<?xml version="1.0" standalone="yes"?>
<Paper uid="A00-2041">
  <Title>A Framework for Robust Semantic Interpretation</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> In order for an approach to robust interpretation to be practical it must be efficient, address the major types of disfluencies that plague spontaneously produced language input, and be domain independent so thatachieving robustness in a particular domain does not require an additional knowledge engineering effort. This paper describes AUTOSEM, a semantic interpretation framework that possesses these three qualities. While previous approaches to robust interpretation have offered robust parsers paired with separate repair modules~ with separate knowledge sources for each, AUTOSEM is a single unified framework that can operate both at parse time and repair time. AUTOSEM is integrated with the LCFLEx robust parser (Ros@ and Lavie, to appear; Lavie and Ros@, 2000). Together AUTOSEM and LCFLEx constitute the robust understanding engine within the CARMEL natural language understanding component developed in the context of the Atlas intelligent tutoring project (Freedman at al., to appear). The evaluation reported here demonstrates that AUTOSEM's repair approach operates 200 times faster than the most similar competing approach while producing hypotheses of better quality.</Paragraph>
    <Paragraph position="1"> AUTOSEM provides an interface to allow semantic interpretation to operate in parallel with syntactic interpretation at parse time in a lexicon driven fashion. Domain specific semantic knowledge is encoded declaratively within a meaning representation specification. Semantic constructor functions are compiled automatically from this specification and then linked into lexical entries as in the Glue Language Semantics approach to interpretation (Dalrymple, 1999). Based on syntactic head/argument relationships assigned at parse time, the constructot functions enforce semantic selectiona\] restrictions and assemble meaning representation structures by composing the meaning representation associated with the constructor function with the meaning representation of each of its arguments.</Paragraph>
    <Paragraph position="2"> AUTOSEM first attempts to construct analyses that satisfy both syntactic and semantic well-formedness conditions. The LCFLEx parser has the ability to efficiently relax syntactic constraints as needed and as allowed by its parameterized flexibility settings. For sentences remaining beyond the parser's coverage, AUTOSEM's repair algorithm relies entirely on semantic knowledge to compose the partial analyses produced by the parser. Each semantic representation built by AUTOSEM's interpretation framework contains a pointer to the constructor function that built it. Thus, each partial analysis can be treated as a constructor function with built in knowledge about how the associated partial analysis can be combined with other partial analyses in a semantically meaningful way. Genetic programming search (Koza, 1992; Koza, 1994) is used to efficiently compose the fragments produced by the parser. The function definitions compiled from the meaning representation specification allow the genetic search to use semantic constraints to make effective use of its search space. Thus, AUTOSEM operates efficiently, free of any hand coded repair rules or any knowledge specifically dedicated to repair unlike other approaches to recovery from parser failure (Danieli and Gerbino, 1995; Van Noord, 1997; Kasper et al., 1999).</Paragraph>
  </Section>
class="xml-element"></Paper>