Knowledge extraction using dynamical updating of representation

We present a system that extracts knowledge from the textual content of documents. The acquired knowledge is represented through an associative network, that is dynamically updated by the integration of a contextualized structure representing the content of the new analysed document. Grounded on the basis of "long term working memory" theory by W. Kintsch and K. A. Ericsson, our system makes use of a scale free graph model to update the final knowledge representation. This knowledge acquisition system has been validated by first experimental results.