MultiTurnCleanup: A Benchmark for Multi-Turn Spoken Conversational Transcript Cleanup
This addresses readability and NLP performance issues in conversational transcripts, but is incremental as it builds on existing disfluency detection work.
The study tackled the problem of multi-turn disfluencies in spoken conversational transcripts by introducing the Multi-Turn Cleanup task and creating the MultiTurnCleanup1 dataset, with experimental benchmarks provided for future research.
Current disfluency detection models focus on individual utterances each from a single speaker. However, numerous discontinuity phenomena in spoken conversational transcripts occur across multiple turns, hampering human readability and the performance of downstream NLP tasks. This study addresses these phenomena by proposing an innovative Multi-Turn Cleanup task for spoken conversational transcripts and collecting a new dataset, MultiTurnCleanup1. We design a data labeling schema to collect the high-quality dataset and provide extensive data analysis. Furthermore, we leverage two modeling approaches for experimental evaluation as benchmarks for future research.