This abstract describes the structure and operation of SOUL (for Semantically-Oriented Understanding of Language), which uses semantic and pragmatic knowledge to correct, reject and/or clarify the outputs of the Phoenix speech and speech transcript caseframe parser in the Darpa air travel or ATIS domain. In this abstract, we summarize both the linguistic phenomena which Soul addresses, and how it works to correct inaccurate parses, and we present the results of three pilot evaluations of independent test sets and the decrease in error rates attributed to the post-processing. In pilot experiments, post-processing by SOUL on spontaneous speech transcripts reduces the error rate of the output of PHOENIX by 54 to 84 percent, depending on the type of utterances that are evaluated. The paper will present results on both spontaneous and clean speech in both recognition and transcript processing modes.
Cite as: Young, S. (1991) Using semantics to correct parser output for ATIS utterances. Proc. 2nd VENACO Workshop - The Structure of Multimodal Dialogue, 61-62
@inproceedings{young91_smmd,
author={Sheryl Young},
title={{Using semantics to correct parser output for ATIS utterances}},
year=1991,
booktitle={Proc. 2nd VENACO Workshop - The Structure of Multimodal Dialogue},
pages={61--62}
}