“I Object!” Modeling Latent Pragmatic Effects in Courtroom Dialogues

Proceedings of the 14th Conference of the European Chapter of the Association for Computational Linguistics, pages 655–663, Gothenburg, Sweden, April 26-30 2014. c⃝2014 Association for Computational Linguistics “I Object!” Modeling Latent Pragmatic Effects in Courtroom Dialogues Dan Goldwasser University of Maryland Institute for Advanced Computer Studies College Park, MD , USA goldwas1@umiacs.edu Hal Daum´e III Department of Computer Science University of Maryland College Park, MD , USA hal@cs.umd.edu Abstract Understanding the actionable outcomes of a dialogue requires effectively modeling situational roles of dialogue participants, the structure of the dialogue and the rele- vance of each utterance to an eventual ac- tion. We develop a latent-variable model that can capture these notions and apply it in the context of courtroom dialogues, in which the objection speech act is used as binary supervision to drive the learning process. We demonstrate quantitatively and qualitatively that our model is able to uncover natural discourse structure from this distant supervision. 1 Introduction Many dialogues lead to decisions and actions. The participants in such dialogues each come with their own goals and agendas, their own perspec- tives on dialogue topics, and their own ways of interacting with others. Understanding the action- able results of a dialogue requires accurately mod- eling both the content of dialogue utterances, as well as the relevant features of its participants. In this work, we devise a discriminative latent variable model that is able to capture the overall structure of a dialogue as relevant to specific acts that occur as a result of that dialogue. We aim to model both the relevance of preceding dialogue to particular action, as well as a binary structured re- lationship among utterances, while taking into ac- count the pragmatic effect introduced by the dif- ferent speakers’ perspectives. We focus on a particular domain of dialogue: courtroom transcripts. This domain has the advan- tage that while its range of topics can be broad, the roles of participants are relatively well-defined. Courtroom dialogues also contain a specialized speech act: the objection. In real court settings (as opposed to fictional- ized courts), an objection is a decision made by the party opposing the side holding the floor, to inter- rupt the flow of the courtroom discussion. While motivation behind taking this decision can stem from different reasons, it is typically an indication that a particular pragmatic rule has been broken. The key insight is that objections are sustained when a nuanced rule of court is being violated: for instance, the argumentative objection is “raised in response to a question which prompts a witness to draw inferences from facts of the case”1, as op- posed to the witness stating concrete facts. The objectionable aspects of the preceding di- alogue can be identified by a well-trained person; however these aspects are quite subtle to a com- putational model. In this work we take a first step toward addressing this problem computationally, and focus on identifying the key properties of dia- logue interactions relevant for learning to identify and classify courtroom objections. …

Content truncated to 500 words