Using BERT Embeddings to Model Word Importance in Conversational Transcripts for Deaf and Hard of Hearing Users

TL;DR

This paper explores using BERT embeddings to better assess word importance in transcripts for deaf and hard of hearing users, aiming to improve caption evaluation metrics by aligning them with user preferences.

Contribution

It introduces a method to model word importance using BERT embeddings and provides a dataset pairing embeddings with human importance annotations, enhancing caption evaluation.

Findings

BERT embeddings correlate better with human importance scores than word2vec.

A new dataset pairing embeddings with importance scores is provided.

A proof-of-concept model achieves an F1-score of 0.57 in classifying word importance.

Abstract

Deaf and hard of hearing individuals regularly rely on captioning while watching live TV. Live TV captioning is evaluated by regulatory agencies using various caption evaluation metrics. However, caption evaluation metrics are often not informed by preferences of DHH users or how meaningful the captions are. There is a need to construct caption evaluation metrics that take the relative importance of words in a transcript into account. We conducted correlation analysis between two types of word embeddings and human-annotated labeled word-importance scores in existing corpus. We found that normalized contextualized word embeddings generated using BERT correlated better with manually annotated importance scores than word2vec-based word embeddings. We make available a pairing of word embeddings and their human-annotated importance scores. We also provide proof-of-concept utility by training word importance models, achieving an F1-score of 0.57 in the 6-class word importance classification task.