Common ground is said to depend on triple co-presence of community membership, physical co-presence, and linguistic co-presence. In thinking about how speakers tailor their language for their audience, researchers like Clark and Marshall (1978) have posited an explicit "reference diary" that speakers keep track of to encode the status of events they experience with others (what the event was, who was there, who has shared knowledge of it, etc.). Under that approach, there's an important role for memory encoding and memory retrieval in describing common ground. An example like the following, from the 'Call Home' corpus, shows just how strategic and explicit reference decisions may be, making it easy to imagine that we do indeed have a "reference diary" that we consult when we are making decisions about how to refer to entities and individuals for a particular audience.
(8) Yeah, I’ve got another buddy who, uh, is a Marine pilot. I'm trying to think if you had ever met this guy. I don't think so. [CallHome Corpus]
However, other aspects of reference production may be much more automatic (and perhaps not even addressee-centric). There is a debate in the literature on how to find evidence that would establish that speakers favour automatic processing (whereby speakers only appear to tailor their language but in reality simply make rough approximations to estimate their listeners' knowledge) over strategic explicit common ground representations (whereby speakers say things like that in (8)).
This debate has been framed around two key questions: Do speakers design utterances with a particular addressee in mind? Do they consider shared vs. privileged information? A study by Horton and Keysar (1996) addressed the issue of shared vs. privileged ground and whether speakers track the relative status of different entities or whether they are egocentric and use their own knowledge set as a rough approximation of the addressee's knowledge. The study used a set up in which a participant had to describe an object (e.g., the triangle on the right in the picture below) to an addressee who either did or did not share some critical contextual information (e.g., a smaller triangle). The hidden triangle is said to be in "privileged ground" whereas the triangle in full view is said to be in common ground.
The premise of the Horton and Keysar study is that if speakers are able to fully optimise their utterances for their addressee, they should take into account what is shared or privileged (what the speaker in the CallHome example in (8) makes explicit) and they should only use context-relevant modifiers when the relevant information is shared. In other words, a speaker should only say "the big triangle" when the existence of the small triangle is shared; otherwise, the expression "the triangle" would be more appropriate if the small triangle is in privileged ground.
The results showed that with unlimited time, speakers only use 'big triangle' when both triangles are shared. Under time pressure, however, speakers fail to show sensitivity to addressee knowledge. This result has been taken to show that effort is required to take into account the information available to addressee (i.e., the process is disrupted by cognitive load). Indeed, a more recent study (Wardlow Lane & Liersch, 2012, Can you keep a secret? Increasing speakers' motivation to keep information confidential yields poorer outcomes. Language & Cognitive Processes) showed that adding incentives such as points or cash for keeping the identity of the hidden object a secret from the addressee actually makes speakers MORE LIKELY to use a context-relevant modifier ("big"). Apparently it's very hard to ignore the elephant in the room, and it becomes even harder when there is more pressure to do so.
So how do we monitor common ground? Clark and Marshall's proposal of a reference diary is based on the idea of explicit strategic searches. This can also be characterised as error detection and correction -- overt monitoring of what we say to check whether it satisfies the reference diary. Under this approach, monitoring is seen as acting separately from other facets of routine language processing. Other researchers have followed this framework, saying that "Common ground is a functionally distinct process that belongs to an 'adjustment' stage of processing, but that imposes no constraint on production or comprehension processes per se" (Barr & Keysar, 2006). Such an approach makes an assumption that part of our ability to monitor common ground relies on special-purpose memory representations. An alternative view of audience design also takes memory as central but posits that there is a split between strategic and automatic lookups. Strategic use (like the reference diary) is triggered by:
Strategic planning for audience design is hence the exception rather than the norm. In the remaining contexts, monitoring common ground is taken to be automatic:
As evidence of non-linguistic context-specific monitoring, Chun and Jiang (2003) report an eyetracking study using the visual world paradigm (where participants look at a "scene" of a handful of objects while their eye movements are tracked). The results showed that searching for visual targets in the context of specific configurations of distractor objects improves the identification of new targets embedded in the same contextual configurations, compared to targets in new configurations. Basically, information learned under particular conditions is better recalled in those same conditions.
The implication of this for a memory-based view of common ground is that if one has a strong-enough pattern of associations between a conversation partner and some information, then that information will be taken as shared. This is akin to the Interactive Alignment posited by Pickering and Garrod (2006) whereby alignment can emerge from low-level processes. They point to data showing that in the context of a particular syntactic construction (e.g., the passive "John was handed a book"), speakers are more likely to produce that construction over others (when prompted with, for example, "teach, John, French"). There is evidence that speakers persist in the constructions they use as they produce language (Bock 1986) and that hearing another speaker in a dialog use a particular construction makes that construction more likely in one's own productions (Branigan, Pickering, & Cleland, 2000). These studies show that alignment is possible without recourse to a reference diary or any independent assessment of common ground. In other words, it is merely the shared experience of having been present for a conversation or event or syntactic construction that is enough to lodge that moment in our memory.
The next section 7.4 is about testing the "Memory-based view of common ground". Click here.