LAUSR

We agree with Clarke regarding the need to understand the mechanisms of labelling effects on visual processing both when a label precedes a visual stimulus – what Clarke calls “primed recognition” – and when the categorisation of a visual stimulus precedes the label, as was the case here – what Clarke calls “unprimed recognition”. As we have shown in past work, hearing a label affects the processing of subsequently presented stimuli in tasks ranging from categorisation (e.g. Boutonnet & Lupyan, 2015; Edmiston & Lupyan, 2015; Lupyan & Thompson-Schill, 2012) to basic visual processing such as simply detecting the presence of an object (e.g. Lupyan & Ward, 2013). Our account for such effects is that the label transiently activates category-relevant features within which the subsequently presented image(s) are processed, helping to distinguish category members from non-members (in previous work we have systematically investigated the timecourse of these effects, e.g. Edmiston & Lupyan, 2015; Lupyan & Spivey, 2010; Lupyan & Thompson-Schill, 2012). In contrast, in unprimed recognition, the label activated by the visual stimulus itself rather than being provided exogenously (we use the term “unprimed” following Clarke, though every visual input is processed within some immediate context, though not necessarily a verbal one). Rather than reflecting two distinct types of labelling effects, we believe primed and unprimed recognition can be usefully incorporated into a common framework (Lupyan, 2012a, 2012b) in which there exist bidirectional connections between labels and percepts (we deliberately remain agnostic as to the nature of these label representations; their format may be phonological, or involve more abstracted lexical representations). The flow of information in the unprimed case is then as follows: a visual input partially activates associated labels and these feed back onto the visual representation, augmenting it as the process of categorisation continues to unfold. On this view, labels should have smaller effects in the unprimed case than the primed case, but the mechanism is the same in both. Clarke raises the important point that the activation of labels from visual inputs “must be probabilistic” in that a given visual stimulus can (and generally does) activate multiple labels in parallel. Clarke then wonders what functional role the label feedback may play given the probabilistic nature of labels activated by the input, and points out that if the visual stimulus was sufficiently unambiguous to activate only one label, then it is unclear what functional role labels would serve since the categorisation problem would appear to be solved. First, we entirely agree that most visual inputs will activate multiple labels at one time – indeed it is useful to think of the input as activating a set of likelihoods for labels (e.g. as discussed by Spivey, 2008). Indeed, recognising the probabilistic nature of this vision-to-label link as multiple labels are activated in parallel helps to clarify a point of confusion concerning the way in which labels can affect visual processing. Researchers sometimes conflate the activation of labels – wherein multiple labels may become active to varying degrees in response to a given stimulus – with overt language production – wherein one label is typically selected for production. A consequence of conflating these two processes is that it becomes difficult to think about how labels might influence perceptual processing in cases where multiple labels are active or in cases where overt naming is impractical owing to e.g. rapid visual presentation or the presentation of multiple