On Combining Language Models to Improve a Text-based Human-machine Interface

摘要

This paper concentrates on improving a text-based human-machine interface integrated into a robotic wheelchair. Since word prediction is one of the most common methods used in such systems, the goal of this work is to improve the results using this specific module. For this, an exponential interpolation language model (LM) is considered. First, a model based on partial differential equations is proposed; with the appropriate initial conditions, we are able to design a interpolation language model that merges a word-based n-gram language model and a part-of-speech-based language model. Improvements in keystroke saving (KSS) and perplexity (PP) over the word-based n-gram language model and two other traditional interpolation models are obtained, considering two different task domains and three different languages. The proposed interpolation model also provides additional improvements over the hit rate (HR) parameter.