Who Says What (WSW): A Novel Model for Utterance-Aware Speaker Identification in Text-Based Multi-Party Conversations

Y. Priyadarshana, Zilu Liang, Ian Piumarta

2023

Abstract

Multi-party conversation (MPC) analysis is a growing and challenging research area which involves multiple interlocutors and complex discourse structures among multiple utterances. Even though most of the existing methods consider implicit complicated structures in MPC modelling, much work remains to be done for speaker-centric written discourse parsing under MPC analysis. On the other hand, pre-trained language models (PLM) have achieved a significant success in utterance-interlocutor semantic modelling. In this study, we propose Who Says What (WSW), a novel PLM which models who says what in an MPC to understand equipping discourse parsing in deep semantic structures and contextualized representations of utterances and interlocutors. To our knowledge, this is the first attempt to use the relative semantic distance of utterances in MPCs to design self-supervised tasks for MPC utterance structure modelling and MPC utterance semantic modelling. Experiments on four public benchmark datasets show that our model outperforms the existing state-of-the-art MPC understanding baselines by considerable margins and achieves the new state-of-the-art performance in response utterance selection and speaker identification downstream tasks.

Paper Citation

in Harvard Style

Priyadarshana Y., Liang Z. and Piumarta I. (2023). Who Says What (WSW): A Novel Model for Utterance-Aware Speaker Identification in Text-Based Multi-Party Conversations. In Proceedings of the 19th International Conference on Web Information Systems and Technologies - Volume 1: WEBIST; ISBN 978-989-758-672-9, SciTePress, pages 26-36. DOI: 10.5220/0012164400003584

in Bibtex Style

@conference{webist23,
author={Y. Priyadarshana and Zilu Liang and Ian Piumarta},
title={Who Says What (WSW): A Novel Model for Utterance-Aware Speaker Identification in Text-Based Multi-Party Conversations},
booktitle={Proceedings of the 19th International Conference on Web Information Systems and Technologies - Volume 1: WEBIST},
year={2023},
pages={26-36},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0012164400003584},
isbn={978-989-758-672-9},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 19th International Conference on Web Information Systems and Technologies - Volume 1: WEBIST
TI - Who Says What (WSW): A Novel Model for Utterance-Aware Speaker Identification in Text-Based Multi-Party Conversations
SN - 978-989-758-672-9
AU - Priyadarshana Y.
AU - Liang Z.
AU - Piumarta I.
PY - 2023
SP - 26
EP - 36
DO - 10.5220/0012164400003584
PB - SciTePress