<?xml version="1.0" standalone="yes"?>
<Paper uid="W03-0607">
  <Title>EBLA: A Perceptually Grounded Model of Language Acquisition</Title>
  <Section position="7" start_page="75" end_page="75" type="concl">
    <SectionTitle>
5 Conclusion
</SectionTitle>
    <Paragraph position="0"> While there have been several systems capable of learning object or event labels for videos, EBLA is the first known system to acquire both nouns and verbs using a grounded computer vision system. In addition, because EBLA operates in an online fashion, it does not require an explicit training phase.</Paragraph>
    <Paragraph position="1"> EBLA performed very well on the entity-lexeme mapping task for both the animations and the videos, achieving success rates as high as 100% and 95.8% respectively. EBLA was also able to generate descriptions for the animations and videos with average accuracies as high as 96.7% and 65.3%. The 65.3% is still quite good when compared to the approximately 15% average success rate obtained by generating three word descriptions at random from the pool of nineteen lexemes processed by EBLA.</Paragraph>
    <Paragraph position="2"> While the initial results from the EBLA system are encouraging, much development and evaluation remains to be done. Adding new attribute calculators along with a mechanism for dropping extraneous attributes would likely make EBLA's entity definitions more robust and facilitate the acquisition of additional nouns and verbs as well as other parts of speech. Since there is nothing in the design of EBLA that prevents it from processing videos with more than three entities/lexemes, it should be thoroughly tested using more complex experiences and/or descriptions As mentioned in the introduction, one of the primary goals of EBLA has been to develop an open system that would be relatively easy for others to use and extend.</Paragraph>
    <Paragraph position="3"> To that end, EBLA was written entirely in Java with a PostgreSQL relational database for storage of all experience parameters, intermediate results, attribute definitions and values, lexemes, entity definitions, and entity-lexeme mappings. EBLA has been released on SourceForge at http://sourceforge.net/projects/ebla/.</Paragraph>
    <Paragraph position="4"> For more information on EBLA, visit http://www.greatmindsworking.com</Paragraph>
  </Section>
class="xml-element"></Paper>