<?xml version="1.0" standalone="yes"?>
<Paper uid="N06-2001">
  <Title>Factored Neural Language Models</Title>
  <Section position="9" start_page="2" end_page="2" type="concl">
    <SectionTitle>
8 Conclusion
</SectionTitle>
    <Paragraph position="0"> We have introduced FNLMs, which combine neural probability estimation with factored word representations and different ways of inferring continuous word features for unknown factors. On sparse-data Arabic and Turkish language modeling task FNLMs were shown to outperform all comparable models (standard backoff 3-gram, word-based NLMs) except FLMs in isolation, and all models when interpolated with the baseline. These conclusions apply to both open and closed vocabularies.</Paragraph>
  </Section>
class="xml-element"></Paper>