<?xml version="1.0" standalone="yes"?>
<Paper uid="W05-0506">
  <Title>A Second Language Acquisition Model Using Example Generalization and Concept Categories</Title>
  <Section position="8" start_page="49" end_page="50" type="concl">
    <SectionTitle>
7 Discussion
</SectionTitle>
    <Paragraph position="0"> We have presented a computational model of second language acquisition. SLA is a central subject in linguistics theory and practice, and our main contribution is in addressing it in computational linguistics. The model's learning algorithms are unique in their usage of a conceptual system, and  its generative capacity is unique in its support for degrees of certainty. The model was tested on a unique corpus.</Paragraph>
    <Paragraph position="1"> The dominant trend in CL in the last years has been the usage of ever growing corpora. We have shown that meaningful learning can be achieved from a small corpus when the corpus has been prepared by a 'good teacher'. Automatic identification (and ordering) of corpora subsets from which learning is effective should be a fruitful research direction for CL.</Paragraph>
    <Paragraph position="2"> We have shown that using a simple conceptual system can greatly assist language learning algorithms. Previous FLA algorithms have in effect computed a CS simultaneously with the syntax; decoupling the two stages could be a promising direction for FLA.</Paragraph>
    <Paragraph position="3"> The model presented here is the first computational SLA model and obviously needs to be extended to address more SLA phenomena. It is clear that the powerful notion of certainty is only used in a rudimentary manner. Future research should also address constraints (e.g. for morphology and agreement), recursion, explicit semantics (e.g. parsing into a semantic representation), word segmentation, statistics (e.g. collocations), and induction of new concept categories that result from the learned language itself (e.g. the Japanese counting system). An especially important SLA issue is L1 transfer, which refers to the effect that the L1 has on the learning process. In this paper the only usage of the L1 part of the examples was for accessing a conceptual system. Using the L1 sentences (and the existing conceptual system) to address transfer is an interesting direction for research, in addition to using the L1 sentences for modeling sentence semantics. null Many additional important SLA issues will be addressed in future research, including memory, errors, attention, noticing, explicit learning, and motivation. We also plan additional applications, such as automatic lesson generation.</Paragraph>
    <Paragraph position="4"> Acknowledgement. We would like to thank Dan Melamed for his comments on a related document.</Paragraph>
  </Section>
class="xml-element"></Paper>