Adapting the NICT-JLE Corpus for Disfluency Detection Models

TL;DR

This paper adapts the NICT-JLE corpus of learner speech into a standardized dataset for disfluency detection, enabling better model evaluation and comparison in the context of non-native speech.

Contribution

It introduces a method to transform the NICT-JLE corpus for disfluency detection, addressing access restrictions and facilitating future research.

Findings

Created a standardized dataset for learner speech disfluency detection.

Compared NICT-JLE with Switchboard to identify key differences.

Provided train, validation, and test sets for future research.

Abstract

The detection of disfluencies such as hesitations, repetitions and false starts commonly found in speech is a widely studied area of research. With a standardised process for evaluation using the Switchboard Corpus, model performance can be easily compared across approaches. This is not the case for disfluency detection research on learner speech, however, where such datasets have restricted access policies, making comparison and subsequent development of improved models more challenging. To address this issue, this paper describes the adaptation of the NICT-JLE corpus, containing approximately 300 hours of English learners' oral proficiency tests, to a format that is suitable for disfluency detection model training and evaluation. Points of difference between the NICT-JLE and Switchboard corpora are explored, followed by a detailed overview of adaptations to the tag set and meta-features of the NICT-JLE corpus. The result of this work provides a standardised train, heldout and test set for use in future research on disfluency detection for learner speech.