Chapter 3: Perspective-taking on-line
8. Questions Moving Forward
The two main models proposed to account for results in the perspective-taking task both assume that there are automatic processes and corrective, monitoring processes. They differ only in what information can be integrated in the automatic phase. The perspective-adjustment model of Keysar and colleagues proposes that initial processes are from an ego-centric perspective and are not affected by information about what is common ground. These processes are monitored by a system that has access to ground information, among other sources, and can be corrected where necessary. Constraint-Based proposals, found in Hanna et al.
(2003), Brown-Schmidt et al. (2008) and elsewhere counter that ground information does get integrated in initial, automatic processes. The idea here is that perspective-taking tasks give rise to a trade-off between common ground constraints and bottom-up constraints that attempt to match linguistic input with targets in the visual context. Corrective processes are necessary where cues to ground information do not sufficiently constrain the set of potential referents to common ground.
It is easy to read the literature on perspective taking and come to the conclusion that Theory of Mind abilities are fundamental to the derivation of ground information. For instance, in a widely cited review of perspective taking literature, Brown-Schmidt & Hanna (2011) say in their introductory section, “In this chapter, we focus on the use of one type of non-linguistic information in incremental language comprehension—information about others’ knowledge and beliefs.”
(Brown-Schmidt & Hanna, 2011 p. 12) but the rest of the paper exclusively
reviews work that explicitly sets out to test whether common ground information influences incremental referential processing. That review paper also explicitly points to the work of Clark and colleagues as providing the theoretical underpinning for ideas about common ground in referential processing. As discussed at length in Chapter 1 of this thesis, Clark (1992, 1996) views common ground as common knowledge of discourse-relevant information, in line with prominent views in linguistics and philosophy (e.g.
Stalnaker 2002). Dating from the work of Clark & Marshall (1981) there is a proper awareness in this work that one cannot establish what is common knowledge in via an infinite series of inferences about ever higher-order beliefs.22 Clark & Marshall propose that common knowledge has to be established via heuristic means. Thus while it may be true that ground information is information about knowledge or beliefs, because it is common knoweldge that is in question, cognitive processes that establish ground information are not liable to call on Theory of Mind abilities.
Thus, both of the widely discussed sets of proposals about behaviour in perspective-taking tasks are compatible with the idea that Theory of Mind information is only available to be integrated at the monitoring and corrective stage of referential processes. It seems fairly clear that it is the view of Keysar and colleagues that Theory of Mind information is integrated at the secondary stage. For instance, Keysar, Lin and Barr (2003) present a version of the perspective-taking task and explicitly argue that the errors incurred in that task are due to the difficulty in integrating Theory of Mind inferences into language processes. As we have seen above, however, other studies show an effect of early integration of Theory of Mind information in language comprehension (Ferguson & Breheny 2011, 2012). We have also seen apparent evidence in Brown-Schmidt et al. (2008) for integration of ignorance inferences in a taks very similar to the standard perspective-taking task, involving questions. However, Brown-Schmidt and colleagues present their results
22 Of course, one can establish that something is not common knowledge by an inference that one’s interlocutor does not believe some proposition. But the focus here is what is in common
as yet more evidence that ground information is integrated on line, extensively citing the work of Clark as providing the theoretical underpinning for their design. This is the case in Brown-Schmidt et al.
(2008), Brown-Schmidt (2009), and Brown-Schmidt & Hanna (2011). The proposal seems to be that interrogative forms are processed using ground information - questions only ask about information that is in privileged ground.
It is of course compatible with Constraint-Based views that Theory of Mind inferences, being just another source of contextual knowledge, are integrated into automatic language processes. However, it was argued above it is in the spirit of Clark’s work on dialogue that Theory of Mind play a peripheral role, perhaps along the lines envisaged by Keysar and colleagues. This is so since common ground (i.e. common knowledge of discourse-relevant facts) is for Clark the focus of dialogue and common ground is maintained in Clark’s work via external cues (see especially Clark
& Wilkes-Gibbs, 1986; Clark, 1996). It is really the forms of behaviour that are exploited in the process of moving pieces of the ‘dialogue game-board’
around.
So, moving forward into the next chapter, we have an open question about the specific role for theory of mind inferences in the perspective-taking task. To date, no research has explicitly studied the role of Theory of Mind inferences in the standard perspective-taking task. More broadly, we are interested in the extent to which processes that maintain common ground play a role in language processing and the extent to which processes that integrate theory of mind inferences play a role. Work in the spirit of Clark sees the lion-share of the constraints coming from common-ground processes. By contrast, my own proposals see a lesser role for common ground and a greater role for Theory of Mind processes.
Specifically, regarding the perspective taking task, we have raised a second question about whether common ground processes in fact create some interference in this task.