<?xml version="1.0" standalone="yes"?>
<Paper uid="H89-2031">
  <Title>PORTING TO NEW DOMAINS USING THE LEARNER *m</Title>
  <Section position="3" start_page="0" end_page="0" type="intro">
    <SectionTitle>
INTRODUCTION
</SectionTitle>
    <Paragraph position="0"> A major problem that restricts the usefulness of natural language processing systems is the cost, in time and effort, of porting a system to a new domain. Typically, the 1 Learner and Parlance are trademarks of Bolt Beranek and Newman Inc.</Paragraph>
    <Paragraph position="1"> system designers are not experts in the application domain in which their system will be used, although they have the knowledge of syntax, semantics, and knowledge representation to configure the system. The end users, on the other hand, are experts in the application domain, but are almost always unversed in computational linguistics. Thus, the people who can most efficiently modify the system do not know what is necessary for a given domain, while the experts in that domain do not have the linguistic knowledge to perform the modifications themselves. Typically, then, the system designers spend long periods of time, on the order of months or even years, trying to acquire expertise in a domain and &amp;quot;tweaking&amp;quot; their systems to perform well in it, or else they try to impart just enough knowledge of syntax and semantics to the end users to allow them to do the configuration themselves. At best, these approaches result in a working version of the system, but at enormous cost; at worst, the end product is a not very usable system.</Paragraph>
    <Paragraph position="2"> This problem is sometimes referred to as the &amp;quot;knowledge acquisition bottle neck&amp;quot;: how can the knowledge of an application area be combined with the knowledge of a working system and its component technologies to produce a useful system without placing heavy burdens on either the end users or the system designers?</Paragraph>
  </Section>
class="xml-element"></Paper>