<?xml version="1.0" standalone="yes"?>
<Paper uid="W03-0202">
  <Title>Learning to Identify Student Preconceptions from Texta0</Title>
  <Section position="3" start_page="0" end_page="0" type="relat">
    <SectionTitle>
2 Related Work
</SectionTitle>
    <Paragraph position="0"> There have been a number of approaches to essay and free-response grading. Burstein et al. (1999) developed a system that uses a per-question lexicon and broad-coverage parser to analyze free-response answers on a sentence-by-sentence basis. It determines whether the responses contain items from a rubric describing specific points a student must touch upon in their answer. This system uses a deeper semantic analysis than does ours and makes explicit use of syntactic structure. On the other hand, it requires the semi-automated construction of a lexicon for each question. Our system only requires labeled responses as training data.</Paragraph>
    <Paragraph position="1"> The LSA group at the University of Colorado at Boulder has developed a system based on Latent Semantic Analysis (Landauer and Dumais, 1997). It uses a text similarity metric and a corpus of essays of known quality.</Paragraph>
    <Paragraph position="2"> The system is primarily intended to identify a student's general level of understanding of a topic and recommend an appropriate text for the student to learn from, but has also been used for essay grading (Wolfe et al., 1998).</Paragraph>
    <Paragraph position="3"> They use the similarity metric to determine whether essays have enough detail in various subtopics that the essay is expected to cover. Because of the statistical properties of the singular value decomposition underlying LSA, this system requires relatively large amounts of data to be trained, and works best on long essay questions, rather than short-answer responses.</Paragraph>
    <Paragraph position="4"> The primary difference between these approaches and ours is that these systems are intended to determine whether or not the student has discussed particular concepts and, in the case of the Wolfe et al. paper, the depth of that discussion. However, neither is aimed at identifying the specific preconceptions held by a student.</Paragraph>
  </Section>
class="xml-element"></Paper>