2.a The standard internalist approach

This subsection first lays out the general internalist approach to epistemology, then explains the standard internalist approach to testimonial epistemology.

Internalists claim that a belief’s justification depends only on factors which are “internal” to a subject’s mind. This is most often cashed out as some version of the idea that a belief is justified by a subset of a subject’s nonfactive mental states (see, e.g., Feldman and Conee Reference Feldman and Conee2001; Wedgwood Reference Wedgwood2002; Huemer Reference Huemer2006; White Reference White, Dodd and Zardini2014). An internalist might hold, for example, that a perceptual belief that P is justified by one’s experience as of P, where experiences are characterized purely in terms of how things perceptually seem for the subject. Similarly, a belief that Q might be justified by the prior beliefs that P and that if P, then Q, where beliefs are characterized purely in terms of what a subject takes to be true.

Internalists (including those cited in the previous paragraph) often speak interchangeably about the mental factors which justify a belief and the basis of a belief. The basis of a belief is the reason (or set of reasons) on the basis of which one believes that P. So, to claim that a belief is justified by some nonfactive mental state(s) is equivalent to claiming that the state(s) in question is/are the reason(s) on the basis of which one formed the belief. One’s perceptual experience as of P, for example, can be the reason on the basis of which one believes that P.

I’ll assume the orthodox, widespread view of epistemic basing which says the basis of a belief is the reason which caused that belief. Combined with the internalist commitments just sketched, this yields the view that the basis of S’s belief is S’s mental state(s) which caused S’s belief. While there are many available accounts of basing that incorporate additional conditions besides causation, it’s very common to include a causal condition at least as a necessary condition, even if not a sufficient one (for overviews of the literature, see Korez Reference Korez1997 and Sylvan Reference Sylvan2016). And, as we’ll see below, proponents of the standard internalist view of testimony discuss basing in terms of causation, in line with this view. Since both my negative and positive arguments in this paper will primarily concern this causal component of basing, they’ll apply to any account that includes causation as a necessary condition even if they also include other necessary conditions.Footnote ³

I turn now to the family of views I’ll call the ‘standard internalist approach’ to the epistemology of testimony, which I take to be the paradigmatic internalist way of understanding the justifying basis of testimonial beliefs. These views hold, in line with the broader internalist tradition, that the basis of a testimonial belief includes only mental states of the listener. In particular, they hold that it includes a mental state which somehow represents the content of the testifier’s utterance. From that starting point, there are various ways of developing this standard internalist approach. Since it’s the main approach I aim to replace with my alternative, externalist account, it’s worth spelling out some versions of this approach in detail.

Perhaps the most common are what we might call ‘doxastic’ versions of the approach. These views say that in the process of acquiring a testimonial belief that P, one first forms a belief that a testifier said that P; one then comes to believe P (at least in part) on the basis of one’s belief that the testifier said that P. In other words, one’s belief that the testifier said that P is among the mental states which causes one to believe P.

Malmgren (Reference Malmgren2006) gives one statement of such a view. She says:

Malmgren also claims that this specifically involves the belief about what the testifier said causing one’s knowledge (sec. 7), and that this talk of “knowledge” is really elliptical for talk of belief (225n56). So, she takes it that a testimonial belief that P is caused (at least in part) by one’s belief that a testifier said that P.

Barnett (Reference Barnett2015) also endorses a doxastic version of the standard internalist approach. He first claims that “when a source of testimony tells you that P, what you learn first is not P itself but instead merely the fact that the source says that P”; subsequently, “on the basis of what you learn first”—i.e., that the speaker said that P—you’re justified in believing that P (356–57). Barnett later writes about the basing relation in terms of causal relations between beliefs (378–79). So, it’s clear his considered view is that the basis of one’s testimonial belief that P includes the belief that S said that P.

Fricker (Reference Fricker1987) is also plausibly read as endorsing a doxastic version of the standard internalist approach. As a general necessary condition for justification, Fricker argues that a “subject’s belief must be causally dependent on his belief in (or some kind of sub-doxastic registering of) propositions … which, if articulated, would constitute a good argument for it” (61). She then argues, in line with the standard internalist approach, that the causes of justified testimonial beliefs include “a belief that the speaker has made an assertion with a particular content” (69–70).

We can also formulate ‘nondoxastic’ versions of the standard internalist approach. On these views, the basis of a testimonial belief includes some nondoxastic grasp of the meaning of an utterance. We can generate versions of the standard internalist approach by substituting in any such view of language comprehension. For example, Longworth (Reference Longworth2018) argues that comprehending an utterance “P” involves, rather than a belief about what a speaker said, a first-order attitude towards P, something akin to entertaining P. He then argues that a listener moves from understanding an utterance to accepting its content, a process of forming a belief “on the basis of having understood” the utterance (829).

There are other nondoxastic views of comprehension to which we could appeal to flesh out the standard internalist approach. For example, perceptual views of comprehension hold that, when we hear an utterance, we literally hear the meaning of that utterance (Fricker Reference Fricker and Barber2003; Brogaard Reference Brogaard2018). Such views aren’t necessarily developed as accounts of the basis of testimonial beliefs, but it’s easy to see how we could position them as such, along the same lines as Longworth’s account.

So, various epistemologists endorse some version of the standard internalist approach.Footnote ⁴ Furthermore, it makes sense to call this the ‘standard’ internalist approach because it’s simply a natural way of applying internalist ideas to testimony. Epistemologists—both internalist and externalist—often claim that, intuitively, it seems natural to cite the fact that a testifier said that P when asked about the reasons for one’s testimonial beliefs (Fricker Reference Fricker1987; Malmgren Reference Malmgren2006; Nagel Reference Nagel, Brown and Simion2021). If internalists want to accept this intuition, they must give a mentalistic interpretation of it. It seems natural to construe it as indicating that one’s testimonial belief that P was caused by a mental state representing what the testifier said. (I agree that this evidence is prima facie intuitively compelling, even though I ultimately argue against this approach.)

Proponents of the standard internalist approach differ as to exactly what else is included in the justifying basis of testimonial beliefs, besides agreeing that it includes a representation of the content of the testifier’s utterance.Footnote ⁵ However, my arguments in section 3 will aim to undermine any view which accepts this point, so I needn’t go into detail about other conditions they endorse.

2.b The other minds approach

This paper defends an alternative to the standard internalist approach, one on which the basis of a listener’s testimonial belief needn’t include only the listener’s mental states. Instead, the listener’s testimonial belief that P can be based (at least in part) on the testifier’s belief that P. In other words, the testifier’s belief can be among the reasons which cause the listener’s belief. I call this view the ‘other minds approach,’ since it says the basis of a listener’s belief can be located in another person’s mind. (Hereafter, for simplicity, I’ll refer to the other minds approach as the view that testifier’s belief can be the basis of the listener’s belief, though I leave open whether other factors are also included in that basis.)

I’ll remain neutral about the precise scope of the other minds approach. I’ll only argue that, contrary to the standard internalist approach, there are at least some cases in which the testifier’s belief is the basis of the listener’s testimonial belief. I therefore won’t take a stance on whether the internalist approach fails as an account of every single case of testimonial belief, as opposed to merely failing in certain cases. Even so, the other minds approach is quite a departure from the internalist way of conceiving the basing relation on which the basis of one’s belief must be a mental state in one’s own mind.

In section 5, I’ll defend the other minds approach by considering cases in which a listener constructs mental imagery which resembles the imagery before the testifier’s mind. I’ll motivate the approach by considering the nature of the causal relationship between each subject’s mental imagery. Considering this relationship reveals that a relationship of basing holds between the listener’s testimonial belief that P and the testifier’s belief that P, since the relationship between their beliefs exhibits certain characteristic properties of basing.Footnote ⁶

It might be tempting to think that the other minds approach is dead on arrival since the testifier’s belief is so far down the causal chain from the listener’s. Philosophers sometimes argue that the basis of a belief includes only proximate causes of that belief (cf. Wedgwood Reference Wedgwood2002; Turri Reference Turri2011). However, it seems there are many causal intermediaries between the testifier’s belief and the listener’s: the testifier’s speech, the listener’s perception of that speech, the listener’s comprehension of what was said, and perhaps the listener’s belief that the testifier said that P. So, one might claim that the testifier’s belief is too far removed from the listener’s to be its basis.

My goal is to close this apparent epistemological gap between the testifier’s and listener’s beliefs, at least in certain cases involving mental imagery. I’ll argue across sections 3–4 that, in certain such cases, no causal intermediary between the testifier’s and listener’s beliefs is part of the listener’s justifying basis. Furthermore, I’ll argue that the causal relationship which holds between the testifier’s and listener’s beliefs has the right properties to count as a basing relationship.

The first step is to argue, in the next section, that the standard internalist approach fails to account for the relevant imagery cases.

3.a Mental imagery in testimonial comprehension

Section 2 described some nondoxastic accounts of testimonial comprehension—accounts on which we comprehend an utterance “P” by, for example, entertaining or perceiving its meaning. Recent empirical work on language comprehension suggests one specific way in which we often nondoxastically comprehend testimony: by constructing mental imagery of the situation described by that testimony. This involves grasping the content the testifier has expressed by constructing mental imagery representing it. In BAD KITTY, for example, Oliver’s utterance “the cat jumped on the table and knocked over the vase” causes Esther to construct a mental image representing the state of affairs he describes, and she thereby grasps the meaning of his utterance.Footnote ⁸

Evidence for this form of comprehension comes from psycholinguistic studies of speech and text comprehension, on the basis of which its widely accepted that comprehension often involves mentally constructing a “situation model” of the scene described. At least in the kind of case I’m focused on, cases like BAD KITTY, such models are typically taken to be imagistic in format (Fincher-Kiefer Reference Fincher-Kiefer2001; Kurby and Zacks Reference Kurby and Zacks2013; Zwaan Reference Zwaan2016). As an interlocutor describes a situation to us, they prompt us to construct an imagistic representation of that situation, which is updated and modified as the conversation continues. We grasp the content a testifier has asserted by representing it in mental imagery.Footnote ⁹

Fincher-Kiefer (Reference Fincher-Kiefer2001) provides classic experimental evidence for thinking mental imagery can be involved in comprehension, as opposed to merely co-occurring with it. She hypothesizes that, if comprehending a story involves constructing mental imagery, then comprehension would be disrupted when subjects simultaneously engage in another task involving imagery—for example, holding some other, unrelated image in mind while reading. That’s because if their capacity to construct imagery is already in use for some other task, it can’t be fully recruited for comprehension. Fincher-Kiefer found that subjects’ comprehension is disrupted significantly more when engaging in simultaneous imagery tasks than when engaging in simultaneous verbal, nonimagery tasks, supporting the idea that imagery is involved in comprehension.

On the basis of such evidence, empirical researchers posit that situation modelling serves various functional roles related to discourse comprehension. For example, it allows one to connect a testifier’s references to particular objects to one’s own stored mental representations of those objects. In BAD KITTY, Oliver’s references to “the vase,” “the “cat,” and “the kitchen” activate imagistic mental representations of those objects that Esther has stored from past experiences, allowing her to comprehend which particular objects Oliver is referring to. Situation models also allow one to mentally organize the individual elements of a scene into the particular spatiotemporal organization a testifier is describing, thus constructing an overall, coherent representation of that scene.

By playing comprehension-related roles like this, situation models ground the distinction between perceiving a testifier’s particular words or utterances and comprehending what the testifier said about the world. In addition to perceptually experiencing an act of testifying, we comprehend that testimony by imaginatively representing the world as described by it.

Now, one might object that we can’t comprehend speech (a propositional representation) by forming mental imagery (often described as an “imagistic” or “iconic” representation rather than propositional). On this objection, since speech has propositional content while imagery has imagistic content, it doesn’t make sense to say we represent the content of an utterance in mental imagery.

However, there are various ways of responding to this objection. For one, we might just respond that mental images really can be vehicles for propositional content. This propositional view of mental imagery content is quite controversial (cf. Nanay Reference Nanay2015; Pearson and Kosslyn Reference Pearson and Kosslyn2015). However, it coheres well with a broader picture many accept, on which other perceptually formatted representations have propositional content, including perceptual experience (e.g., Dretske Reference Dretske1997; Byrne Reference Byrne2009) and episodic memory (e.g., Fernández Reference Fernández2006; Byrne Reference Byrne2010). If one accepts this broader picture, then this response seems promising.

Alternatively, one might argue that the propositional/imagistic distinction merely concerns the format of representations rather than their content. As Nanay (Reference Nanay and Zalta2021, sec. 1.5) argues, it seems representations in these distinct formats can represent the same content—that, for example, the utterance “the cat broke the vase” can represent the same state of affairs as a mental image of the cat breaking the vase. If that’s true, a mental image could be a way of comprehending the same content a testifier conveyed in their speech.

In what follows, I remain neutral between these two responses to the foregoing objection.

3.b Mental imagery as format of testimonial belief

In addition to their role in comprehension, situation models are also often the representations we commit to long-term memory when retaining information about the world gleaned from testimony. In other words, rather than remembering a testifier’s specific words or utterances, we often instead store a situation model of the situation they described to us (Zwaan and Radvansky Reference Zwaan and Radvansky1998; Zacks, Mar, and Calarco Reference Zacks, Mar, Calarco, Schober, Rapp and Britt2018; McClelland et al. Reference McClelland, Hill, Rudolph, Baldridge and Schütze2019).

Evidence for this comes from classic studies such as those by Bransford, Barclay, and Franks (Reference Bransford, Barclay and Franks1972). In one such study, subjects first listened to a set of sentences being read to them, including one of the following:

1. (1) Three turtles rested on a floating log, and a fish swam beneath them.

2. (2) Three turtles rested beside a floating log, and a fish swam beneath it.

Later, subjects were asked to identify which sentences on a second list they recognized as having heard previously. This second list included one of the following:

1. (3) Three turtles rested on a floating log, and a fish swam beneath it.

2. (4) Three turtles rested beside a floating log, and a fish swam beneath them.

Notice that while all of these sentences are superficially very similar (cf. Glenberg, Meyer, and Lindem Reference Glenberg, Meyer and Lindem1987), the scenes described by (1) and (3) are the same, while (2) and (4) each describe scenes distinct from the rest. When asked to recall which sentences they recognize, subjects who’d earlier heard (1) are highly likely to mistakenly say they recognize (3), while they can much more easily discriminate between all other pairs of sentences.

The standard explanation of such results is that subjects don’t tend to retain and then recollect a representation of a speaker’s particular words or utterances. Instead, they store a situation model of the scene described by those utterances. When listening to the second set of sentences, they then retrieve this stored situation model as a means of determining whether they previously heard any of those sentences.

The kind of case just described involves committing to memory information about a toy, fictional scene—one doesn’t actually take the world to be such that three turtles sat on a log while a fish swam beneath them. However, we can apply the same kind of distinction between storing a representation of the speaker’s words and storing a situation model of the scene these words describe to cases like BAD KITTY, where what’s described is instead the actual world. We can distinguish Esther storing a representation of Oliver’s utterances, such as “the cat knocked over the vase,” from Esther storing a representation of the scene Oliver’s utterances describe, i.e., the cat knocking over the vase. The difference between this case and the turtle/fish case is that Esther isn’t storing a representation of a fictional scenario. Instead, Esther is storing a representation of the way the world is, as learned from Oliver’s testimony. Committing a situation model to memory is therefore a way for her to remember the fact that the cat knocked over the vase, information acquired when Oliver’s testimony prompted her to imagine this scene.

This suggests that, in addition to its role in comprehension, such mental imagery can also be the format of a listener’s testimonial belief. That’s because this imagery plays the same functional role that we’d assign to a belief acquired from testimony. Acquiring a testimonial belief is a process of acquiring some information about the world from a testifier, after which one can store this belief and have it available for later retrieval. And, in the kind of case I just described, one doesn’t comprehend testimony via mental imagery and then subsequently form and store a distinct representation or belief. Instead, the situation model constructed in response to testimony is the format in which one acquires and retains information from testimony—it’s the means by which one adds information gleaned from testimony to one’s existing set of information about the world. So, we should take such imaginative representations to be a listener’s testimonial beliefs.Footnote ¹⁰

My argument here assumes that beliefs can be imagistic. This is controversial but does have various adherents (Kaplan Reference Kaplan1968; Van Leeuwen Reference Van Leeuwen2013; Langland-Hassan Reference Langland-Hassan2020). My view is that once we recognize that imagistic situation models can play the functional role of testimonial beliefs, this gives us a positive reason to accept that beliefs can be imagistic, at least in relevant cases.

One might object that beliefs are propositional attitudes while mental images are nonpropositional. There are at least two ways we might respond. If the propositional/imagistic distinction is taken to concern content, then we might argue mental images do in fact have propositional content (see the end of Section 3.a for more on this view). Alternatively, if the propositional/imagistic distinction concerns format rather than content (see again the end of Section 3.a), we might argue that the considerations I’ve given in this section reveal some beliefs are nonpropositional after all. I’ll remain neutral between these responses.Footnote ¹¹

3.c Comprehension and belief formation as simultaneous

When the very same instance of mental imagery is involved in both comprehending testimony and forming a testimonial belief, can we both comprehend and acquire a belief from testimony simultaneously via a single process of constructing that imagery? Or, must we always first comprehend testimony by constructing mental imagery and then, subsequently, accept or assent to the content of this imagery as true, only after which it becomes a belief? I think it’s plausible that the two are often simultaneous due to the role of trust in acquiring testimonial beliefs.

I’ll adopt Nguyen’s (Reference NguyenForthcoming) account of trust as an unquestioning attitude. Nguyen argues that trust needn’t involve an explicit attitude towards a testifier, nor a conscious decision to trust a testifier. Instead, trusting a testifier often involves a disposition to accept what she says as true without deliberating about it or pausing to evaluate her testimony’s merits. In other words, accepting trusted testimony simply involves a lack of actively questioning or deliberating. In this sense, Nguyen argues that our relationship to those we trust is very much like our relationship to our own cognitive faculties. Unless we’re aware of a reason not to, we’re disposed to simply treat the information delivered by these faculties, such as perception and memory, at face value. Likewise, trusting a testifier involves being disposed to immediately accept the content of her testimony as true with no intervening checking or deliberation.Footnote ¹²

Sometimes, we initially distrust a testifier, or we’re initially unsure about whether to trust her. Often, though, we already trust a testifier prior to receiving testimony. I’ll assume that this is the case in BAD KITTY: that Esther already trusts Oliver before receiving his testimony. (It’s controversial how often we trust testifiers prior to receiving their testimony; for my purposes, all that matters is that we do so at least sometimes, which seems accurate at least in the context of close personal relationships.)

If a listener already trusts a testifier—i.e., is predisposed to simply take her at her word—then we should expect her to automatically add the content of that testifier’s word to her existing store of beliefs about the world (cf. McDowell Reference McDowell, Matilal and Chakrabarti1994; Westra and Nagel Reference Westra and Nagel2021). This seems to fit with the phenomenology of accepting trusted testimony. Imagine being Esther in a case of trust like BAD KITTY in comparison with this case:

Here, Esther could still comprehend Oliver’s testimony by engaging in situation modelling for the purpose of mere comprehension. But what distinguishes this case from BAD KITTY is that she uses mental imagery merely to entertain what Oliver tells her rather than trusting him and therefore automatically accepting what he says at face value.

Now, if testimonial comprehension and belief formation were distinct states or processes, we might expect automatic, trusted acceptance to involve the former process automatically causing the latter. That would be a version of the standard internalist approach on which comprehension of a speaker’s utterance causes a testimonial belief, thus serving as its basis. So, on their own, these considerations about the nature of trust don’t suggest anything about the causal relationship between comprehension and belief formation.

However, in cases involving imagery, these considerations about trust give us reason to think comprehension and belief formation can occur simultaneously. In BAD KITTY, Esther’s construction of mental imagery can be both the process by which she comprehends Oliver’s testimony and the process by which she forms a testimonial belief. If Esther is predisposed to take Oliver’s word at face value, then there’s no reason to think she must first comprehend his testimony and then form a testimonial belief. That’s because if the same imagery serves both roles, it would be psychologically redundant to first entertain the content of his testimony and then, subsequently, accept it. Instead, we should expect that Esther’s construction of imagery is simultaneously a process of comprehending Oliver’s testimony and of forming a belief. In other words, since Esther is already disposed to accept what Oliver says before he even says it, the imagistic process by which she comprehends his testimony will already be a process of adding the information he transmits into her existing store of information about the world.

Of course, it’s ultimately an empirical question whether such construction of mental imagery ever simultaneously functions as both the process of comprehension and the process of belief formation. But my point is that we have theoretical reason to think the two can occur simultaneously based on philosophical considerations about trust. If a listener trusts a testifier, it would be psychologically redundant to engage in two, distinct steps, since both comprehension and belief formation involve the same mental imagery. Unless we have good reason to favor a more complex theory, we should opt for a more simplistic theory over one that posits multiple, psychologically redundant processes.

To sum up this section so far: there are cases of testimonial belief acquisition in which the construction of mental imagery serves simultaneously as both the means of comprehension and the formation of a listener’s belief. I next argue that the standard internalist approach to conceptualizing the basis of testimonial beliefs cannot account for such cases.

3.d Against the standard internalist approach

The standard internalist approach says that a listener first represents the content of a testifier’s utterance either via a belief that the testifier said that P or via some nondoxastic way of grasping that content. This mental state then serves as (part of) the basis of the listener’s testimonial belief. On an orthodox, casual picture of the nature of basing, this means the listener’s mental state causes her testimonial belief. However, I now argue that this internalist approach cannot apply to cases on which imagery simultaneously serves as the process of both comprehending testimony and of forming a testimonial belief.

It should be clear why these cases preclude a picture on which a listener nondoxastically comprehends what was said and this causes her testimonial belief. In the relevant cases, the construction of mental imagery is the process by which we simultaneously comprehend testimony and acquire a testimonial belief. But if this same process serves both of these roles simultaneously, then there’s no point where a listener’s comprehension causes her belief. Since the basis of a belief is its cause, it can’t be that one’s nondoxastic comprehension of what was said is part of the basis of one’s belief.

Such cases also pose counterexamples to doxastic versions of the standard internalist approach, where a listener’s belief that a testifier said that P causes her testimonial belief that P. That’s because a listener has to comprehend an utterance before she forms a belief about what was said—one can’t form a belief that a testifier said that P without comprehending the testifier’s utterance “P.” But, again, the process of comprehending testimony is, in the cases I’ve considered, simultaneously the process of forming a testimonial belief. If comprehension must come before a belief about what was said, and if comprehension and testimonial belief formation occur simultaneously, then testimonial belief formation also comes before a belief about what was said. It therefore can’t be that one’s belief about what a testifier said causally precedes one’s testimonial belief.

The various versions of the standard internalist approach thus assume a picture of the psychology of testimonial belief formation which doesn’t apply to the imagery cases I’m considering.

4.a Belief that the testifier believes that P

The other minds approach holds that the listener’s belief that P can be based on the testifier’s belief that P. However, one might object that the listener’s belief that the testifier believes that P is more causally proximate to her testimonial belief. So, an alternative view says that in forming a testimonial belief, the listener first learns about what the testifier believes. Then, this belief that the testifier believes that P is among the causes of the testimonial belief that P.

However, in the kind of imagery cases I considered in the previous section, the listener’s belief about what a testifier believes cannot be among the causes of her testimonial belief. Before the listener can form a belief about what the testifier believes, she first has to comprehend the testifier’s utterance: since an utterance serves to express the testifier’s belief, it’s only by comprehending the utterance that the listener learns what the testifier believes. So, comprehending an utterance must be causally prior to forming a belief about what the testifier believes. In the cases I’m considering, though, comprehension is simultaneous with forming a testimonial belief. Thus, in these cases, forming a testimonial belief that P must be prior to forming a belief that the testifier believes that P. So, it can’t be that a belief about what the testifier believes causes the listener’s testimonial belief.

4.b Perceptual experience of the testifier’s utterance

There’s a causal line that proceeds from the testifier’s belief, through the testifier’s utterance, through the listener’s perception of that utterance, to the listener’s imagery (i.e., the listener’s belief). So, it seems that the listener’s perceptual experience of the testifier’s utterance is a mental state that’s causally proximate to her testimonial belief. Thus, another potential alternative to the standard internalist approach is to give an account of the listener’s basis in terms of this experience. However, I now argue that in the imagery cases I’m focused on, the listener’s experience isn’t part of her justifying basis.

First, note that if this experience is part of the listener’s justification for her belief, it must be so in the sense that it’s part of an inference from this experience to her belief. That’s because the contents of one’s experience of a testifier’s utterances are quite distinct from the contents one ends up believing. In BAD KITTY, for example, Esther auditorily perceives Oliver uttering strings of words that form sentences (e.g., “the cat knocked the vase to the floor”), while she acquires a testimonial belief not about words, sentences, or utterances, but about the situation Oliver describes (the cat knocking the vase to the floor).

A justified transition from the perception of an utterance “P” to a belief that P would therefore require something in addition to the mere perception of the utterance—something more like an inference that includes additional premises connecting the content of one’s perception to the content of one’s testimonial belief. Such premises might include, for example, one’s linguistic knowledge about what the testifier’s words mean or evidence from the meanings of past similar utterances one experienced that would allow one to justifiably infer one’s testimonial belief. Suppose, for example, that one knows from past experience that an utterance “P” has the function of presenting proposition P as true; based on this knowledge, in conjunction with one’s present experience of a testifier uttering “P,” one might then justifiably infer P.

Now, I take it as intuitively clear that Esther’s belief in BAD KITTY is justified as long as she has no reason to think Oliver is untrustworthy or that his testimony is misleading.Footnote ¹³ So, if the justifying basis of Esther’s testimonial belief is to include her perception of Oliver’s utterance, this basis must also include other premises that allow her to justifiably infer her testimonial belief from her perception of this utterance. However, I’ll now argue that, in the kind of imagery case on which I’m focused, it’s implausible that we go through an inferential process to transition from perceiving a testifier’s speech to acquiring a testimonial belief.

As Burge (Reference Burge1993) points out about testimony in general, introspective evidence suggests we don’t go through any kind of inferential process to transition from perceiving a testifier’s speech to comprehending it; rather, such transitions seem quite fast and automatic.Footnote ¹⁴ According to my account of relevant imagery cases in section 3, comprehending testimony is simultaneous with acquiring a belief from it via a single process of imagery construction. So, if Burge is right that the transition from hearing a testifier’s speech to comprehending it is merely causal rather than inferential, the same would go, in these cases, for the transition from hearing the testifier’s speech to forming a testimonial belief.

Now, Burge’s opponents have pointed out that this introspective evidence isn’t airtight: it’s plausible that many of our everyday inferences are based on tacit or implicit premises rather than performed in an explicit, consciously introspectable way (Christensen and Kornblith Reference Christensen and Kornblith1997; Malmgren Reference Malmgren2006). However, empirical facts about the relation between perceiving speech and perceptual processing help support Burge’s view in my imagery cases.

Both behavioural and brain-imaging evidence suggest that perceiving speech rapidly and automatically modulates perceptual processing such that the brain automatically begins constructing perceptual representations of objects referred to in speech (Lupyan and Clark Reference Lupyan and Clark2015; Boutonnet and Lupyan Reference Boutonnet and Lupyan2015). Importantly, this includes the modulation of very low-level perceptual areas typically thought to be encapsulated from conscious reasoning or inferential processes. So, for example, hearing the word ‘cat’ cues the visual system to expect to see a cat even before one has been visually presented with a cat. This amounts to the brain automatically beginning to construct a visual representation of a cat prior to perceiving a cat, which then has measurable effects in terms of speed and performance on object recognition and perceptual discrimination tasks.

This sort of mechanism plausibly also explains why the brain automatically constructs mental imagery in response to perceiving speech, as in the kinds of testimony cases I’m considering (Clark Reference Clark2016, chap. 3; Jones and Wilkinson Reference Jones, Wilkinson and Abraham2020). Sensory imagining involves constructing perceptual representations in absence of stimuli. So, if our brains automatically construct perceptual representations in response to speech, we should expect this to result in conscious mental imagery in cases where one isn’t presented with a relevant stimulus (e.g., where one hears a description of some object or scene but that object or scene isn’t currently present to be perceived). In the kinds of cases I’ve been discussing, this seems intuitive and phenomenologically apt: when listening to a testifier triggers mental imagery, this imagery often occurs automatically and unreflectively. This is explained by the fact that, when speech triggers perceptual processing, this is an automatic and noninferential process rather than a personal-level, inferential one.

So, in my imagery cases, the transition from perceiving a testifier’s speech to acquiring a testimonial belief is a noninferential one, a merely causal transition that occurs when the perception of speech automatically activates mental imagery. I argued above that if the listener’s experience is part of what justifies her belief, this would require additional, inferential steps linking the content of the experience to the content of the testimonial belief. So, a listener’s experience of a testifier’s utterances isn’t part of the justifying basis of her testimonial belief.

4.c The testifier’s utterance

Again, there’s a causal line that proceeds from the testifier’s belief, through the testifier’s utterances, through the listener’s experience of those utterances, to the listener’s imagery. So, the remaining causal intermediary that might count as part of the listener’s basis is the testifier’s utterance itself. (This is already quite a departure from the internalist approach, since it cites a cause that isn’t one of the listener’s mental states.)

However, the same arguments I just gave in section 4.b regarding the listener’s experience of the testifier’s utterance can easily be extended to the testifier’s utterance itself. If the content of the listener’s experience of the utterance doesn’t justify the listener’s belief, then neither does that utterance itself, since the content of one’s experience just is that utterance. So, assuming Esther’s testimonial belief is justified in BAD KITTY, we can’t appeal to Oliver’s utterance in an account of the basis of her belief.

5.a Imagery in the production of testimony

Section 1 noted that mental imagery often occurs in the mind of the testifier while producing descriptions of some scene. In BAD KITTY, we’d expect Oliver to episodically remember the events he perceived firsthand as he describes them, as well as imagine the events he mentally reconstructed. Neuroimaging evidence suggests such imagery does often occur: producing a narrative or descriptions of a scene often involves visualization and other imagery-construction processes, either episodic remembering or imagining (Mar Reference Mar2004; Zadbood et al. Reference Zadbood, Chen, Leong, Norman and Hasson2017).

What role, exactly, does this imagery play in the process of producing testimony? Plausibly, a testifier like Oliver “reads off” descriptions of a scene from the mental imagery before his mind. Episodic memory allows one to bring to mind a rich representation of an event via a mental image which recreates one’s experience of the event and contains various information about it. One can then select particular elements of a memory image’s contents to report, describing them in one’s utterances (cf. Koriat and Goldsmith Reference Koriat and Goldsmith1996; Mahr and Csibra Reference Mahr and Csibra2018). Sensory imaginings can play a similar role given their similar format to episodic memory: one can imagine something and select aspects of one’s mental image to report.

I’ll now argue that when Oliver’s imagery plays this role in the production of his testimony, it’s because his imagery is the format in which he brings to mind his beliefs about the events he’s testifying about. He then shares these beliefs with Esther via testimony.

In order for one’s utterances to constitute genuine testimony about a particular event, one must be thinking about that event with one’s utterances causally connected in the right way to these thoughts. One could merely recite some words one has memorized and appear to be producing testimony about that event, but actually testifying about that event seems to require mentally representing it. In Oliver’s case, he does this by bringing to mind mental imagery where the content of this imagery causally determines the content of his testimony.

More specifically, though, when testimony is a means of sharing a belief about an event, it seems that a testifier’s utterances need to be causally connected to her belief: the testifier must bring her belief to mind and generate her testimony based on this belief. In other words, one’s belief must play the very functional role I just attributed to Oliver’s mental imagery in BAD KITTY. This suggests that Oliver’s mental imagery is the format of his occurrent beliefs about the events he’s describing.

So, when one perceived an event firsthand, one can bring to mind beliefs about it in the form of an episodic memory of the event. One can then share one’s beliefs by expressing the contents of that memory in testimony. Similarly, sensorily imagining can be a way to bring to mind beliefs about an event. As Oliver does, one can imaginatively “simulate” an event one didn’t witness firsthand in order to arrive at beliefs about it. One can then share one’s beliefs by reporting the contents of that imagining. In both memory and imagination cases, one’s imagery of an event is the format of one’s beliefs about that event.

5.b Imagery across testimonial transmission

So far, I’ve separately covered the roles imagery plays in the minds of the listener (section 3) and testifier (section 4.a). However, it’s not merely that imagery happens to occur on both sides of the knowledge-transmission process. Rather, there’s a more intimate relationship between both subjects’ imagery.

Consider how, as Esther imagines in response to Oliver’s descriptions, we’d expect Esther’s imagery to resemble Oliver’s imagery—i.e., the way Esther imagines the events described should resemble the way Oliver remembers and imagines them. The neuroscience of speech production and comprehension is consistent with this: brain imaging shows that the neural patterns underlying situation modelling in the listener’s mind typically mirror those underlying the testifier’s imagery (Hasson et al. Reference Hasson, Ghazanfar, Galantucci, Garrod and Keysers2012; Silbert et al. Reference Silbert, Honey, Simony, Poeppel and Hasson2014), while the degree of similarity correlates with the degree to which a listener comprehends and can later recall the contents of testimony (Stephens, Silbert, and Hasson Reference Stephens, Silbert and Hasson2010; Zadbood et al. Reference Zadbood, Chen, Leong, Norman and Hasson2017).

But what, precisely, does it mean to say that the listener’s imagery “resembles” the testifier’s? We can cash this out in terms of the content of each’s imagery, or the information contained in each’s imagery: if Oliver’s imagery represents the cat knocking over the vase, then Esther’s imagery resembles Oliver’s when it represents the same content (of course, we wouldn’t expect these images to perfectly resemble one another; I’ll gloss over this complication for now, returning to it momentarily). Since I’ve argued that the mental imagery in each subject’s mind is the format of their beliefs, this means that for a listener to form imagery that mirrors the testifier’s is for her to form a belief with the same content as the testifier’s belief.

So, in relevant imagery cases like BAD KITTY, testimonial belief transmission is a process in which the listener constructs imagery whose content mirrors the content of the testifier’s imagery. Furthermore, I already argued in section 4.b that a testifier’s utterances cause a listener to construct mental imagery in an automatic, noninferential way. With this in mind, we can see that there’s a tight causal relationship between the testifier’s and listener’s respective beliefs, which are connected via the testifier’s speech. As per section 5.a, the testifier’s belief that P in the form of an occurrent mental image causes the production of his utterances; these utterances then automatically cause the listener to form a belief, constructing a mental image whose content mirrors that of the testifier’s image.

Now, in BAD KITTY, it seems obvious that there would be many discrepancies in the specifics of each subjects’ mental images. It’s not as if they’d represent all the details of the scene the same way—exactly where the cat was on the table, the exact arrangements of the vase shards on the floor, etc. These discrepancies might seem to problematize my claim that the content of Esther’s imagery comes to mirror the content of Oliver’s. However, my account in the paper thus far can accommodate this with some slight refinements.

First, consider Esther. Notice that we wouldn’t expect her to believe all the very specific details of her mental imagery beyond those which correspond with the content of Oliver’s testimony. Esther might believe the more general contents of her imagery, such as that the cat jumped onto the table and knocked the vase onto the floor. However, if we asked her whether she’s imagining more fine-grained details accurately, we’d expect her to acknowledge that she isn’t, and that she’s withholding belief in those fine-grained contents. We can think of the contents of her imagery which she doesn’t believe as details that are merely imaginatively “filled in” to fill out the scene. So, it’s not strictly speaking true that Esther’s imagery is overall a single belief, which is how I’ve glossed things in the paper so far. Instead, her imagery is a mix of both belief and mere imagining.

While this might sound strange, it seems less so when we recognize that similar phenomena arise in other domains. For example, it’s a widespread fact that episodic memories are imaginatively filled in in various respects with only certain contents corresponding to past experiences (De Brigard Reference De Brigard2014). This doesn’t seem to prevent us from genuinely remembering the contents that are accurate (cf. Munro Reference Munro2021), which makes one’s mental image a mixture of remembering and imagining. Similarly, if one thinks that perception is a distinct mental state from hallucination, an experience that involves a hallucinated object in the foreground with a veridically perceived background seems to be a mixture of perception and hallucination. Since this sort of “mixture” of mental states can arise elsewhere, it’s not so strange to say that Esther’s imagery is a mix of belief and mere imagining.

Next, consider Oliver. We typically don’t remember all the fine-grained details of an event accurately since various details are imaginatively filled in (De Brigard Reference De Brigard2014). However, we also don’t typically believe those inaccurate, fine-grained details. Empirical study shows that subjects tend to generate memory-based testimony in ways that accommodate inaccuracies. When asked to describe a scene from memory, subjects tend to generate descriptions in a way that comes very close to optimizing accuracy; they do so by only reporting the particular details of a memory image in which they have high confidence, as well as by describing a scene at a high enough level of generality that their descriptions gloss over inaccuracies in precise details (Koriat and Goldsmith Reference Koriat and Goldsmith1996). It thus seems that testifiers like Oliver withhold belief in specific contents of their mental imagery. Thus, like Esther, Oliver’s imagery is a mix of belief and mere imagining.

Now, consider the relationship between Esther and Oliver’s imagery. The contents of Oliver’s utterances correspond only to contents of his imagery which he believes—for example, that the cat knocked over the vase. Furthermore, when Esther constructs her imagery, she forms beliefs in the contents that correspond to the contents of Oliver’s testimony rather than in the contents which are imaginatively filled in. Accordingly, the “resemblance” that’s necessary for belief transmission between Oliver and Esther is a matching between (a) the contents of Oliver’s imagery that he believes and expresses in testimony, and (b) the contents of Esther’s imagery which are directly caused by Oliver’s utterances.

5.c In favor of the other minds approach

My account in this section so far reveals that in the imagery cases on which I’m focused, there’s a tight causal relationship between the testifier’s and listener’s beliefs. Furthermore, across section 3 and section 4, I argued that no causal intermediary between these beliefs is part of the basis of the listener’s belief. In this subsection, I now argue in favor of the other minds approach by arguing that the causal relationship between each subject’s beliefs has the characteristics of the basing relation. I’ll begin with some more general discussion of the basing relation, then situate my imagery cases in relation to this discussion.

A belief can be based on another mental state either inferentially or noninferentially. Perceptual beliefs are often thought to be noninferential: one has a perceptual experience as of P, and this noninferentially causes one to believe that P (e.g., Pryor Reference Pryor2000; Huemer Reference Huemer2006). It’s also often thought that introspection is noninferential: that mere introspective awareness of one’s own mental states serves as the noninferential basis for beliefs about one’s mental states (e.g., Armstrong Reference Armstrong and Cassam1994). It’s controversial whether perception and introspection are indeed noninferential, but I’ll assume they are in what follows. (This is merely so I have some illustrative examples of noninferential beliefs so readers who disagree can substitute their favorite examples.)

A paradigmatic case of inferential basing is forming the belief that Q on the basis of one’s belief that P and one’s belief that if P, then Q. As in this example, the contents of inferential beliefs are typically distinct from the contents of the bases on which they’re formed. (Perhaps one could also infer that P from a belief that P; my point is just that it’s a widespread feature of inferential beliefs that their contents are distinct from the contents of their bases.)

In contrast, as long as nothing has gone epistemically awry, the content of noninferential beliefs matches the content of the mental states on which they’re based. In perception, one typically forms the belief that P because the content of one’s belief is causally sensitive to the content of one’s experience as of P. However, when one has an experience as of P but forms the belief that Q—for example, if an experience as of a tomato on the table causes the belief that there’s a watermelon on the table—this is because something has gone wrong. It could be, for example, that some psychological malfunction causes one’s perceptual belief-forming system not to be properly sensitive to one’s experience, or that one just wasn’t properly paying attention to one’s experience.Footnote ¹⁵

Similarly, in introspection, one typically forms the belief that one possesses mental state M because the content of one’s belief is causally sensitive to the content of one’s introspective awareness as of M. When one has an introspective awareness as of M but forms the belief that one possesses mental state N, this is because something has gone wrong. It could again be that some psychological malfunction has occurred in one’s belief-forming process, or one could be somehow misled—for example, if a therapist one takes to be reliable has misled one to think an introspective awareness as of anxiety is caused by indigestion.

So, sensitivity of the content of a belief to the content of its base such that the two end up matching is a hallmark of noninferential basing. Unlike inferential cases, the two only fail to match when things go wrong epistemically.

Now, my arguments above in section 4.b show that imagery cases of testimonial transmission are noninferential. In that subsection, I appealed to empirical evidence to argue that hearing a testifier’s speech causes a listener to construct mental imagery in a noninferential way. So, if the listener’s testimonial belief is based on the testifier’s belief, we should expect the causal relationship that holds between their imagery to be structurally like the relationship between base and belief in paradigmatic examples of noninferential belief, such as perception and introspection. In other words, we’d expect that the content of the listener’s belief is causally sensitive to the content of the testifier’s belief such that the two end up matching, unless something goes wrong. I’ll now argue that the relationship between the two beliefs does exhibit these properties.

First, consider when everything goes well in the belief transmission process, as described in section 5.b. In relevant cases, testimonial transmission is a process in which the testifier’s imagery causes his speech, which then directly causes the listener to construct imagery whose content recapitulates the content of the testifier’s imagery. In these cases, each subject’s imagery is their belief. So, when everything goes well and a belief that P is successfully transmitted from testifier to listener, the listener forms a belief with content P because the content of her belief is causally sensitive to the content of the testifier’s belief. This looks much like the relationship between base and belief in perception and introspection.

Furthermore, the content of the listener’s belief only fails to match the content of the testifier’s belief when things go wrong. This could be because of a failure on the part of the listener, as in this case:

A lack of sensitivity of the content of Esther’s belief to Oliver’s could also stem from a failure of Oliver to successfully transmit his belief:

Both these kinds of cases involve a failure in the testimonial transmission process, the result being that the content of Esther’s belief fails to be causally sensitive to the content of Oliver’s belief. Barring such malfunctions in the transmission process, this sensitivity obtains. This again makes the relation between beliefs much like the relations in other cases of noninferential belief.

So, in imagery cases of testimonial transmission like BAD KITTY, the causal relationship between the testifier’s and listener’s beliefs has the properties of paradigmatic cases of noninferential basing. The content of the listener’s testimonial belief is causally sensitive to the content of the testifier’s belief such that the listener forms a belief whose content matches that of the testifier’s. And, when this matching fails to occur, it’s because something has gone wrong (unlike cases of inferential belief, where the contents of beliefs are typically distinct from the contents of their bases, and where this doesn’t necessarily mean anything has gone wrong).

This gives us inductive support for the hypothesis that the listener’s belief is based on the testifier’s belief, which in turn supports the other minds approach. Because the listener’s belief is noninferential in relevant imagery cases, this hypothesis predicts that the relationship between each subject’s beliefs exhibits the characteristic properties of noninferential basing. As I just argued, this prediction is confirmed both in cases where belief transmission goes well and in cases where it malfunctions.

The other minds approach thus gives a plausible account of imagery cases of testimonial transmission. Furthermore, sections 3–4 have already argued that no causal intermediary between the testifier’s and listener’s beliefs is included in the listener’s basis. So, given the absence of a plausible alternative view, we should accept the other minds approach. Thus, in at least some cases, the basis of the listener’s testimonial belief that P is the testifier’s belief that P.Footnote ¹⁶