Cornell University
Towards Actionable Understandings of Conversations: A Computational Approach
Abstract
dc:description.abstractConversations are central to many consequential settings. Understanding how conversationalists navigate through them could unlock great improvements in domains like mental health, where the provision of social support is crucial. Such domains also present a promising opportunity for research: many interactions are recorded in large collections of transcripts, facilitating systematic analyses. In this dissertation, we take up this opportunity: we consider computational approaches to analyzing conversations, that can arrive at descriptively rich and prescriptively informative accounts of how conversationalists interact. We start by proposing methodology to model two particular conversational phenomena. In the British House of Commons, we consider the wide range of rhetorical roles encompassed by the questions that legislators ask, and develop an unsupervised method to infer types of rhetorical roles given a dataset of questions and answers. In the context of a crisis counseling service, we develop a method to model how counselors orient the flow of complex and high-stakes interactions with people in mental health crises. We apply these methods to analyze the respective domains, drawing correspondences between interactional dynamics and broader aspects of the setting, such as a legislator's political standing or the effectiveness of a counseling conversation. We then describe a general approach, the Expected Conversational Context Framework, for modeling utterances in terms of their roles in a conversation. The framework's key idea is that we can derive a range of characteristics of an utterance by accounting for its expected conversational context---i.e., the distribution of preceding or subsequent utterances that could occur next to it in a conversation. Via a series of empirical explorations, we illustrate how the framework is generative of a variety of characterizations and analyses, including and beyond those proposed in our initial studies. We end with a critical appraisal of the extent to which such approaches can arrive at actionable understandings. Drawing on a broad range of literature, ranging from sociological studies of interaction to causal inference, we consider the various complexities of conversations and the challenges they raise for methods such as ours.
Degree
thesis:*- Name thesis:degree_name
- Ph. D., Information Science
- Level thesis:degree_level
- Doctor of Philosophy
- Discipline thesis:degree_discipline
- Information Science
- Grantor
- Cornell University
- Year dc:date.issued
- 2021
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Zhang, Justine
- Chair dc:contributor.chair
-
- Danescu-Niculescu-Mizil, Cristian
- Committee members dc:contributor.committeemember
-
- Lee, Lillian
- Kleinberg, Jon M.
Subjects
dc:subject × 6Rights
dc:rights- Statement dc:rights
-
- Attribution 4.0 International
- Licence dc:rights.uri
- Language dc:language.iso
- en
Identifiers
dc:identifier.*- Dc Identifier Bibid
- 15160004
- Dc Identifier Other
-
ProQuest Submission ID: 12625
ProQuest Publication ID: 28648438