The next generation in conversational tools
Developed under the DUMAS (Dynamic Universal Mobility for Adaptive Speech Interfaces) project funded by the EC, new speech driven applications have been enhanced through a multi-lingual application. Providing further enhancements such as AI intuition that customises itself to user profiles. However language is a complex system of relations, made more complicated by the way in which people communicate. For a dialogue-based application to respond to vocal commands intuitively various tools are required. One such tool is a web-based editor for dialogue corpora to be marked up in an Annotation-Graph notation. This annotator allows for the selection of attribute values of a corpus of a previously transcribed spoken language data. Hence, the necessary flexibility is provided, allowing the annotator to select one or multiple languages at any given turn. It can also ascribe additional attribute values. This aspect of the technology uses Java Servlet and a relational Access database. Part of the projects' objective was to develop a tool that not only understood the spoken commands but also could provide a conversational interface/feedback. The WEKA machine-learning toolkit has been used in this instance, based on a second tool; the editor, which facilitates the construction of a dynamic classification module. The editor allows users to select variables and to specify their attributes. Using a parser based on these attributes allows the tool to format syntactically correct interrelations. This provides for a text-parsing methodology which has the ability to function with language errors in a robust, fault tolerant manner. The DUMAS project may not yet be the ideal conversationalist but it is a leap forward in providing mankind a multilingual, intelligent and articulate vocal interface. As such, it could be used for a number of applications, from email-voicemail services to cross lingual translations.