Distinguishing Repetition Disfluency from Morphological Reduplication in Bangla ASR Transcripts: A Novel Corpus and Benchmarking Analysis

2025-11-17

AI Summary

This study addresses the challenge of distinguishing “repetitive disfluencies” (unintentional speech errors) from “morphological reduplication” (intentional, semantically motivated word formation) in Bangla automatic speech recognition (ASR) transcripts. To overcome the scarcity of annotated data, we introduce the first publicly available, linguistically fine-grained corpus of 20,000 Bangla utterances exhibiting repetition phenomena and define a linguistics-informed benchmark task. Methodologically, we integrate linguistic constraints with deep learning: (i) a few-shot prompting strategy leveraging multilingual large language models, and (ii) task-specific fine-tuning of BanglaBERT. Experiments show that the fine-tuned model achieves 84.78% accuracy and an F1-score of 0.677—substantially outperforming few-shot LLM baselines—and establishes the first strong baseline for this task. Our core contribution lies in the first systematic linguistic characterization and computational modeling of the semantic distinctions between these two repetition types, enabling high-fidelity ASR post-processing.

Technology Category

• Natural Language Processing: Lexical Semantics and Morphology
• Machine Learning: Large Multimodal Models (LMMs)
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Automatic Speech Recognition (ASR) transcripts, especially in low-resource languages like Bangla, contain a critical ambiguity: word-word repetitions can be either Repetition Disfluency (unintentional ASR error/hesitation) or Morphological Reduplication (a deliberate grammatical construct). Standard disfluency correction fails by erroneously deleting valid linguistic information. To solve this, we introduce the first publicly available, 20,000-row Bangla corpus, manually annotated to explicitly distinguish between these two phenomena in noisy ASR transcripts. We benchmark this novel resource using two paradigms: state-of-the-art multilingual Large Language Models (LLMs) and task-specific fine-tuning of encoder models. LLMs achieve competitive performance (up to 82.68% accuracy) with few-shot prompting. However, fine-tuning proves superior, with the language-specific BanglaBERT model achieving the highest accuracy of 84.78% and an F1 score of 0.677. This establishes a strong, linguistically-informed baseline and provides essential data for developing sophisticated, semantic-preserving text normalization systems for Bangla.

Problem

• Distinguishing repetition disfluency from morphological reduplication in Bangla ASR
• Solving ambiguity in low-resource language speech recognition transcripts
• Providing annotated corpus and benchmarks for Bangla text normalization

Innovation

• Created first annotated Bangla corpus distinguishing disfluency from reduplication
• Benchmarked multilingual LLMs and fine-tuned encoder models
• Achieved best results with fine-tuned BanglaBERT model