Human-Centered Compression for Collaborative Text Input

越来越多的信息设备用户——移动电话、个人数字助理等——正在使用它们来交流文本信息，并且发现很难与它们进行交互。使设备首先变得有用的属性（尺寸、移动性、嵌入性）恰恰是使输入文本变得困难的属性。首席研究员正在开发利用自然语言固有冗余的方法，以显着提高自然语言文本输入到此类设备的效率。例如，这种冗余解释了我们阅读压缩文本的能力，例如“Cn u rd This ?”。如果信息设备也能高可靠性地读取这种高度缩写的文本，那么就可以使用缩写文本来加快文本输入。开发以这种“以人为中心”的方式解码文本压缩的系统，需要在压缩方法的建模和用户界面技术方面取得进展，这些技术允许交互式地纠正计算机对压缩文本的理解中的任何错误。本研究通过利用统计语言建模技术来解决最小化用户认知负荷所涉及的挑战，从而解决了所需的进展。这种能力将极大地简化和扩展与使用文本的计算机和电信系统进行通信的过程，并扩展可能发生文本输入和用户可以参与的潜在环境的基础。同样，残疾用户和其他在恶劣条件下工作的人也会受益。

Increasingly, users of information appliances -- cell phones, personal digital assistants, and the like -- are using them to communicate textual information, and finding them difficult to interact with. The properties that make the devices useful in the first place (size, mobility, embeddedness) are exactly the properties that make inputting text difficult. The principal investigator is developing ways to make use of the intrinsic redundancy of natural language to dramatically improve the efficiency with which natural language text can be entered into such devices. This redundancy accounts, for instance, for our ability to read compressed text such as "Cn u rd ths?". If information appliances could also read such highly abbreviated text with high reliability, abbreviated text could be used to speed up text input.Developing systems to decode text compressed in such a "human-centered" manner requires advances in modeling of the compression method and in user interface techniques that allow interactive correction of any mistakes in the computer's understanding of the compressed text. This research addresses the required advances by exploiting statistical language modeling techniques to address the challenges involved in minimizing cognitive load on the user. Such an ability would dramatically simplify and expand the process of communicating with and through computer and telecommunications system using text, and expand the base of potential environments where text input could occur and users who could participate. Similarly, benefits would accrue to disabled users and others operating under degraded conditions.