Addressee and Response Selection in Multi-Party Conversations With Speaker Interaction RNNs

Abstract

In this paper, we study the problem of addressee and response selection in multi-party conversations. Understanding multi-party conversations is challenging because of complex speaker interactions: multiple speakers exchange messages with each other, playing different roles (sender, addressee, observer), and these roles vary across turns. To tackle this challenge, we propose the Speaker Interaction Recurrent Neural Network (SI-RNN). Whereas the previous state-of-the-art system updated speaker embeddings only for the sender, SI-RNN uses a novel dialog encoder to update speaker embeddings in a role-sensitive way. Additionally, unlike the previous work that selected the addressee and response separately, SI-RNN selects them jointly by viewing the task as a sequence prediction problem. Experimental results show that SI-RNN significantly improves the accuracy of addressee and response selection, particularly in complex conversations with many speakers and responses to distant messages many turns in the past.

Is ChatGPT a Good Multi-Party Conversation Solver?

“…Existing methods on building MPC systems can be generally categorized into retrieval-based approaches (Ouchi and Tsuboi, 2016;Zhang et al, 2018;Wang et al, 2020;Gu et al, 2021Gu et al, , 2023 or generation-based (Hu et al, 2019;Gu et al, 2022;Li and Zhao, 2023). On the one hand, Ouchi and Tsuboi (2016) and Zhang et al (2018) proposed to update speaker embeddings with conversation streams dynamically and role-sensitively. Wang et al (2020) proposed to track the dynamic topic in a conversation.…”

Learning WHO Saying WHAT to WHOM in Multi-Party Conversations

“…Thus, the ability to predict which agent in the conversation is the most likely to speak next, and conversely, when an agent must wait before interacting, is important for conducting engaging and social conversations (Pinhanez et al, 2018;de Bayser et al, 2019). Furthermore, detecting who is being addressed, i.e., who the current speaker is talking to, is also non-trivial in these conversation scenarios (Ouchi and Tsuboi, 2016;Zhang et al, 2018;Le et al, 2019;Gu et al, 2021Gu et al, , 2023. Last but not least, only after knowing a speaker and an addressee at the current dialogue state, can the system return an appropriate response following the conversation Wang et al, 2020;Gu et al, 2022a;Li and Zhao, 2023).…”

“…Explicit Addressee Recognition This task explicitly determines the intended addressee of an utterance. Previous studies mainly focus on predicting the addressee of only the last utterance of a conversation (Ouchi and Tsuboi, 2016;Zhang et al, 2018), while recent studies pay more attention to predicting the addressees of all utterances of a conversation (Le et al, 2019;Gu et al, 2021Gu et al, , 2023. Le et al (2019) propose a who-to-whom (W2W) model to recognize and complete the addressees of all utterances in a conversation to help understand the whole conversation, given an MPC where part of the addressees are unspecified.…”

“…Ouchi and Tsuboi (2016) propose jointly modeling the tasks of response selection and addressee selection to capture what is being said to whom at each time step in a context. Zhang et al (2018) follow this framework and improve it by updating the interlocutor embeddings role-sensitively. Gu et al (2021) propose jointly learning who says what to whom in a unified framework by considering addressee-and speaker-related tasks as complementary signals for response selection during the pre-training stage.…”

“…Thanks to breakthroughs in sequence modeling (Sutskever et al, 2014;Vaswani et al, 2017) and pre-trained language models (PLMs) (Radford et al, 2019;Devlin et al, 2019;Lewis et al, 2020), researchers have proposed various effective models for conversations between two participants (Serban et al, 2016;Wen et al, 2017;Zhang et al, 2020). Recently, researchers have paid more attention to a more practical and challenging scenario involving more than two participants, which is well known as multi-party conversations (MPC) (Ouchi and Tsuboi, 2016;Zhang et al, 2018;Le et al, 2019;Kummerfeld et al, 2019;Gu et al, 2021Gu et al, , 2022aGu et al, ,b, 2023Li and Zhao, 2023). Figure 1 presents instances of twoparty and multi-party conversations, corresponding to the common scenarios of one-to-one chat and group chat respectively in daily life.…”