Font Size: a A A

A knowledge-based message generation system for motor and speech disabled persons: Design methodology and prototype testing

Posted on:1989-04-02Degree:Ph.DType:Dissertation
University:Northeastern UniversityCandidate:Sy, Bon KiemFull Text:PDF
GTID:1478390017954809Subject:Engineering
Abstract/Summary:
A computer-based nonvocal communication device is an aid to assist nonvocal, motor disabled persons in generating written and spoken messages through an "interrogation" process. This process is executed by suggesting a list of probable messages for the user to affirm, deny or select from. The message search strategy used in existing devices is usually based on fixed statistical information (such as the occurrence frequency of a message). This existing approach is ineffective when a user attempts to use the device to generate infrequently used messages (i.e. messages with low likelihood or low occurrence frequency).; This ineffectiveness is overcome, at least in part, through the development of a Knowledge-based Message Generation System (KMGS) which can "reason out" a desired message. The reasoning process is based on the embedded knowledge concerning the likelihood and the grammatical structure of the message, as well as any sources of information (such as partially understandable speech) provided to the system. The new approach can effectively seek out the desired message even if it is infrequently used, and thus improves the communication rate of the device.; The current design of KMGS makes use of an entropy measurement in the selection of message elements for the knowledge base, which optimizes both the articulateness and the fluency of the system. The knowledge base of the system is a language graph which is encoded with English grammatical rules and message elements. The search for the message elements is conceptualized as a path search in the language graph, and a special frame architecture is used to construct and partition the graph. Bayesian belief reasoning from the Dempster-Shafer Theory of Evidence is augmented to cope with the time varying evidence so that both the information from external sources (such as a speech recognition system), and the embedded knowledge, can be used to optimize the process of message search. An "information fusion" strategy is introduced to integrate various forms of external information. Experimental testing results of the prototype system are reported.
Keywords/Search Tags:Message, System, Information, Speech
Related items