The physics of conversation: illocutionary forces driving common ground stabilisation
In the previous post, we discussed how, while having conversations, we are not simply producing tokens or sounds but we are actually doing something together with our interlocutors. What are we actually doing, though? And how can we model the reasons why we communicate and the perceived consequences of the actions we take through speaking?
In a dedicated post in this series, we also discussed how graph based representations can help describe various situations in which speakers use different speech acts to keep the interaction going. We can take this metaphore further and find links between cognitive theories of communication and mathematical measures characterising the role of nodes in conversation dynamics. Take the case of plausible arguments, for example, intended as powerful data to present to support a claim. In [Paglieri & Castelfranchi, 2004], the concepts of self-evident datum and explanatory datum were presented using the following Figure.
The schema used in [Paglieri & Castelfranchi, 2004] can be put in direct relationship with the mathematical measures associated with authoritative and hub nodes in network analysis, as described in [Kleinberg, 1999]. The concepts are depicted in the following Figure.
It is beautiful to see how such different disciplines used the same graph structures to describe, in separate domains, structurally similar principles and this gives us the chance to use these mathematical descriptors to represent cognitive aspects of data usage for dialogue management. Considering all the aspects described in [Paglieri & Castelfranchi, 2004], a set of possible descriptors can be found that helps formalise a model of argument selection [Di Bratto et al., 2024], as shown in the following Table.
In the Table, entropy measures the confidence a system has in the collected information while hard evidence represents information that has been collected during the dialogue about the interlocutor. The correspondence between graph structures and mathematical descriptors for concepts like arguments and speech acts inspires a view of conversation dynamics as a kind of physical system, represented as a graph, on which speakers apply forces aimed at modifying its structure, which effectively represents an internal representation of the common ground.
Following the concepts presented in this post, we can now relate graph configurations representing dialogue states with the speakers’ interventions, in the form of speech acts. These are aimed at pushing the graph towards a stable configuration that exhibit patterns that are desirable from the points of view of the different interlocutors. More than that, the hierarchy defined in the previous post can be put in relationship with different levels of strength for illocutionary forces! Speech acts aimed at solving communication problems (asking to repeat, asking for missing information) are, then, characterised by low illouctionary strength, meaning that the intentional force is low because the user’s goals are not relevant and the act is almost automatic. On the other hand, when speech acts aimed at pursuing interactional goals are produced, these are characterised by high illocutionary strength. The following Figure shows how the continuum defined by the increasing illocutionary strength can be put in direct relationship with the graph configurations described before. Moreover, sub-axes can be defined to describe the hierarchical order of speech acts characterised by a higher degree of intentionality.
The concepts presented in this post describe multiple aspects of dialogue management in a linguistic framework that can be mapped on the technological infrastructure defined by FANTASIA, which will be the topic of the next post in the series.
References
[Di Bratto et al., 2024] Di Bratto, M., Origlia, A., Di Maro, M. Mennella, S. (2024). Linguistics-based dialogue simulations to evaluate argumentative conversational recommender systems. User Model User-Adap Inter. https://doi.org/10.1007/s11257-024-09403-3
[Kleinberg, 1999] Kleinberg, J.M. (1999). Authoritative sources in a hyperlinked environment. Journal of the ACM (JACM), 46 (5), 604–632.
[Paglieri & Castelfranchi, 2004] Paglieri, F., & Castelfranchi, C. (2004). Argumentation and data-oriented belief revision: On the two-sided nature of epistemic change. In Proc. of the 4th workshop on computational models of natural argument (pp. 5–12).
