<?xml version="1.0" standalone="yes"?>
<Paper uid="W05-0201">
  <Title>Applications of Lexical Information for Algorithmically Composing Multiple-Choice Cloze Items</Title>
  <Section position="9" start_page="7" end_page="7" type="concl">
    <SectionTitle>
8 Concluding Remarks
</SectionTitle>
    <Paragraph position="0"> We believe that NLP techniques can play an important role in computer assisted language learning, and this belief is supported by papers in this workshop and the literature. What we have just explored is limited to the composition of cloze items for English vocabulary. With the assistance of WSD techniques, our system was able to identify sentences that were qualified as candidate cloze items 65% of the time.</Paragraph>
    <Paragraph position="1"> Considering both word frequencies and collocation, our system recommended distractors for cloze items, resulting in items that had unique answers 90% of the time. In addition to assisting the composition of cloze items in the printed format, our system is also capable of helping the composition of listening cloze items. The current system considers features of phonemes in computing distances between pronunciations of different word strings.</Paragraph>
    <Paragraph position="2"> We imagine that NLP and other software techniques could empower us to create cloze items for a wide range of applications. We could control the formats, contents, and timing of the presented material to manipulate the challenging levels of the test items.</Paragraph>
    <Paragraph position="3"> As we have indicated in Section 7, cloze items in the listening format are harder than comparable items in the printed format. We can also control when and what the students can hear to fine tune the difficulties of the listening cloze items.</Paragraph>
    <Paragraph position="4"> We must admit, however, that we do not have sufficient domain knowledge in how human learn languages. Consequently, tools offered by computing technologies that appear attractive to computer scientists or computational linguists might not provide effective assistance for language learning or diagnosis. Though we have begun to study item comparison from a mathematical viewpoint (Liu, 2005), the current results are far from being practical. Expertise in psycholinguistics may offer a better guidance on our system design, we suppose.</Paragraph>
  </Section>
class="xml-element"></Paper>