Implicit and Indirect: Computational Identification of Ambiguous Conversational Actions in Asynchronous Crisis-Related Conversations
Abstract: This paper presents a digital conversation analysis based approach to the computational detection of ambiguous actions in asynchronous online conversations. Action detection has been widely studied for synchronous chats. However, models or datasets for asynchronous conversations are scarce, and have not sufficiently considered the special characteristics of asynchronous discussion, most importantly the tendency for comments to involve multiple actions and multiple valid interpretations of actions. We provide a theory-driven annotation scheme for crisis-related asynchronous conversations, and an annotated dataset for Finnish. We show that considering the multi-action characteristics of asynchronous data statistically improves classification performance, and that an ensemble of best models can represent the ambiguity of actions, which is especially characteristic of face-threatening actions in controversial conversations.
Paper Type: Long
Research Area: Discourse and Pragmatics
Research Area Keywords: conversation, discourse and multilinguality, datasets for low resource languages, resources for less-resourced languages
Contribution Types: Model analysis & interpretability, Approaches to low-resource settings, Data resources
Languages Studied: Finnish
Submission Number: 93
Loading