The treachery of words: Agent-based modelling of pragmatic conversations
Keywords
No Thumbnail Available
Authors
Issue Date
2020-06-29
Language
en
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The capacity of human referential language use is seen as an important adaptation, contributing to human abilities to communicate ideas and coordinate actions regarding the surrounding world. Conversational partners reaching mutual understanding is of fundamental importance for successful referential communication. Asymmetry of lexical knowledge between partners, ambiguity of language signals and lexical uncertainty pose challenges to this ability. Hawkins et al. (2017) propose a theory of how humans can achieve mutual understanding, but their theory assumes explicit feedback about referents during conversation. The present thesis argues that explicit feedback is not a psychologically plausible assumption, as effective referential linguistic communication is possible even in situations where explicit feedback is unavailable. The aim of this study was to come up with alternatives for explicit feedback, that are more plausible.
First a convention building model, strongly inspired by Hawkins et al. (2017), was built, and simulations of the model were used to gain insights in the problematic nature of the theory. Reflection on the convention building model led to the proposal of three tactics for replacing explicit feedback: dialogue, other-initiated repair and exclusion. Conversational partners use the dialogue tactic when the listener refers to an inferred referent with a new signal, and the speaker uses this to conclude whether understanding was reached. Other-initiated feedback is used when the listener communicates a lack of understanding, requesting more informative signals from the speaker. Speakers use exclusion to eliminate a referent that interferes with successful communication. Three adapted models were build, cumulatively adding tactics. These three models were compared to a solely pragmatic model with no replacement for the absence of explicit feedback, using simulations.
The simulation results showed that pragmatic reasoning already explains a lot of communicative success in referential language use. The dialogue, other-initiated repair and exclusion tactic can be valuable given the right conditions. Other-initiated repair is valuable when listener ambiguity is high by preventing misunderstanding from happening, while the dialogue tactic resolves misunderstandings due to high asymmetry. Exclusion can be valuable when the speaker has only ambiguous signals available for the intention, while the listener believes that an unambiguous signal exists, making pragmatic reasoning difficult. Other-initiated repair and exclusion can avoid misunderstandings, while the dialogue tactic can resolve misunderstandings directly after they occurred. However, the tactics are not capable of resolving misunderstandings that happened in the past. When the speaker has used all informative signals available, but still no understanding was reached, the conversation stagnates. I argue that the occurrence of stagnation is related to the limited size of the lexica in simulation, an effect of the intractability of the theory by Hawkins et al. (2017).
In conclusion, the tactics proposed can be valuable given the right conditions, by avoiding or resolving misunderstandings where pragmatic reasoning fails. Considering the costs of such tactics, further research could investigate how and to what extent humans are capable of detecting the right set of tactics given the circumstances.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen