An associative semantic model for text processing

Natural language texts have an underlying structure that conveys an essential part of their information content. In order to better exploit text resources, this structure must be rendered explicit, which requires an automatic analysis based on local context and general world knowledge. The analysis must closely match the expectations of a typical reader. The paper presents a computational model that is able to emulate some fundamental aspects of human semantic processing and preference heuristics. It is based on a psycholinguistic motivated associative network that highlights the role of memory as a predictive context for the interpretation of natural language utterances.