Pickering & Garrod (P&G) call for closer attention to the mechanisms underlying coordination in dialogue. There are good empirical grounds for accepting many of the basic assumptions of the interactive alignment model. Specifically, the strong egocentrism of speakers and listeners that we have uncovered in our own studies (cf. Barr & Keysar [in press], for a recent review) makes much sense within a context of strong representational overlap, and the interactive alignment model provides an appealing explanation for how such overlap comes about.
A central assumption of P&G is a categorical distinction between language processing in monologue and in dialogue. Clearly, monologue is different from dialogue because in monologue there is no feedback and no opportunity for interactive repair. But P&G go further than this, asserting that in monologue, "the automatic mechanisms of alignment are not present" (sect. 3.2, para. 3) and that "there is no opportunity to call on aligned linguistic representations" (sect. 8.1, para. 5). Under this view, dialogue involves processes that are fundamentally distinct from those present in monologue. We disagree. We suggest that only the strategic mechanisms of feedback and interactive repair are absent in monologue, not the automatic mechanisms of alignment. Just as there is a "dialogic continuum" defining different kinds of interactive activities, we argue that there is a continuum of processing and that alignment will be observed even under non-interactive circumstances.
Differences in processing may simply be a matter of degree, not of kind, especially given that monologue-like episodes are common even in naturalistic conversation. Consider, for example, the first utterance of a conversation, which in many cases is a mini-monologue before a full-fledged interactive exchange develops. The speaker will need to go through the various stages to produce the first utterance (e.g., Levelt 1989), and the comprehender must parse the utterance in order to appreciate its significance. Later on in the exchange, enough of shared but nonmutual information (what P&G unfortunately term "implicit common ground")1 may have built up to short-circuit many aspects of these processes. But there is no reason to believe that different theories would be required to explain the processing of the first utterance of a dialogue versus the hundred-and-first. Indeed, P&G claim that the alignment processes are automatic and resource-free, and processes of these sorts cannot simply be switched off. Finally, the "dialogic continuum" cited by the authors does not just provide a means for classifying whole conversations but actually represents a domain of activity that can be fully traversed even within the span of a single conversation. A conversationalist can at any point secure turn space in order to engage in an extended monologue (Sacks et al. 1974) – for example, about her trip to India – followed by close coordination with her interlocutor in order to arrange a time for a future meeting. There is no reason to expect alignment processes and bridging inferences to toggle on and off as the interactivity of the discourse changes.
Studies that find differences between the quality of comprehension of interlocutors and of overhearers (e.g. Schober & Clark 1989; Wilkes-Gibbs & Clark 1992) might be taken as evidence to support the idea of radically different cognitive processes in interactive discourse. They need not be. In such studies interlocutors had opportunities for feedback and repair that the overhearers lacked. Because different people will misunderstand different things, those who can ask for clarification will receive feedback that is relevant to them and consequently might understand better. Therefore any difference between such noninteractive and interactive comprehension could be fully attributable to strategic, effortful feedback but not necessarily to automatic alignment. In fact, Barr and Keysar (2002) found that even when such feedback is removed, listeners who believed themselves to be overhearers automatically aligned their semantic representations with the speaker's to the same degree as listeners who believed themselves to be addressees.
In closing, far from qualitatively changing the nature of processing, it is likely that dialogue provides a radically different context in which the same processes operate. The context includes an interlocutor and mechanisms for feedback and interactive repair. For us there is no question that it is important to study conversation in vivo, but it remains to be seen whether this would reveal automatic processes that are truly unique to dialogue.
Footnotes (1)
1 In their seminal work on common ground, Clark and Marshall (1981) clearly make the case that common ground is a form of meta-knowledge that is conceptually distinct from shared knowledge. What P&G are referring to by “implicit common ground” is really just shared knowledge, not common ground, because interlocutors need not represent the fact that their representations are shared. Such usage is certain to contribute to the legacy of confusion that has plagued discussions of mutual knowledge and common ground (see Keysar 1997 and Lee 2001 for discussion).