Encapsulation, inference and utterance interpretation

While utterance interpretation is standardly understood as context-sensitive inference, there are prima facie reasons to doubt it is fully unencapsulated. First, utterance interpretation is normally fast and automatic, while it has been argued that unencapsulated processes are slow and reflective. Secondly, certain illusions appear to show that the processing of utterances is unrevisable in light of further information. I argue that these reasons are not conclusive, and utterance interpretation relies on indefinite tracts of background knowledge and contextual information. Examples show that interpretation requires the use of information that is not in the input and is neither purely linguistic nor purely stereotypical knowledge. I show that this applies to arriving at the proposition expressed (e.g. the proposition asserted) as well as implicatures. ‘Grammatical’ and ‘semantic’ illusions show less than one might think about encapsulation of utterance interpretation. Some do not depend on reaching any stable interpretation. The inversion illusion is due in part to background knowledge. While the Moses illusion is revisable, such revision may be performed by a distinct capacity, so is not clear evidence for cognitive penetrability of the normal utterance interpretation process.