1. THE SUPPLY DOMAIN – INSTANCES OF TESTIMONY

In this section I propose a conception of a distinctive epistemic source, a joint-carved epistemic-cum-socio-linguistic kind, testimony. My main focus in this paper is not to defend this way of carving out testimony, but to develop some consequences. However I will say some things in defence of this way of picking out the epistemic kind once I have described it.

‘Testimony’ as a term is ambiguous between individual acts of testifying, giving her word on some topic, by an agent, and the content testified to in such an act. The epistemic kind I am concerned with is the former, a distinctive type of communicative act. It may be defined thus:

T(iii) in fact implies T(i) and T(ii); however, for perspicuousness I spell the weaker conditions out. I shall not attempt to define ‘presents P’, nor ‘presents P as true’, confident that it suffices to gesture at an intuitively graspable phenomenon. When one articulates in language to an audience one's wondering whether P or wishing that P, one presents P, but does not present it as true; when one conjectures or guesses that P, one presents P as true, but does not present oneself as assuming responsibility for P's truth. When one asserts P, one both presents P as true, and oneself as taking responsibility for P's truth to one's audience. ‘Presents’ is a success concept – so there is such a thing as attempting to present a proposition to one's audience but failing due to lack of uptake. Recognition by one's audience that this is one's project suffices for its success. However it may not be quite enough to guarantee success in assuming responsibility for the truth of P to one's audience, that one is recognised as attempting to do so – if one's credibility in this respect is too low. I am not sure whether the notions of ‘assertion’ and ‘testifying’ should be conceived as success concepts in this respect, and so have not included the requirement that the speaker succeeds in assuming responsibility for the truth of P to her audience in the definition above. But it may be that, as with promises, those who acquire a reputation of giving false testimony or making assertions with epistemic recklessness may become unable to bring off asserting or testifying at all – by this I do not mean that they are not believed, a much stronger condition, but that they are unable even to bring off a performance that effects the commitment to be held responsible for truth by their act. In the absence of a bad reputation however, successfully presenting oneself to one's audience as assuming responsibility for the truth of P to her suffices for doing so – incurring this normative commitment.

T(i)–T(iii) – assuming a linguistic vehicle for the act – characterises assertions.Footnote ⁵ In any assertion the speaker S, bad reputations aside, assumes responsibility for the truth of P to her intended audience H. In many, though not all, conversational situations (see end of section) this amounts to making available to H the opportunity to form belief on her (the speaker's) say-so. But assertions are made for many reasons.Footnote ⁶ To home in on testifying we also specify T(iv) above. (Defence of this restriction is given at the end of this section.)

So: our epistemic kind of interest, acts of testifying, are acts aimed at communicating information, aimed at enabling the hearer to believe with epistemic propriety and so come to know the fact testified to on the speaker's say-so. Our definition allows that ill-grounded testimony – this including testimony that is false either through deceptive intent or honest error, plus true testimony that is nonetheless insincere (that is, the speaker does not believe what she states), as well as testimony where the speaker believes irrationally or otherwise unjustifiedly what she states – is indeed testimony. These are cases where the speaker apparently offers to the audience the entitlement to believe on her say-so, but in fact has no such entitlement to offer. A further word on this point is needed. In any act of testifying the speaker takes on responsibility for the truth of what she tells to her audience. This includes cases where her testimony is ill-grounded, that is she is not epistemically so placed as to properly offer entitlement to believe on her say-so. So I can properly complain to the speaker about her ill-grounded testimony. But as I am theorising it, there is an entitlement to believe on her say-so created by the speaker's testimony only when it is well-grounded. Ill-grounded testimony only apparently offers an entitlement – as when I offer to pay for your purchase, but in reality have no money in the bank. There is an existential presupposition that is false in both cases. If you had money in the bank your offer would entitle me to pay with it; since you have none, there is no such entitlement. The same goes for believing on someone's say-so. Cases of ill-grounded testimony thus may produce blameless belief in a justifiedly trusting recipient, and perhaps justified belief; but never knowledge. Ill-grounded testimony really is testimony; but it is also fake – I fake an offer to you of entitlement to believe on my say-so when in fact I have no such entitlement to offer. This description is most apt for testimony with deceptive intent; but it also holds of otherwise ill-grounded testimony.

The classic and easiest way to testify to the truth of a proposition P is by means of a speech act of assertion made to one's intended audience, by means of an apt sentential vehicle of a shared language. Any assertion satisfies T(i)–T(iii). T(iii) distinguishes assertions from conjecturings and guessings which satisfy only T(i) and T(ii). Assertions whose primary conversational aim is that the hearer should believe on one's say-so are acts of testifying. These correspond roughly to the ordinary language notion of telling.

T(i)–T(iv) give my proposed definition of the supply domain of testimony, instances of testifying. Our definition has rightly not stipulated that this distinctive communicative act only occurs by means of language, but leaves that a further question. I have argued elsewhere (Fricker Reference Fricker2012) that the full assumption of responsibility for the truth of the presented proposition P that testimony involves can only be made by means of speech acts of telling, which are governed by a social norm requiring that one tells that P only when one is epistemically so placed properly to guarantee the truth of P to one's audience. Just as marrying someone is a type of act one can achieve only when the needed surrounding institutions are in place to give one's utterance the required social and legal significance; so equally, uttering some words can constitute assuming responsibility for the truth of some proposition, and offering entitlement to believe on one's say-so, only when the needed linguistic regularities in usage, and social norms, are in place to bestow the needed performative significance and correlated normative and expectational consequences on one's utterance. Even if my arguments on this point are not accepted, the possibilities for effecting the communication of messages without recourse to language are clearly very limited, and testimony is in practise a linguistic phenomenon – namely, speech acts of telling, and various forms of written testimony (see Fricker Reference Fricker, Greco and Henderson2015: section 6).

Our definition also leaves open as further issues to be explored these questions: Can a proposition P be testified to when malapropism, slip of the tongue, or other linguistic error engenders a gap between what is literally said, and the intended message? Given that the testimonial act is a speech act of assertion, are presuppositions and conversational implicatures of the assertion also testified to, i.e. are T(i)–T(iv) satisfied with respect to them? On linguistic errors, I am inclined to think the answer is positive, so long as mutual knowledge of the intended message is achieved. On the second, I have argued (see Fricker Reference Fricker2012) that conversational implicatures are not testified to – the speaker does not present herself as assuming responsibility for their truth, and so does not do so, in contrast with her assumption of responsibility for truth of the content explicitly asserted.Footnote ⁷

For the rest of the present discussion I will set aside the issues just raised about linguistic errors, presupposition and implicature. This gives us the following account of instances of testifying to work with:

Informing someone that P means letting them know that P. ‘… apparently informing’ is included to cover ill-grounded testimony, which cannot do this even if true. Only a minor regimentation of the ordinary language term ‘telling’ is needed to appropriate this term for our concept. I have observed that assertions are made for many purposes. But even tellings, it may be objected, include cases where the teller does not care whether her audience believes what she says, or whether they already know it. I concede that there are cases we should classify as tellings, and as testimony, where this is so. But the cases in question, I maintain, are ones where the teller effects her act because it is her duty in virtue of some professional or social role she has. For instance, someone introducing a speaker at a seminar may first tell the audience where the fire exits are, and the procedure in the event of a fire. She does not much care whether they attend, or believe her; but she is performing her duty. I think these tellings are still ‘aimed at’ the audience believing them, once we take into account their institutional setting. Once we have agreed that acts of testifying – acts with the required performative significance – can only occur by means of acts of assertion exploiting a human language practise including that socio-linguistic act-type, we can gloss T(iv) thus:

It is not the speaker's private motives that determine whether her assertion is an act of testifying, but rather the overt aim and point in the ongoing conversation or communicative context of her act of assertion.

Before I move on to my main concerns in this discussion, I will briefly defend this demarcation of the epistemic source testimony, our supply domain. I comment on three features. First: we have stipulated that testimony is intentional acts aimed at communication. Specifically, for an utterance to be an act of testifying there must be an intended audience to whom the speaker is testifying – to whom she offers the right to believe on her say-so. This audience may not be present, and may not be a specific individual or collection of individuals; the speaker may have in mind only some vague conception of her target audience. But the essence of testifying is relational – the speaker must have at least a vague general idea of some potential audience to whom she offers the right to believe on her say-so. This does not preclude testifying with oneself as audience; whether this is a coherent concept is a further matter I will not debate here. However I am certain that linguistic productions with no intended audience – for instance a doodled sentence inscribed while one is bored in a lecture, perhaps expressive of one's current sentiment – does not constitute testimony. Though it has the syntax of an assertion it does not have the epistemic force that would enable someone reading it properly to form belief on the say-so of what is written – since it was not produced by the speaker under the constraint of such an intended assurance.

Second: we have stipulated that testimony is intentional acts aimed at communication. So we do not count thermometers, fuel gauges and so forth as testifying to the temperature, or to how much fuel is left in the tank. Nor do we count tree rings as testifying to the age of the tree. Nor are you to be counted as testifying to me that it is cold, when I see you – unaware of my presence and not mindful of my reaction – reaching for your thick winter coat before you go out. In Fricker (Reference Fricker, Greco and Henderson2015) I defend the idea that there is a fundamental contrast between the way in which we can get information from two kinds of sources of it, that based in natural correlations between phenomena, and that based in the intentional communicative acts of an agent seeking to let one know about some matter. Grice (Reference Grice1957) first drew attention to this contrast between ‘natural’ versus ‘non-natural’ or, as I prefer to label it, agential meaning. The key point is that human acts of communication do not bear their intended message, the item of agential meaning, as an item of merely natural meaning. There are, generally speaking, no natural correlations to be found between pre-interpretatively identifiable utterance-types occurring in human language-use and worldly states of affairs; not, at any rate, with the right ones – the speaker's intended message, what she states to be so. This being so, the only way in which one can extract the information offered to one in an intentional speech act aimed at telling one about some topic, is by appreciating that this is what it is: realising that this is an intentional act of communication, and successfully understanding something about the speaker's motivation and background of belief in order to interpret her and in doing so identify her intended message. In short: the difference in the nature of the phenomena – natural correlation, versus intentional act aimed at communication – maps into an epistemic difference of kind, not merely of degree, in how one may obtain information from tree rings and so forth on one hand, and from people telling one things, on the other. (Fricker Reference Fricker, Greco and Henderson2015 goes into more detail on these matters. See also the pioneering discussion in Moran Reference Moran, Lackey and Sosa2006.)

(This contrast of natural and epistemic kind between natural versus agential meaning is muddied by the existence of what I think of as fake testimony: announcements at railway stations of train times, or automated messages one receives on telephone connections, that sound like a live human voice making statements, but are no such thing. I say a little about this phenomenon in my final section.)

Third: why specify as the supply domain only those assertions whose primary conversational aim is to offer the audience one's word on a topic, with the aim of being believed? If indeed any assertion presents the speaker as assuming responsibility for the truth of what she asserts, surely what her exact motivation is for doing so does not matter to the epistemic usefulness as source of her act? This thought has some force, and needs to be addressed.

First, note that it is not the speaker's private motives that matter here, but the overt conversational aim of her utterance. This is what T(iv)L constrains. Tellings have as their overt point in the conversational context the aim of informing their audience. (Thus we aptly also include deceptive tellings, which no less are presented as having the conversational role of informing.) Many other assertions do not do so, fulfilling a different conversational role.

I have two reasons for including T(iv)L in the definition of our supply domain. First reason: we are trying to understand the nature of testimony as an epistemic source, and as a socio-linguistic phenomenon. And I think that assertions not aimed at offering the audience the opportunity to believe on the speaker's say-so are a secondary phenomenon: they exploit an existing repertoire whose existence turns on a certain purpose, for what is an ab-use, another purpose. It is only possible to make assertions with the motive of annoying someone because the institution of asserting exists; and it exists only because a core subset of assertions are tellings, made in order to communicate information. This being so, I think we will understand the phenomenon of testimony better if we consider as supply domain the type of linguistically mediated intentional act that is the core phenomenon, tellings.

My second reason is that, while all assertions are acts in which the asserter presents herself as assuming responsibility for the truth of what she asserts, it may be that there are nuances between assertions in different contexts, and for different purposes, as to what precisely this amounts to. In all tellings the asserter offers the audience the right to believe on her say-so, implying a high epistemic standard – in my view, knowledge (see section 6 and Fricker Reference Fricker, Greco and Henderson2015). In different discourse contexts, for instance where one is contributing to a theoretical argumentation, say in a philosophy seminar, the standard is plausibly rather different. Assuming responsibility for truth is not quite the same thing as offering the right to believe on one's say-so, though in tellings they coincide.Footnote ¹⁰ This is why, though I am sure all tellings are governed by the norm: ‘Tell that P only if you know that P’, I am doubtful this norm holds for all assertions with other conversational aims and contexts. This is why one should, after all, be somewhat wary of forming belief from assertions not made with the aim of offering an audience the right to believe on the asserter's say-so, and is my second reason for adding T(iv)L to our conception of the supply-domain of testimony. (However the series of persuasive considerations by means of which I develop my case for my preferred conception of the Core Mechanism in the next sections goes through equally if the supply domain for testimony is taken to be assertions in general, i.e. speech acts for which T(i)–T(iii) holds, without the additional restriction of T(iv)L – so long as these are ones where the conversational context determines that ‘assuming responsibility for truth’ amounts to making available the right to believe on the speaker's say-so.)

Not all instances of testimony, as defined by T(i)–T(iv)L, are apt to yield knowledge.Footnote ¹¹ Clearly, testimony that P when P is false cannot yield knowledge that P. Intuitively neither will accidentally true non-knowledgeable testimony that P yield testimonially based knowledge in the recipient.Footnote ¹² But there are other cases that are theoretically more problematic, which provoke my enquiry. What of insincere testimony – by which I here mean: all cases where the speaker presents P as being so, and herself as assuming responsibility for the truth of P, though she does not herself believe P? Can this yield testimony-based knowledge of P, if due to some quirky configuration of facts the insincere testimony is reliably true? – as is stipulated to be so in Jennifer Lackey's well known ‘Creationist Teacher’ example (Lackey Reference Lackey2008: Ch. 2). Are such cases apt supply to yield knowledge in the proprietary manner distinctive of testimony – the Core Mechanism? In what follows I seek to provide a theoretically motivated answer to this question. I develop an account of both the Core Mechanism, and as corollary the subset of instances of testifying that provide apt supply for its operation to yield knowledge – that is, I provide an answer to Q2, developed in light of my answer to Q1, and this entails an answer to Q3.

I first in section 2 state an account of the Core Mechanism that is intuitively plausible in light of my proposed conception of the supply domain. In the rest of the paper I develop material that provides further theoretical support for this account. In section 3, progressing through consideration of some cases, I arrive at a statement of two contrasting rival conceptions of how testimony works to spread knowledge in a linguistic community. The two conceptions engender rival answer-pairs to Q2 and Q3. The first conception has it that any testifier who reliably supplies only true testimony provides apt supply to yield testimony-based knowledge. The second conception appeals to richer materials to make invidious distinctions within the class of reliable testifiers. In the rest of the paper I develop a persuasive case for the second conception: only those reliable testifiers who are so because they express their knowledge provide apt supply. Each conception has some intuitive appeal. But I argue that the thinner reliabilist conception is blind to key features of testimony as speech act and epistemic source.

2. THE CORE MECHANISM: AN INTUITIVE ACCOUNT

There are lots of ways one can get justified belief apt to be knowledge as a result of observing an instance of testimony. Not all of these instance the proprietary manner in which testimony can yield knowledge to its recipient. This is so even if we constrain the case by specifying that the belief the hearer forms has the same content as is testified to – the testifier asserts that P, and the hearer as a result forms belief that P.Footnote ¹³ This is not sufficient, as this example illustrates: someone states ‘Ich bin Deutsch’ and, while not understanding her utterance, I recognise it as of a German sentence, and thence conclude and come to know that she is of German nationalityFootnote ¹⁴.

Intuitively, there is a proprietary manner, a Core Mechanism, by which testimony functions to spread knowledge in a community. In this discussion my project is to characterise it, and as a corollary to offer a view as to which instances of the supply domain are apt to furnish knowledge by this Core Mechanism. Let's introduce terminology for the contrast: belief as upshot of observing testimony versus belief through accepting testimony. The former is a broad category, while the latter is reserved for belief acquired via the Core Mechanism. In this section I set out my proposed conception of the Core Mechanism, showing it to be intuitively plausible as the natural corollary to our account of the supply domain of testimony.

In an instance of testimony the speaker S presents herself to her intended audience H as assuming responsibility for the truth of the asserted content P to H (T(iii)), and the primary conversational aim of S's act of assertion is that H shall or may form belief that P through accepting the epistemic entitlement to believe P on her say-so offered by S's utterance (T(iv)L ).Footnote ¹⁵ Given this characterisation of the nature and epistemic force of an act of testifying, the salient way to conceive the proprietary response to this act is as the complementary response to it: taking up what is offered or purportedly offeredFootnote ¹⁶ by the speaker, the epistemic entitlement to form belief that P on her say-so; and indeed doing so. The speaker S by her act of assertion overtly renders herself responsible to her intended recipient for the truth of P, thereby rendering the recipient so placed that she can take it on S's authority that P.Footnote ¹⁷ The Core Mechanism operates when the recipient does just that. She forms belief in what the speaker has presented herself as assuming responsibility for truth of, P, on the basis that the speaker has done this – she takes the speaker's word for it that P. In an act of testifying the speaker presents herself as assuming responsibility for the truth of P to her intended audience; so the audience only has to accept the speaker's testimony, accept the purported import of her act, taking it at face value, to form belief in the proprietary way made available by testimony, the Core Mechanism.

OthersFootnote ¹⁸ have noted as a corollary of this that such knowledge through acceptance of testimony exhibits the feature of buck-passing: when I have formed belief that P on another's say-so, via acceptance of her testimony, I will and may refer challenges back to her. And, we may add, my belief in what I have thus accepted on another's authority exhibits epistemic dependence on the testifier: I have relied on her epistemic and character virtues to ensure that she indeed gives me the truth about P when she presents P as true, and presents herself as assuming responsibility for P's truth.Footnote ¹⁹

A testifier ‘giving her word’, and a recipient ‘taking a person's word for it that P’, believing on someone's ‘authority’ or ‘say-so’ are everyday locutions I have here employed. When a person tells one something her word is offered, and when the testimony is accepted her offered word on the matter is accepted. But what exactly does it mean to say the recipient has taken the speaker's word for it that P, accepted her testimony? What is, as it were, the epistemic cash value of this idea? It is not primarily a matter of how precisely the recipient's belief was caused, and is sustained –though there may be corollaries regarding the psychology of her belief-formation – but is a matter of how her belief is epistemically based, what justification for it she has and would be able to offer if challenged.Footnote ²⁰

A speaker in testifying that P offersFootnote ²¹ to her intended recipient the opportunity to form belief and come to know that P on her say-so. Thus she presents herself as epistemically so placed properly to make that offer. A recipient who accepts her testimony may be said to take the speaker's act at ‘face value’ – that is to say, she takes things to be as the testifier presents them to be, that she is indeed epistemically so placed as to properly give her word that P. Hence the style of defence apt from a recipient who has formed belief on this basis of accepting the speaker's offered word for what she states, is some form of words to the effect: ‘S gave me her word, assured me that P, and not easily would she do so unless she was epistemically placed properly to do so.’ What a recipient would probably in practise say is: ‘S told me that P, and she would not easily do so unless she knew that P.’ In section 6 I will argue that knowledge is indeed the standard for a testifier to be epistemically so placed properly to give her word that P. But we can separate the two notions, and others may contest this equation.Footnote ²² So our first point is that it is constitutive of what it is for a recipient to take a speaker's word for it that P, that she incurs as a rational commitment of doing so the requirement to accept as true that the speaker was indeed epistemically so placed as to properly offer her word that P. Our second point is that, given the equation to be argued for in section 6 of being epistemically so placed as to properly give one's word as to P with knowing that P, a recipient who takes the speaker's word for what she tells is rationally committed to accept as true that the speaker knows what she stated. A recipient might say: ‘S told me that P, and I believe her’, or ‘… I trust her.’ These unpack as involving the commitment just articulated. One trusts a testifier to give her word only when she is epistemically so placed properly to do so.

Our account of the supply domain of testimony makes the account given above of the complementary response to it, the Core Mechanism, the salient candidate. In what follows I defend it further. In the next section consideration of some cases leads us to formulate two rival accounts of how testimony works to spread knowledge across a community with a shared language. One has the conception of the Core Mechanism stated above, the other has a less restrictive conception. Each account has some intuitive appeal. In the rest of the paper I develop further theoretical motivation, via an analogy with perception, to boost the case for my favoured account.

3. TWO RIVAL CONCEPTIONS OF HOW TESTIMONY WORKS

Given our description of the Core Mechanism, the following does not instance it:

GIRLFRIEND does not instance the Core Mechanism. Clearly, GIRLFRIEND is not a case where H takes S's word, relies on his epistemic authority to form belief on his say-so. This cannot be so, since what S offers his word about is a different proposition from the one H comes to know as a result of observing S's testimony. There is no presentation of L as true in S's act, so there is simply no opportunity for H to accept L on S's say-so. S's speech act is for H, given what H knows about S, evidence that L, but S does not by his act present L as being so, a fortiori S does not by his act assume responsibility for L being so to H. Instead, H infers from the fact of D being asserted to L via an empirical correlation that is not mutually known.

GIRLFRIEND makes vivid through contrast this fact: when the Core Mechanism operates, the hearer forms belief in a proposition presented by the speaker as true and for which she overtly assumes responsibility for truth; doing so on the speaker's say-so. Clearly, H cannot form belief through accepting S's word that P, if P is not what S has given her word about. So we can certainly conclude: a necessary condition of the Core Mechanism is that the content in which the recipient forms belief as upshot of receiving the testimony coincides with content that the speaker presents as true, and presents herself as assuming responsibility for truth of. Putting aside, as we have done, the issue of malapropisms, and of whether conversational implicatures and presuppositions are overtly presented as true and responsibility for their truth assumed, we can formulate a necessary condition for the Core Mechanism thus:Footnote ²³

However MATCHING alone is not sufficient to narrow down to the Core Mechanism, since we must consider also cases like that mentioned earlier:

In GERMAN the feature of the utterance that is explanatorily crucial to the hearer's coming to know that the speaker is German is not the content and force of the utterance (though the proposition asserted coincides with what she comes to know), but merely the fact that it is recognisably an utterance of a sentence of German. Any utterance recognisably of a sentence of German would have enabled the inference, what particular speech act the sentence was apt to effect is inessential.Footnote ²⁴

GERMAN does not count intuitively as a case of knowledge through acceptance of testimony, showing we need a further necessary condition. Intuitively, the following is missing in GERMAN: the specific content and force of the utterance must be explanatorily crucial for how the hearer forms knowledgeable belief that P from it. That is, the means through which the hearer comes to know that P as a result of observing the speaker's utterance is such that the fact that it was an assertion that P is critical; and the fact that it was an assertion that P is critical. This condition on how the recipient is able to obtain knowledge from the testimony splits into two conditions – one regarding the manner in which the recipient forms belief in response to the testimony, and a second on how the event of the speaker's testifying relates to the putative fact testified to. We can formulate the interim conclusion that instances of belief formed via the Core Mechanism are occasions of testifying on which:

ASR means approximately that, if S asserts that P H will form the belief that P; while if she instead were to assert that Q H would form the belief that Q, and so forth; and if S had not asserted, but say questioned whether P, H would not have formed a belief from the content of her utterance at all.Footnote ^25, Footnote ²⁶ Only approximately, since there are some propositions such that the recipient would not believe them if told to her, for instance if she independently knows them to be false; and the recipient may and should be sensitive to defeaters for the reliability of the testimony.

ASR is one half of a two-part condition relating to instances of testimony such that its holding generally of instances of testimony in a community C ensures that testimony as a multi-instance process operative between testifiers and recipients in C nearly always yields beliefs that are true, and moreover are reliably so, satisfying a safety condition on knowledge. The other needed half is that the following condition must hold of speakers on almost every occasion:

In the previous section we developed an intuitively apt characterisation of the Core Mechanism as accepting the speaker's offered word for the proposition she asserts, thereby assuming responsibility for its truth; and forming belief on her say-so. This process of acceptance of the speaker's word clearly satisfies ASR. Thus the general holding of instances of RT (pace complications noted in footnote 26) is necessary for the Core Mechanism as characterised to be a process-type whose operation over many occasions reliably yields reliably true beliefs. Moreover in any instance the recipient's acceptance of the speaker's word rationally commits her, as a consequence of doing so, to the truth of the proposition that the testifier's testimony was true – which, given that she asserted that P, is equivalent to the proposition that P. Given that we are considering a situation in which the recipient gains knowledge from accepting the speaker's testimony, she cannot know that P independently of accepting the testimony. So the only way she can know P is true is if she knows the speaker's testimony that P indicates its truth. Thus she is rationally committed, as the minimal needed justifying basis of her belief, to the proposition that the speaker S has this property: not easily on this occasion would S assert that P unless P – that is, to the relevant instance of RT. By forming belief that P in response to being the audience of an assertion that P, in a way that is sensitive to its content and force, she is rationally committed to this.Footnote ²⁸

If a speaker S satisfies the relevant instances of RT for some range of topics on which she testifies – that is, on each occasion she reliably speaks truth – then a hearer who forms belief in response to instances of S's testimony by taking S's word for what she states, conforming to ASR and to our intuitive conception of the Core Mechanism, will by doing so acquire beliefs that are reliably true. Does this mean that all there is to being a suitable testifier, one from whom knowledge can be gained by the Core Mechanism, is being such as to satisfy RT? This is the issue I will explore in the rest of my discussion. My answer will be No. I aim to discredit a certain reliabilist conception of how testimony functions to spread knowledge across a society with a shared language. This conception has initial plausibility, being supported via this thought:

‘So long as there is a community of testifiers who satisfy relevant instances of RT, together with recipients who form belief from encountering their testimony in accordance with ASR, testifying will engender in recipients reliably-true belief, apt to be knowledge.Footnote ²⁹ Hence we may conclude that the core mechanism of getting knowledge from testimony occurs whenever a hearer acquires a belief by forming belief in what was asserted, in response to the fact that it was asserted, from any speaker who satisfies RT.’

Our target reliabilist conception consists of two components, complementary accounts of the reception and supply sides of testimony:

Note that the characterisation of the Core Mechanism given in RCMVT is subtly but crucially weakened as compared with the account of it developed in section 2 as taking the speaker's word for what she states. This is intended. I finish this section shortly by describing two cases where the recipient forms belief from a testifier's assertion in a manner that satisfies RCMVT, but does not instance the Core Mechanism as described in section 2. Correlatively as we shall see, there are possible cases of testifiers who satisfy RT, but who do not satisfy the intuitive conception of a well-grounded instance of testimonial supply as characterised it in section 1, namely that the testifier is epistemically so placed as to properly assume responsibility for the truth of what she asserts to her audience. The inadequacy of RCMVT as capturing our initial description, and associated inadequacy of RSSVT, will be brought out in sections 6 and 7.

RSSVT and its associated conception of the Core Mechanism as RCMVT might be further defended thus:

‘In the philosophy of language, and of mind, there may be explanatory things to say about why there are reliable testifiers in a community; but within epistemology, all that matters is that there are such people, and being so they are an epistemic source for other people, recipients of their reliable testimony. Epistemology theorises sources of belief according to their reliability; and all reliable testifiers are equally good qua reliable source; so that further distinctions amongst reliable testifiers are irrelevant to the epistemology of testimony.’

Against RCMVT-plus-RSSVT I shall contrapose and argue for a rival view. On this view our account of the epistemology of testimony cannot be properly developed without paying attention to overlapping domains – the nature of human agents, and the general nature of the variety of intentional speech acts made possible by the institution of human natural language. Once we pay full attention to the nature of the speech act of assertion, and why it has epistemic force at all, I shall argue, we see that there are implications for the nature of the Core Mechanism – RCMVT is too thin; and grounds to discern epistemically relevant distinctions within the class of reliable testifiers. Specifically, I will develop a persuasive case that only those who are reliable because they speak from knowledge of what they assert are sources from which knowledge can be gained via the Core Mechanism – the proprietary way in which speech acts of telling as such are able to spread knowledge. In the next section I develop an analogy with perception intended to be rhetorically persuasive. (The point of the analogy is mainly rhetorical, so you don't have to accept everything about how perception is there characterised to allow its force.)

I end this section by giving two examples that illustrate that RT does not coincide with a more stringent property of speakers:

Since ‘A knows P’ entails ‘P’, any speaker that satisfies KT satisfies RT; but not conversely. Here are two cases of a speaker that satisfies RT, but not KT:

Would-be liar Simonetta satisfies RT, but not in virtue of satisfying KT. She intends to deceive, to tell a false story; but inadvertently reliably speaks truth in relation to her topic not because she expresses knowledge, but due to a freak combination of circumstances. Undiscerning mimic George also satisfies RT by a deviant route: his statements express his beliefs, and his beliefs about the quality of the art at the show that evening are true, and he would not easily have false beliefs about it. But they are formed through a deviant route which does not render them knowledge, since he has no sensitivity to the quality of the art, and does not even have a reliable indirect informational link to the quality of art at exhibitions generally. George, like Simonetta, reliably speaks truth not through having knowledge of his topic, but due to a coincidental one-off combination of circumstances.

The cases just described are far-fetched anomalies unlikely to occur in real life. This does not prevent them serving my current purpose. They are possible, and that is enough to reveal the fact that KT is crucially stricter than RT.

Simonetta is a reliable testifier about the events in this village; but she does not know what she asserts, since she does not believe it. Someone, say Paul, who is wise to the whole situation has a basis to know of Simonetta that what she asserts on this topic is reliably true (despite her insincerity); and so can and we suppose does reliably and with justification form knowledgeable beliefs in what she asserts. His manner of forming these beliefs conforms to RCMVT but, I will argue, it is not a case of forming beliefs in accordance with the Core Mechanism. Instead it has significant analogy to GIRLFRIEND. Paul does not take Simonetta's word about matters in the village. Paul infers from the facts of what Simonetta asserts to truths about the village, due to an empirical basis he has for taking Simonetta's statements as a reliable sign of what in fact happened; but he does not take her word for what she asserts, does not form beliefs on her authority. A similar account holds for how someone aware of the relevant facts about the set-up may get knowledgeable beliefs from George's testimony. These are beliefs that are reliably formed in a way that is sensitive to the content and force of George's utterances, satisfying RCMVT; but they do not instance the Core Mechanism as we intuitively and aptly described it in section 2. So I shall try to demonstrate in what follows.

It may be that no actual philosopher holds or would subscribe to RSSVT + RCMVT.Footnote ³² It is nonetheless worth articulating and critiquing because it does, as I have sought to exhibit, have strong intuitive appeal. Seeing how and why it is an inadequate conception of how testimony functions to spread knowledge in a community is thus enlightening.

According to RSSVT any reliable testifier is an equally good supplier for getting reliably-true, knowledgeable beliefs from testimony by the Core Mechanism, conceived as RCMVT. Thus, RSSVT has it, RT, not KT, picks out the good supply domain for testimony. In section 6 I will argue, against RSSVT, that the Core Mechanism correctly conceived is more restrictive than RCMVT, and operates with good outcome only on testifiers who satisfy KT: there are epistemologically key distinctions within the class of reliable testifiers. But first, as announced, I develop an analogy intended to be rhetorically persuasive.

4. AN ANALOGY: PERCEPTION VERSUS NON-PERCEPTUAL RELIABLE MATCHING

In visual perception one has visual experiences with objective representational content and somehow one gains corresponding knowledge of the external world. These experiences present one's environment as being a certain way– for instance that there is a lighted candle on a table in front of one; or that there is a robin in the garden seen through the window. How and under what conditions do one's visual experiences yield one experience-based knowledge?Footnote ³³

Here is one possible answer to this question:

On the Simple Reliabilist Account what underwrites the epistemology of experientially-based knowledge generally is the fact that our experiences have matching co-variance with how things objectively are, so that their content reliably indicates this. The only epistemologically relevant property of a subject's experience in some situation is whether or not there is modally reliable matching of what her experience represents to be so, and what is so in the subject's environment. So we can formulate:

The reader will have spotted the analogy with our simple reliabilist account of the good supply domain for testimony, RSSVT. Our Simple Reliabilist account of perceptual or, as one must say on this view, experientially-based knowledge, is a straw-philosopher target. My reason for formulating RSSVP is not that it has defenders, nor that it has prima facie plausibility. On the contrary, as already advertised, its role in my argument is, by suggestive analogy, to undermine the attractiveness and reveal the deficits of the analogous account of testimonial knowledge, RCMVT-plus-RSSVT. According to RSSVP, any case where a subject enjoys an environment-matching visual experience generated in a manner ensuring that any token visual experience the subject has on that occasion reliably matches how things are in S's environment, is a case where she gains knowledge via the epistemological category of experientially-based belief; and this is all there is to the epistemology of this source of knowledge. But this is not what we ordinarily think. For the cases that satisfy RSSVP include, but are not exhausted by, what we ordinarily consider to be perception.

Assume, for purposes of argument, that visual perception is a causal process producing visual experiences, but that similar experiences can also be produced by other means, including when the experience does not match how things are in the subject's environment. Specifically, assume that these experiences caused by other means can have the same general and quantified content as bona fide perceptual experiences.Footnote ^36, Footnote ³⁷ Now consider these two contrasting possible situations. In the first situation a person is seeing (visually perceiving) a lighted candle on a table in front of her. This is the second situation:

I think our everyday judgement is decisively against the view that the subject in CANDLE is seeing, visually perceiving, the candle burning on the table in front of her and obscured by the interposed screen.Footnote ⁴⁰ My present thesis of interest is this: CANDLE is not only not a case of perception, it is also epistemologically entirely different from any case of perception. This is so despite the fact that the set-up ensures that the qualitative quantified content of the subject's ongoing candle-experiences reliably matches how things are on the table in front of her – given the set-up, it is non-accidental that the ongoing experiences she enjoys as of a candle in front of her gradually burning down have matching variation with the current condition of the obscured candle behind the screen on the table, as it burns down. My main aim is not particularly to insist that the person seeing matching candle images projected on the screen cannot get knowledge of what is in behind the screen on the table front of her.

As with Paul and Simonetta, if she is wise to the entire set-up she has available a basis for inferential knowledge of what is behind the screen. What I insist is that if the total situation is such that she does get such knowledge, this is by an entirely different epistemic route from what occurs in perception. CANDLE is not to be grouped indifferently with genuine cases of perception from the standpoint of epistemology. Perception is an epistemic kind. Howsoever-mediated reliably-matching experience is not.

If we want to explain how it is that we are able to get knowledge from our experiences at all, we have to explain this as, in general, made possible because we are subjects of the epistemic link of perception. Perceiving that P involves having a matching experience as of P that is caused in the right kind of way –the way needed for perception. Whether or not this admits of non-circular characterisation, it admits of counter-examples; and CANDLE is definitely not a case of normal perception. On the simple reliabilist account of experientially-based knowledge formulated above in RSSVP there is no epistemological difference discernible between CANDLE and a normal case of perception, as sources of justified belief and knowledge, since both satisfy the relevant instance of RE. But there is a huge difference between them. There are epistemologically key distinctions within the class of reliably matching experiences. A simple reliabilist view has no plausibility applied to perception.

In the next section I spell out these epistemologically relevant differences between CANDLE and cases of normal visual perception. Then in section 6 I return to our contest of RSSVT-plus-RCMVT versus KT plus the intuitive account of the Core Mechanism as taking the speaker's word for what she states. I show how the arguments made regarding perception in section 5 have analogues establishing the parallel result regarding testimony.

5. EPISTEMOLOGICALLY RELEVANT DIFFERENCES BETWEEN PERCEPTION AND NON-PERCEPTUAL RELIABLE MATCHING

It may be we should allow a category of ‘animal knowledge’ – knowledge gained by simple creatures who lack any capacity for reflection on the status of their beliefs. But any user of language must, perforce, be more sophisticated than this. Language-users will have the capacity for some basic level of reflection about the status of their beliefs. (I return to animal knowers briefly later.) Human language-users possess a background conception of the world and their various means of epistemic access to it. They make sense of their beliefs in terms of this background theory, including having a basic explanation of how, for any item of knowledge, they were able to access the known fact.

So when a cognitively normal human forms a perceptually based belief that there is a robin in the garden because she sees it through the window, her ground for her belief, and a rational commitment of it is: ‘I saw a robin.’ Part of the background justification for her belief is that she knows about the epistemic source of visual perception or seeing – that this is a fertile source of knowledge of what is going on in one's immediate environment. She is a master of this part of our folk epistemology. This being so, if she somehow learns that she did not see a robin, this is a normative and doxastic defeater for the belief. Her basis for holding the belief is undermined, since its folk epistemological justifying ground and rational commitment is defeated.Footnote ⁴¹

(If the situation is a one-off such that there is an alternative story she can be told about how her non-perceptual experiences are nonetheless reliable in virtue of some other deviant reliable link-up – as in CANDLE – this is a further, incidental feature. Learning these additional facts about one's situation makes available another, completely different, route to knowledge that there's a robin in the garden.)

Given the background theory of herself and her epistemic access to the world in terms of which a normal human theorises her epistemic situation, she will make sense of a perceptual belief – for instance that there is a robin in the garden – explaining how she has come to know this fact, by answering the question (posed by another, or herself in her own epistemic housekeeping) ‘How do you know that there is a robin in the garden?’ with the reply: ‘I saw one there’.Footnote ⁴² She certainly will not answer this challenge by citing an instance of RE, Robin-RE: ‘Well, I know that not easily would I have a visual experience as of a robin in the garden unless there actually were a robin in the garden.’

That she saw a robin ensures that Robin-RE holds, on this occasion.Footnote ⁴³ So she has grounds to believe Robin-RE because, but only because, she has grounds to believe she saw a robin. She has absolutely no other basis for thinking that Robin-RE holds! In any normal situation of perception, one has grounds to believe the relevant instance of RE only to the extent that, and because, one believes that one is perceiving. This being so our target Simple Reliabilist conception of what matters in the epistemology of experiential knowledge, namely that all that matters is that RE holds, formulated at the start of section 4 in RSSVP, fails to capture a key feature of how experiences are able to ground belief as they do – they do so not in virtue of being somehow-or-other reliably correlated with how things are; but in virtue of being perceptions!

There are two separate complementary points to be disentangled in what has just been set out, to argue my case against RSSVP. First is the observation that we in fact theorise our epistemic relation to the world mediated through our experiences in terms of the folk-epistemological concepts of perception – seeing, hearing, touch and so forth. We theorise ourselves as having experiential access to how things are via these perceptual links, so that learning one has not perceived some fact one took oneself to have experience-based knowledge of is a defeater, given one's background theory. But this observation alone does not establish that perceptual experience is the true epistemic kind, rather than the broader category reliably-matching experience. What our folk categories currently are would cut little epistemological ice, if these folk categories were a mere arbitrary contingency; if an alternative broader category, ‘however-mediated reliably-matching experience’, would serve equally well or better. If this were so, it could legitimately be objected that our current view is mere folk ideological prejudice, and we should move to the more perspicuous broader category.

So this first point, the observation about our folk-epistemological categories, has force only when supplemented by a second one. Namely the thesis that our folk-epistemological categories are not arbitrary, but do indeed carve our epistemic relations to the world at their joints. This is so. Perceptual experiences (of the various perceptual modalities) are a causal-explanatory kind. Lawlike generalisations can be made about them, and regularities concerning their occurrence are projectible. In contrast, ‘reliably-matching’, is not a projectible property of experiences. There are few, if any, lawlike generalisations to be made about this collection; it is not a unified psychological or physical causal explanatory kind. It is comprised of the causal-explanatory kind ‘perceptual experiences’, together with a rag-bag of one-off unlikely possibilities that have nothing important in common with either genuine perceptions, or each other.Footnote ⁴⁴ Because this is so one will generally, special information about unlikely anomalous scenarios like CANDLE aside, have grounds to think that one has a reliably matching experience just when and to the extent that one has grounds to think one is perceiving. The broader category reliably-matching experiences does not pick out a kind, nor a projectible confirmable relation. Consequently it is only in a very strange and unusual set-up that one might have grounds to believe oneself the subject of reliably-matching experiences that were not so in virtue of being perceptions. This second point shows that conceiving our epistemic relation to our environment in terms of the categories of the various modalities of perception is not a happenstance piece of theorising that we could easily drop. Our current ideology, our folk epistemology, cuts our epistemic situation at its true joints.

These two points together demonstrate that there are epistemologically relevant distinctions within the class of reliably-matching experiences. Perceptual experiences are a distinct and epistemologically central subcategory within these. Our epistemic relation to the world when we perceive, our ability to get knowledge, turns on the fact that we are indeed perceiving, not on the broader non-projectible disjunctive fact that we are enjoying somehow-or-other reliably-matching experiences.

There is a deep and crucial third point to be made on this matter. Space constraint allows me here only to assert it without defence. In thinking about the epistemology of perception we cannot just take it as a given that our experiences have the objective content they do. The correct account of how this objective content accrues to them invokes the fact that they are mainly veridical perceptions of features of our environment that their content represents. So the case where experiences are reliably matching to the world in virtue of being indeed perceptions of that world is basic in what makes us able to have experiences of that world at all. Reliably-matching but non-perceptual experiences with objective content are an essentially derivative and secondary phenomenon. This third point adds further support to our already strong case against the Simple Reliabilist Account of Perceptual Knowledge expressed in RSSVP. We have found a further way in which the core case of reliably-matching experience which is perceptual is privileged: this case is what makes experience with objective content available to the perceiver at all.

My preceding argument invoked the fact that the epistemic subject is an occupant of the ‘space of reasons’. Her perceptually based beliefs are formed and maintained against the background of her folk epistemology. Thus she incurs certain rational commitments of her belief consequent on how she supposes herself to have had epistemic access to the fact believed. I invoked as one part of my case against the Simple Reliabilist Account of experientially-based belief, the fact that she will treat the news that she was not in fact seeing as a defeater for a belief formed from seeing. This prompts the thought that it is only within an internalist epistemology of this genre that my case can be made. My precise arguments turn on the internalist conception of justification and knowledge that I favour. However, as was noted, our target Simple Reliabilist Account of experientially based knowledge is only a philosophical straw man. Current reliabilist theories are more refined, and have richer resources. Most of these will have ways to make the same invidious distinctions amongst reliably-matching experiences that I have drawn using the resources of my internalist framework. Process reliabilism, for instance, if it invokes a substantial notion of a real-time process can certainly do so. Perception is a real causal process, whereas the other ways in which one might enjoy non-perceptual reliably-matching experiences are a shapeless disjunction.

All three of our points about perception – that learning one is not perceiving is a defeater for an experience-based belief; that while ‘perceptual’ is a unified causal-explanatory property of experiences, ‘reliably-matching’ is not, so our folk epistemology that takes perception as its category is deeply rooted in the nature of our epistemic link-up to our environment; and that the core case of genuine perception is privileged, since it plays a foundational role in what makes experience have objective content, hence afford a way of getting objective knowledge at all – all three have analogues in the case of primary interest to me in this discussion, namely testimony. The next two sections develop the analogous points regarding testimony and RT versus KT.

6. TESTIMONY AGAIN: THE CASE FOR KT

In section 1 I developed and defended a definition of the supply domain of testimony in terms of T(i)–T(iv)L. Events of testifying are tellings: acts in which the speaker presents herself as assuming responsibility for the truth of the proposition P she presents as true, thereby offering her intended recipient the entitlement to believe that P on her say-so, and with the conversational aim of allowing the recipient to do this. In section 2 I then developed an account of the complementary response to this offer on the part of the recipient: taking up the epistemic assurance as to the truth of P offered by the speaker in her speech act, taking her word for what she states, and forming belief on her say-so. This, I suggested, is the proprietary manner of forming belief in response to testimony, the Core Mechanism. What exactly this amounts to was further spelled out in terms of what justifying grounds a belief formed through taking the speaker's word is based on. We saw that a recipient who takes the speaker's word for it that P incurs a rational commitment to this supposed fact: that the speaker was epistemically so placed as to be able properly to give her word that P. Only so does the entitlement to believe on her say-so apparently offered by the speaker indeed exist (see section 1). Testimony that P offered by a speaker who is not epistemically so placed as to properly give her word that P is ill-grounded testimony.

I think the standard for being epistemically so placed as to properly give one's word that P to another is that one knows that P.Footnote ⁴⁵ I argue for this in Fricker (Reference Fricker, Greco and Henderson2015). I there offer an argument inspired by Edward Craig's work (Craig Reference Craig1990) that knowledge being the required epistemic standard for tellings is a basic fix on the epistemic standards we operate with in everyday language-use for someone's counting as satisfying ‘knows’, so that our epistemic standards for knowing are regulated by this link.Footnote ⁴⁶

It is a contingentFootnote ⁴⁷ fact that we have a speech act type, telling, that is regulated by the following social norm:

As was already argued in section 1 it is not merely contingent however, that acts of testifying as characterised by T(i)–T(iv) are effected by means of speech acts of telling. One can effect the performance of giving one's word that P only when a suitable social-normatively regulated act-type exists enabling one to do so. Suitable normative regulation is that one is sanctioned for offering one's word when one was not epistemically placed properly to do so – that is, that one uses a speech act governed by the K-norm.Footnote ⁴⁸

Given our equation of the standard for being epistemically so placed properly to give one's word that P with knowing that P, it follows in light of what we have already set out above that one who takes the speaker's word for it that P is rationally committed to accepting as true the proposition that the speaker knows what she stated.Footnote ⁴⁹ As was spelled out in section 2, that the speaker satisfies this condition, knows what she stated, is part of the justifying grounds for a belief formed on this basis. We also saw in section 2 that the recipient will generally only be able to know this about the speaker S if she knows something stronger about her: that S satisfies the relevant instance of (KT) – not easily would S tell that P on this occasion unless she knew that P. Hence the typical defending justification a recipient will offer for her belief formed through acceptance of a speaker's testimony is: She told me that P, and not easily would she do that unless she knew. This is the recipient's justifying grounds, her basis, for her belief that the speaker indeed knows what she, by telling that P, presents herself as knowing.

From our identification of the supply domain of testimony as tellings, and our account of the complementary response to a telling, taking the speaker's word for what she tells, which we identify as the Core Mechanism, it follows that (KT) is the property of speakers that underpins the good supply domain for testimony. Only if speakers, testifiers, in a community C nearly always satisfy (KT), will instances of testimony that occur in C nearly all be made by speakers who are epistemically so placed properly to give their word –that is, they know – what they state. We have delivered a package providing motivated answers to our three initial questions about how testimony as an epistemic source should be individuated and theorised.

Tellings where the speaker knows what she states comprise the good instances of testimony-supply in two ways. First, given our characterisation of the speech act, tellings where the speaker does not know what she states are ill-grounded: she presents herself as offering entitlement to believe on her say-so when she in reality has none to offer. This already gives us a sense in which they are inapt as testimony-supply. Second (relating directly to our framing question Q3), tellings where the speaker does not know what she states cannot engender knowledge in the recipient by the Core Mechanism, on the account of it we have developed. Here is why not:

When a recipient forms belief in what she is told, P, by accepting the teller's word for it that P, she is rationally committed to: S knows that P. If she learns that S does not know that P (for instance, by finding out that S could not possibly have come to know that P), this is a normative and doxastic defeater for her belief. She no longer has any reason to believe what S told her on the basis she took herself to have for this. That basis is undermined by finding out that S does not know what she offered her word about. H's reason for accepting S's testimony as true was that she supposed it to be well-grounded, expressing S's knowledge. Once H learns that S does not know what she states, her available justifying ground for her belief is defeated. And, crucially, once H learns that S's testimony does not express her knowledge H has lost her original ground for expecting it to be true – namely, that S would not offer testimony unless she was epistemically so placed properly to do so. (Some more will be said in defence of this second claim below.) Of course, there may be some circumstances in which H has available other independent grounds to expect S's testimony to be true, regardless of whether it is epistemically well-grounded. But this, like justificatory over-determination generally, is not the basic case. I say a little more about where such cases are likely to come up in my final section. But if finding out the speaker does not know what she tells is a normative and doxastic defeater for her testimony, then even if the recipient never finds out that the speaker lacks knowledge of what she states, the recipient's testimony-belief has a factual defeater, and so – on a standard view of conditions for knowledge – is not knowledge, even if as it happens what the speaker told her is true; and even if her testimony was reliably true, but non-knowledgeable.

The reader will already have realised that our favoured conception of the Core Mechanism as taking the speaker's word for what she states is stricter than that formulated in section 3 as one half of the reliabilist conception of how testimony works to transmit knowledge, namely:

To see that RCMVT is weaker, consider again the scenarios described at the end of section 3, Would-Be Liar and Undiscerning Mimic. We noted that someone wise to the whole set-up, Paul, can get knowledge from would-be liar Simonetta's assertions inferentially via his knowledge of the specific one-off mechanism ensuring correlation of her assertions with the facts asserted. This way of getting beliefs as a result of observing T conforms to RCMVT. But Paul is clearly not taking Simonetta's word for what she states. He does not form belief in what she says because he takes her to be expressing her knowledge – he knows she is not doing so. Instead, he makes an inference from what Simonetta states to what is so via his knowledge of a nomological correlation between her assertions and what is the case supported by the specific facts of the situation. This way of getting beliefs from observing someone's assertions is analogous to GIRLFRIEND. It does not constitute taking the speaker's word for what she states. Correspondingly, the good supply domain for the Core Mechanism characterised by KT is stricter than that characterised by RSSVT, as Undiscerning Mimic and Would-Be Liar illustrate – both George and Simonetta satisfy RT, but not KT.

Once our account of the supply domain of testimony as intentional acts of communication in which the speaker offers her word on her topic is accepted, the account of the Core Mechanism developed in section 2 flows from this, and with it the conception of the good supply domain for testimony associated with (KT), not (RT). So it seems we have solid reason to make invidious distinctions amongst reliable testifiers – Undiscerning Mimic and Would-Be Liar can engender knowledge of what they testify to in a suitably wised-up recipient; but their ill-grounded tellings cannot yield knowledge by the Core Mechanism, the way of getting knowledge from tellings that these are constituted as communicative social-norm-governed speech-act-type to make available. Simonetta's and George's testimony is ill-grounded, they do not know what they state. So taking their word for it is equally ill-grounded. Once it is learned they do not know what they state, this defeats belief formed by doing so; and even if the recipient never learns this, her belief is subject to a defeater and so is not knowledge, even if true.

I have sketched a neat picture here. But the persuasive case for it is not completed until we rebut objections parallel to those discussed in section 5 regarding perception.

7. FURTHER DEFENCE OF KT AND CM AGAINST OBJECTIONS

In section 5 we observed that human knowers reflect on the epistemic status of their beliefs, making sense of them in terms of their own folk theory of their place in the world and means of epistemic access to it. When someone forms a belief that P through visual perception she justifies it with the folk-epistemological defence: I saw that P. This is an essential part of her justifying basis for it. We further argued that this carving up of her epistemic terrain is well-grounded in underlying facts. Visual perception is a unitary causal and explanatory category, so that ‘genuinely perceptual experiences’ is also such; whereas ‘reliably-matching experiences’ is no such thing. It comprises the causal-explanatory kind of perceptual experiences plus a rag-bag of other one-off items with no natural unity to them. This, we argued, means there is good reason in the underlying epistemic facts to stay with the theorisation of experientially-based knowledge in terms of the category of perception. There is good reason to make epistemological distinctions within the class of reliably-matching experiences. We noted that a broadly reliabilist account of the epistemology of perception that appeals to richer resources than the Simple Reliabilist Account formulated as our straw-man target could accommodate this distinction, since perception is a substantial unitary process of belief-formation, whilst a broader category of ‘reliably-matching experiences’ corresponds to no such substantial process.

Similar considerations apply when we consider our rival conceptions of the good supply domain for testimony, RT versus KT. Speakers who testify easily only to what they know are a kind with a considerable degree of causal-explanatory unity. They satisfy this condition due to their conscientious success in conforming to the K-Norm that governs tellings, this being based in a combination of their good will and their reluctance to incur the social sanctions of transgression of this social norm, along with their competence in estimating what they do and do not know. In contrast speakers who, like Would-Be Liar and Undiscerning Mimic, conform to RT but not in virtue of conforming to KT are not a unitary kind. There is no generally applicable explanation of why they do so, only a variety of serendipitous combinations of facts each of which results in the unlikely phenomenon of someone whose assertions are reliably true though not in virtue of her conscientious conformity to the K-Norm governing tellings. Correlatively, the fact that a testifier on a particular occasion O due to some unlikely combination of circumstances conforms to RT though not KT on O is no reason to predict that on some other occasion she will conform to RT.

These facts mean that categorising speakers in terms of whether or not they satisfy KT is cutting epistemic reality at its joints, while categorising them in terms of RT is not. The fact that a speaker reliably tells only what she knows, satisfying KT, entails that she reliably tells only what is true, satisfying RT. So a recipient of testimony is very often in a position to know that the speaker satisfies RT; but this is only because she has grounds supporting her knowledge that the speaker satisfies KT. It is only in a very unlikely one-off set-up that a speaker will satisfy RT but not KT, and correlatively it is unlikely, even in that situation, that a recipient will be in a position to know this.Footnote ⁵⁰ Since this is so, once a recipient H's original basis for accepting a speaker S's testimony is defeated, by H somehow learning that S does not, as H originally supposed, know what she stated, there is in the normal run of things no weaker basis available to H to justify her in believing S's non-knowledgeable testimony to be nonetheless true. I expect someone's testimony to be true because I take her to be consciously conforming to the K-Norm governing tellings, and consequently to know what she states. Once this basis is defeated, I have – unlikely one-off set-ups aside – no other grounds, no other evidence, to expect her testimony to be true.

A further related fact is that, when I am seeking evidence that someone's testimony on some topic is to be trusted, a track record of past true testimony is only projectible to predict future true testimony via the hypothesis that her past testimony was all true because she as a speaker conscientiously conforms to the K-Norm. Other explanations of why her past testimony has all been true are unlikely to admit of extrapolation to future cases. (Consider Would-Be Liar: the explanation of why her testimony about the village is reliably true clearly does not project – it turns on a particular false belief she has about the book in question, that it is fictional when in reality it is an accurate historical account. A projection of the same motives to future cases would predict future false testimony, not a continuation of true testimony.)

Our account of the supply domain of testimony in terms of T(i)–T(iv)L led to our conception of the Core Mechanism as taking the speaker's word for what she states, and correlatively of KT as specifying the good supply domain for testimony. The considerations just adduced further support this conception. They do so by revealing that the following is not an epistemically viable option for testimony recipients: stop thinking of speakers in terms of whether they know what they state, and instead regulate your responses to their testimony by reference to whether or not what they tell is reliably true, whether they satisfy RT – confirm or disconfirm that hypothesis about them instead, to determine whether or not to believe what they tell you. RT, in contrast with KT, is not a projectible property of speakers, so this alternative strategy for responding doxastically to testifying will fail to tap into the regularities that are out there, as to whose testimony on what topics is and is not to be trusted. This is not surprising, since in failing to be informed by the fact that a speaker in testifying assumes responsibility for the truth of what she tells, and so presents herself as knowing what she tells, it misses out on the essence of what a human communicative act of testifying is.

This last remark introduces a final further consideration in favour of KT and our preferred conception of the Core Mechanism as taking the speaker's word for what she states, as against RSSVT and RCMVT. To get reliably true belief from testimony a recipient must be able successfully to identify assertions and their contents. And – I hereby assert the conclusion of a long argument elaborated in Fricker (Reference Fricker, Greco and Henderson2015) – there is just no chance of doing this in any way short of fully interpreting the speaker – seeing her act as the speech act it is, upshot of her motives and beliefs. This includes seeing it as what it is, an intentional act of Gricean communication: letting someone else know what one knows, through making public one's own knowledge. This being so, there is no epistemic possibility of successfully identifying the reliable testifiers, except via identifying them as what they in fact are – agents with typical human motives and communicative purposes, who conform to the norms of language governing the speech act of assertion. This further confirms my central claim that classifying speakers according to whether or not they know what they are talking about, rather than whether they reliably speak truth on a topic, is not incidental, but essential, to being a suitably discriminating recipient of testimony.

8. ORIGINATING VERSUS PASSING-ON TESTIMONY AND A LIMITATION TO THE ROLE OF KT

I have now set out my persuasive case for my favoured package of answers to the three questions that generated my enquiry. In this last section I flag a significant limitation in scope of the account I have set out, and area for further work.

I have described the nature of acts of testifying and the proprietary response to such acts by recipients. But to characterise the Core Mechanism by which testimony facilitates the spreading of knowledge within a society with a common language is not fully to determine one's account of how testimony works. In considering whether a speaker who engenders knowledge in an accepting recipient must herself know what she tells, there is an important difference to be drawn between originating testimony, as we will call it; versus passing-on testimony. An originating testifier reports on matters she has found out for herself by dint of her own individual epistemic resources and situation. The epistemic situation when a recipient is offered originating testimony is very different from that when she is offered merely passing-on testimony – when a speaker relays what is already known by others, and she herself has learned from their testimony. In the previous section I argued that once the belief that the speaker knows what she tells is defeated, the recipient typically has no other basis on which to expect what she states to be true. This is true of originating testimony. But it is much less so of passing-on testimony. Very often, when a testifier is passing on what is known, the belief that the speaker herself knows what she states is not essential to the recipient's basis for believing what she is told is true. What matters, very often, is the recipient's belief that what is told represents what is known; not necessarily that the speaker herself knows or even believes it.

Similar considerations must be applied to explain what goes on when a recipient hears, for instance, plane departure times being announced over a loudspeaker at an airport. Finding out that the utterances are produced by an automated artificial speaking mechanism does not, in this case, undermine the basis she has for believing what she hears. What matters is that the apparent speech act she perceives relays what is known; not that it is being produced by a speaker who herself knows what she states. These considerations indicate how the intuition can be accommodated that the schoolchildren in Jennifer Lackey's well-known Creationist Teacher example (Lackey Reference Lackey2008) can acquire knowledge from their teacher despite her not knowing, since she does not believe, what she teaches them about evolutionary theory, whilst maintaining the account of how testimony works described in the preceding sections. Further consideration of these issues I regretfully leave to another occasion.

I have defended the idea that to understand how testimony works as an epistemic source we cannot take the theoretically minimalist stance: all reliable testifiers are as one, from the standpoint of epistemology. Instead I have tried to show that to offer a perspicuous characterisation of how knowledge is spread by means of testimony, when it works in its proprietary fashion, we have to look more broadly and deeply at the nature of testimony as socio-linguistic institution, and the nature of the human agents who make these intentional Gricean communicative acts, in order fully to understand the phenomenon, and get a perspicuous taxonomy. When we achieve this we come to appreciate that the class of instances of originating testimony from which knowledge can be got in the proprietary way is not the relatively broad one: all reliable testifiers, but the narrower one: testifiers who are reliable in virtue of speaking from knowledge – not RT, but KT. This narrower class picks out the good supply domain from which knowledge can be gained by the Core Mechanism. So there are epistemologically invidious distinctions to be made amongst reliable testifiers.Footnote ⁵¹