<?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="3" start_page="1" end_page="45" type="intro">
    <SectionTitle>
2 Background
</SectionTitle>
    <Paragraph position="0"> We use the term 'second language acquisition' to refer to any situation in which adults learn a new language  . A major concept in SLA theory [Gass01, Mitchell03] is that of interlanguage: when learning a new language (L2), at any given point in time the learner has a valid partial L2 language system that differs from his/her native language(s) (L1) and from the L2. The SLA process is that of progressive enhancement and refinement of interlanguage. The main trigger for interlanguage modification is when the learner notices a gap between interlanguage and L2 forms. In order for this to happen, the learner must be provided with com- null Some SLA texts distinguish between 'second' and 'foreign' and between 'acquisition' and 'learning'. We will not make those distinctions here.</Paragraph>
    <Paragraph position="1">  prehensible input. Our model directly supports all of these notions.</Paragraph>
    <Paragraph position="2"> A central, debated issue in language acquisition is whether FLA mechanisms [Clark03] are available in SLA. What is clear is that SL learners already possess a mature conceptual system and are capable of explicit symbolic reasoning and abstraction. In addition, the amount of input and time available for FLA are usually orders of magnitude larger than those for SLA.</Paragraph>
    <Paragraph position="3"> The general linguistic framework that we utilize in this paper is that of Construction Grammar (CG) [Goldberg95, Croft01], in which the building blocks of language are words, phrases and phrase templates that carry meanings. [Tomasello03] presents a CG theory of FLA in which children learn whole constructions as 'islands' that are gradually generalized and merged. Our SLA model is quite similar to this process.</Paragraph>
    <Paragraph position="4"> In language education, current classroom methods use a combination of formal rules and communicative situations. Radically different is the Pimsleur method [Pimsleur05], an audio-based self-study method in which rules and explanations are kept to a minimum and most learning occurs by letting the learner infer L2 constructs from translations of contextual L1 sentences. Substantial anecdotal evidence (as manifested by learner comments and our own experience) suggests that the method is highly effective. We have used a Pimsleur corpus in our experiments. One of the goals of our model is to assist the authoring of such corpora.</Paragraph>
  </Section>
class="xml-element"></Paper>