Language Generation for a Speech Prosthesis

这个项目正在开发一种自然语言生成的新方法，将其应用于身体残疾人士的计算机辅助文本和语音生成。许多由于身体残疾而不能说话的人使用文本到语音发生器作为假肢设备。然而，使用语音假肢的人通常会更普遍地失去运动控制，尽管有单词预测等辅助工具，但文本输入速度缓慢且困难。对于普通用户来说，目前的语音假肢的输出速度还不到正常语音速度的十分之一。这阻碍了自然的社交对话，因为它完全扰乱了通常的轮流过程，并可能对听者对假体使用者的态度产生负面影响。这项研究的主要焦点是研究能够在不牺牲内容灵活性的情况下，充分提高更自然对话率的技术。这种新方法结合了广泛覆盖的语法、基于语料库的词频数据和会话模板。应用于语音假肢，它能够以上下文敏感的方式从最小的用户输入产生完整的句子。这种方法也可以更普遍地应用于有效地制作公式化文本，例如在商业和政府中广泛使用的结构化报告。

This project is developing a novel approach to natural language generation, applying it to computer-aided text and speech generation for people with physical disabilities. Many people who cannot speak because of physical disability utilize text-to- speech generators as prosthetic devices. However, users of speech prostheses often have more general loss of motor control, and despite aids such as word prediction, text entry is slow and difficult. For typical users, current speech prostheses have output rates less than a tenth of the speed of normal speech. This prevents natural social conversation, since it completely disrupts the usual processes of turn-taking, and can lead to negative effects on the listener's attitude to the prosthesis user. The main focus of this research is the investigation of techniques which can improve rates sufficiently for more natural conversation to be possible, without sacrificing flexibility of content. This new approach employs a combination of a wide- coverage grammar, corpus-based word frequency data, and conversational templates. Applied to speech prosthesis, it enables the production of full sentences from minimal user input in a context-sensitive way. The approach can also be applied more generally for efficient production of formulaic text like the structured reports used widely in business and government.