Universität Tübingen
Adaptive word prediction and its application in an assistive communication system
Abstract
This thesis investigates the capacities of adaptive methods for word prediction. We present and evaluate several adaptation methods: First, we consider strategies enabling to adapt to the lexical and syntactic preferences of the user of an AAC system. Here we investigate the cache model, an auto-adaptive user lexicon and the dynamic user model (DUM), which integrates every input of the user. The second class of methods aims to adapt to the semantic context. Here we focus in particular on Latent Semantic Analysis (LSA), a vectorial model establishing semantic similarity from distributional properties. In the last part an assistive communication system is presented that implements the previously investigated adaptation methods. After a description of the user interface we report results from the application of this system in a rehabilitation center.
Author and committee
dc:creator, dc:contributor.*- Author
-
- Wandmacher, Tonio
Identifiers
dc:identifier.*- Identifier
- hdl:10900/46325