<?xml version="1.0" standalone="yes"?>
<Paper uid="W03-0426">
  <Title>Named Entity Recognition with Long Short-Term Memory</Title>
  <Section position="5" start_page="0" end_page="0" type="concl">
    <SectionTitle>
5 Conclusion
</SectionTitle>
    <Paragraph position="0"> A LSTM network was trained on named entity recognition, yielding an fscore just above the baseline performance on English and significantly above baseline for German. Whilst the just-above-baseline performance for English is disappointing, it is hoped that further work will improve on these results. A number of ways of boosting performance will be looked at including: a0 Increasing the size of the hidden layers will increase the power of the networks at the risk of overfitting.</Paragraph>
    <Paragraph position="1"> Increasing training times may also increase performance, again at the risk of overfitting.</Paragraph>
    <Paragraph position="2"> a0 Increasing the informativeness of the lexical representations. Given that the number of elements used here is less than the number of characters in the character sets, there should be some scope for boosting performance by increasing the size of the SARD-NETs. The representations of different words will then become more distinct from each other.</Paragraph>
    <Paragraph position="3"> The lexical space vectors were derived from a context of +/- 1 word, where in earlier work on clause splitting a context of +/- 2 words was used. Using the larger context and/or using more than 25 of the top principal components may again boost performance by incorporating more information into the vectors.</Paragraph>
    <Paragraph position="4"> a0 Further exploitation of the word lists. Whilst the networks are made aware of which categories of named entity the current word can belong to, it is not made aware of how many named entities it belongs to or of what positions on the named entities it could occupy. null</Paragraph>
  </Section>
class="xml-element"></Paper>