Automatic Processing of Spoken and Written Lecture Material

在过去的十年里，我们看到在线学术讲座材料的可用性急剧增加。然而，值得注意的是，与许多其他交际活动相比，到目前为止，演讲加工几乎没有从人类语言技术的发展中受益。本提案的目标是使快速，准确和方便地访问讲座内容。我们将在语音识别、结构归纳和摘要领域开发新技术。我们的工作将有助于更好地理解书面语言和口语之间的关系，这是语言学中一个长期存在的问题，但实证研究有限。我们的研究将通过从词汇到话语变化等不同层面对这种关系进行广泛的基于语料库的研究来纠正这种情况。我们建议开发的工具将在麻省理工学院开放课件计划的框架下进行整合和测试，该计划是一个大型的公开在线存储库，包含了麻省理工学院500门课程的教学材料。此外，我们还将把我们的工具纳入解放学习倡议，该倡议致力于将残疾学生融入主流高等教育。

In the past decade, we have seen a dramatic increase in the availability ofon-line academic lecture material. It is conspicuous, however, that in contrastto many other communicative activities, lecture processing has until nowenjoyed little benefit from the development of human language technology. Thegoal of this proposal is to enable fast, accurate and easy access to lecturecontent. We will develop new technologies in the area of speech recognition,structure induction and summarization.Our work will contribute to a better understanding of the relationship betweenwritten and spoken language, a long standing issue in linguistics which hasseen limited empirical research. Our research will rectify this situation byan extensive corpus-based study of this relationship at different levels,ranging from vocabulary to discourse variations.The tools we propose to develop will be integrated and tested in the frameworkof the MIT Open CourseWare Initiative, a large publicly available on-linerepository of teaching material from 500 MIT courses. In addition, wewill also incorporate our tools in the Liberated Learning Initiative, whichworks on the integration of students with disabilities in mainstream highereducation.