Recognizing irony in speech and text can be challenging even for humans. For natural language processing (NLP) applications, irony recognition presents a unique challenge as irony alters the sentiment and meaning of the words themselves. Combining phonological insights from past literature on irony prosody and deep learning modeling, this research presents a new approach to irony classification in naturalistic speech data. A new corpus consisting of nearly five hours of irony-annotated, naturalistic, conversational speech data has been constructed for this study. A wide array of utterance-level and time-series acoustic features were extracted from this data and utilized in the training and fine-tuning of a series of deep learning approaches for irony classification. The best-performing model achieved an area under the curve of 0.811 in the speaker dependent condition, and 0.738 in the speaker independent condition, outperforming most irony classification models in the existing literature. In addition to the myriad real-world applications for this approach, its contribution to the understanding of prosodically-encoded augmentation of semantic content constitutes a significant step forward for research in the fields of linguistics and NLP.
Cite as: Gent, H., Adams, C., Tang, Y., Shih, C. (2022) Deep Learning for Prosody-Based Irony Classification in Spontaneous Speech. Proc. Interspeech 2022, 3993-3997, doi: 10.21437/Interspeech.2022-10978
@inproceedings{gent22_interspeech,
author={Helen Gent and Chase Adams and Yan Tang and Chilin Shih},
title={{Deep Learning for Prosody-Based Irony Classification in Spontaneous Speech}},
year=2022,
booktitle={Proc. Interspeech 2022},
pages={3993--3997},
doi={10.21437/Interspeech.2022-10978}
}