Emergence and adaptation of referential conventions in dialogue
Item statusRestricted Access
Embargo end date02/07/2020
When speakers in dialogue are faced with the need to repeatedly refer to the same items, they usually use the same references they or their partners had used before. These previously-used references act as precedents, standing in speakers’ memory as successful ways of solving that particular communicative need. From mechanistic models explaining this reuse as a consequence of low-level priming (Pickering & Garrod, 2004), to models assuming sophisticated partner-modelling processes (Clark & Marshall, 1981; Clark & Wilkes-Gibbs, 1986), most theories assume speakers will maintain their referential choices throughout the dialogue. While references might be modified to discard superfluous elements (as in referential reduction, Krauss & Weinheimer, 1964; or simplification, Bard & Aylett, 2005), conceptualisations will be preserved. However, there are several reasons why speakers might need to change their previous referential choices: a context change might render the old reference insufficient to identify the target, or overly detailed, prompting the listener to wonder if additional meanings are implied. It might also be that the repetition of the task highlights a better referential alternative, or that additional information makes other alternatives more salient for the speakers. In this thesis, I present an investigation of the dynamics of reference repetition and change in dialogue, bringing together a theoretical analysis of the existing literature and 5 experiments that aim to clarify how, when, and why speakers might change their referential choices. Experiment 1 explores the dynamics of reference change when the repetition of a task creates additional pressures that were not evident in an initial exposure. The experiment compares pairs and individual speakers describing positions in two spatial contexts (regular or irregular mazes, as in Garrod & Anderson, 1987; Garrod & Doherty, 1994) that cue participants into using different referential descriptions. As the task is repeated over 3 rounds, an additional pressure for efficiency is created, pushing participants across contexts into using one of the two initial descriptive choices. Crucially, only interacting pairs of participants adapted to this additional pressure by switching to a more efficient alternative, while participants completing the task individually maintained their initial choices. Additionally, this chapter reports a pilot study of the same experiment in 4-person groups that showed a switching pattern similar to the one found in interacting pairs. Experiments 2 and 3 further investigate the drivers of linguistic change in referential choice and the relationship between interaction and adaptation found on experiment 1. Experiment 2 explores the relationship between context change and linguistic change. Using the maze game paradigm, the experiment presented individuals and pairs of participants with either the same maze in each round of the task, or different mazes in the first and the second halves of the experiment. The results of Experiment 2 offer some support to the conclusions of Experiment 1, as participants switched to Abstract descriptions as they gained experience in the task; however there were no significant differences between Interaction conditions, nor between Same or Different mazes. Experiment 3 was aimed at exploring which features of interaction were relevant for reference change. The experiment used the maze game in different interactive setups, in which participants played a first round of the game either as Matchers (in direct interaction with the Director) or Overhearers (having access to another pair’s dialogue), and three successive rounds as Directors with either the same partner as in the first round, or a new partner. Participants showed higher levels of adaptation as they gained experience in the task; while the different interactive setups did not significantly influence their reference choices. Experiments 4 and 5 further explore the relationship between reference change and participant role in interaction. Using a picture matching paradigm (Brennan & Clark, 1996), Experiment 4 tested participants interacting with either the Same Partner throughout, a New Partner in the second half of the experiment, or an Overhearer (who had witnessed the first half of the experiment) in the second half. Participants in all conditions maintained previously used overspecific picture descriptions even when those detailed descriptions were not needed to identify the referents, pointing towards a predominance of speaker-centred factors in reference choice. Experiment 5 used a similar interactive setup, testing participants on a larger set of pictures. Participants maintained their overspecific descriptions only if interacting with the Same Partner they had on the first half of the experiment, or with the Overhearer, switching to context-appropriate basic-level descriptions if interacting with a New Partner. Taken together, both experiments suggest a complex balance between speaker-centred and audience-design factors in the potential change of reference choices, where speakers need to weigh their own effort against the communicative needs of their partner. These experiments highlight the crucial role of interaction in the adaptation of reference choices to changes in context, and show that individuals’ ‘conservative bias’ that leads them to maintain their own previously used references can be overturned in the search for better communicative alternatives in interactive dialogue.