Fragment Processing in the DELPHI System

Fragment Processing in the DELPHI System David Stallard, Robert Bobrow BBN Systems and Technologies 10 Moulton St. Calnbfidge, MA 02138 ABSTRACT This paper presents the fallback understanding component of BBN's DELPHI NL sysystem. This component is invoked when the core DELPHI system is unable to understand an input. It incorporates both syntax- and frame-based fragment combination sub-components, in an attempt to provide a smoother path from accurate but fragile conventional parsers on the one hand to the robust but less accurate schema-based methods on the other. The frame-based sub-component is fully integrated with the DELPHI's core grammar and parser, and represents an advance over previous proposals. The complete fallback understanding component, incorporating both sub-components, was used in the February 1992 NL and SLS evaluations of the DELPHI system and we report on its contribu- tion to these results, and those of its two separate sub-components. For SLS, use of the frame-based sub-component alone resulted in a figure 39.2% Weighted Error--signifigantly lower than our lowest official score of 43.7% Weighted Error. 1. INTRODUCTION We describe the fallback understanding component of the DELPHI Natural Language component of BBN's Spoken Language System. This component is invoked when DEL- PHI's regular chart-based unification grammar parser is un- able to parse an input; it attempts to come up with a parse and semantic interpretation, or a semantic interpre- tation alone, based on a fragmentary analysis of the input. The fallback understanding component consists of three sep- arate stages, which are invoked successively. First, the Frag- ment Generator produces a sequence of fragmentary sub- parses from the chart state left over from the unsuccess- ful parse. Next, two different combination modules---the Syntactic Combiner and Frame Combiner - employ alterna- tive and complementary strategies for combining these frag- ments. The Syntactic Combiner uses extended grammar rules that can sldp over intervening material to combine constituents in an attempt to re-construct a plausible parse of the input. This parse can be a clause or some other useful constituent such as an imperative VP. A semantic interpretation for this reconstructed parse is automatically provided through the 305 action of the grammar rules. The Frame Combiner is invoked when the Syntactic Com- biner is unsuccessful. It utilizes a set of pragmatic slot-filling schemata that embody the goals that ATIS users most com- monly have, such as finding a flight or fare that satisfies some set of constraints, or asking about ground transporta- tion between an airport or a city. As such, it detennines only a semantic interpretation and not a parse. The intent of this multi-step approach to fallback process- ing is to provide a smoother path between the accuracy but fragility of regular parsing on the one hand, and the robust- ness but possible inaccuracy of schemata-based methods on the other. The remainder of the paper is taken up with detailed de- scription of each component. The next section, Section 2, describes the Fragment Generator. Section 3 describes the Syntactic Combiner and Section 4 the Frame Combiner. Fi- nally …

Content truncated to 500 words