Continuous Understanding

持续理解

• 批准号: 0328810
• 负责人: James Allen
• 金额: --
• 依托单位: University of Rochester
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2003
• 资助国家: 美国
• 起止时间: 2003-09-01 至 2007-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0328810&HistoricalAwards=false
• 关键词: Continuous; Understanding

Current spoken language understanding systems operate in a layered fashion, where utterances are processed entirely at one level (e.g., speech recognition) before hypotheses are passed to the next level (e.g., parsing). Such a layered organization creates significant problems in developing robust highly-interactive spoken language systems because expectations and feedback that arise from the context of the dialogue are not used by the lower levels of language processing. In collaboration with SRI international, we are developing a new architecture for continuous understanding, in which all levels of processing operate concurrently, sharing information throughout the system, enabling understanding on a word-by-word basis as the utterance progresses. This new architecture should have three main advantages: (1) it should improve the accuracy and efficiency of real-time spoken language understanding; (2) it should better support the processing of spontaneous speech; and (3) it should enable more effective spoken language interfaces that could interact with the user in a free-form, natural manner. The SRI group is focusing on speech recognition and analysis of prosody. This project will have a large impact on the field, and introduce a new paradigm for language understanding systems that could lead to significant leaps in robustness and accuracy, and produce a new generation of highly interactive human-computer interfaces. On a more general scale, as computers become an ever more pervasive part of the fabric of our society, our lives become increasingly more complex. If successful, this project would contribute towards creating much more human-friendly, easy to use, computer interfaces.

当前的口语理解系统以分层的方式运行，在假设被传递到下一层（例如解析）之前，话语在一个层次（例如语音识别）被完全处理。这种分层组织在开发健壮的高交互性口语系统时产生了重大问题，因为对话上下文产生的期望和反馈不用于较低层次的语言处理。在与SRI国际合作中，我们正在开发一种用于持续理解的新架构，在这种架构中，所有层次的处理都同时进行，在整个系统中共享信息，随着话语的进展，实现逐字理解。这种新架构应该有三个主要优势：(1)提高实时口语理解的准确性和效率；(2)要更好地支持自发言语的加工；(3)它应该使更有效的口语界面能够以自由、自然的方式与用户交互。SRI团队专注于语音识别和韵律分析。该项目将对该领域产生重大影响，并为语言理解系统引入一种新的范式，这可能导致鲁棒性和准确性的重大飞跃，并产生新一代高度交互的人机界面。在更广泛的范围内，随着计算机成为我们社会结构中越来越普遍的一部分，我们的生活变得越来越复杂。如果成功，这个项目将有助于创造更人性化、更易于使用的计算机界面。