The BBN Spoken Language System

Abstract

We describe HARC, a system for speech understanding that integrates speech recognition techniques with natural language processing. The integrated system uses statistical pattern recognition to build a lattice of potential words in the input speech. This word lattice is passed to a unification parser to derive all possible associated syntactic structures for these words. The resulting parse structures are passed to a multi-level semantics component for interpretation. 6 19a. NAME OF RESPONSIBLE PERSON

Development of a spoken language system

“…WE appears on both ROOT-S and S, since they may be either statements or questions, but does not appear on REL-S, since it has no option. REL-S contains a single TRACE flag, to bind the relativized position within the clause to the head of the relative, while S has two, to provide the correct links between extracted WH phrases and their traces via the well-known difference list [73] or "trace-threading" [24] mechanism. ROOT-S has no need for this feature, since, by definition, there is no place external to a matrix clause to bind a trace.…”

“…As was mentioned in [24], VP in our former grammar war used to handle the following types of constituents: …”

“…Chapter 1 provides a general 5 overview of the current architecture of HARC and details the changes from the version described in [24]. Chapter 2 describes the grammar formalism, concentrating on the aspects of our approach to syntax that provide for maximum flexibility and coverage.…”

“…In this chapter we describe the developments in semantics during the course of the project A major change from our earlier work is that interpretation in DELPHI has shifted from a Montague Grammar [65] style rule-for-rule approach to one that carries out semantic interpretation directly during the course of the parse, rather than as a post-process on a finished parse tree [24,25]. Meaning representations are thereby constructed, and semantic filtering constraints applied, as part of parsing the utterance.…”

“…In order to eliminate spurious candidates as the antecedents of pronouns, we use syntactic constraints on intra-sentential anaphora, such as those described in [46] and [24], as part of our anaphora resolution mechanism. Centering algorithms …”