Where ‘it’ is an ordinary physical object like my favourite pen, each of these three identifies a process by which knowledge could be acquired.

(This is not to say that talk about reasoning invariably picks out a process (cf Alvarez), just that there is a process.)

There seems to be a clear contrast the first and last ways of coming to know about it. (Philosophers debate about whether the first and second are genuinely different routes to knowledge.)

Even if, as do I, you think perception involves inference-like processes, there is still a contrast between inferring and perceiving.

For instance, suppose you know that I am all but inseparable from my favourite pen. Then when you see me arrive you can infer that my pen is here too. Contrast this with simply seeing my pen on the table. In one case we see something other than the pen, namely me, and inferring that the pen is here; in the other case you are seeing the pen itself.

This is the contrast I need in what follows.

Are mental states the kinds of thing we can come to know about through perceiving them? I will focus on emotions like joy.

So we can’t straightforwardly take what people say as a guide to what they perceive.

Here you might object that we can perceptually experience things like winning or losing, so there is no reason to question verbal reports.

But note that taking this line amounts to rejecting the claim I started with.

I claimed that there is a contrast between seeing an indicator and inferrring the presence of an object

If there is a contrast here, what we colloquially call seeing someone win must be a case of perceiving indicators and inferring winning. So to say that we can straightforwardly infer that people perceive emotions from their verbal reports is to reject the existence of the very contrast that allows us to make sense of the question.

Part of the evidence relevant to the perceptual hypothesis is evidence that humans can categorically perceive expressions of emotion.

But what does the evidence from studies of categorical perception tell us about the truth or falsity of the perceptual hypothesis ?

But if we reject this assumption, what is the alternative?

Let me explain the increasingly bold commitments involved in accepting this conjecture.

First, the things categorised in categorical perception of expressions of emotion are events rather than configurations or anything static. (Note that this is consistent the fact that static stimuli can trigger categorical perception; after all, static stimuli can also trigger motor representations of things like grasping \citep{borghi:2007_are}.)

Second, these events are not mere physiological reactions (as we might intuitively take blushing to be) but things like frowning and smiling, whose performance involves motor expertise. \footnote{ To emphasise, one consequence of this is that not everything which might intuitively be labelled as an expression of emotion is relevant to understanding what is categorised by perceptual processes. %For example, in the right context a blush may signal emotion without requiring motor expertise. }

Third, these events are perceptually categorised by the outcomes to which they are directed. That is, outcomes represented motorically in performing these actions are things by which these events are categorised in categorical perception.

Should we accept the wild conjecture? It goes well beyond the available evidence and currently lacks any reputable endorsement. In fact, we lack direct evidence for even the first of the increasingly bold commitments just mentioned (namely, the claim that the things categorically perceived are events). A further problem is that we know relatively little about the actions which, according to the wild conjecture, are the things categorical perception is supposed to categorise (\citealp[p.\ 47]{scherer:2013_understanding}; see also \citealp{scherer:2007_are} and \citealp{fernandez-dols:2013_advances}). However, the wild conjecture is less wild than the only published responses to the problems that motivate it.% % (which, admittedly, are wilder than an acre of snakes).% \footnote{ See \citet[p.\ 15]{motley:1988_facial}: ‘particular emotions simply cannot be identified from psychophysiological responses’; and \citet[p.\ 289]{barrett:2011_context}: ‘scientists have created an artifact’. } And, as I shall now explain, several considerations make the wild conjecture seem at least worth testing.

Consider again the procedure used in testing for categorical perception. Each experiment begins with a system for categorising the stimuli (expressions). This initial system is either specified by the experimenters or, in some cases, by having the participants first divide stimuli into categories using verbal labels or occasionally using non-verbal decisions. The experiment then seeks to measure whether this initial system of categories predicts patterns in discrimination. But what determines which category each stimulus is assigned to in the initial system of categories? You might guess that it is a matter of how likely people think it is that each stimulus---a particular facial configuration, say---would be associated with a particular emotion. In fact this is wrong. Instead, each stimulus is categorised in the initial system according to how suitable people think such an expression would be to express a given emotion: this is true whether the stimuli are facial \citep{horstmann:2002_facial} or vocal \citep{laukka:2011_exploring} expressions of emotion (see also \citealp[pp.\ 98--9]{parkinson:2013_contextualizing}). To repeat, in explicitly assigning an expression to a category of emotion, people are not making a judgement about the probability of someone with that expression having that emotion: they are making a judgement about which category of emotion the expression is most suited to expressing. Why is this relevant to understanding what perceptual processes categorise? The most straightforward way of interpreting the experiments on categorical perception is to suppose that they are testing whether perceptual processes categorise stimuli in the same ways as the initial system of categories does. But we have just seen that the initial system categorises stimuli according to the emotions they would be best suited to expressing. So on the most straightforward interpretation, the experiments on categorical perception of expressions of emotion are testing whether there are perceptual processes whose function is to categorise actions of a certain type by the outcomes to which they are directed. So the wild conjecture is needed for the most straightforward interpretation of these experiments. This doesn't make it true but it does make it worth testing.

So far I have focussed on evidence for categorical perception from experiments using faces as stimuli. However, there is also evidence that perceptual processes categorise vocal and facial expressions alike (\citealp{grandjean:2005_voices,laukka:2005_categorical}; see also \citealp{jaywant:2012_categorical}). We also know that %judgements about which emotion an observed person is expressing in a photograph can depend on the posture of the whole body and not only the face \citep{aviezer:2012_body}, and that various contextual factors can affect how even rapidly occurring perceptual processes discriminate expressions of emotion \citep{righart:2008_rapid}. There is even indirect evidence that categorical perception may concern whole bodies rather than just faces or voices \citep{aviezer:2008_angry,aviezer:2011_automaticity}. In short, categorical perception of expressions of emotion plausibly resembles categorical perception of speech in being a multimodal phenomenon which concerns the whole body and is affected by several types of contextual feature. This is consistent with the wild conjecture we are considering. The conjecture generates the further prediction that the effects of context on categorical perception of expressions of emotion will resemble the myriad effects of context on categorical perception of speech so that `every potential cue ... is an actual cue' (\citealp[p.\ 11]{Liberman:1985bn}; for evidence of context effects see in categorical perception of speech, for example, \citealp{Repp:1987xo}; \citealp{Nygaard:1995po} pp.\ 72--5; \citealp{Jusczyk:1997lz}, p.\ 44).

Now I think we can answer this question. Or rather, we know that the answer depends on what the objects of categorical perception are.

If the objects of categorical perception are merely facial expresssions, then I think the evidence does not support the claim that we see aspects of each others’ mental lives.

But if, as I’ve conjectured, the objects of categorical perception are actions directed to the expression of particular emotions, then I think the evidence does provide modest support for the claim that we perceive aspects of each others’ mental lives.

This claim will need qualifying in various ways.

First, the evidence suggests that categorical perception of expressions of emotion, like categorical perception of speech, is not tied to single modality in any very robust way.

Second, I suggest elsewhere that categorical perception isn’t exactly like perception as philosophers standardly understand it. In particular, its phenomenology is to be understood in terms of \textbf{PHENOMENAL EXPECTATIONS} rather than by analogy with perception of shapes or textures.

But let me put these aside because I want to mention a second challenge.

The challenge is to specify a model of anger and other mental states which captures how these mental states appear to the perceivers. This is a hard challenge to meet because available models arguably render mental states imperceptible.

But what is a model of the emotions? Since emotions are mental states, here I think we can step back and think about mental states generally. On a widely accepted view, mental states involve subjects having attitudes toward contents (see figure). Possible attitudes include believing, wanting, being happy that, and being angry that. The content is what distinguishes one belief from all others, or one desire from all others. The content is also what determines whether a belief is true or false, and whether a desire is satisfied or unsatisfied. There are two main tasks in constructing a model of mental states. The first task is to characterise some attitudes. This typically involves specifying their distinctive functional and normative roles.% \footnote{ For examples, see \citet{Bratman:1987xw} on intention or \citet[][chapter 11]{Velleman:2000fq} on belief. } % The second task is to find a scheme for specifying the contents of mental states. This typically involves one or another kind of proposition, although some have suggested other abstract entities including map-like representations.% \footnote{ See \citet[p.\ 163]{Braddon-Mitchell:1996ce}: ‘what is inside our heads should be thought of as more like maps than sentences.’ }

FIRST LIMIT: Determinate emotions (the particular evaluation implied by my anger in this case) vs the determinable (anger, in this case). Categorical perception provides evidence only for the determinable. (Joel said in discussion that he made this point with an analogy with CP of colour in Smith (2010)).

SECOND LIMIT: Static emotions vs emotions unfolding (see notes below from ‘sharing a smile’). It seems that CP doesn’t capture dynamics of emotion.

So only a very crude model of the mental would capture emotions as we categorically perceive them. What we need here is a minimal theory of the emotions.

Motor theory of speech perception and goal ascription ...