| Sponsor | Gregory Abowd
abowd@cc.gatech.edu 380 CRB |
| Area | HCI, Software Engineering and Future Computing Environments |
Problem
Experience shows that currently a goal of 100% recognition rate in
natural input systems (e.g. speech, handwriting) is an unattainable. However,
it is well-known that smaller vocabularies result in higher recognition
rates. Although most natural input tasks (e.g. writing a paper) cannot
take advantage of this fact, a few interesting ones are more well structured.
We are designing a natural input system for writing source code for use by programmers with disabilities such as RSI. Since programming code is highly structured, we hope to use this structure to improve recognition. This project involves 2 parts: writing a simple interface for the user to write natural ink in, and writing a parser which uses information about the grammar of the input language to limit the vocabulary. As proof of concept, we are using a simple arithmetic grammar. The follow sets of its elements can be used to learn the possible input symbols at any moment. Write a parser that will take the current state and generate a vocabulary for the next input. From there you have several choices: hook it up to a gesture recognizer we provide (gdt) and an editor. Or extend theparser to handle uncertainty (what's the follow set of 1 if I'm not sure whether 1 is ) or 1.
Background