<?xml version="1.0" standalone="yes"?>
<Paper uid="W05-1604">
  <Title>Real-Time Stochastic Language Generation for Dialogue Systems</Title>
  <Section position="10" start_page="7" end_page="7" type="concl">
    <SectionTitle>
7 Conclusion
</SectionTitle>
    <Paragraph position="0"> Stochastic approaches to natural language processing are often criticized for being too slow, particularly in recent attempts in language generation. This paper describes Acorn, a system that generates dialogue utterances in an average of 0.37 seconds. The approach and its additional advances in word forest creation were described, such as a technique called trickle-down features that allow a grammar to pass head/foot features through a generation input, enabling language phenomena such as wh-movement to be represented.</Paragraph>
    <Paragraph position="1"> The grammar syntax and an evaluation of the coverage in an unknown domain were presented. The coverage is comparable and the runtime drastically out-performs previous approaches. null A Example Semantic and Lexical Input Below is an example utterance from the Monroe corpus and its purely semantic and lexical input to Acorn. In this example, only the words have, helicopter, and Strong Memorial are absent in the semantic input. The resulting generation output from Acorn is also shown.</Paragraph>
    <Paragraph position="2"> Original utterance: 'and i also have a helicopter at strong memorial'  Below are some target and generated utterances from Acorn, illustrating several common errors, and are not examples of success. The first utterance is the real target one, and the  second is the Acorn generated utterance.</Paragraph>
    <Paragraph position="3"> 1. &amp;quot;i think i have a disability with maps&amp;quot; &amp;quot;i think i have disability with maps&amp;quot; 2. &amp;quot;they should have stayed in front of the tv&amp;quot; &amp;quot;in a front of the tv should stay they&amp;quot; 3. &amp;quot;and i also have a helicopter at strong memorial&amp;quot; &amp;quot;and i have a helicopter also at strong memorial&amp;quot; 4. &amp;quot;i can't see it on the map&amp;quot; &amp;quot;i can not on the map see it&amp;quot; 5. &amp;quot;probably all of them are hospitals&amp;quot; &amp;quot;probably hospitals are all them&amp;quot; 6. &amp;quot;are you talking to me&amp;quot; &amp;quot;are you talking me&amp;quot; 7. &amp;quot;and there are three people on a stretcher at the airport &amp;quot; &amp;quot;and three people on a stretcher are at the airport&amp;quot; 8. &amp;quot;then there's one stretcher patient at the mall&amp;quot; &amp;quot;then stretcher one patient is at the mall&amp;quot; 9. &amp;quot;so that guy should just walk to the hospital&amp;quot;  &amp;quot;so that guy should walk to the hospital just&amp;quot; 10. &amp;quot;i think that's a very good plan&amp;quot; &amp;quot;i think that is very good plan&amp;quot; C Example Good Output Below are a list of target utterances that Acorn matched ex- null actly, word for word. It is obviously not a complete list. 1. &amp;quot;i'm not doing this on purpose&amp;quot; 2. &amp;quot;we can bring it to strong memorial&amp;quot; 3. &amp;quot;it's on elmwood and mount hope &amp;quot; 4. &amp;quot;so the heart attack person can't go there&amp;quot; 5. &amp;quot;and bring them to saint mary's&amp;quot; 6. &amp;quot;do you have any suggestions?&amp;quot; 7. &amp;quot;we can put him in one ambulance&amp;quot; 8. &amp;quot;because we have only six wounded&amp;quot; 9. &amp;quot;i think that's a good idea&amp;quot; 10. &amp;quot;and the other one is at the airport&amp;quot; 11. &amp;quot;what can i say?&amp;quot;</Paragraph>
  </Section>
class="xml-element"></Paper>