Summary

The development and coverage of NL-Soar has been driven mainly by the various applications of the system. Each application has linguistic (i.e. grammatical, semantic, and pragmatic) characteristics which may be ideosyncratic to it or which may be shared by other applications. Our approach to handling the language required in a particular domain is to collect a corpus of utterances in that domain and to extend NL-Soar to cover that corpus.

This is an index into the corpora that the NL9702 release of NL-Soar handles.

• Regression This corpus contains a list of sentences which exemplify the core syntactic structures that NL-Soar covers. The form of sentences do not reflect a particular domain, but since its goal is to exemplify frequently occuring structures in English we would expect similar sentences to occur in several domains.
• Simultaneous Interpretation. This corpus is a selection of sentences simplified from the Dillinger simultaneous translation corpus.

• TacAir This corpus contains utterances in the tactical air domain.
• Dialogue This corpus is a collection of dialogues which occur in the Tactical Air domain.
• Processing Breakdown This corpus contains sentences testing various problematic and unproblematic sentences, including garden paths and other syntactic ambiguities.
• Second Language Learning This corpus contains a collection of sentence pairs in English and in Polish which require the production of definite and indefinite NPs in various contexts.
• Speech These corpora are from the ATIS Air Traffic Controller task, and from traffic reports recorded from the radio. They were chosen because of the accompanying speech files.

Maintenance

NL-Soar's coverage of these corpora is maintained from release to release by use of a suite of regression testing tools .