NOÛS 43:4 (2009) 581–598

Experts, Semantic and Epistemic

SANFORD GOLDBERG
Northwestern University

In this paper I argue that the tendency to defer in matters semantic is

rationalized by our reliance on the say-so of others for much of what we
know about the world. The result, I contend, is a new and distinctly epistemic
source of support for the doctrine of attitude anti-individualism.

1.
Sally is not a scientist. She depends on others for what she knows about
electrons and ion chambers, platinum and heliobacters. They tell her such
things as that electrons have spin, that ion chambers detect ionizing radiation,
that the melting point of platinum is over 1700◦ C, and that the ecological
niche of heliobacters (a bacteria that can cause digestive illnesses, including
gastritis and peptic ulcer disease) is the lower intestine.
That all of us depend for much of what we know on the testimony of others
is not news. Many recent epistemologists have used this sort of dependence to
argue for claims regarding the social nature of knowledge. Thus Wellbourne
(1986) and Kutsch (2002) have appealed to the phenomenon of testimony
to argue that there are cases in which the subject of knowledge is not a
single individual, but rather a community; Burge (1993) and Faulkner (2000)
use testimony cases to argue that the warrants enjoyed by one’s testimonial
beliefs are ‘extended’ (Burge’s term) or ‘social’ (Faulkner’s term) in that they
outstrip one’s reasons for trusting one’s source and the reliability of one’s
own cognitive processing; and Schmitt (1994) and (2006) has used testimony
cases to argue that one’s reasons for believing as one does, when one believes
through testimony, extend to include the reasons one’s source had (even as
these reasons are inaccessible to one oneself). It is a familiar (if not entirely
uncontroversial) point that testimonial knowledge is importantly social in
some sense or other.


C 2009 Wiley Periodicals, Inc.

581
582 NOÛS

The aim of this paper is not to advance a thesis of this sort. Rather,
my aim is to show that the social nature of testimonial knowledge extends
beyond epistemology proper. In particular, it will be my contention that
the fact that we pick up much of what we know about the world from the
say-so of others rationalizes a particular social practice regarding language
use—namely, the practice whereby a speaker typically defers to experts to
explicate the meanings of her own words and the application conditions of
her own concepts. This practice of semantic deference (as it is known) is
often used to motivate the doctrine of anti-individualism about language
and thought (see e.g. Burge 1979). Consequently, if my argument is sound,
it would show that the conditions on knowledge communication provide a
heretofore underexplored source of motivation for anti-individualist views in
the philosophy of mind and language.1

2.
As my argument will turn on the role experts play in the acquisition of
testimonial knowledge, it will be helpful to have a clear characterization of
the notions involved. To a first approximation, I will count H’s knowledge
that p as a case of testimonial knowledge when it is acquired through, and
depends for its status as knowledge on, H’s comprehension and acceptance
of a piece of testimony that p.2 It should be clear that most testimonial
knowledge is acquired from people who are not themselves experts on the
topic at hand—at least not if ‘expert’ is to retain its standard meaning.
Illustrations are easy to come by. I did not go to last night’s baseball game,
but come to know through your testimony that the Yankees won. This does
not make you an expert on baseball, or on the Yankees, or even on last night’s
game. To be sure, I regard your testimony as authoritative, trustworthy, and
so forth; the point is rather that authoritative and trustworthy testimony is
not the exclusive property of experts. You don’t have to excel in the theory
and practice of baseball to be in a position to report knowledgeably on the
outcome of last night’s game.
What, then, is the role—or, more accurately, the roles—experts play in
testimonial knowledge? One such role was presented above: there are cases
in which one’s testimonial knowledge depends on one’s having received tes-
timony from someone who is an expert on the topic at hand. I will call such
experts knowledge domain experts: these are people who have what we might
call ‘expertise’ in the relevant domain.3 Let ‘expertise’ designate the state of
having specialized background knowledge (or at least justified belief) in a
given domain, where the knowledge in question is organized in a manner
that allows for easy access and use in appropriate circumstances. One role
experts play in testimony cases, then, is when they testify as knowledge do-
main experts, on some topic that falls within their expertise. My claim above,
to the effect that most testimony cases do not involve experts, is simply the
 Experts, Semantic and Epistemic 583

claim that most cases of testimony involve reports of conditions that can
be discerned without need of any specialized (well-organized) background
knowledge—e.g., that the Yankees won last night.
But the role of experts in testimony cases goes beyond the cases in which
they testify on matters within their expertise. They also play an important—
if potential and implicit, and rarely manifested—role in ordinary cases of
testimony by non-experts. And it is this role that I want to dwell on in the
remainder of this paper. My claims here will be two. One is that our epistemic
reliance on our peers, in ordinary testimony cases in which the testifiers them-
selves are not experts, rationalizes our semantic deference to experts in our
community. The other is that in particular cases the semantic experts (if I can
call them that) are none other than the relevant knowledge domain experts. If
I am correct about this, then the phenomenon of knowledge communication
through speech appears to provide a new, heretofore underexplored source
of support for anti-individualist views about mind and language.

3.
My thesis is that our semantic deference to experts is itself rationalized by
our epistemic reliance on the say-so of other (expert or non-expert) speakers.
The basic motivation behind this thesis can be captured in two ideas. First,
there is a cost associated with one’s entry into the practice that enables one to
acquire a good deal of substantive and specific pieces of knowledge through
accepting the say-so of one’s peers. I will be arguing below that the cost is that
one is ‘answerable to’ the linguistic norms that make such knowledge trans-
mission practically feasible. Second, these norms themselves are provided by
the relevant knowledge domain experts. If these two ideas are correct, then
we should see our reliance on our peers for much of what we know about
the world, and our reliance on (some subset of) our peers for explicating the
meanings of our terms and the application conditions of concepts, as two
sides of a single coin. In what follows I develop this line of argument.
Consider the conditions on acquiring testimonial knowledge—the sort of
knowledge that is acquired through, and depends for its status as knowl-
edge on, one’s recognition that one has been told that p. To acquire knowl-
edge of this sort, one must be competent in recovering the content of the
testimony-constituting speech acts one observes: one can’t acquire testimo-
nial knowledge that p, unless one is reliable in comprehending the testimony
one has observed. If testimonial knowledge is (as I said above) knowledge
that depends for its status as knowledge on the hearer’s comprehension and
acceptance of the testimony, why must the comprehension be reliable (in
a sense to be identified below)? The point can be made in terms of the
epistemological truism that, in order to be knowledge, a belief must be non-
accidentally related to the truth of what is believed. Imagine a subject, S,
who forms the belief that p through a process of comprehension in which
584 NOÛS

the proposition that p is recovered through a lucky guess as to the content of

the observed testimony. Then even if p is true and the testimony S observed
(to the effect that p) was reliable, S’s coming to believe that p in this way
makes it accidental that her testimonial belief is true. After all, S could easily
have formed the belief she did, in the way she did (guessing as to the content
in question), under conditions in which the belief is false. It is for this rea-
son that believing what one guesses to be what another said is generally an
unreliable, insensitive, and unsafe way to form beliefs. So if S is to acquire
the testimonial knowledge that p, it must be the case that she has a reli-
able way to recover the propositional content of the testimony, in something
like the following sense: she would apprehend a piece of testimony as having
the content p, only if it did in fact have the content p.4
The foregoing considerations support the idea that reliable comprehen-
sion is a necessary condition on the acquisition of testimonial knowledge.
However, one might object that this necessity claim is the result of an overly-
narrow characterization of testimonial knowledge itself. The objection is as
follows. If we regard testimonial knowledge as a matter of a hearer’s coming
to know the (truth of the) very proposition attested to, then of course the
hearer has to recover the propositional content of the source testimony if she
is to acquire testimonial knowledge. But (the objection continues) this is only
because we have conceived of testimonial knowledge in an overly-restrictive
fashion. For suppose that we do not assume that testimonial knowledge is a
matter of a hearer’s coming to know the very proposition attested to. In that
case we can say that, so long as the content S takes the testimony to have
had is likely to be true given the testimony S observed, then this apprehen-
sion can underwrite the acquisition of testimonial knowledge—whether or
not the content she apprehended is the same as the content her interlocutor
attested to.
But it is not as if my characterization of testimonial knowledge—as knowl-
edge that p acquired through, and depending for its status as knowledge on,
the hearer’s comprehension and acceptance of a piece of testimony that
p—is arbitrary (or otherwise question-begging). First, this characterization,
or something like it, is widely assumed by those working on testimony.5
Second, this characterization emerges from any position which holds the
following combination of views, each of which is independently plausible:
testimonial knowledge involves testimonial belief; testimonial belief is belief
formed through the comprehension and acceptance of a piece of testimony;
and accepting a piece of testimony involves (among other things) accept-
ing the attested proposition. In addition, my characterization of testimonial
knowledge emerges from two other related features of such knowledge. One
is that, in forming a belief through accepting another’s say-so, a hearer H
is being guided in belief-fixation by how her interlocutor has (linguistically)
represented things to be: H believes that things are so because H’s interlocu-
tor told her that things are so. But this gloss makes sense only if what H
 Experts, Semantic and Epistemic 585

believes is what her interlocutor told her to be so. Relatedly, in forming a

belief through accepting another’s say-so, H is relying on her source to have
(reliably) gotten things right. But H’s reliance on this score requires that she
has recovered how her source has “gotten” things in the first place—and
this involves, at a minimum, recovering the content presented-as-true in the
testimony. And so we see that my characterization of testimonial knowledge,
and with it the thesis that reliable comprehension is a necessary condition
on testimonial knowledge, are well-motivated.
Consider now what is involved in the relevant sort of ‘reliable comprehen-
sion process’. Above I said that a comprehension process would be reliable
only if it conforms to the following specification: the process would repre-
sent a piece of observed testimony as having had the content p, only if the
testimony did have the content p. Now I do not want to go into details
regarding what this requirement comes to.6 But whatever it comes to, this
requirement must be regularly satisfied by actual hearers. If this requirement
is not regularly satisfied by actual hearers, then testimonial knowledge is a
much less prevalent phenomenon than we would pretheoretically take it to
be. The trouble is that our pretheoretic views regarding the scope of testimo-
nial knowledge are not negotiable: they are surrendered at the cost of inviting
a substantive form of skepticism. This is owed to the widely-acknowledged
role testimonial knowledge plays in the fabric of our knowledge more gener-
ally. Given this role, a failure to see the reliable comprehension condition as
regularly satisfied will raise the risk of an unacceptable sort of skepticism.7
I now want to use the claim, that the reliable comprehension condition
is regularly satisfied, to argue against a popular view about the nature of
linguistic understanding. I will call the targeted view the ‘doctrine of complete
grasp’:

CG In order to understand a speech act, one must completely grasp each of

the concepts that compose the content of the speech act.

The notion of complete grasp can be understood as follows. For any concept
C, C’s application conditions can be given in a statement of the form, ‘Nec-
essarily, C applies to an object o if and only if . . .’. A subject S completely
grasps C at a given time t just in case at t S can recognize via reflection
alone8 some correct and informative completion of this statement-form.9
Now (CG) or something like it is a doctrine that has been the object of a
good deal of familiar criticism by Tyler Burge.10 Burge’s case against (CG)
turns on considerations pertaining to the semantics of attitude-ascriptions.11
My case, however, is different. I want to argue against (CG) by appeal to
considerations pertaining to the nature of the testimonial exchange. I submit
that, together with some plausible subsidiary assumptions (to be enumerated
and defended below), (CG) is incompatible with the claim that the reliable
comprehension condition is regularly met. If so, we have the makings of a
586 NOÛS

reductio argument against (CG)—and one whose main premises concern epis-
temic considerations (regarding the conditions on testimonial knowledge).
In outline form, the argument is as follows. Suppose (1) that the doctrine
of complete grasp is true; (2) that we speak a public language;12 and (3) that,
at least when it comes to the concepts that our public language assigns to the
lexicon, complete grasp (in the sense of (CG)) is not a common phenomenon.
Given (1)-(3), it will not be a common phenomenon that we satisfy the re-
liable comprehension condition, and so will not be a common phenomenon
that we count as acquiring testimonial knowledge. But I already noted the
skeptical consequences that await anyone who undermines our pretheoretic
views regarding the extent of our testimonial knowledge. It would thus seem
that anyone who wants to avoid skepticism has a motive to reject the com-
bination of (1)-(3). My claim will be that it is (1) that ought to go.
Before we can discuss (1)-(3) we need to be clear about what a public
language is. Let a language L be an assignment of meanings to each of
the elements in a specified lexicon (the vocabulary of L). The ‘meanings’ L
assigns to its expressions may not be identical to the expressions’ semantic
values on occasions of use: we have to allow for indexicals and other seman-
tically context-sensitive expressions. But for our purposes we can ignore this
complication and suppose that the meaning of an expression is its semantic
value. We can suppose further that, for the expressions of interest here, the
meaning of an expression e will be the concept(s) that L assigns to e.13 L
is a public language, then, when there is a community of speakers whose
utterances are correctly regarded as utterances (as we say) ‘of L’.
As illustrated in Lewis (1981), it is a complicated matter to determine the
conditions for regarding speakers as producing utterances ‘of’ some common
language. Happily, we need not address this matter to make the point I wish
to make in defense of (2), the claim that speakers speak public languages. For
the point to be made in defense of the hypothesis of public languages is simply
this: if people do not speak public languages, then the process of compre-
hension would appear to involve a massive coordination problem—one that
would make it very unlikely that the reliable comprehension condition is regu-
larly satisfied. Since I have defended this claim elsewhere,14 here I will be brief.
In standard cases, comprehension proceeds in a homophonic fashion:
when the hearer H articulates what she takes speaker S to have said, H re-uses
the very same word-forms she takes S to have used.15 (Context-sensitivity
requires some adjustments on H’s behalf; but I will ignore these in what
follows.) Now, whether or not H and S speak a common, public language,
H’s homophonic comprehension is a reliable way to comprehend S’s speech
act only if the relevant portion of the semantics for H’s and S’s respective
idiolects is identical, assigning the same concepts to the same word-forms. (It
is only in that case that H’s re-use of the same word-forms will ensure that
H’s comprehension of S’s statement, as manifested in H’s re-use of the same
word-forms, does in fact correctly represent the content of S’s statement.16 )
 Experts, Semantic and Epistemic 587

Moreover, the point is perfectly general, along two different dimensions:

the semantics for the relevant portion of a speaker’s and hearer’s respective
idiolects would have to be identical for any utterance by any speaker that the
hearer could reliably comprehend in this homophonic way. Thus it would
appear that the standard sort of comprehension process is a reliable one only
if the semantics of virtually all speakers in a given language community is
identical (assigning the same concepts to the same word-forms). Such an
overlap would appear to be sufficient to regard the members of a speech
community as speaking a common language. What is more, it is arguable
that the hypothesis of a common language is necessary to explain how this
state of more-or-less total coordination between the semantics of the idiolects
among all members of a speech community is attained in the first place.
Admittedly, the need for a public language (or public standards for indi-
viduals’ idiolects) would dissipate if it could be shown how reliable compre-
hension could be attained in the absence of hearers’ reliance on homophonic
comprehension. One might think to approach this sort of project by appeal to
Davidson’s suggestions about radical interpretation. The radical interpreter,
after all, makes no prior assumptions about the meanings of others’ expres-
sions, even when the other speaker appears to be speaking the same language
as the radical interpreter herself. But even Davidson himself came to think
that ordinary cases of comprehension do not involve the sort of interpreta-
tive excess he described under the method of radical interpretation.17 What
is more, there are good reasons in support of the view Davidson came to
hold. The coordination problem that would have to be solved by a single in-
dividual, if that individual is to remain open to the acquisition of knowledge
expressed by speakers in her community, would be too great for any single
individual to solve. Not only would the hearer have to keep track of as many
different idiolects as there were speakers (potential sources of knowledge)
in her community; what is more, she would have to keep track of changes
within a single idiolect over time. The difficulty of this problem seems to be
much, much greater than the difficulty of comprehending one’s colinguals in
ordinary contexts of knowledge communication. This suggests that radical
interpretation is not the right model for comprehension in ordinary, everyday
cases.18
In sum, it would seem that only the postulation of a public language
will vindicate the reliability of the ordinary process of comprehension, and
so claim (2) would appear secure. With (2) defended, the task of defending
claim (3)—asserting the rarity of complete grasp—becomes correspondingly
easier. (3) appears contentious only against the backdrop of a theory on
which a speaker’s concepts are individuated in terms of (some subset of) the
speaker’s own beliefs and/or her inferential practices with the corresponding
word. But if a speaker’s concepts just are the public language concepts, (3)
appears downright obvious. For it would appear patent that each speaker is
such that, for many of the public language concepts in her repertoire, she
588 NOÛS

cannot fully explicate these concepts or articulate the concepts’ application

conditions; and for most of these she may not even recognize when she has
been presented with such an explication. To be sure, any speaker can offer
and accept ‘trivial’ explications, of the sort expressed by the following:

(T1) The concept DOG applies to all and only dogs.

(T2) The predicate ‘x is a dog’ applies to all and only dogs.

But these are hardly explications, as they do not articulate the application
conditions of the concept/predicate, but instead re-employ the very predicate
whose application conditions are in question.19 And once we recognize that,
when it comes to public language concepts, it is often the case that subjects
are not in a position to produce or recognize an explication of the concepts
or their application conditions, we have seen our way to recognizing that
complete grasp is rare.
We now have a dialectical case against (1), the doctrine of complete grasp
itself. Given the implausible implications that come with any position that
conjoins (1)-(3), together with the independent plausibility of both (2) and
(3), we ought to reject (1). But at this point it might be wondered what
comprehension could consist in, or how comprehension could be as deter-
minate as it is, if the doctrine of complete grasp is false. Since a failure to
answer these questions will leave us with an inclination to accept (1) despite
the troubles that we would land in for doing so, I want to address them
head-on. My claim here is that one can reliably comprehend a statement
without possessing complete grasp of the concepts composing its content, as
reliable comprehension is merely a matter of reliably representing observed
speech as having the content it in fact has.20 So long as one possesses the
relevant concepts—something that falls out of our assumption that speaker
and hearer alike speak a public language—one can reliably represent the
content of an observed utterance merely by redeploying those very concepts,
whether or not one completely grasps them.
Of course at this point we are close to a position that is familiar from the
literature on attitude anti-individualism. I will return to this in the section
following, but before doing so it is worth underscoring how we reached this
point. So far our dialectic has assumed nothing more than the following three
things: first, that testimonial knowledge is a prevalent phenomenon; second,
that the acquisition of such knowledge requires reliable comprehension of the
source testimony; and third, that complete grasp of public language concepts
is a rare phenomenon. My claim is that once these three claims are endorsed,
one is forced to accept that the sort of understanding presupposed by the
acquisition of testimonial knowledge does not, and indeed cannot (consistent
with these claims), require complete grasp. Thus the denial of complete grasp
is not a premise of my argument; it is a conclusion.
 Experts, Semantic and Epistemic 589

4.
Suppose that the foregoing argument succeeds in showing that the prevalence
of testimonial knowledge depends on a sort of understanding that stops short
of complete grasp. Even so, the relevant sort of understanding must be such
that through it hearers ascribe determinate propositional contents to the
speech acts they observe. Otherwise it is left mysterious how it is that hearers
acquire the very piece of knowledge expressed in the testimony they observed.
All of this raises a question: how can the hearer’s understanding both be
specific enough to underwrite the acquisition of the very piece of knowledge
expressed in the speaker’s testimony, and yet fall short of a complete grasp
of the concepts in the attested content? My claim will be that, in cases
of incomplete grasp, the determinacy of the hearer’s comprehension of the
content of a testimony-constituting speech act is grounded in her semantic
deference to the relevant experts.
To bring this out, we can begin with a question: in virtue of what does
the hearer count as having comprehended—and so as having come to be-
lieve (and in the favorable case, as coming to know)—the very propositional
content attested to? This question is especially pressing in cases involving
incomplete grasp. This is because, given a hearer’s incomplete grasp of the
public language concepts composing the attested content, there will always be
other—indeed, many other—hypotheses regarding the content of her com-
prehension (or what she understood to be said in the testimony). Suppose
that S tells H that arthritic knees are painful (this is the content of the
telling), but that H does not completely grasp the concept ARTHRITIS or
its derivatives. What does H take S to be saying, when H’s comprehension
is homophonic? One hypothesis—the standard one—is that H takes S to be
saying that arthritic knees are painful (where that arthritic knees are painful
is the content H takes S to be asserting). But there are others. Although
H’s understanding is homophonic, it might nevertheless be the case that H’s
understanding of these same word-forms is idiosyncratic: she might take S
to be saying that thartritic knees are painful (where thartritis is a disease of
the joints and ligaments, in the spirit of Burge 1979). Alternatively, H’s un-
derstanding might be metalinguistic: she might understand S to be asserting
that the condition known around here as “arthritic knees” is painful. And
there are various other interpretative possibilities as well.
The important point here is that H’s re-use of the very words S used, in
H’s attempt to manifest her (homophonic) understanding of S’s testimony,
underdetermines which of these hypotheses correctly represents how H has
understood S. Perhaps some of these hypotheses will be ruled out on grounds
of overall coherence with H’s other speech behavior. But it is easy to imagine
cases in which this basis is not rich enough to discriminate among several
(or more) different hypotheses. This is easy to see in cases in which the
various competing hypotheses ascribe states of comprehension to H that
590 NOÛS

differ in fine-grained ways. Yet if the case is one in which H acquires the very
knowledge S aimed to communicate, we must have a way to discriminate one
of these hypotheses—typically the standard one21 —as correctly capturing
how H has understood S’s speech. My claim is that the needed determinacy
here is provided by H’s semantic deference regarding ‘arthritic’: H takes S
to be saying that arthritic knees are painful (where this is the content of
H’s state of comprehension) in virtue of H’s deference to (some subset of
speakers in) her linguistic community.
How does semantic deference underwrite this sort of determinacy in com-
prehension? We just saw that, taken by itself, H’s homophonic compre-
hension of S’s testimony does not fix the content of the comprehension
H has attained. But H’s homophonic comprehension, together with H’s
(perhaps implicit) intention to be ‘speaking the same language as’ S, does
fix that content. (At least it does so bracketing issues of generic forms of
context-sensitivity, ambiguity, and polysymy; I will continue to disregard
these phenomena in what follows.) The need for semantic deference arises
once a hearer recognizes both that she herself cannot exhaustively articulate
the application conditions of the public language terms she is using, and
that nevertheless these terms (in her own and others’ mouths) have deter-
minate application. Given that these terms are terms of a public language,
their determinate application is given by the standards of the public lan-
guage. Because these standards articulate the terms’ application conditions
to worldly entities and properties, the standards themselves are best articu-
lated by those who are most expert in the nature of the relevant entities and
properties themselves—the relevant knowledge domain experts.22 We see, then,
that deference to experts figures centrally in the determinacy one achieves
in comprehending these terms—at least in cases involving the hearer’s own
incomplete grasp.
To be sure, semantic deference is not sufficient for determinate comprehen-
sion of an incompletely-grasped concept. A semantically deferential subject
who thinks that ‘cup’ means the number 6, and who has no non-standard
theory that would rationalize the ascription to her of the concept CUP in
the face of this idiosyncracy, is not properly regarded as having the concept
CUP (or as comprehending the English word-form ‘cup’). Some minimal
competence with the word is needed.23 What is more, it is arguable that
there are cases in which semantic deference is not necessary for determinate
comprehension either. Presumably there are cases of complete grasp. Or so
I can grant; my point is that it is semantic deference by a minimally compe-
tent speaker that ensures the determinacy in comprehension required by the
acquisition of testimonial knowledge.
Still, we might wonder whether semantic deference is the only way, in cases
involving incomplete grasp, to attain the sort of determinacy in comprehen-
sion required by the acquisition of testimonial knowledge. Although I have
no knock-down argument to show that it is, the following considerations are
 Experts, Semantic and Epistemic 591

suggestive. First, speakers can communicate very specific pieces of knowl-

edge to one another even when their respective understanding of the various
expressions differ greatly. (I take this to be patent; were this not so, testi-
mony would hardly be the pervasive source of knowledge that it is.) What
is more, comprehension routinely proceeds in an effortless manner. This is
true even under conditions in which speaker and hearer know nothing about
each other’s speech and interpretative dispositions save what is manifest in
their brief conversational exchange. Given the degree of diversity in speakers’
and hearers’ background beliefs and explicational abilities—a diversity that
is compatible with the successful communication of very specific pieces of
knowledge—it is a challenge to show how comprehension can be generally
reliable. This challenge is met on the assumption that hearers standardly
exhibit semantic deference; but it is hard to see how it could be met if this
assumption is false. Surely the burden is on those who disagree.
What should be said, then, about a hearer who incompletely grasps the
concepts involved in an attested content, but who has no tendency towards
semantic deference? Can such a hearer still posses the minimal sort of com-
petence with the concept that would enable her to comprehend the attested
content, and so to acquire testimonial knowledge of the attested content? If
a hearer with an incomplete grasp of a concept really has no tendency to-
wards semantic deference, then it is hard to see how she might be minimally
competent with respect to the concept in question. To be sure, such a subject
could employ a metalinguistic concept with precisely the same extension—as
when, instead of employing the concept ELECTRON, a subject employs the
concept WHATEVER PEOPLE AROUND HERE MEAN BY ‘electron’.
For many purposes of communication, comprehension that involves a co-
extensive metalinguistic concept of this sort will do.24 But it is hard to see
how our subject could count as minimally competent regarding ELECTRON
itself. To repeat, the difficulty is that there seems to be no grounds for picking
out a single concept, of the various concepts she might be employing, as the
concept she is employing. Had she been semantically deferential, we would
have had such grounds: she intends to be using the concept employed by her
peers. Lacking semantic deference, and failing to possess a complete grasp
of the public language concept associated with the word, a subject would
appear to give us no grounds to ascribe that concept, as opposed to others
that she might be associating with the word, to her. So it would seem that
in any case involving incomplete grasp, determinate comprehension (of the
sort that underwrites content-preserving testimonial knowledge) requires at
least some disposition to semantic deference.
The same point can be reinforced from another direction. As we saw
above, anyone who aims to acquire knowledge through accepting how an-
other speaker linguistically represents things to be must capture how her
source linguistically represents things to be. Such a task already points in the
direction of the hearer’s having a deferential intention: the hearer must intend
592 NOÛS

to be using her words as the speaker had (at least if the hearer’s comprehen-
sion was homophonic). This is already a minimal sort of semantic deference:
here one is semantically deferring to one’s source. The semantic deference of
which I spoke in the previous paragraph is the generalization of this minimal
sort of semantic deference: the hearer defers, not (or not necessarily) to the
particular speaker whose speech she just observed, but rather to those in the
best position to explicate the concepts in question. The move to this more
generalized sort of semantic deference arises along with the hearer’s (perhaps
implicit) recognition of the speaker’s perspective. Just as the hearer may not
know anything about the speech dispositions of the speaker (save what is
manifested in the brief conversational exchange), so too the speaker may not
know anything about the interpretative dispositions of her audience. Each
one needs to rely on a set of standards common to both, and indeed to any
member of their shared language community. It is only in this way that two
arbitrary members of this community can share knowledge through speech,
without knowing anything regarding each others’ speech and interpretative
dispositions (save what is manifest in their brief communicative exchange).
So it would appear that semantic deference is a natural part of any com-
munity that aims to share knowledge through speech in situations in which
speaker and hearer may have no prior knowledge of one another. A speaker
who fails to exhibit any tendency towards semantic deference would thus
seem to be one who renders herself incapable of participating in this sort
of knowledge acquisition. It might even be argued that, in the absence of
evidence to the contrary, an arbitrary member of a speech community may
be presumed to intend to participate in the practice of knowledge exchange
through speech, and so may be presumed to defer semantically to the rel-
evant experts.25 (Such a presumption would be defeasible, but it would be
positive-presumptive nevertheless.) Perhaps this is the basis of our crediting
children with comprehension, long before they have any clear disposition to
semantic deference.
As I have been presenting matters, semantic deference is generic in the
sense that hearers typically defer, not to particular individuals taken as ex-
perts, but to the relevant experts whoever they happen to be. This is appro-
priate in any situation in which the deferring individual cannot distinguish
the experts as such. But presumably there are cases where deference is to
particular individuals taken to be experts. And this raises the possibility of
cases in which different hearers defer to different experts, where the experts
themselves disagree regarding the proper explication of a key term.26 Must
we treat these hearers as comprehending the term (and the utterances involv-
ing the term) in different ways? This would be an unfortunate conclusion.
However, it need not be forced on us.
For one thing, we can distinguish between various deferential dispositions.
Thus a hearer might be disposed to defer to a particular individual, but might
have the higher-order disposition to surrender this first-order disposition in
 Experts, Semantic and Epistemic 593

certain circumstances (the expertise of the so-called expert is called into

question; the matter of diversity of expert explications is made manifest to
the hearer; and so forth). The ruling disposition—what we might call the
fundamental deferential disposition, which is the disposition that determines
how the hearer is comprehending the term in question—is the disposition
that guides the hearer’s reactions to various counterfactual scenarios (were
she to be presented with them).27
Another way to resist the conclusion that hearers who defer to different ex-
perts comprehend the term differently is to recognize that expert explication
is itself a part of a process that involves reflective equilibrium on paradigm
applications and accepted explications.28 So far I have been speaking as if it
is expert explications, and these alone, that fix the concepts comprehended
by a semantically deferential hearer. But this is an oversimplification. Experts
can get things wrong. The process by which concept individuation proceeds
is one of reflective equilibrium aiming to maximize coherence of existing
theory with paradigm applications. If no single concept emerges from such
a process, then the case might be one in which no determinate concept is
in play; or, alternatively, it may be one in which more than one determinate
concept is in play (in which case hearers deferring to different experts may
well comprehend the term differently). I suspect that there is no saying in ad-
vance which of these descriptions is appropriate. The important point is that
the mere fact of disagreement in expert explications need not always signal
disagreement in how deferring hearers comprehend the term in question.
Although the route has been circuitous, we have finally reached the doc-
trine known as anti-individualism. According to one standard formulation,
anti-individualism is the thesis that some of a subject’s propositional attitudes
depend for their individuation on features of the subject’s social or physical
environment. What we now see is that such a doctrine gets some support
from considerations pertaining to the conditions on the testimonial trans-
mission of knowledge. To acquire such knowledge, a hearer must reliably
comprehend the source speech act. In cases in which she only incompletely
grasps the concepts composing the attested content, her comprehension is
reliable only if she exhibits semantic deference; and ultimately such deference
succeeds in individuating the concepts figuring in her state of comprehension
only as part of a process of reflective equilibrium aiming to maximize the
coherence of expert explications and paradigm applications. In such cases,
the individuation of the content that the hearer took the speaker to be attest-
ing to—which is identical to the content of the belief that the hearer forms
on the basis of her acceptance of that testimony—depends on features of
her social and physical environment. Concept individuation depends on the
hearer’s social environment insofar as she is deferring to experts: there can
be cases of two hearers who are intrinsic duplicates but who differ in the
content of their comprehension, and hence in the content of their testimony-
based belief, owing to the fact that they have deferred to different experts
594 NOÛS

whose explications differ. And concept individuation depends on the hearer’s

physical environment insofar as she is deferring to experts: concept individu-
ation is the result of the process of reflective equilibrium aiming to maximize
coherence between expert explication and the paradigm applications of a
term. There can be cases of two hearers who are intrinsic duplicates but who
differ in the content of their comprehension, and hence in the content of
their testimony-based belief, owing to the fact that the objects or properties
to which the term is paradigmatically applied in the two cases differ in kind
in ways that have escaped notice of even the experts themselves.
It would thus seem that what Putnam called the “division of linguistic
labor” (Putnam 1975) is itself an essential component of the “epistemic
division of labor,” where this includes such phenomena as domain expertise
as well as other forms of epistemic authority implicated in standard testimony
cases. The picture is this: we find ourselves in need of relying on others
for what we know about the world; but if we are to acquire knowledge in
this way from the host of potential sources that make themselves and their
knowledge available to us, we need a reliable way to recover the contents
transmitted in speech; and since it is not the case that we all possess equal
knowledge of the application conditions of the concepts that constitute the
communicated contents, we find ourselves in need of relying on others in
our speech community; only in this case, this reliance is manifested in our
semantic deference to the relevant experts. In this way our epistemic reliance
on others gives rise to the need to rely semantically on others as well; and
given that the upshot of our epistemic reliance is the possibility of acquiring
lots of knowledge ‘on the cheap’, our epistemic reliance rationalizes the sort
of semantic deference on which the acquisition of such ‘knowledge on the
cheap’ depends.

5.
The central aim of this paper has been to draw connections between our
epistemic reliance on others for what we know about the world, and the sort
of semantic deference Putnam and Burge have called to our attention. If my
attempt to forge these connections is novel, its novelty lies in my attempt to
use epistemic premises, asserting the conditions on and prevalence of testi-
monial knowledge, to argue against a semantic thesis asserting the doctrine
of complete grasp. But in arguing in this fashion I hope to have cleared the
way for a more general point to be made: epistemic considerations of this
sort can be used to support anti-individualistic views about the meanings of
our words and the contents of our thoughts.
It should be clear that such an argument for anti-individualism is novel.
Standardly, anti-individualistic views in the philosophy of mind and lan-
guage are taken to be supported by considerations such as the semantics of
speech- and attitude-reports, the possibility of non-standard theorizing, or
 Experts, Semantic and Epistemic 595

the objectivity of perceptual representations. Here, however, I have suggested

that such views can be supported by appeal to the conditions on knowledge
communication: the claim is that a proper account of the semantic dimen-
sion of linguistic communication (pertaining to the hearer’s comprehension
of the source speech act), together with humdrum facts about the sorts of
circumstance under which hearers acquire a communicated piece of knowl-
edge, yield anti-individualistic results regarding linguistic meaning, speech
content, and the propositional attitudes.29 What is more, the present argu-
ment reveals that the phenomena often taken to be at the core of the case
for attitude anti-individualism—in particular, incomplete grasp and semantic
deference—can themselves be traced to the conditions on knowledge com-
munication, and in particular to our epistemic reliance on the other members
of our linguistic community.
One might doubt whether the present argument really does depend on
epistemological premises, as advertised. After all, the argument turns on
claims about comprehension, and comprehension would appear to be more
a semantic phenomenon than an epistemological one. But this reaction misses
what I regard as the key novelty of the present argument: the notion of com-
prehension itself, at least as it figures in the present argument, is thoroughly
epistemic. For one thing, the sort of comprehension in question is epistem-
ically reliable comprehension: comprehension of testimony must be reliable
on pain of not supporting the hearer’s knowledge-underwriting connection
to the fact attested to. For another, the thesis that hearers regularly attain
such comprehension is itself supported by our intuitions regarding the preva-
lence of testimonial knowledge. It is because we take testimonial knowledge
to be a pervasive feature of our cognitive lives, that we have no choice but to
regard hearers as satisfying all of the conditions on acquiring such knowl-
edge, including the condition on reliable comprehension. The hypothesis that
hearers regularly satisfy this condition is thus an inference from a prior epis-
temic premise asserting the prevalence of testimonial knowledge. Given the
pervasiveness of our epistemic reliance on others for what we know of
the world, we have little choice but to rely on others semantically as well: the
former rationalizes the latter. This point should have the effect of expanding
the discussion regarding the motivation for anti-individualism in the philos-
ophy of mind and language, to make room for considerations pertaining to
the testimonial transmission of knowledge.30

Notes
1 I say ‘underexplored’. The point I am developing here has been suggested in Burge (1999:
243), where he writes that one’s own concepts are acquired through “having comprehended
thoughts (one’s own) that were shaped and expressed through the words of others” (italics mine).
However, Burge does not develop this point at any length.
2 ‘Acceptance’ here involves the endorsement of a content on the basis of an endorsement

of the testimony. See Chapter 1 of Goldberg 2007 for further details.

596 NOÛS

3 Obviously, the utility of this notion depends on treating domains as extensive. Otherwise,

one who reports knowledgeably about the outcome of last night’s Yankees’ game would count
as a knowledge domain expert on the restricted domain of the victor of last night’s Yankees’
game.
4 The epistemology of comprehension seems to me a curious lacuna in the epistemology

of testimony literature: it is clearly relevant to a complete account of the epistemology of

testimony, yet few people have attended to it. Two noteworthy exceptions are Jack (1994) and
Fricker (2002). (I have also devoted a good deal of attention to this matter in my Goldberg
2004.)
5 See Fricker 1987 and 2006. Admittedly, theorists who work on testimony are rarely as

explicit as Fricker in making this assumption. But it is noteworthy that the vast majority of
discussions of ‘the epistemology of testimony’ focus exclusively on content-preserving cases,
where the content of the testimony is precisely the content of the knowledge the hearer acquires
through accepting the testimony.
6 See Goldberg 2004 and Chapter 2 of Goldberg 2007 for a discussion.
7 One might suppose that, in order to avoid this skeptical result, it suffices to show that

testimony is a source of knowledge, period—whether or not that knowledge is testimonial

knowledge, and so whether or not it is acquired through reliable comprehension. But elsewhere
I have argued at length that, so long as one does not appeal to considerations of comprehension,
showing what needs to be shown will present a virtually insurmountable task. See Chapter 2 of
Goldberg 2007 for details.
8 The modality in ‘can recognize via reflection alone’ is meant to be reminiscent of the

modal condition on apriority. Thus we speak of a proposition as a priori knowable to a subject

S just in case there is a method M, presently available to S, which is such that, if S comes to
know that p through M, then S knows a priori that p. Similarly, we can say that S can recognize
via reflection alone that p, where p is some correct and informative proposition of the form
‘Necessarily, C applies if and only if . . .’, just in case there is a reflective method M, presently
available to S, which is such that, if S comes to know that p through M, then S knows through
reflection alone that p.
9 A completion has to be informative, or else producing something like ‘Necessarily, DOG

applies to an object if and only if the object is a dog’ would count as manifesting a complete
grasp of the concept DOG. More on this below.
10 See especially Burge 1979 and 1986.
11 It would be more correct to say that Burge’s case against (CG) is often taken (by others)

to depend on the semantics of attitude-ascriptions. Burge himself has denied that his case is to
be understood in this way; and I have defended Burge on this score (see Goldberg 2002). But
in any case Burge has not offered a testimony-based case against (CG), which is what I am
attempting here.
12 Below I will argue that this assumption is denied at the cost of jeopardizing reliable

comprehension.
13 Concept(s): there may be more than one. (I have in mind cases of polysymy, not cases

of lexical ambiguity; the latter could be handled by regarding a single expression-form as

corresponding to two distinct lexical items.) I will ignore the complications arising from such
phenomena.
14 See Goldberg 2002 and 2007.
15 This point is emphasized by Burge (1999).
16 I speak here of “representing the content” of the source’s statement. To avoid misun-

derstanding, I repeat here something I said above: I am not assuming that the phenomenology
of testimonial belief-fixation must pass through a state of mind in which the hearer explicitly
represents the speaker as having said such-and-such. The only condition I am insisting upon
here is one of content-preservation. In particular, hearer H’s re-use of the same sentence-form S
used—whether as manifesting her (H’s) understanding of S’s testimony-constituting speech act,
 Experts, Semantic and Epistemic 597

or as expressing the belief that she (H) acquired through accepting that testimony—must “say
the same thing” (strictly speaking) as S’s use of that sentence-form in her testimony-constituting
speech act.
17 See Davidson (1999). Admittedly, it is unclear whether Davidson ever really intended

radical interpretation to be a model of everyday comprehension. Although there are suggestions

in some early papers that he did so (see e.g. Davidson 1973/1984: 125), in his (1999) he denies
ever having had such an intent.
18 This line of argument is developed at length in Goldberg 2004.
19 See Goldberg 2002 and 2006 where I discuss this sort of triviality at length.
20 Once again, this talk of “reliably representing observed speech as having the content it

in fact has” should be understood to be neutral on the phenomenology of testimonial belief-

fixation; it merely amounts to a requirement on (reliable) content-preservation. See endnote
17.
21 ‘Typically’: we should allow that there are cases in which a hearer’s comprehension of a

source speech act will be idiosyncratic.

22 Here I must acknowledge that proper names would appear to constitute an exception

to this: in many cases there are no ‘experts’ regarding the bearer of the name (but consider
names for historical figures). At the same time, perhaps it is no surprise that proper names are
exceptional: there is some debate regarding whether they should even be included as part of
the public language (such as English). In any case I hope to return in later work to discuss
the use and comprehension of sentences involving proper names in circumstances involving the
transmission of knowledge through speech.
23 Characterizing this sort of minimal competence is a very difficult matter. For some

comments on this, see Brown 2000 and Goldberg 2002.

24 Although it is deeply implausible to suppose that such metalingustic comprehension is

the norm: surely the knowledge that we acquire through accepting other’s say-so is knowledge
of the world, not of others’ language.
25 What might underwrite such a presumption? I hope to answer this on another occasion.
26 This is the semantic analogue of the sort of disagreement among experts discussed in

Goldman (2001).
27 There is a parallel here in the theory of reference, in the matter of reference-determination

in cases of ‘divided reference’ (see Wettstein 1984; and Kvart 1989, 1992, and 1994).
28 I borrow this idea from Burge (1986).
29 See Goldberg 2007.
30 I would like to thank audiences at the University of Mississippi, Bar Ilan University,

Haifa University, Hebrew University, and a special symposium at the 2006 APA Pacific Division
Meeting, where I have given versions of this paper. I would also like to thank an anonymous
referee for this journal for comments on an earlier version of this paper.

Work Cited
Burge, T. 1979: “Individualism and the Mental.” Reprinted in The Nature of Mind, D. Rosenthal,
ed. (Oxford: OUP, 1991), 536–67.
Burge, T. 1986: “Intellectual Norms and the Foundations of Mind.” Journal of Philosophy 83,
697–720.
Burge, T. 1993: “Content Preservation.” Philosophical Review 102:4, 457–488.
Burge, T. 1999: “Comprehension and Interpretation.” In Lewis Hahn, ed., The Philosophy of
Donald Davidson (Chicago: Open Court Press), 229–50.
Brown, J. 2000: “Critical Reasoning, Understanding, and Self-Knowledge.” Philosophy and
Phenomenological Research, pp. 659–77.
Davidson, D. 1973: “Radical Interpretation.” Dialectica 27, pp. 313–28. Reprinted in Davidson
1984.
598 NOÛS

Davidson, D. 1984: Inquiries into Truth and Interpretation. (Oxford: Oxford University Press.)
Davidson, D. 1999: “Reply to Tyler Burge.” In Lewis Hahn, ed., The Philosophy of Donald
Davidson (Chicago: Open Court Press), 251–54.
Faulkner, 2000: “The Social Character of Testimonial Knowledge.” Journal of Philosophy 97:11,
581–601.
Fricker, E. 1987: “The Epistemology of Testimony.” Proceedings of the Aristotelian Society,
Supplemental Vol. 61, 57–83.
Fricker, E. 2003: “Understanding and Knowledge of What is Said.” In A. Barber, ed. The
Epistemology of Language. (Oxford: Oxford University Press), 325–66.
Fricker, E. 2006: “Testimony and Epistemic Autonomy.” In J. Lackey and E. Sosa, eds., The
Epistemology of Testimony. (Oxford: Oxford University Press), 225–52.
Goldberg, S. 2007: Anti-Individualism: Mind and Language, Knowledge and Justification. Cam-
bridge University Press.
Goldberg, S. 2006: “An Anti-Individualistic Semantics for ‘Empty’ Natural Kind Terms.” Grazer
Philosophische Studien 70: 55–76.
Goldberg, S. 2004: “Radical Interpretation, Understanding, and Testimonial Transmission.”
Synthese 138:3, 387–416.
Goldberg, S. 2002: “Do anti-individualistic construals of the attitudes capture the agent’s con-
ceptions?” Noûs 36:4, pp. 597–621 (December 2002).
Goldman, A. 2001: “Experts: Which Ones Should You Trust?” Philosophy and Phenomenological
Research 63:1, 85–110.
Jack, J. 1994: “The Role of Comprehension.” In B.K. Matilal and A. Chakrabarti, eds., Knowing
from Words (Amsterdam: Kluwer Academic Publishers), 163–93.
Kutsch, M. 2002b: Knowledge By Agreement: The Programme of Communitarian Epistemology.
(Oxford: Oxford University Press.)
Kvart, I. 1989: “Divided Reference.” Midwest Studies in Philosophy 14: 140–80.
Kvart, I. 1992: “Mediated Reference and Proper Names.” Mind 102, 611–28.
Kvart, I. 1994: “A Theory of Thinker Reference.” Philosophical Studies 74:3, 291–323.
Lewis, D. 1975: “Language and Languages.” Reprinted in Martinich, A. P., ed., The Philosophy
of Language, 4th Edition (Oxford: Oxford University Press, 2001), pp. 562–581.
Putnam, H. 1975: “The Meaning of ‘Meaning’.” In Mind, Language, and Knowledge: Philosoph-
ical Papers, Volume II. (Cambridge: Cambridge University Press.)
Schmitt, F. 1994: Socializing Epistemology. (Maryland: Rowman and Littlefield).
Schmitt, F. 2006: “Testimonial Justification and Transindividual Reasons.” In Lackey, J. and
Sosa, E., eds., The Epistemology of Testimony (Oxford University Press), 193–224.
Welbourne, M. 1986: The Community of Knowledge. (Aberdeen: Aberdeen University Press.)
Wettstein, H. 1984: “How to bridge the gap between meaning and reference.” Synthese 58:1,
63–84.