How context and semantic information can help a machine learning system?

Abstract

In Natural Language Processing there are different problems to solve: lexical ambiguity, summarization, information extraction, speech processing, etc. In particular, lexical ambiguity is a difficult task that nowadays is still open to new approaches. In fact, there is still a lack of systems that solve efficiently this kind of problem. At present, we find two different approaches: knowledge systems and machine learning systems. Recent studies demonstrate that machine learning systems obtain better results than knowledge systems but there is a problem: the lack of annotated contexts and corpus to train the systems. In this work, we try to avoid this situation by combining a new machine learning system with a knowledge based system. © Springer-Verlag Berlin Heidelberg 2007.