<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-2011">
  <Title>Knowledge Extraction Using Dynamical Updating of Representation</Title>
  <Section position="7" start_page="0" end_page="0" type="concl">
    <SectionTitle>
6 Conclusions
</SectionTitle>
    <Paragraph position="0"> A new system for the automatic acquisition of the knowledge has been presented. It is based on the concept of long term working memory developed by Kintsch and Ericsson.</Paragraph>
    <Paragraph position="1"> The system updates an associative network (LTM) whose structure varies dynamically in time on the basis of the textual content of the analyzed documents. During the analysis of each new document the LTM can be queried by the simple procedure of the diffusion of the activation signal developed by Kintsch and Ericsson. In this way the context of the document can be easily and exactly identified.</Paragraph>
    <Paragraph position="2"> To reduce the computational time we have implemented the WM block with a scale free graph model. The obtained network is used to update the content of the LTM.</Paragraph>
    <Paragraph position="3"> Some analyses have been performed over the WM model developed. The results have confirmed that the network evolves as a scale free graph. Also the LTM graphs seems to keep the scale free features, and their coherence rate indicates that the system conceptualizes the terms according to a precise inner schema.</Paragraph>
    <Paragraph position="4"> Now we are considering alternative models for the WM that use much more information present in the LTM and that guarantee more plasticity to its structure. We are also going to compare the LTM graphs with the knowledge structures obtained by the Pathfinder analysis computed over the associations provided by a group of human subjects.</Paragraph>
  </Section>
class="xml-element"></Paper>