Content planning and generation in continuous-speech spoken dialog systems∗
Content planning and generation in continuous-speech spoken dialog systems∗
复制标题
连续语音对话系统中的内容规划和生成
DOI：
--
复制
发表时间：
2000
期刊：
影响因子：
--
通讯作者：
Amanda Stent
中科院分区：
文献类型：
--
作者：
Amanda Stent
对构建能够在相对复杂的领域中自然交互的会话代理感兴趣的研究人员面临着一组独特的约束。Generation必须发生在真实或接近真实的时间里。语言覆盖范围必须广泛，语言使用必须多样化。基于语法的方法可能既缓慢又笨拙。另一方面，使用模板很难提供所需的语言覆盖。在本文中，我们提出了一种结合这两种方法的生成架构，利用它们的优点并最大限度地减少它们的缺点。在这个过程中，我们试图回答这个问题：“模板能让我们走多远？”
Researchers interested in constructing conversational agents that can interact naturally in relatively complex domains face a unique set of constraints. Generation must take place in real, or near-real, time. The language coverage must be extensive, and language use must be varied. A grammar-based approach can be both slow and awkward. On the other hand, it is difficult to provide the required language coverage using templates. In this paper we propose an architecture for generation that combines these two approaches, capitalizing on their strengths and minimizing their weaknesses. In the process, we attempt to answer the question, “How far can templates take us?”