Carol's Song is a program to help the speech impaired communicate. The distinguishing
feature of the program is that it quickly retrieves simple messages while also assisting
in the development of more complex and original statements. The program adapts to
usage, predicting future words and messages based on selections made in the past.
Current speech aids require too much input for an ALS patient to participate in a
conversation without lengthy delays while text is entered. The goal for Carol's Song
is to reduce input by 80%. This will be accomplished without resorting to lengthy
abbreviation lists that are hard to memorize or severely restricting the vocabulary
and sentence structure of the final message.
At the heart of the program is the Bayesian Adaptive Predictor Algorithm. This algorithm
takes the most relevant information from an input string and generates complete words,
phrases, and even entire sentences. This is augmented by a sophisticated context generator
that creates grammatical structures based on the words presented so far.