Is there an empirical case for semantic perception?

ABSTRACT

I argue that results in perception science do not support the claim that there is semantic perception or that typical, unreflective utterance comprehension is a perceptual process. Phenomena discussed include evidence-insensitivity, the Stroop effect, pop-out, and adaptation – as well as how these phenomena might relate to the function, format, and structure of perceptual representations. An emphasis is placed on non-inferential transitions from perceptual to conceptual representations, which are important for debates about the admissible contents of perception more generally.

KEYWORDS:

• perception
• meaning
• pop-out
• adaptation
• iconicity

Acknowledgements

Thanks to Peter Achinstein, Ned Block, Tyler Burge, Chaz Firestone, E. J. Green, Maegan Kaczmarek, Ian Phillips, Georges Rey, McKenzie Young, and audiences in Bogota and Oslo and at Johns Hopkins and the Society for Philosophy and Psychology. Thanks also to the referee. Special thanks to the organizers of the workshop on which this special issue is based—Anna Drożdżowicz and Kim Pedersen Phillips—and to the other workshop participants: Nick Allott, Brendan Balcerak Jackson, Brit Brogaard, Elvira Di Bona, and Anders Nes. Extra thanks to Brit for her work on this topic and for conversation.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 In what follows, I mostly elide (or, attempt to remain largely neutral on) some important issues. Just what aspects of content are supposed to be perceived in utterance comprehension: what is said, explicature, some implicatures as well? (I will assume that lower-level semantic features cognitively inaccessible except via reflective theorizing and empirical investigation, such as Pietroski’s (2018) instructions to construct a concept, are not at issue.) To what is the content supposed to be attributed in perception: a speaker, a sentence in context, an utterance? How finely is the attributed content individuated? What speech acts are supposed to be attributable in perception – presumably at least asserting, asking, and ordering, but also more, or more finely delineated, speech acts?

2 Block (forthcoming) calls such states perceptual judgments. I avoid this terminology for two reasons. First, it courts confusion: such judgments are not part of perception, but perception-based. Second, some understand judgments to be occurrent beliefs and thus committal. Also, Block seems to assume that perceptual judgments causally precede beliefs. There is room to question this – for example, from a Spinozan perspective on belief formation (Gilbert 1991; Mandelbaum 2014). A remark also on automaticity: There are various notions one can distinguish – for example, a process can be involuntary but require a certain kind of attention (see Section 2.2 on Stroop interference below). I won’t take the matter further here. See Palmeri (2003) for discussion.

3 While being meaningful requires having a meaning, it’s a distinct question whether perceiving something as meaningful requires perceiving it as having some specific meaning. The general Humean principle that perceiving a determinable requires perceiving a determinate should be rejected. But restricted versions could hold, perhaps as a nomological matter. In the case at hand, however, a perception of meaningfulness (if there is such) could be driven, for example, by the perceptual representation of lexicality or of lower-level semantic features. (That said, it is also plausible that the ‘More people have been to France’ case involves representations of higher-level meaning fragments. The proponent of semantic perception could try arguing that they are perceptual and drive the perception of meaningfulness. One would need to see what the argument is to know if it escapes the replies presented in this article.)

4 The conceptual state to the effect that it is wood could be mediated by a distinct, evidence-insensitive conceptual state to the effect that is brown, with such-and-such shape and texture, etc. (i.e. that it has the ‘wood gestalt’). But see the remarks on basic conceptualization in Section 5.

5 The question ‘Why some evidence-sensitivity and some evidence-insensitivity?’ arises just as much for proponents of semantic perception as for a view that adverts to post-perceptual conceptualization. Indeed, evidence-sensitivity, insofar as it’s a cognitive top-down effect, is often raised as an objection to semantic perception (e.g., Stanley 2005). (I say ‘cognitive top-down effect’ to side-step debates about what should and shouldn’t count as cognitive penetration – see Gross (2017b).) The presence of cognitive top-down effects is not in itself inconsistent with semantic perception, as Brogaard (2017, 2020) correctly points out; uncontroversial cases of perception are also subject to such effects. However, many of the top-down effects involved in utterance comprehension seem different in kind from those well-established in uncontroversial cases of perception, which are mainly attentional effects. The possibility of post-perceptual conceptualization might open up as well the possibility of further kinds of top-down effect.

6 There is an on-going debate whether perceptual color attribution is categorical (Witzel 2019). The point in the text is neutral on this. Even if perceptual color attribution is categorical, the task might still require a transition from the perceptual attribution of a color category to a conceptual attribution (e.g. from a perceptual representation as of that’s being red to a conceptual representation that it is red – where I’m taking license to use ‘red’ here, first, for a perceptual attributive, and then for a conceptual attributive).

7 Renumbered figures and captions reproduced from Brogaard (2020). Brogaard calls the distractors ‘pseudowords’. ‘Non-words’ might be a better label because ‘pseudoword’ is often used for meaningless but phonologically acceptable (pronounceable) strings. Many of Brogaard’s non-words are not pseudowords in this sense. Orthographic constraints are violated as well.

8 To be clear, Tian & Huber do not themselves claim that their results support a perceptual view of utterance comprehension. Indeed, they explicitly refer to semantic satiation as a non-perceptual process (Tian and Huber 2010, 271).

9 There is also evidence of neural adaptation in the visual word form area. But the relation of this phenomenon to behavioral adaptation remains unclear (Larsson and Smith 2011). Two further notes: First, for ease of exposition, in the text, I do not pause to distinguish visual word forms, auditory word forms, and lexemes, though these distinctions and others are important for a fuller discussion of what’s exhibiting adaptation. (Lexemes are representations stored in long-term memory that link simple word forms with other linguistic features. Complex words (Pat’s) are constructed from them.) Second, on terminology: The term ‘adaptation’ is used in the psycholinguistics literature for a different phenomenon: adjusting one’s speech and comprehension strategies to bring them into conformity with others’. This is closer to priming and the opposite of a repulsive effect. It’s been argued that syntactic adaptation, in this sense, explains a kind of satiation (Lu, Lassiter, and Degen 2021). But ‘satiation’ is also being used in a different sense here! This kind of satiation is when a string is judged less unacceptable after repeated exposure.

10 The lack of repulsive aftereffects could be owing to differences in the organization of semantic space in comparison with the quality spaces of lower-level features. But this difference might itself again suggest that the phenomena of adaptation and semantic satiation are different.

11 There is also Burge’s (2010, 413) view that perceptual constancy mechanisms are sufficient and perhaps necessary for perception. But, as Burge notes, because the constancy mechanisms need to be perceptual, one cannot simply use the presence of constancy mechanisms to settle hard cases. In addition, Burge allows some features to be perceptually represented in virtue of being ‘harnessed’ to a process that involves perceptual constancy mechanisms for other features (cf. Gross 2017a). It’s thus unclear whether consideration of constancy mechanisms would advance debates about semantic perception.

12 It is a crucial part of Burge and Block’s views that perception (non-conceptually) attributes properties. Such attributions are general in the sense that attributives can hold of multiple entities, and it is fundamental to their function that this is so – for example, in enabling the generation of expectations. But – the claim is – perceptual contents do not exhibit quantification. There is an issue here. This precludes accounts of perceptual content in terms of existential quantification: there is green there. Block (forthcoming), contra Burge (2010), is open to such perceptual contents. Perhaps, though, it suffices for the non-propositionality of perception that its logical complexity is greatly restricted. Cf. Burge (2010, p. 540).

13 My intent here is to explore the question of semantic perception through the lens of Burge and Block’s views, with which I am broadly sympathetic, not to argue for these views. As indicated, their views in some ways differ. Relevant to the above: Burge (2020) seems more open than Block to Bayesian computations in perception, but would not consider such transitions inferential since the probabilistic representations are not propositional. Block’s (forthcoming) argument against Bayesian inference in perception centers more on a defense of instrumentalism regarding the Bayesian models. (See also Gross 2020.)

14 I am assuming that utterance understanding involves more than in some sense entertaining the proposition expressed: there is a representation as well of force and an attribution to something (speaker, utterance, sentence-in-context). But if utterance understanding involves only entertaining the proposition expressed, then the propositionality point is even more obvious. A different strategy would be to drain the perceptual indirect speech attribution of propositionality and logical structure in neo-Davidsonian fashion (Nes, this issue). Perhaps the perceptual content is something like: <<that₁, that₂>, samesays>, where the first demonstrative refers to the utterance and the second to an appropriate mental representation. A problem is that the second demonstrative is not perceptual and so neither is the attribution of samesaying.

15 Perhaps a neo-Fregean about indirect speech attributions may object that the matrix-representations represent Sinne that stand in inferential relations mirrored by the representations themselves. But there would remain the representational constituents outside the matrix.

16 At least the rejection will seem unmotivated if the proponent of semantic perception wants to base her case on empirical arguments of the sort we’ve been examining. Other proponents might try rejecting the claim that perception is a natural kind or try arguing for a kind of pluralism according to which there are various notions of perception useful for different purposes (cf. Nes, Sundberg, and Watzl forthcoming).

17 Nor is the claim that iconicity plays no role in post-perceptual cognition. Indeed, Burge (2020) maintains that all propositional beliefs immediately formed from perceptions are iconic. This is relevant in considering the possibility of iconic representation in perception with non-iconic elements (cf. Clarke forthcoming). There are certainly iconic representations with non-iconic elements – e.g. maps with a ‘You are here’. Perhaps there are mental representations of this sort as well. But it wouldn’t follow that there are perceptual representations of this sort. It could be that any non-iconic elements would be introduced post-perceptually. The arguments of this paper suggest that this would be so at least with meaning attributions.

18 Kazanina, Bowers, and Idsardi 2018 provide a general defense of phonemes’ central role in speech perception. Phonemes do not exhaust what is perceived in typical utterance comprehension: there is stress, duration (which, like stress, provides cues to word boundaries), larger intonational contours, etc. There are also the various non-linguistic items and features – perceived features of utterance context – that can be relevant to utterance comprehension. Moreover, the phonemes may be perceived as ordered and grouped.

19 For the sake of simplicity, I am ignoring the possible mediation of post-perceptual, but still non-conceptual states, such as perhaps those in working memory. Cf., e.g. Burge (2014).