Year: 2024
Type: conference-paper
Abstract: In dialogue systems, utterances do not occur in isolation. One conversation might involve interactions between several speakers. It's crucial to determine the intentions behind utterances in multi-party conversations when more than two interlocutors are interacting. Beyond directly capturing the speaker's intention, our proposed model first focuses on identifying speakers from utterances, and based on this knowledge, it classifies the corresponding dialogue acts. For the speaker identification process, the study extracted linguistic features related to speakers from conversations and incorporated them during the fine-tuning process, which is particularly beneficial in dealing with multiple speakers. After that our model aims to improve dialogue act recognition baselines on shorter utterances by implementing a pipe-lining approach based on speaker model predictions. The effectiveness of our approach is demonstrated using two benchmark datasets, MRDA and SwDA, which are based on multiparty and twofold conversations, respectively.
Institution: Robert Gordon University
Language: English
Topic: Speech and dialogue systems
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
Funder: Robert Gordon University
OpenAlex ID: W7018622788
FWCI:
Cites:
Cited by:
Related to: