Text Retrieval via Semantic Forests
Patrick Schone, Jeffrey L. Townsend, Thomas H. Crystal, Calvin Olano · 1997
We approached our first participation in TREC with an interest in performing retrieval on the output of automatic speech-to-text (speech recognition) systems and a background in performing topic-labeling on such output. Our primary thrust, therefore, was to participate in the SDR track. In conformance with the rules, we also participated in the Ad Hoc text-retrieval task, to create a baseline for comparing our converted topic-labeling system with other approaches to IR and to assess the effect of speech-transcription errors. A second thrust was to explore rapid prototyping of an IR system, given the existing topic-labeling software. Our IR system makes use of software called Semantic Forests which is based on an algorithm originally developed for labeling topics in text and transcribed speech (Schone & Nelson, ICASSP 96). Topic-labelling is not an IR task, so Semantic Forests was adapted for use in TREC over an eight-week period for the Ad Hoc task, with an additional two weeks for SDR. In what follows, we describe our system as well as experiments, timings, results, and future directions with these techniques.