Content internalism and testimonial knowledge

ABSTRACT

It is commonly assumed that content preservation is required for success in testimonial exchanges. Many content internalists, however, cannot endorse this assumption. They must claim instead that testimonial exchanges can often succeed when the content grasped by the hearer is not the content of the speaker’s testimony, p, but some merely similar content, p*. Goldberg (2007. Anti-Individualism: Mind and Language, Knowledge and Justification. Cambridge: Cambridge University Press) argues that this internalist approach is epistemically problematic: it cannot maintain certain features thought to be characteristic of testimonial exchanges. I argue that, contrary to appearances, the internalist’s account is just as epistemically respectable as the traditional ‘same content’ approach favoured by externalists.

KEYWORDS:

• Testimony
• content internalism
• Goldberg
• descriptivism

1. Introduction

Much work in the philosophy of testimony centres around determining the epistemic conditions on the success of a testimonial exchange. To make this task easier, the semantic dimension of testimony is typically held fixed: it is stipulated that, in the testimonial exchanges under consideration, the hearer grasps the same content as that expressed by the speaker. Epistemologists debate which conditions must obtain in order for a hearer to gain knowledge that p from a speaker’s testimony that p, but they do not often debate whether p really is what the hearer must grasp. Prima facie, this ‘Same Content’ approach to testimony looks pretty plausible, at least for the majority of cases. However, if the stipulation of sameness of content cannot be dispensed with, then many (and perhaps all) internalist views of mental content are in serious trouble. These internalist views cannot sensibly endorse the Same Content approach to testimony. This is because they claim that subjects do not often share concepts or speak a communal language but, instead, speak and think in more or less similar idiolects; as such, in most testimonial exchanges, the best a hearer can do is grasp a merely similar content to that proffered by the speaker. If this kind of internalist wants to maintain that testimonial exchanges are often successful, she must claim that these exchanges often succeed even when the content grasped by the hearer is not the content of the speaker’s testimony, p, but some merely similar content, p*. I call this view the ‘Similar Content’ account of knowledge through testimony (‘SimTest’, for short). My aim in this paper is to defend content internalism by arguing that it can plausibly maintain this Similar Content approach.

SimTest may seem obviously problematic: it maintains that the paradigmatic cases of knowledge through testimony are those in which hearers fail to grasp the content that the speaker expressed. At the very least, this picture is strikingly different from the traditional Same Content approach that underpins (more or less explicitly) most work in the epistemology of testimony. However, few authors have explored precisely what might be epistemically problematic about the Similar Content alternative. One notable exception is Goldberg (2007), who presents a sustained critique of both SimTest and its partnership with internalism.¹ There are two problems that he identifies. The first is a problem for SimTest itself: Goldberg argues that SimTest cannot uphold an attractive account of epistemic reliance, where this phenomenon is thought to be a distinctive feature of knowledge through testimony. The second problem concerns the plausibility of combining SimTest with content internalism. Goldberg argues that any internalist who endorses SimTest must concede that knowledge is rarely gained through testimony. I argue that neither of these objections succeeds; the internalist can maintain a Similar Content account of knowledge through testimony that is just as epistemically respectable as the Same Content accounts favoured by content externalists.²

The paper proceeds as follows. In Section 2, I introduce SimTest. In Sections 3 and 4, I introduce, and respond to, Goldberg’s objection to SimTest itself. In Section 5, I respond to Goldberg’s objection that endorsing SimTest fails to render the internalist’s approach to testimony plausible.

2. Semantic conditions on testimony

In a testimonial exchange, a hearer accepts the content that she grasps through her interpretation of the speaker’s assertion. The semantic conditions on testimony are conditions that specify the relationship that must hold between the content expressed by the speaker and that recovered by the hearer. It is typical for epistemologists to talk as if, in a successful testimonial exchange, the content that the hearer grasps will be the same content as that which the speaker asserts.³ For example, Lackey writes, ‘In explaining how we acquire knowledge via the testimony of others, we are interested in offering an account of how hearers can come to know that p through a speaker’s statement that p’ (Lackey 1999, 488). Similarly, Fricker writes, ‘[T]here must be a proposition which the teller intends by her action to present as true, and this must be identical with the one grasped by her audience as so presented, and accepted by her’ (Fricker 2006, 229).⁴ For some who work on testimony this might be a stipulation introduced to restrict attention to paradigm or simple cases; however, Goldberg (2007) argues for the necessity of a same content condition on testimony. I present this as follows:

Same Content Condition: A hearer can acquire knowledge through a speaker’s testimony only if the content of the hearer’s testimonial belief is the same as the content of the speaker’s testimony.

There are caveats that one might want to make to this condition; Goldberg (2001), for example, suggests that a hearer can come to know that q from testimony that p when q is ‘conveyed’ by p. However, this simple version of it will do for our purposes. I will call any account of testimony that endorses this condition a ‘Same Content’ view (or ‘SameTest’, for short).

The Same Content view tends to be popular amongst content externalists (See, e.g. Fricker 2006; Burge 1993). Content externalism is the view that content is individuated, in part, by factors external to the individual, such as her social and/or physical environment (e.g. Burge 1979; Fodor 1994; Millikan 1984). Goldberg endorses social externalism. Social externalists claim that colingual subjects share a public language. This means that, with respect to non-indexical expressions (and setting aside phenomena like ambiguity), utterances of the same word-forms by different speakers in the same language community typically express the same mental contents. On this view, any condition that requires sameness of content is relatively easily satisfied by colinguals because they defer to the same language community when it comes to the determination of the content of their thoughts and utterances.⁵ According to Goldberg (2007, 53ff), this makes the comprehension process quite straightforward.⁶ Hedescribes the picture of comprehension required by an adequate theory of testimony as one on which, typically:

modulo considerations of context-sensitivity, the hearer is correct to interpret the speaker as expressing the propositional content that she (the hearer) would express with the same words. (2007, 63–64)

In opposition to SameTest, I present the ‘Similar Content’ account (‘SimTest’). This view rejects the Same Content Condition and maintains the following weaker condition:

Similar Content Condition: A hearer can acquire knowledge through a speaker’s testimony only if the content of the hearer’s testimonial belief is appropriately similar to the content of the speaker’s testimony.

How similar must these contents be? How we answer this question will generate different versions of the approach.⁷ As I argue later, certain versions of the view are better equipped to maintain a traditional epistemology of testimony than others. One division amongst views of content that will be important to my later defence of SimTest concerns fineness of grain. Fine-grained, or ‘Fregean’, views of content claim that two contents can be type-distinct without differing with respect to what is represented (e.g. Bezuidenhout 1997). Whereas coarse-grained views might identify contents with sets of possible worlds, fine-grained views claim that two contents can be modelled by the very same set of possible worlds, and yet fail to be type-identical due to things like differences in the inferential role of these contents in two subjects’ respective cognitive economies or in two different public languages. Both internalists and externalists alike can endorse fine-grained contents; in Section 4, I argue that doing so will enable the internalist to maintain a plausible epistemology of testimony: the version of SimTest that she should endorse is one that requires that the hearer grasp a content that is co-representational with the content expressed by the speaker.

Both SameTest and SimTest state only necessary conditions on testimony. To provide a complete analysis of testimonial knowledge acquisition, one must add one’s preferred epistemic conditions. For example, perhaps the speaker must express her own knowledge with her testimonial utterance; and/or perhaps the hearer must possess positive reasons for accepting the testimony (Fricker 1994). In this paper, I remain neutral as to which epistemic conditions are required for success. With these epistemic considerations bracketed, it will be easier to focus on the viability of the semantic constraint that I defend. Having said that, one epistemic condition is worth flagging here: both SimTest and SameTest will add that the content grasped by the hearer must be grasped reliably.⁸ (Goldberg 2007, 36ff) This condition is included to rule out cases in which one recovers the appropriate content by luck (e.g. through mishearing or a lucky guess).⁹ I return to this issue in Section 5 – as we will see, it may appear that this epistemic constraint causes problems for SimTest.

Before turning to problems for SimTest, however, let us first consider why one might want to reject SameTest in favour of it. Certain views of content cannot sensibly endorse SameTest. There are various content internalist theories that may face this difficulty. Internalism about mental and utterance content is the view that content is individuated solely by factors internal to the individual – for example, by phenomenal states, or by conceptual role. Whilst some internalist views may be able to posit shared content, it is quite common for internalists to claim instead that there is considerable conceptual variation across individuals. This is because individuals tend to vary significantly with respect to their internal states. To take an extreme example, radical holists about conceptual role claim that any difference, however minute, between the conceptual networks of two subjects entails differences between all of their concepts (Rapaport 2003). On this approach, two subjects will rarely, if ever, entertain concepts that are type-identical – even two doctors who specialise in the treatment of arthritis may still express different concepts with the word ‘arthritis’ due to differences in distant parts of their respective conceptual networks. This kind of approach entails that SameTest’s semantic condition is rarely, if ever, satisfied: when a speaker asserts ‘Arthritis is painful’, a hearer can never grasp precisely the content that she expressed, although she may grasp something highly similar.¹⁰ Thus, if the holist endorsed SameTest, she would have to claim that testimonial exchanges are rarely successful. Most epistemologists believe that we gain a great deal of knowledge through testimony; thus, it would be unwise for an internalist to endorse a condition that rules this out. In addition to the holist, many internalist approaches (to the extent that they deny that concepts are widely shared) will struggle for similar reasons. Goldberg (2007, 36ff) has argued that internalists of all stripes will face this problem. For the purposes of this paper, I will grant that Goldberg is right about the implausibility of combining content internalism with SameTest (although, I discuss a related issue in Section 5). My thesis is that, even if a theory precludes that content is often shared between subjects, that theory can nonetheless maintain an epistemically attractive theory of testimony. Goldberg also argues that SimTest is untenable. It is this claim which I will dispute. I turn next to Goldberg’s critique of SimTest.

3. Problems for SimTest

Goldberg considers a version of SimTest, which he presents as follows: ‘the sort of understanding involved in testimonial knowledge does not require recovery of the very content attested to; close enough is good enough, epistemically speaking.’ (2007, 45) Most generally, Goldberg’s objection to SimTest is that it must maintain a revisionary and implausible conception of the distinctive features of testimonial knowledge. To demonstrate this, Goldberg first outlines what these features are supposed to be. Testimonial knowledge is knowledge that is gained ‘through’ testimony, where knowledge through testimony is importantly distinct from other kinds of knowledge that a hearer might gain from observing a speech exchange (Audi 1997). Knowledge through testimony, according to Goldberg, is special in that it exhibits what he calls ‘epistemic reliance’. A hearer’s epistemic reliance on a speaker consists in her relying on the speaker to have lived up to whatever norms govern testimony. He writes:

… what is distinctive of cases of knowledge through speech is precisely the element of epistemic reliance: in aiming to acquire knowledge through S’s testimony, H is relying on that testimony to have the epistemic features that, in virtue of its being a case of testimony, it ought to have. (Goldberg 2007, 16)

Precisely which norms these are will depend on one’s broader commitments in epistemology but, to follow Goldberg, let’s assume that these norms require that the speaker’s testimony is reliable: she would have proffered the testimony only if it were true. Thus, we can present the epistemic reliance thesis as follows:

Epistemic Reliance: A hearer, H, is epistemically reliant on a speaker, S, just when H, in forming her testimonial belief, is relying on S to have produced reliable testimony.

As Goldberg (2007, 17) explains, exchanges that exhibit epistemic reliance are distinctive in that the hearer, in virtue of appropriately accepting the content she grasps through the exchange, is entitled to regard the speaker as being in an epistemically privileged position with respect to the content of her resultant testimonial belief (where, for present purposes, we should remain neutral over whether this content must be the same as the content of the speaker’s testimony).

It may be that this characterisation of what is distinctive of testimonial exchanges is too strong and, as such, a content internalist is under no pressure to comply with it. However, I will take a different approach in this paper. My aim is to show that the internalist can indeed endorse even this more demanding thesis. If the internalist can meet her critics on their own terms, there should be little doubt that she can endorse an attractive epistemology of testimony. The problem for SimTest that Goldberg (2007, 46) presses is that (regardless of whether one requires that the hearer has positive reasons to trust the speaker) it’s unclear how a hearer can be described as epistemically reliant on a speaker when the hearer grasps a content that is different to that which the speaker expressed. In exchanges which conform to the Epistemic Reliance thesis, the hearer is relying on the speaker to have represented the world correctly and reliably. Where the hearer grasps a merely similar content, however, Goldberg claims that she is not relying on the speaker to have represented the world in the right way: the hearer does not recover, nor does she accept, the speaker’s representation of the world in this case. If Goldberg is right, then exchanges that satisfy SimTest (but not SameTest) do not exhibit epistemic reliance. Goldberg (2007, 47) suggests that, on SimTest, at best, the hearer is relying on the likelihood of the truth of the content she grasps given that it is similar to the content proffered by the speaker. He considers that his opponent might try to pitch this as a surrogate notion of epistemic reliance that can nonetheless play the same role as the Epistemic Reliance thesis in an account of testimony. Unfortunately, Goldberg argues, this surrogate notion forces us to give up several important features of knowledge through testimony.

To see this, recall that exchanges that conform to the Epistemic Reliance thesis are supposed to be such that hearers are committed to regarding speakers as being in an epistemically privileged position with respect to the content of their testimonial beliefs. According to Goldberg, many of the exchanges that SimTest countenances lack this feature: a hearer should not regard the speaker as being in an epistemically privileged position with respect to the content of her testimonial belief, p*, if what the speaker asserted was some merely similar content, p – the speaker may even believe that p* is false. There are further problems that follow from this. For example, Goldberg (2006, 2007, 49) argues that it is part of our conception of testimonial knowledge that, if a hearer is challenged as to the reliability of her testimonial belief, it is epistemically appropriate for her to ‘pass the buck’ to the testifier, who possesses further epistemic support for the belief’s content. Relatedly, exchanges in which it is appropriate for a hearer to pass the buck to the testifier are supposed to be exchanges in which it is appropriate for her to blame this testifier if her belief turns out to be unreliable (Goldberg 2007, 83). However, if Goldberg is correct, then neither of these practices make sense for cases in which a hearer recovers a merely similar content. If a hearer who gained a testimonial belief that p* from testimony that p is asked to provide support for p*, it would be epistemically inappropriate for her to pass the buck to the testifier, as the testifier never presented herself as an authority with respect to p*. For the same reason, it would not be appropriate for the hearer to blame the speaker if her testimonial belief turns out to be false. Given these problems, it seems that the surrogate notion of epistemic reliance available to SimTest is not really epistemic reliance at all. Instead, SimTest offers a highly revisionary conception of testimonial exchanges that jettisons several related features that are plausibly taken to be distinctive of knowledge through testimony.

To recap: I am assuming that Epistemic Reliance is distinctive of knowledge through testimony. Given this, it looks like content internalism is in serious trouble. It seems that only exchanges that satisfy SameTest’s semantic condition on testimonial knowledge will exhibit genuine epistemic reliance; SimTest offers an implausible and revisionary surrogate. As such, only exchanges that satisfy SameTest are genuine testimonial exchanges. But internalism entails that SameTest is rarely satisfied and, thus, the internalist must concede that testimonial knowledge is a very rare phenomenon indeed. How should the internalist respond? In the next section, I argue that she can avoid this problematic result by dropping an assumption that has been present throughout this exposition of Goldberg’s objection: the objection relies on the assumption that merely similar contents cannot be (or, typically, are not) co-representational. Endorsing a fine-grained view of content will allow the internalist to reject this assumption.

4. Retaining epistemic reliance

Above, I introduced fine-grained, ‘Fregean’, views of content as those which claim that contents can be type-distinct despite no difference in what is represented. In what follows, I shall describe such contents as representing the same ‘state of affairs’. I use this terminology in a somewhat loose sense:¹¹ in saying that two contents represent the same state of affairs, I simply mean to say that they represent the same thing, where what a sentential content represents will depend on what is represented by its component concepts and the manner in which they are combined.¹² States of affairs are coarse-grained: for example, ‘Hesperus is bright’ and ‘Phosphorus is bright’ represent the same state of affairs yet, for the Fregean, can differ in content. The assumption that merely similar contents represent distinct states of affairs is implicit in Goldberg’s presentation of SimTest; as he describes the approach: ‘so long as the hearer’s belief is a belief in a content that is ‘close enough’ to what was actually said, then the belief in that “close enough” content will likely be true’ (Goldberg 2007, 45). That the content grasped by the hearer is merely likely to be true suggests that it represents a distinct state of affairs from the speaker's testimony. For, if it represented the same state of affairs, it would be guaranteed to be true given the truth of the testimony proffered. Goldberg also writes that SimTest ‘denies that the hearer needs to recover how the speaker is representing the world to be.’ (ibid, 46) Again, it is only true that SimTest denies this if it maintains that the similar (yet distinct) contents appealed to by the view represent different states of affairs. If two distinct contents can be co-representational, however, then SimTest need not deny that the hearer must recover how the speaker is representing the world to be. This version of SimTest is not the only version available to the internalist. We can defend SimTest by rejecting this assumption and adopting a Fregean version of the account that pins down one dimension along which contents must be similar:

Similar Content Condition – States of Affairs: A hearer can acquire knowledge through accepting a speaker’s testimony only if the content of the testimony proffered by the speaker and the content (reliably) grasped by the hearer are similar to the extent that they represent the same state of affairs.

Call a version of SimTest that endorses this condition ‘Revised SimTest’ (henceforth, ‘RSimTest’). Because RSimTest requires that the content expressed by the speaker and recovered by the hearer are co-representational, the approach can maintain the Epistemic Reliance thesis: the content grasped by the hearer is no longer merely likely to be true given the reliability of the testimony, but guaranteed to be true. As such, the hearer need not rely on additional epistemic support for her testimonial belief over and above what is required on SameTest. In forming her testimonial belief, she is relying on the speaker to have represented a particular state of affairs as obtaining. And, importantly, she is relying on the speaker to produce testimony that is not just true but reliable. If the speaker’s testimony is unreliable, this unreliability will transfer to the hearer’s testimonial belief: given that the hearer’s belief is true just when the speaker’s belief is true, any close possible worlds in which the speaker’s belief is false will be worlds in which the hearer’s belief is also false. Similarly, any close possible world in which the speaker testifies by expressing a false belief formed using the same (unreliable) method she employs in the actual world will be a world in which the hearer gains a false belief if she accepts this testimony. Thus, in accordance with the Epistemic Reliance thesis, the hearer is relying on the speaker to have produced reliable testimony (and no more than this).

If we can recapture the traditional Epistemic Reliance thesis by appeal to fine-grained contents, the worries introduced earlier do not arise for RSimTest. Given that the hearer’s testimonial belief represents the same state of affairs as that which is represented in the speaker’s testimony, the hearer can regard the speaker as being in an epistemically privileged position with respect to her testimonial belief. Because of this, the hearer can pass the buck to the speaker, deferring to her authority regarding support for her testimonial belief. Equally, should the testimony turn out to be false, the speaker can be considered blameworthy for testifying to a content that represents a state of affairs that does not obtain, or for offering testimony that is otherwise unreliable. Rejecting the assumption that distinct contents must (or usually) represent distinct states of affairs, then, allows RSimTest to maintain Epistemic Reliance. However, endorsing the more demanding RSimTest might appear to be an option which is unavailable to the content internalist. In his criticism of SimTest, Goldberg advances an additional argument that is not addressed by appeal to RSimTest. This argument is not aimed against SimTest itself but at the plausibility of combining SimTest with content internalism. As we will see, endorsing RSimTest appears to make this problem worse.

5. RSimTest and content internalism

Goldberg’s (2007, 48ff) original objection (aimed at SimTest rather than RSimTest) runs as follows. Suppose it is granted that the content of a hearer’s testimonial belief is likely to be true if it is similar to the content of the testimony proffered. Even granting this, on content internalism, our beliefs are simply not often similar enough to facilitate knowledge exchange. Consider that, on content externalism, our dispositions to employ terms can be quite diverse whilst we still succeed in talking (and thinking) about the same objects. On many internalist theories, by contrast, differences between subjects’ dispositions to employ a given term are taken to determine quite significant differences in what is represented by that term (Crane 1991; Segal 2000). At this point, then, an obvious problem faces any internalist wanting to endorse RSimTest. RSimTest demands that two contents be so similar that they represent the same state of affairs but, if Goldberg is right, we should already be concerned that internalism couldn’t underpin enough conceptual similarity to meet the less demanding formulation of SimTest with which we began. If internalists are to offer an account of testimony that rivals the externalist’s account, it is not enough to maintain that it is possible for an exchange to satisfy RSimTest's condition. Rather, it must be the case that exchanges often – and reliably – satisfy it. Otherwise internalism will not be able to underpin the widespread success of our testimonial exchanges. What is needed, then, is reason to think that there is some mechanism for determining reliable, widespread co-reference for distinct idiosyncratic concepts and contents on internalism. For any view of content that includes an external determiner of meaning, securing co-reference appears to be easy; externalists can appeal to individuals’ relationships to a shared social and/or physical environment. In what follows, I explain how an internalist can also accomplish this task by adopting a strategy that is parasitic on the resources employed by the externalist.

The strategy that the internalist should employ is one that is appealed to in contemporary defences of descriptivism about reference – from simple causal descriptivism to deflationary and two-dimensional approaches. Whilst this will be familiar territory to many, it will be useful to say a bit more about the sort of descriptivist approach I have in mind and explain how it can be extended to mental content. This is so for two reasons. Firstly, that the internalist about mental content can usefully appeal to this kind of resource might not be immediately clear: descriptivism is traditionally a thesis about expressions in a language, rather than mental content and, moreover, descriptivists need not endorse internalism about either language or thought.¹³ Secondly, the descriptivist resources that I will employ have an important epistemic upshot concerning testimonial reliability that will be useful in defusing Goldberg’s argument – I explain this upshot in Section 5.2. One further concern one might have with the present strategy is that descriptivism is currently an unfashionable view; it seems that the majority of philosophers remain persuaded by Kripke’s (1980) critique of the position, as well as related arguments from Putnam (1975) and Burge (1979). It is beyond the scope of this paper to mount a comprehensive defence of descriptivism.¹⁴ However, it is worth noting that the version of descriptivism that I appeal to is descriptivism about reference determination; this view is weaker than descriptivism about the meaning of an expression and escapes some of the problems with the latter view.¹⁵My thesis is that, if otherwise defensible, descriptivism about reference determination can be used to underwrite a plausible epistemology of testimony for the content internalist.

5.1. Descriptivism and externalism

Descriptivism about reference determination is the view that the reference of an expression (such as a name), for a speaker, is determined by a description that is associated with the expression by that speaker. For example, the reference of the name ‘Gödel’, for a speaker, would be determined by the description ‘the prover of the incompleteness of arithmetic’ if this is the description that the speaker associates with the name. As Kripke (1980) has famously argued, simple versions of descriptivism suffer from a range of serious problems. One such problem is that the view gives the wrong intuitive verdicts for a great many cases. Kripke (1980, 84) presents a thought experiment in which we discover the following fact:

1. The incompleteness theorem was proved by a man baptised ‘Schmidt’ and who never called himself anything other than ‘Schmidt’; a person who called himself ‘Gödel’ stole the proof from him.

Kripke thinks it’s most natural to say that, contra descriptivism, we refer to Gödel, even when we are wrong about his having proved the incompleteness theorem: that is, even when he does not satisfy the associated description. This is supposed to support the view that names refer to whomsoever they are appropriately causally related – not (necessarily) to the person who is picked out by any associated description(s) (Kripke 1980, 91). Kripke’s arguments have been influential because most people share his judgments about cases.

This naïve descriptivism would certainly be of no use to the content internalist. However, contemporary defences of descriptivism maintain that expressions like ‘Gödel’ do indeed have their intuitive referents in the mouths of different individuals regardless of their mistaken beliefs and associations. And, crucially for our purposes, this is something that objectors like Kripke must accept on pain of undermining their own arguments. There are different versions of this response in the literature. Field (1994), for example, advances a deflationary theory of the reference of proper names, according to which there is no more to reference than what is given in a disquotation schema. He presents the schema for singular terms as follows: ‘If b exists then “b” refers to b and nothing else; if b doesn’t exist then “b” doesn’t refer to anything’ (261). Field points out that we can take Kripke’s observations about our linguistic intuitions to show merely that subjects regard (1) as grounds for inferring (2),

• (2) Gödel didn’t prove the incompleteness theorem.

rather than inferring (3),

• (3) Gödel was actually baptised as ‘Schmidt’ and never called himself ‘Gödel’.

Thus, Field thinks that Kripke’s thought experiment can be construed merely as evidence that speakers tend to treat certain beliefs about the properties of objects as non-negotiable. As Field writes, ‘[I]t is just part of our inferential procedure to regard claims of roughly the form “The dominant causal source of our beliefs involving ‘b’ is b” as pretty much indefeasible’ (ibid, 261).¹⁶ The semantic intuitions that are taken to support the causal theory of reference are precisely those which motivate the claim that speakers have the inferential tendencies that Field identifies.

Jackson (1998b) makes related points in defence of his (non-deflationary) descriptivism. He argues that when we solicit speakers’ judgments about possible cases – those that are taken to support various externalist theses about meaning – we are actually learning what the relevant identifying descriptions are for those speakers. He writes,

If you say enough about any particular possible world, speakers can say what, if anything, words like ‘water’, ‘London’, ‘quark’, and so on refer to in that possible world. […] Our ability to answer questions about what various words refer to in various possible worlds, it should be emphasised, is common ground with critics of the description theory. The critics’ writings are full of descriptions (descriptions) of possible worlds and claims about what refers, or fails to refer, to what in these possible worlds. Indeed, their impact has derived precisely from the intuitive plausibility of many of their claims about what refers, or fails to refer, to what in various possible worlds. But if speakers can say what refers to what when various possible worlds are described to them, description theorists can identify the property associated in their minds with, for example, the word ‘water’ […]. (1998b, 212, emphasis in original)

It is common ground between most parties to the debate, then, that speakers can discern which objects are picked out by expressions in possible scenarios. Thus, these parties should agree both that, as Jackson claims, speakers are in possession of implicit identifying descriptions (or properties) and that, as Field claims, speakers have particular inferential practices with expressions which single out these descriptions as indefeasible. For the same sorts of reasons, if we agree that names are rigid designators, we should accept that the descriptions that subjects treat as non-negotiable are rigidified descriptions (Jackson 1998b, 213). The foregoing is not intended to demonstrate that descriptivism is true. Rather, the point is that, if we think that most speakers would agree with opponents of descriptivism about what various expressions refer to (in both the actual and counterfactual worlds), we must think that these speakers share the kinds of dispositions or beliefs that contemporary descriptivists claim are responsible for determining reference. And if speakers do indeed share these beliefs and dispositions, descriptivists can maintain that speakers tend to refer to the same objects (or properties, or kinds, etc.) with their expressions of the same word-forms (context-sensitivity aside). It is worth noting that this approach does not require that subjects possess type-identical non-negotiable beliefs in order to secure co-reference. All that is required is that different speakers’ reference-determining beliefs, for a given expression, single out the same object (property, kind) as the referent of that expression – if externalists are correct that co-reference is widespread, we must think that speakers’ dispositions are indeed sufficiently similar to do this.

The approach advocated here should carry over straightforwardly to the case of mental content. An internalist can agree with Burge (1979) about what objects are represented by subjects’ concepts. But, to the extent that we think that the majority of language users share his claims about what is represented by a given concept, we should accept that these subjects have the sorts of inferential dispositions and/or beliefs that the descriptivist strategy posits. Those of us who do not possess full linguistic competence with a concept are typically (though not always) disposed to treat beliefs of roughly the form c is whichever object the (actual) experts talk about with tokenings of ‘c’,¹⁷ as essential to determining which object is represented by a concept.¹⁸ The ways in which other speakers use expressions are indeed relevant to content determination on this internalist approach, but only insofar as they figure in the individual’s own reference-fixing criteria.

With the preceding in mind, let’s return to Goldberg’s objection. The objection was that content internalists can’t provide the kind of close conceptual similarity which RSimTest demands. For the combination of RSimTest and content internalism to be plausible, it must be the case that the internalist can claim that RSimTest's condition is often satisfied. This requires that subjects with distinct idiosyncratic concepts can often represent the same states of affairs with their thoughts and utterances. It should now be clear that the descriptivist solution demonstrates that co-reference of idiosyncratic concepts will be pretty much exactly as widespread on internalism as is concept-sharing on externalism. This is because the dispositions appealed to are precisely those which are supposed to motivate externalist theses about meaning and content. This means that externalist and internalist approaches to reference (of the kinds considered here) are linked: if we think that most speakers would agree with ‘externalist’ intuitions about reference, we must accept that these different speakers possess sufficiently similar dispositions to secure widespread co-reference for the internalist. If we deny that speakers tend to share these dispositions, then neither the internalist nor the externalist can posit widespread co-reference. Thus, if it is true that a Same Content Condition would be often satisfied on externalism, RSimTest will be similarly often satisfied even on content internalism.¹⁹ This connection between the two approaches will be important when considering the reliability of testimony in the next subsection.

5.2. Content internalism and linguistic uniformity

I have argued that the internalist can maintain that satisfaction of RSimTest is as common on internalism as is satisfaction of the Same Content Condition on externalism. However, an objector might claim that my account must covertly appeal to externalist resources. If testimonial knowledge acquisition is indeed a widespread phenomenon, it must be the case that there is widespread uniformity in speakers’ dispositions. But this uniformity in dispositions is crying out for an explanation. The objector might insist that the most plausible explanation is the existence of something like Goldberg’s ‘public linguistic norms’ (cf. Goldberg 2007, 69): in this case, norms that determine assignments of non-negotiable application-conditions to concepts – how else could speakers with idiosyncratic grasp of a language achieve such widespread agreement? Indeed, Goldberg’s reason for thinking that internalists cannot endorse SameTest is precisely that internal facts could not determine a sufficiently high degree of cross-speaker linguistic uniformity with respect to concepts. If Goldberg’s argument works against combining internalism with SameTest (as I have said I will grant), why shouldn’t we think that it will also work against internalism and RSimTest? Fortunately, I think that Goldberg’s argument cannot be extended in this way; I have three things to say in response. Firstly, public linguistic norms cannot explain this particular kind of linguistic uniformity. Secondly, the kind of uniformity that I posit is more minimal (and more easily achieved) than the sort that Goldberg objects to. Thirdly, if the objection undermines my internalist approach, it also undermines externalism – this is the epistemic upshot of the descriptivist strategy employed in Section 5.1. I will explain each of these points in turn.

Public linguistic norms, of the sort argued for by Goldberg (2007, 56), are norms that determine the contents of our thoughts and utterances in accordance with some external, social standard. On externalism, norms play this role even when our grasp of the concepts that these norms determine is incorrect: a subject can be correctly attributed a concept (arthritis, say) even when she is mistaken about how this concept applies. For the externalist, public linguistic norms do not explain our actual internal dispositions to employ the shared concepts that we grasp. The standards for correct usage are opaque to us in the sense that we cannot learn them through introspection; we can only discover them through consulting expert members of our language community, and we often get them wrong.²⁰ All this is to say that, even if there were public linguistic norms, these norms would not explain the widespread uniformity in our psychological dispositions to retain the particular kind of non-negotiable belief identified above.²¹ That we have these dispositions, as argued above, is something that all sides of the debate agree on. The content externalist, however, is not in the business of explaining how any of our dispositions to employ concepts are acquired or maintained; rather, her job is to tell us how concepts can be shared even when many of our dispositions to employ these concepts are not. The foregoing, of course, still leaves our question unanswered: if not public linguistic norms, then what explains uniformity in non-negotiable beliefs? Turning to the second part of my response, I will present one promising explanation in what follows.

One explanation for the widespread uniformity in our non-negotiable dispositions is that this uniformity results from social approbation. When we are taught to use language, we are encouraged to use it in ways that cohere with the practices of our teachers and discouraged from maintaining idiosyncratic practices. Goldberg (2007, 64ff) argues that it is not plausible to think that social approbation could result in the widespread sharing of concepts if internalism is true; and I am inclined to agree with him about this. This is one of the reasons why Goldberg thinks that internalists cannot sensibly endorse SameTest. However, social approbation can explain widespread uniformity in non-negotiable beliefs. To see this, notice that this latter kind of uniformity is far less demanding: all that is required to get the internalist approach off the ground is that young language users learn to apply something like the schema introduced above with respect to each new concept that they acquire: c is whichever object the (actual) experts talk about with tokenings of ‘c’. Although it would be miraculous if most members of a language community agreed on the definitional meanings of each of their concepts, it is not hard to see how social approbation could result in language users acquiring this sort of schema and applying it to all or most of their concepts. The approach does not require that language users explicitly endorse (or are explicitly taught) this schema, nor, as mentioned above, does it require that they each acquire precisely the same schema. Rather, the fact that they do endorse it (or something like it) is evident in their inferential practices. Social approbation is one explanation of how they could have acquired it.

There is one objection that Goldberg raises against the appeal to social approbation with respect to concept sharing which may also apply to my current defence of social approbation as an explanation of uniformity in non-negotiable beliefs. This is that, if social approbation does explain uniformity, this renders the obtaining of this uniformity hostage to chance in a way that may undermine the reliability of testimony. Goldberg writes (with reference to concept sharing):

[O]n the present view it is simply a contingent feature of one’s linguistic environment whether or not the idiolect of a ‘‘co-lingual’’ interlocutor overlaps in relevant respects with one’s own – something that depends on whether one’s interlocutor has had the relevant portions of her idiolect shaped by the forces of social approbation. (2007, 64)

The worry, transposed to non-negotiable beliefs, is that, even when a hearer gains a testimonial belief from a speaker who does in fact share the relevant dispositions, if there are many speakers in her community who don’t share these dispositions (or if the speaker’s own dispositions could easily have been different), it is lucky that the hearer acquires a testimonial belief that represents the same state of affairs as the speaker’s testimony. Is the contingency of these practices sufficient to undermine the reliability of testimony for the internalist? I think it’s plausible that these dispositions are sufficiently widespread and deeply embedded in our linguistic practices that testimonial reliability will not be undermined. However, rather than attempt to demonstrate this, I will make a different manoeuvre – taking us to the third part of my response to the uniformity challenge. As emphasised above, the dispositions that I appeal to in my account are precisely those dispositions that externalists like Burge appeal to in motivating their views. For the externalist, subjects can possess communal concepts, but only if they possess these dispositions. That is, a subject should be attributed a public rather than idiosyncratic concept only if she has (for example) the disposition to accept correction from members of her language community with respect to the application of that concept, or otherwise takes herself to be answerable to communal standards. This means that, for the externalist, the fact that individuals speak a communal language is also a contingent phenomenon²² – moreover, it is precisely as contingent as the phenomenon of uniformity in non-negotiable beliefs. The extent to which we are lucky to find ourselves talking to someone who shares our non-negotiable beliefs (or possesses beliefs that are sufficiently similar to secure co-reference), for the internalist, mirrors the extent to which we are lucky to find ourselves talking to someone who shares our concepts, for the externalist. As stated above, my inclination is that the amount of luck involved here is not sufficient to undermine the reliability testimony for either theory; but my point is that if it is, this is a problem for both my internalist view and my externalist rival alike.

6. Conclusion

The aim of this paper was to defend content internalism. In doing so, I presented an account of testimony, ‘SimTest’, which claims that a hearer can gain knowledge through testimony that p from testimony that p*. This view is designed for those content internalists who cannot posit widespread shared content. I considered a collection of objections to the combination of SimTest and content internalism from Goldberg. Although Goldberg’s objections were problematic for one version of SimTest, I demonstrated that the view could be reformulated to avoid the objections. The revised version of SimTest claimed that a hearer can gain knowledge that p from testimony that p* provided that p and p* represent the same state of affairs. Finally, by appeal to defences of descriptivist theories of reference, I showed how a content internalist could maintain the revised version of SimTest. If my arguments succeed, even those internalists who think that we rarely entertain type-identical concepts can endorse an account of knowledge through testimony that is no more problematic than the account endorsed by her externalist rivals. Thus, considerations pertaining to testimonial knowledge acquisition do not give us reason to favour content externalism over internalism.

Acknowledgements

I am grateful to Jesper Kallestrup, Anders Schoubye, Åsa Wikforss, Sanford Goldberg, Tim Kenyon, and Andrew Peet for comments on previous drafts of this work. Thank you also to audiences at the University of Copenhagen and University of Edinburgh.

Disclosure statement

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

Additional information

Funding

This work was supported by the Arts and Humanities Research Council (AH/H019529/1), and by Norges Forskningsråd (250654).

Notes

1 See also Goldberg (2015).

2 One might think that, not just testimony, but communication in general, requires shared content. This paper does not address this more general worry. See Pollock (2015) and Bezuidenhout (1997) for arguments in favour of a similar content approach to communication. For opponents, see Cappelen and Lepore (2007).

3 For present purposes, I restrict my attention to cases in which the content of the assertion does not diverge from the thought content it is used to express.

4 Although, see Fricker (2012, fn. 8), which suggests a context-sensitive approach to same-saying that could be considered a similar content approach.

5 This is not to say that grasping the concepts expressed by the terms evident from the surface form of an utterance (and their combination) is all there is to understanding utterance content. A hearer may also need to recover the explicature of an utterance.

6 Goldberg writes that there is little written on the role of the comprehension process in testimonial knowledge acquisition, although he notes exceptions, including McDowell (1981), Fricker (2003), and Goldberg (2004). For more recent work, see Peet (2016) and (2019).

7 Peet (2019) develops a view that could be considered a similar content account. However, his focus is the more general category of ‘knowledge-yielding’ communication, rather than knowledge through testimony, which is my concern in this paper. Knowledge through testimony is knowledge that characteristically involves the hearer’s epistemic reliance on the speaker.

8 On SimTest, this requirement must be stated differently: there is a range of contents that the hearer can recover, each of which would be permissible interpretations. On SimTest, it must be the case that the hearer would not easily have recovered a content outside of this permissible range.

9 For example, the reliability condition should rule out the sorts of problematic cases presented in Heck (1995). In their paper, Heck is not arguing against SimTest, or SameTest. However, similar considerations to those that they present motivate including a condition that rules out communication by luck when considering testimony.

10 For and account of conceptual similarity in holistic networks, see Pollock (2020).

11 In particular, I do not wish to commit to any view regarding the metaphysics of states of affairs. For examples of such views, see Plantinga (1974) and Fine (2012).

12 There are more sophisticated theoretical frameworks that one may employ here. For example, if one endorses two-dimensional semantics, one may appeal instead to sameness of 2-intension (Chalmers 2004; Jackson 1998a; Chalmers and Jackson 2001). However, I think the simpler appeal to contents that are co-representational is sufficient for the purposes of this paper.

13 Jackson (1998b), for example, explicitly does not extend his descriptivism to mental content.

14 For further objections to descriptivism, see Soames (1998), Everett (2005), Martí (2020). For defences, see Stanley (1997), Jeshion (2002), Kallestrup (2012), Jackson (1998a).

15 This includes those that Goldberg (2007, 72) raises against the related 'metalinguistic interpretant' proposal.

16 For a non-deflationary causal descriptivist approach, see Kroon (1987).

17 A descriptivist need not spell-out any particular reference-fixing criterion for expressions. The strategy is to claim that such reference-fixing criteria are manifest in an individual’s patterns of judgements concerning reference in counterfactual scenarios. It is also worth noting that an individual’s reference-fixing criteria need not be captured in the form of a propositional content. As Schroeter and Schroeter (2016) note, the internalist’s reference-fixing criteria may take the form of recognitional or inferential dispositions, exemplars, prototypes, mental models, etc.

18 Burge (1979) does offer arguments against reinterpretation strategies. However, these arguments are directed at the claim that the subject has a true object-level belief and makes a merely metalinguistic error. The present proposal does not claim this. The proposal agrees with Burge’s claim that his arthritis patient, Alf, really is talking about arthritis (and Alf is indeed wrong that he has arthritis in his thigh). What I aim to show is that this doesn’t require that Alf shares concepts with his doctor.

19 One could construe my argument as demonstrating that internalists can posit shared content. This would also block Goldberg’s argument, which claims that no version of internalism could underpin the widespread success of testimonial exchanges. However, I don’t think an internalist is forced to adopt this interpretation of the approach.

20 This is not to say that we lack privileged access to the contents of our thoughts (Burge 1988).

21 A similar point is made in Begby (2014). Begby argues that we can explain why individuals speak in reliably similar ways without appeal to norms of the sort that Goldberg posits.

22 Begby (2014) presents a similar argument. He argues that the internalisation of norms for correct usage is a contingent matter. My claim here is slightly different: I claim that the possession of dispositions to accept correction – those that enable concept possession (when one has failed to internalise the norms for correct usage) – are equally contingent.