Copy the page URI to the clipboard
Plüss, Brian
(2014).
DOI: https://doi.org/10.21954/ou.ro.00009d1c
Abstract
A common assumption in the study of conversation is that participants fully cooperate in order to maximise the effectiveness of the exchange and ensure communication flow. This assumption persists even in situations in which the private goals of the participants are at odds: they may act strategically pursuing their agendas, but will still adhere to a number of linguistic norms or conventions which are implicitly accepted by a community of language users.
However, in naturally occurring dialogue participants often depart from such norms, for instance, by asking inappropriate questions, by avoiding to provide adequate answers or by volunteering information that is not relevant to the conversation. These are examples of what we call linguistic non-cooperation.
This thesis presents a systematic investigation of linguistic non-cooperation in dialogue. Given a specific activity, in a specific cultural context and time, the method proceeds by making explicit which linguistic behaviours are appropriate. This results in a set of rules: the global dialogue game. Non-cooperation is then measured as instances in which the actions of the participants are not in accordance with these rules. The dialogue game is formally defined in terms of discourse obligations. These are actions that participants are expected to perform at a given point in the dialogue based on the dialogue history. In this context, non-cooperation amounts to participants failing to act according to their obligations.
We propose a general definition of linguistic non-cooperation and give a specific instance for political interview dialogues. Based on the latter, we present an empirical method which involves a coding scheme for the manual annotation of interview transcripts. The degree to which each participant cooperates is automatically determined by contrasting the annotated transcripts with the rules in the dialogue game for political interviews. The approach is evaluated on a corpus of broadcast political interviews and tested for correlation with human judgement on the same corpus.
Further, we describe a model of conversational agents that incorporates the concepts and mechanisms above as part of their dialogue manager. This allows for the generation of conversations in which the agents exhibit varying degrees of cooperation by controlling how often they favour their private goals instead of discharging their discourse obligations.
CORE (COnnecting REpositories)
CORE (COnnecting REpositories)
