Advanced Prediction Methods for Augmentative Communication

Augmentative and Alternative Communication (AAC) devices can provide an effective avenue of communication for individuals with communicative disabilities. A core component of many AAC systems is dynamic word prediction, in which a list of probable words is autonomously generated and presented for immediate user selection. Current word prediction systems employ algorithms which do not take full advantage of the language context, and are therefore suboptimal. Enkidu Research proposes to apply advanced statistical language models to AAC word prediction. In this approach, an extensive source corpus is analyzed to extract meaningful statistics concerning the probability of all possible sequences of words. By then considering the past sequence of words entered by the user, the database can be cross-indexed to provide context-dependent predictions of subsequent words. This paradigm can be extended to arbitrary language components, such as, syntactic and semantic classes, to exploit additional contextual information. The combinatorial nature of a statistical approach implies that there will often be missing or unreliable entries in the predictive database. These inconsistencies will be avoided using a variety of established and innovative techniques. Performance measures for a large collection of predictive algorithm variants will be compiled and then contrasted with traditional AAC prediction method.