ABSTRACT
Utterance interpretation is widely seen as an inferential achievement, and utterance interpretation resembles ‘system 2’ reasoning in some respects: the inferences are generally warranted and unencapsulated. But in normal, smooth communication, they are quick and seemingly effortless and thus more akin to paradigm unconscious ‘system 1’ inferences. Resolving this tension is a goal for any cognitively realistic account of utterance interpretation. This chapter argues on theoretical and empirical grounds for a minimalist kind of metacognition whereby a mental process is unconsciously monitored and controlled by another, perhaps without the latter metarepresenting the former. Cognitively realistic inferential theories of utterance interpretation require there to be such feedback, even in normal smooth communication. Also, two separate sets of experiments, reviewed here, show that feedback occurs in comprehension without hearers being aware of it. A potential objection concerns levels of explanation (in David Marr’s sense): that some process is inferential seems to be a claim at Marr’s computational level, while claims about the way that feedback works in utterance interpretation are at the algorithmic level. This chapter argues that facts at the algorithmic level constrain facts at the computational level and suggests that processes for implementing abductive inference must have monitoring and control.
