CODA: COherent Dialogue Automatically generated from text

CODA（从文本中自动生成的连贯对话）项目将有助于实现联合王国政府科学和技术理事会的愿景，即“在人们需要的时候、地点和方式向他们提供服务和信息”。下一代个人数字助理（PDA）[…]还有很多其他的人类导向的方法还没有被预见到。CODA将通过开发从独白形式的文本自动创建对话内容的理论和技术来帮助实现这一目标。有充分的经验证据表明，在某些情况下，以对话的形式提供信息可能比独白更有效（例如，辅导和有说服力的沟通）。然而，由于大多数信息都被锁定在文本（书籍、传单、网页等）中，文本到对话生成技术可以发挥重要作用，以最能满足人们对易于处理和引人入胜的信息的需求的形式提供信息。对话的有效性被放大的事实是，它非常适合新的多媒体演示风格-例如，对话可以由数字计算机动画角色来执行。因此，以对话形式呈现信息不仅有望提供有效的信息呈现，而且还能娱乐和吸引人们，如在传统媒体中广泛使用对话，如新闻简报、商业广告、教育娱乐和游戏。拟议的研究建立在与国家信息学研究所（东京）的Prendinger博士合作进行的初步可行性研究的基础上。这项研究导致了第一个原型，它需要一个病人的信息传单与文本，如：你可以使用阿司匹林，如果你有头痛。虽然阿司匹林确实有副作用：它会损害血液循环。，并自动生成虚拟药剂师和客户之间的对话：C：如果我头痛怎么办？你可以用阿司匹林。C：但是它有副作用吗？是的，它会损害血液循环. . Prendinger博士被提名为本项目的客座研究员。该项目将把上述第一个原型开发成一个独立于领域的系统，用于从文本生成对话，从而保留输入文本的含义，并使生成的对话既连贯又有凝聚力。据我们所知，它还将产生第一个广泛的文本跨度与对话片段配对的集合，这些对话片段在意义上是等同的（一个平行的文本对话语料库）。该语料库将在项目中用于学习从文本到对话的转换，然后系统将实现这些转换。在该项目的后半期，将对该系统进行彻底评价，以确定所产生对话的内容和组织的质量。它将被应用于输入文本从各种领域，把它的鲁棒性/域独立性的测试。我们预计，如果成功的话，这个项目将导致潜在的商业开发的中间件，用于弥合锁定在文本中的内容与通过最先进的多媒体演示工具有效和吸引人的信息演示之间的差距，并在教育中应用（介绍教科书材料）、电子保健（以吸引人的方式呈现医疗信息）和严肃/教育游戏（自动生成非玩家角色的对话内容）。

The CODA (COherent Dialogue Automatically generated from text) project will make a contribution to realizing the UK Government's Council for Science and Technology's vision of `providing people with services and information when, where and how they need it [...] Interaction will be through next generation personal digital assistants [...] and doubtlessly a variety of other human-oriented methods as yet unforeseen'. CODA will help achieve this by developing the theory and technology for automatically creating dialogue content from text in monologue form. There is ample empirical evidence that presentation of information in the form of a dialogue can be more effective than monologue in certain settings (e.g., tutoring and persuasive communication). Since most information is, however, locked up in text (books, leaflets, webpages, etc.), text-to-dialogue generation technology can play an important role in making information available in a form that best meets people's needs for easily processible and engaging information. The effectiveness of dialogue is magnified by the fact that it is eminently suitable for new multimedia presentation styles - e.g., a dialogue can be performed by digital computer-animated characters. Thus, presentation of information in dialogue form promises to not only deliver effective information presentation, but also entertain and engage people, as evidenced by the widespread use of dialogue in conventional media, such as news bulletins, commercials, educational entertainment and games. The proposed research builds on a preliminary feasibility study that was undertaken in collaboration with Dr. Prendinger at the National Institute of Informatics (Tokyo). That research led to a first prototype that takes a patient information leaflet with text such as: You can use aspirin, if you have a headache. Though aspirin does have side effects: it can harm the circulation. , and automatically generates a dialogue between a virtual pharmacist and client: C: What if I have a headache? P: You can use aspirin. C: But does it have side effects? P: Yes, it can harm the circulation. . Dr. Prendinger is proposed as Visiting Researcher for the current project.The project will develop the aforementioned first prototype into a domain-independent system for the generation of dialogue from text such that the meaning of the input text is preserved and the resulting dialogue is both coherent and cohesive. It will also produce the, to our knowledge, first extensive collection of text spans paired with snippets of dialogue that are equivalent in meaning (a parallel text-dialogue corpus). This corpus will be used in the project to learn transformations from text to dialogue that the system will then implement. During the second half of the project, a thorough evaluation of the system will take place to determine the quality of the content and organization of the generated dialogues. It will be applied to input texts from a variety domains to put its robustness/domain-independence to the test. We anticipate that if successful, this project will lead to potentially commercially exploitable middle-ware for bridging the gap between content locked up in text and effective and engaging presentations of information through state-of-the-art multimedia presentation tools, with applications in education (presentation of textbook materials), E-health (presenting medical information in an engaging way), and serious/educational games (automatic generation of dialogue content for non-player characters).