<?xml version="1.0" standalone="yes"?>
<Paper uid="W02-0102">
  <Title>An Interactive Spreadsheet for Teaching the Forward-Backward Algorithm</Title>
  <Section position="12" start_page="0" end_page="0" type="concl">
    <SectionTitle>
13 Availability
</SectionTitle>
    <Paragraph position="0"> The spreadsheet (in Microsoft Excel) and assignment are available at http://www.cs.jhu.</Paragraph>
    <Paragraph position="1"> edu/~jason/papers/#tnlp02.</Paragraph>
    <Paragraph position="2"> Also available is a second version of the spreadsheet, which uses the Viterbi approximation for decoding and reestimation. The Viterbi algorithm is implemented in an unconventional way so that the two spreadsheets are almost identical; there is no need to follow backpointers. The probability of the best path through state H on day 3 is 3(H) 3(H), where and are computed like and but maximizing rather than summing path probabilities. The Viterbi approximation treats p3(H) as 1 or 0 according to whether 3(H) 3(H) equals max( 3(C) 3(C); 3(H) 3(H)).</Paragraph>
    <Paragraph position="3"> Have fun! Comments are most welcome.</Paragraph>
  </Section>
class="xml-element"></Paper>