<?xml version="1.0" standalone="yes"?>
<Paper uid="A00-1015">
  <Title>J avox: A Toolkit for Building Speech-Enabled Applications</Title>
  <Section position="8" start_page="109" end_page="110" type="evalu">
    <SectionTitle>
6 Discussion and Future Work
</SectionTitle>
    <Paragraph position="0"> In practice, building a JAvox-based, speech interface - for limited-functionality applications - is straightforward and reasonably quick. To date, we have used three diverse applications as our test platforms. Speech-enabling the last of these, an image manipulation program, took little more than one person-day. Though these applications have been small; we are beginning to explore JAvOX's scalability to larger applications. We are also developing a library of JAVOX grammars for use with the standard Java classes. This resource will shorten development times even more; especially compared to building a SLS from the ground up.</Paragraph>
    <Paragraph position="1"> One of the existing challenges is to work with applications consisting entirely of dynamic objects, those that cannot be identified at load time. Some typical dynamic-object applications are drawing programs or presentation software; in both cases, the user creates the interesting objects during runtime. We have implemented a system in JSL which allows objects to be filtered based on an attribute, such as color in the utterance: Move the blue square.</Paragraph>
    <Paragraph position="2"> In situations where there is a one-to-one correlation between a lexical item in the grammar and an object in the program, it is often the case that the lexical item is very similar to the element's identifier. It is quite often the same word or a direct synonym. Since JAVOX is primarily performing upcalls based on existing functions within the program, it also can be predicted what type of objects will co-occur in utterances. In the Mr. Potato Head applio  cation, we can assume that objects representing a Point or integers will occur when the user speaks of moving a BodyPart. We are developing a system to exploit these characteristics to automatically generate JAVOX grammars from an application's compiled code. The automatically-generated grammars are intended to serve as a starting point for developers - though they may certainly require some hand crafting. Our current, grammar-generation tool assumes the program is written with Java's standard naming conventions. It is imaginable that additional data sources - such as a sample corpus - will allow us to more accurately generate grammars for an application. Though in its infancy, we believe this approach holds vast potential for SLS development.</Paragraph>
  </Section>
class="xml-element"></Paper>