States and Europe to combine complementary talents and share research resources in areas of strong mutual interest and competence. Broader impacts include early career introduction of U.S. graduate students to an international professional network of leading linguists, computational theorists, and experts in human language technology.