On The Case for Contextualism*
Jason Stanley
The University of Michigan
DRAFT of September 9, 2002
My purpose in this paper is to evaluate the case for contextualism, the doctrine
that the relation expressed by “know” relative to a context is determined in part by the
standards of justification salient in that context. As Keith DeRose, one of its chief
proponents, writes:
...according to contextualist theories of knowledge attributions, how strong
a subject's epistemic position must be to make true a speaker's attribution of
knowledge to that subject is a flexible matter that can vary according to features
of the speaker's conversational context. Central to contextualism, then, is the
notion of (relative) strength of epistemic position.1
According to the contextualist, the relation expressed by the word "know" relative to a
context is determined in part by the degree of epistemic strength relevant in that context.
So, the contextualist allows that in some context c, the relation expressed by "know"
relative to c is such that Hannah may stand in that relation to a proposition despite only
having weak inductive evidence for the truth of that proposition. Relative to another
context c', Hannah would then not stand in the relation expressed by "know" in c' to that
proposition, because in c', the salient degree of epistemic strength is higher than the
degree of epistemic strength in c.
There are two main arguments for contextualism. The first, which I shall call the
indirect argument, is due to David Lewis. It is that contextualism is the only alternative
to fallibilism and skepticism, and that there is a good argument for rejecting fallibilism.
In the first section of this paper, I argue that the indirect argument is not compelling. The
I owe a lot of people a lot of thanks. First, discussion with Timothy Williamson led me
to the points in the first section; if one could co-author a section of a paper, that section
would be co-authored with him. Secondly, three people talked continuously with me
about all the points in this paper: my two colleagues Thomas Hofweber and Jim Joyce,
and Jeff King. Each of these people was willing to drop whatever they were doing to
help me sort out the arguments in the paper. Finally, less frequent discussions with Delia
Graff, Richard Heck, and Peter Ludlow helped at key points, for which they deserve
special thanks.
1
Section 8 of "Solving the Skeptical Problem", in The Philosophical Review 104.1
(1995): 1-52.
*
1
second, which I shall call the direct argument, is emphasized more by Stewart Cohen and
Keith DeRose. It is that there is a set of clear intuitions which only a contextualist
semantics can capture in a charitable way. In the second section of the paper, I argue that
the direct argument is compelling. The rest of the paper is devoted to arguing that,
nevertheless, we should reject contextualism about "know".
Section I. The Indirect Argument
In his paper "Elusive Knowledge", David Lewis writes, concerning the
epistemological doctrine of fallibilism:
If you claim that S knows that P, and yet you grant that S cannot eliminate a
certain possibility in which not-P, it certainly seems as if you have granted that S
does not after all know that P. To speak of fallible knowledge, of knowledge
despite uneliminated possibility of error, just sounds contradictory.
Lewis concludes that fallibilism is uncomfortable, though preferable to skepticism.
However, he believes that contextualism about knowledge allows us to "dodge the
choice" between fallibilism and skepticism. For the contextualist semantics for "know"
(and epistemic possibility) can explain the oddity of fallibilism, without landing us into
skepticism. This is Lewis's indirect argument for contextualism about "know".
However, as several commentators have argued, there are explanations of the
oddities Lewis correctly notes that are consistent with fallibilism. Here is my own
favored explanation. Following Patrick Rysiew, let us call statements of the form "S
knows that p, but it's possible that q" (where q entails not-p) concessive knowledge
attributions.2 Here are a few such cases:
(1)
2
a. I know that Harry is a zebra, but it's possible that Harry is a painted mule.
b. John knows that Harry is a zebra, but it's possible that Harry is a painted mule.
c. John knows that Harry is a zebra, but it's possible for John that Harry is a
painted mule.
"The Context-Sensitivity of Knowledge Attributions", Nous 35:4 (2001): 477-514.
2
It should be uncontroversial that all of the sentences in (1) are odd. What is controversial
is the additional claim that the oddity impugns fallibilism.
The most natural reading of "possible" in these sentences is as epistemic
possibility. The correct characterization of epistemic possibility is a vexed matter. But in
interpreting the sentences in (1), the first thing to bear in mind is that the epistemic use of
"possible" involves implicit anaphora; a proposition may be epistemically possible for
one person, but not for another. One plausible explanation for its relationality comes from
the following principle:
(Epistemic Possibility) It is possible that p is true if and only if nothing is known that
obviously (metaphysically) entails not-p.
Since knowledge requires a knower, epistemic possibility does as well. Claims of
epistemic possibility are relative to knowers. So the right statement of (Epistemic
Possibility) is:
(Epistemic Possibility*) It is possibleA that p is true if and only if what A knows does not,
in a manner that is obvious to A, entail not-p.3
Assuming (Epistemic Possibility*), and recognizing the relational nature of
epistemic possibility, allows for an explanation of the data in (1). The sentences in (1) are
equivalent to:
(2) a. I know that Harry is a zebra, but what I know does not entail, in a manner that is
obvious to me, that Harry is not a painted mule.
b. John knows that Harry is a zebra, but what I know does not entail, in a manner that
is obvious to me, that Harry is not a painted mule.
c. John knows that Harry is a zebra, but what John knows does not entail, in a manner
that is obvious to him, that Harry is not a painted mule.
Keith DeRose (in "Epistemic Possibilities", Philosophical Review Vol. C.4 (1991): 581605) has argued that it is epistemically possible that P if and only if (1) no member of the
relevant community knows that P is false and (2) there is no relevant way by which
members of the relevant community can come to know that P is false. Treating this as an
invariantist version of a Relevant Alternatives definition of epistemic possibility would
also allow me to make the points that follow.
3
3
Now that we are clearer on the contents of the sentences in (1), it is apparent that (1a) and
(1c) are nearly contradictions. In both cases, one asserts of someone that he knows
something that obviously entails that q, and that person does not know anything that, in a
manner obvious to them, entails that q.4
(2b) is more complicated. (2a) and (2c) are odd because they are assertions of
virtual contradictions. (2b), in contrast, is not. Its oddness is rather "Moore's
paradoxical". (2b) is odd because of two facts. First, "knows" is factive. Secondly, it is a
pragmatic contradiction to utter sentences like (3), where it is obvious that p entails q:
(3) p, but what I know does not entail q.
The factivity of "know" entails that asserting (2b) commits one to:
(4) Harry is a zebra, but what I know does not entail, in a manner obvious to me, that
Harry is not a painted mule.
And (4) is an instance of the Moore's paradoxical schema (3) (where it is obvious that p
entails q). Given that it is widely agreed that Moore's paradox is not semantic, but
pragmatic in nature, the oddity of (1b) therefore does not call out for a contextualist
semantic solution. The oddity of (4) (and hence (2b)) is rather due to the link between
assertion and knowledge.5
Let us now return to Lewis's claim that "If you claim that S knows that P, and yet
you grant that S cannot eliminate a certain possibility in which not-P." Any instance of
this schema will be an example of the kinds given by the sentences in (1). Furthermore,
there is no obstacle blocking the fallibilist from endorsing these explanations of the
oddity of concessive knowledge attributions. For example, consider a non-contextualist
version of Relevant Alternatives Theory, one according to which the relevant alternatives
are determined by the subject of the knowledge ascription together with the
These would be metaphysical contradictions when added to the assumption that the
entailments are obvious to the relevant parties.
5
In forthcoming work ("Assertion, Knowledge, and Context"), Keith DeRose argues that
the knowledge account of assertion, that explains this link easily, presupposes
contextualism. This is another new argument for contextualism, one that unfortunately I
cannot address in this paper.
4
4
circumstances of evaluation.6 Nothing in non-contextualist Relevant Alternatives Theory
excludes the above explanations for the oddity of the sentences in (1). Yet noncontextualist Relevant Alternatives Theory is a version of fallibilism.
There are other odd sounding concessive knowledge claims. Consider, for
example, the sentences in (5):
(5) a. I know that I have hands, but I haven't ruled out that I'm just a brain in a vat
with no body at all.
b. I know that Harry is a zebra, but I haven't excluded the possibility that Harry is a
painted mule.
These sentences certainly sound odd. But they do not overtly contain the term "possible".
So one might think that fallibilism is committed to the acceptability of these sentences.
However, as Timothy Williamson has pointed out to me, the reason for the oddity
of the sentences in (5) is similar to the explanation for the oddity of the sentences in (1).
Consider the phrase "excluded". What is it for someone to exclude a possibility?
According to Fred Dretske, "In saying that [someone] is in a position to exclude these
possibilities I mean that his evidence or justification for thinking these alternatives are
not the case must be good enough to say that he knows they are not the case."7 This is a
natural reading both of "exclude" and "rule out". But if this is the right rendering of
"exclude" and "rule out", then the sentences in (5) express:
(6) a. I know that I have hands, but I don't know that I'm not a brain in a vat with no body
at all.
b. I know that Harry is a zebra, but I don't know that Harry is not a painted mule.
Now there are some versions of fallibilism that do entail the truth (and presumably the
acceptability) of the sentences in (6). For example, as Keith DeRose has pointed out,
Robert Nozick's account of knowledge has this consequence.8 But it is certainly not the
On this view, A knows that p at <w,t> iff p is true at <w,t>, A believes that p at <w,t>,
and A's evidence eliminates all relevant alternatives to p for A at <w,t>, where the
extension of "relevant" is determined by A and the facts at w and t. I discuss the position
in more detail in Section V.
7
P. 57 of the "The Pragmatic Dimension of Knowledge", in Fred Dretske, Perception,
Knowledge, and Belief: Selected Essays (Cambridge: Cambridge University Press,
2000).
8
See Section 7 of DeRose (1995).
6
5
case that any non-contextualist fallibilist account of knowledge predicts that the
sentences in (6) can be both true and acceptable.
So, I don't find Lewis's motivation for contextualism compelling. The oddity of
concessive knowledge attributions poses no problem for fallibilist non-contextualist
accounts of "know". A better motivation for contextualism is required.
Section II. The Direct Argument
The most powerful consideration in favor of contextualism is that it explains
certain relatively clear intuitions about the truth-conditions of various knowledgeascriptions. Here is one example of the kind of argument at issue, due to Stewart Cohen
(2000, p. 95):
Mary and John are at the L.A. airport contemplating taking a certain flight to New
York. They want to know whether the flight has a layover in Chicago. They
overhear someone ask if anyone knows whether the flight makes any stops. A
passenger Smith replies, "I do. I just looked at my flight itinerary and there is a
stop in Chicago." It turns out the Mary and John have a very important business
contact they have to make at the Chicago airport. Mary says, "How reliable is that
itinerary, anyway. It could contain a misprint. They could have changed the
schedule since it was printed, etc." Mary and John agree that Smith doesn't really
know that the plane will stop in Chicago on the basis of the itinerary. They decide
to check with the airline agent.
Here is a second kind of case, of the sort discussed by Keith DeRose:
Hannah and her husband are driving home on a Friday afternoon. They plan to
stop at the bank on the way home to deposit our paychecks. But as they drive past
the bank, they notice that the lines inside are very long, as they often are on
Friday afternoons. Thinking that it isn't very important that their paychecks are
deposited right away, Hannah says "I know the bank will be open tomorrow, since
I was there just two weeks ago on Saturday morning. So we can deposit them
tomorrow morning." But then Hannah's husband reminds her that a very
important bill comes due on Monday, and that we have to have enough money in
our account to cover it. He says, "Banks do change their hours. Are you certain
that's not what is going to happen tomorrow?" Hannah concedes, uttering "I guess
I don't really know that the bank will be open tomorrow."
6
Here is a third kind of case, familiar to all of those that have taught introductory
philosophy classes:
Prof. X asks her student whether she knows that she is in a classroom, to which
she replies "Of course I know!" Prof. X responds by asking the student whether
she knows that she is not a brain in a vat. After thinking about it for a moment, the
student concludes "I guess I don't really know I'm in a classroom."
These sorts of examples support contextualism about knowledge ascriptions, because
they suggest that the propositions expressed by one and the same knowledge-attribution
may differ with respect to two different contexts of use, even though the two contexts are
identical in all respects relevant for fixing the values of obvious indexicals. Let's go
through an example in detail.
Consider a sentence like (7), as uttered by Hannah in the second situation:
(7) I know that the bank is open tomorrow morning.
Before she realizes the importance of having a bank account flush with resources by
Monday, she utters (7). What she utters expresses a proposition that seems perfectly true.
The proposition concerns a particular time, namely the next morning. She is then
informed about the pressing need for a full bank account. She then utters:
(8) I guess I don't really know that the bank is open tomorrow morning.
Again, it looks like Mary has expressed a proposition that seems perfectly true, one that
concerns the same time as the proposition expressed by her previous utterance of (7). But
(8) looks to be the denial of (7). If we take these intuitions at face-value, we obtain a
contradiction.
Here are the two basic options one has to respond to this problem:
(a) One can reject the semantic significance of one of the two intuitions. For example,
one could argue that the intuition that the proposition semantically expressed by Hannah's
utterance of (7) is true is really incorrect. Alternatively, one can argue that the intuition
that the proposition semantically expressed by Hannah's utterance of (8) is true is
incorrect; that is, the intuition is an inaccurate indicator of the actual truth-value of the
proposition semantically expressed.
7
(b) One can deny that the proposition expressed by (8) is really the denial of the
proposition expressed by (7).
Contextualism about knowledge is a version of strategy (b).
The contextualist maintains that the situation vis a vis (7) and (8) is exactly like a
situation in which Hannah utters (9) and John utters (10):
(9) I am tired.
(10) I am not tired.
There is no threat of contradiction here, because the proposition semantically expressed
by John's utterance of (10) is not the denial of the proposition semantically expressed by
Hannah's utterance of (9), despite the fact that the same words are used. For the word 'I' is
an indexical expression; it contributes potentially different semantic contents to the
semantic content of sentences containing it relative to different contexts of use. Similarly,
according to the contextualist, the word "know" is an indexical expression. Relative to
different contexts of use, it expresses different relations between persons and
propositions. So, (8) is not the denial of the proposition semantically expressed by (7), for
the same reason that (10) is not the denial of the proposition semantically expressed by
(9). The word "know" has a different content in Hannah's utterance of (7) than it does in
Hannah's utterance of (8).
The most obvious way to pursue option (a) is to appeal to a certain view about the
relation between semantics and pragmatics. According to this view, we are often wrong
in even our clearest judgements about the truth of propositions semantically expressed
relative even to perfectly clear hypothetical circumstances like the ones described above.
The reason we are wrong is that we confuse the pragmatic content conveyed by the act of
asserting a sentence, with the semantic content expressed by the sentence relative to the
context at issue. For example, one might argue that we are wrong to think that Hannah's
utterance of (7) in the envisaged context is true, because "know" expresses a relation that
holds between a person and only a very few select propositions, those for which (say) she
has deductive valid arguments from a priori premises. But knowledge-ascriptions may
pragmatically convey that the subject stands in some epistemically looser relation with
the proposition. One could then "explain" the mistaken intuition on the hypothesis that
8
we often confuse what an assertion of a sentence pragmatically conveys with the
semantic content of that sentence relative to a context.
This sort of error theory concerning our intuitions about semantic content has
achieved great popularity in philosophy of language and cognitive science over the last
twenty years. For example, some philosophers are led to it by some of their antecedent
semantic commitments concerning certain linguistic constructions. Others are led to it by
a general desire to minimize semantic explanation in favor of pragmatic explanations. Fo
example, philosophers such as Francois Recanati, Herman Cappelen, and Ernie Lepore
have argued that we are unable to distinguish our intuitions about the pragmatic content
of the use of a sentence from its semantic content, and our intuitions about semantic
content are therefore thoroughly untrustworthy. In Relevance Theory, Robyn Carston,
Dan Sperber and Deirdre Wilson have devoted their careers to arguing for largely
pragmatic explanations of apparently semantic intuitions.
So, giving pragmatic explanations of apparently semantic intuitions is very
popular. However, I share the contextualist's negative attitude towards such explanations,
both generally, and for the examples that motivate contextualism. No one has been more
explicit about the dangers of this strategy than Keith DeRose. In several of DeRose's
papers, he considers "crazed theorists" who use pragmatic explanations to defend
ludicrous accounts of meaning. For example, in one of his papers (DeRose (1999)), he
considers a crazed theorist who defends the view that "bachelor" just expresses the
property of being a man, with pragmatics, rather than semantics, explaining our intuition
that a bachelor must also be unmarried. In another paper (DeRose (1995)), he considers a
crazed account of the meaning of the term "physician", according to which it has as its
extension only those who can cure any conceivable disease, with pragmatics, rather than
semantics, explaining our "loose use" of the term.
DeRose’s point is that the tendency philosophers have to give pragmatic rather
than semantic explanations of apparently semantic intuitions threatens to undermine the
whole enterprise of giving semantic explanations. As DeRose writes (1999, p. 198),
concerning pragmatic explanations of speakers' apparently semantic intuitions about the
cases that motivate contextualism:
9
It's an instance of a general scheme that, if allowed, could be used to far too easily
explain away the counterexamples marshaled against any theory about the truthconditions of sentence forms in natural language. Whenever you face an apparent
counterexample --where your theory says that what seems false is true, or when it
says that what seems true is false-- you can very easily just ascribe the apparent
truth (falsehood) to the warranted (unwarranted) assertability of the sentence in
the circumstances problematic to your theory. If we allow such maneuvers, we'll
completely lose our ability to profitably test theories against examples.
By undermining the data for semantic theory, this kind of strategy threatens to undermine
the semantic project.
I agree with the contextualist that denying the semantic significance of apparently
semantic intuitions is a significant cost, one which we should be reluctant to bear. So the
direct argument for contextualism is a powerful motivation for it. However, I do not think
that the direct argument is ultimately successful. According to the contextualist, the
semantics of “know” is linked to a scale of degrees of epistemic strength, much like the
predicate “is tall” is linked, on the dominant treatment of comparative adjectives, to a
scale of height. In the next section, Section 3, I argue that the case for this semantic link
is very weak.
My purpose in third section is to show that, aside from the intuitions motivating
the 'direct argument', there is none of the evidence one would expect there to be, if the
contextualist semantics were correct. So, this shows that the only evidence for the
contextualist semantics is the direct argument. In the fourth section, I give three empirical
arguments against the contextualist semantics. Together, these two sections provide a
powerful circumstantial case against contextualism. However, the anti-contextualist must
still provide an account of what is going on with the intuitions that motivate the 'direct
argument'. In the fifth section, I sketch two distinct ways of capturing a large portion of
the intuitions motivating the direct argument. I conclude this section with a discussion of
the intuitions that only a contextualist semantics for "know" can capture. So, the
contextualist can give a more charitable interpretation of a small set of cases. The
remaining question is whether this is sufficient to establish contextualism.
DeRose's argument against merely pragmatic accounts of the data that support
contextualism is methodological in nature. The argument is that the methodology
employed by those who seek to give pragmatic accounts of a wide range of apparently
10
semantic intuitions undermines our ability to evaluate semantic proposals. This is one of
the most important points made within the contextualism literature, one which has not
attracted the attention it deserves within philosophy of language.9 However, it is not in
the end an objection the contextualist can give. For, as I argue in the sixth and final
section, the contextualist semantic strategy itself raises the very same methodological
worries that its proponents rightfully raise against those of its detractors that attempt to
give pragmatic accounts of the data that support it.
Section III. The prima facie plausibility of contextualism about ‘know’
Contextualists typically tell us, when introducing the thesis, that it wouldn’t be at
all surprising if predicates such as “knows that Bush is president” turned out to be
context-sensitive in the ways they describe. After all, we are told, many natural language
predicates are context-sensitive. As Stewart Cohen (1999) writes:
Many, if not most, predicates in natural language are such that the truth-value of
sentences containing them depends on contextually determined standards, e.g.
‘flat’, ‘bald’, ‘rich’, ‘happy’, ‘sad’…. These are all predicates that can be satisfied
to varying degrees and that can also be satisfied simpliciter. So, e.g., we can talk
about one surface being flatter than another and we can talk about a surface being
flat simpliciter. For predicates of this kind, context will determine the degree to
which the predicate must be satisfied in order for the predicate to apply
simpliciter. So the context will determine how flat a surface must be in order to be
flat.
There is a great deal of evidence for Cohen’s claim that there is a kind of predicate, of
which ‘flat’, ‘bald’, ‘rich’, ‘happy’, and ‘sad’ may occur as a constituent, the semantics of
which involve degrees or scales. I suppose the reason that contextualists appeal to such
predicates is so that the claim that ‘know’ is a predicate of this kind will be unsurprising.
Is it? How good is this prima facie case for contextualism about ‘know’?
The predicates of which Cohen writes –the “kind” of which Cohen speaks in his
second to last sentence, are not a motley disjunctive sort. They are predicative uses of
For discussion of this, see Jeffrey C. King and Jason Stanley, "Semantics, Pragmatics,
and the Role of Semantic Content", on http://www-personal.umich.edu/~jasoncs.
9
11
comparative adjectives. These words are almost certainly context-sensitive.10 In talking
about buildings, “is tall” may express a property that it doesn’t express when talking
about people. Comparative adjectives have also been the focus of intensive linguistic
research, especially in recent years, as Christopher Kennedy, among others, has helped to
revive degree theoretic accounts of comparatives generally.11 The main point I want to
establish in this section is that the relation between ‘know’ and predicative uses of
comparative adjectives is not sufficiently close to appeal to the latter sort of predicate to
justify context-dependence in the former.
It’s worth noting initially that it’s not even remotely clear how the appeal to
comparative adjectives is supposed to provide evidence for contextualism about ‘know’.
Words like ‘small’, ‘flat’, and ‘tall’ are adjectives; in contrast, ‘know’ is a verb. The
syntax of comparative adjectives is wildly different than the syntax of ‘know’. In a
predicate such as “is small”, the copula, as Partee has emphasized, merely expresses
functional application; the semantics of the predicate is determined by the semantics of
the adjective together with a contextually salient degree. It is a matter of some
controversy what the semantic type of the predicative use of the comparative is, and how
the contextually salient degree enters into the compositional semantics of the whole
construction. But none of the various hypotheses about the semantic function of the
predicative adjective, and its relation to the contextually salient degree or interval on the
relevant scale, seem remotely relevant to the syntax and semantics of ‘know’. In contrast,
there does seem to be hope that, say, degree theoretic accounts can provide a uniform
semantic analysis to comparative adjectives, including on their predicative uses. In short,
predicates such as ‘flat’, ‘bald’, ‘rich’, ‘happy’, and ‘sad’ do seem to form a syntactic and
semantic natural kind, one to which, prima facie, ‘know’ does not belong.
On a more detailed level, there are numerous differences between predicative uses
of comparative adjectives, on the one hand, and the verb ‘know’, on the other. Predicative
uses of comparative adjectives allow for modification, as in:
More precisely: I am convinced that they are context-sensitive. Herman Capellen and
Ernie LePore have, however, argued that they are not context sensitive, and that there is
just one property e.g. of smallness, that both buildings and people can have.
11
See his Projecting the Adjective: The Syntax and Semantics of Gradability and
Comparison, New York, Garland Press, 1997.
10
12
(11) a. That is very flat.
b. John is very tall.
c. That is very small.
In contrast, ‘know’ does not allow for this sort of modification:
(12) * John very knows that Bush is president.
Another distinction is that ‘positive’ comparative adjectives, such as (in English) ‘tall’,
‘wide’, and ‘old’, co-occur with measure phrases, as in “Five feet tall”, “two feet wide”,
and “30 years old”. There is no natural measure phrase with ‘know’, even though it
doesn’t seem to be like the ‘negative’ comparative adjective, like ‘flat’, ‘small’, and
‘young’.
Another distinction between comparative adjectives and the verb “know” is that
the former class of words are conceptually related to comparatives.12 So, for “flat”, “tall”,
and “small” we have “flatter than”, “taller than”, and “smaller than”. This is to be
expected, if underlying their use is a semantics involving degrees or intervals on a scale.
In contrast, there is no natural comparative conceptually related to ‘know’. Even the
following sort of locution is deeply strained:
(13) ??John knows that Bush is president more than Sally knows it.
If the semantics of ‘know’ did involve scales, it would be mystery why there wouldn’t be
a comparative form of ‘know’ available to exploit the scale.13
There are certain verbs that do seem to admit of degree. For example, one speaks
of the degree to which someone likes someone else, which is presumably what is
involved in the semantics of a construction such as:
This is a rather obvious distinction between "flat" and "tall, on the one hand, and
"know", on the other, but the only place I know of where it is noted is on the first page of
Fred Dretske's "The Pragmatic Dimension of Knowledge".
13
Some of my informants find “John knows N better than Bill does” acceptable, when N
is replaced by a noun, like “logic ” (‘John knows logic better than Bill does’). However,
in this construction, ‘know’ takes a noun rather than a clause. So its acceptability is
irrelevant to the evaluation of the contextualist thesis that the semantics of ‘know’, when
it takes a clause, involves scales.
12
13
(14) Hannah likes Mary more than she likes Bill.
Here, the verb ‘like’ suggests a natural scale, which smoothly allows for such
comparisons. Furthermore, ‘like’ allows for certain kinds of modifiers:
(15) Hannah likes Mary very much.
In contrast, compare (14) and (15) with:
(16) a. ?? Hannah knows that Bush is president more than she knows that Clinton was
president.
b. *Hannah knows that Bush is president very much.
c. *Hannah knows very much that Bush is president.
These facts strongly suggest that while ‘likes’ does invoke a scale that allows degree
comparisons, ‘know’ does not.
It is worth mentioning that ‘know’ can marginally occur with “very much” or less
marginally with “very well”, as in:
(17) a. ? I very much know that Bush is president.
b. I know very well that Bush is president.
However, it is doubtful that these occurrences of “very much” and “very well” are
genuine semantic modifiers of the knowing relation, rather than pragmatic indicators.14 In
this sense, these constructions are similar to:
(18) 2 is very much an even number.
Decisive evidence for this comes from several sources. First, note the unacceptability of
negating the constructions in (17):
(19) a. *I don’t know very much that Bush is president.
b. *I don’t know very well that Bush is president.
The unacceptability of the sentences in (19) contrasts with the naturalness of negating the
verb phrase in a case in which “very much” is clearly modifying the verb:
14
"Much" and "well" typically occur in complementary distributions as degree modifiers.
14
(20) I don’t like Bill very much.
Secondly, "know" is only with great awkwardness combined with "very well" in nonassertoric speech acts. Contrast the sentences in (21) with (22):
(21) a. ?? Do you know very well that Bush is president?
b. *Do you know very much that Bush is president?
(22) Do you like Bush very much?
So, the sentences in (17) are clearly not cases where the degree of knowing is operated on
by “very much” or “very well”.
Defenders of contextualism might hold that “better than” rather than “more” is the
comparative relevant to “know”, as in:
(23) Hannah knows better than anyone that she is poor.
But here again, the construction means that Hannah is familiar with the fact more than
anyone else – e.g. she lives with the consequences.15 More importantly, “better than
anyone” is idiomatic. For example, consider the oddity of:
(24) a. ?? Hannah knows better than three people that she is poor.
b. *Hannah doesn’t know better than anyone that she is poor.
So, “better than anyone” is simply an idiomatic construction, one from which we can
infer little about the semantics of “know”.16
Furthermore, none of the non-philosopher informants I asked found the following
acceptable, though they disagreed amongst themselves which was worst:
(25) a. ??John knows that Bush is president better than Mary does.
As Fred Dretske ("The Pragmatic Dimension of Knowledge", footnote 1) writes,
concerning such examples, "I take such constructions to be describing not better
knowledge, but more direct, more compelling kinds of evidence." Of course, if evidence
is just knowledge, as Timothy Williamson has argued, Dretske's explanation is less
compelling.
16
As Nunberg, Sag, and Wasow ("Idioms", Language 70: 491-538) emphasize,
idiomaticity is a matter of degree, ranging from completely frozen idioms such as “kick
the bucket” to somewhat less frozen idioms. “Knows better than anyone” is intermediate
on the scale of idiomaticity.
15
15
b. ??John knows that Bush is president better than Bill knows that Clinton is a
Democrat.
Furthermore, all of my informants reported a strong difference in acceptability between
these sentences, on the one hand, and the perfectly acceptable:
(26) a. John likes Bill more than Mary does.
b. John likes Bill more than Mary likes John.
So “better than” is not a natural way to express comparisons between levels of epistemic
position with “know”. If the semantics of “know” did involve scales of epistemic
strength, then there should be uncontroversially non-idiomatic comparisons and
modifications.
One might think that these facts about ‘know’ have syntactic rather than semantic
explanations. Perhaps sentences like (12) and (13) and those in (16) are deviant because
verbs that take sentential complements grammatically do not allow for comparisons or
intensifiers. But consider “regret”, a factive verb in the same syntactic category as
“know”:
(27) a. Hannah very much regrets that she is unemployed.
b. Hannah doesn’t regret very much that she is unemployed.
c. Hannah regrets very much that she is unemployed.
d. Hannah regrets that she is unemployed very much.
Here, the degree of regret clearly seems to be modified by “very much”. Furthermore,
“regret” easily allows for comparisons:
(28) Hannah regrets that she is unemployed more than she regrets that she is unpopular.
This shows that the lack of straightforward comparatives or degree modifiers has nothing
to do with the syntax, or even the factivity, of “know”. There are syntactically similar
expressions whose link to degrees and scales is far more plausible.
Furthermore, even in the case of verbs such as “like” and “regret” that allow for
comparisons and modifications, the case that they themselves are indexical is not
straightforward. Though Hannah might regret p more than q, in both cases there may be a
baseline level of regret that is stable across contexts. The same seems to be true of ‘likes’.
16
In neither case is there obvious need for supposing that the semantic value of “regret”
changes with a contextually salient degree of regretfulness, as there is with a predicative
use of a comparative adjective. So even with a factive clausal complement verb that,
unlike ‘know’, does seem to be correlated with a scale, it is not obvious that it is to be
treated as an indexical.
The evidence is more complicated when one considers the deverbal adjective
'known'. But even here, there does not appear to be a good case for a semantics involving
a scale of epistemic strength. This adjective, unlike its verbal relative, does give rise to
comparisons and modifications. But they are not ones of the relevant sort. So, for
example, consider:
(29) a. That broccoli is low-fat is better known than that broccoli prevents cancer.
b. That broccoli is low-fat is well known.17
(29a) does not mean that there is more evidence that broccoli prevents the flu than that
broccoli prevents cancer; rather, it means that the fact that broccoli prevents the flu is
more widely known than the fact that broccoli prevents cancer. Similarly, (29c) means,
not that there is a lot of evidence that broccoli prevents the flu, but that it is widely known
that broccoli prevents the flu. Evidence for this hypothesis comes from the fact that while
(30a) is perfectly acceptable, (30b) sounds quite odd:
(30) a. That broccoli prevents the flu is well-known, but ill-understood.
b. ?That broccoli prevents the flu is well-known, though few people know it.
This data is explicable on the assumption that the only available reading for "wellknown" is widely known. So, while the data is more complex here, the adjectival relative
of ‘know’, on the rare use of it where it expresses propositional knowledge, does not to
be an obvious candidate for analysis via degrees on a scale of epistemic strength. 18
There is a complication involved in examples with 'well', since some maintain that
'well' is polysymous between a manner reading and a 'high degree reading'.
18
Christopher Kennedy and Louise McNally, in a series of excellent papers on deverbal
adjectives, discuss 'known' in passing as a case of a gradeable adjective modified by
'well'. However, their discussion of constructions like 'well-known' focuses on the
‘acquaintance’ meaning of ‘known’, rather than its less natural propositional reading (e.g.
in Section 5.1.2 of "Scale Structure and the Semantic Typology of Gradable Predicates"
(ms.), they speak of the extension of the adjectives arguments being objects that can be
17
17
Contextualists often write as if the prima facie case for the contextual sensitivity
of ‘know’ is strong, citing the uncontroversial context-sensitivity of adjectives such as
‘tall’, ‘flat’, and ‘small’. Their purpose in so doing is to shift the burden of proof from
their shoulders to their opponents; if “many if not most predicates of natural language are
context-dependent”, then someone who claims that ‘know’ is context-dependent does not
suffer from a large burden of proof. My point in this section has been to emphasize that
these arguments do not suffice in the least to lift the burden of proof from the shoulders
of the contextualist about ‘know’. The fact that the semantic content of comparative
adjectives are sensitive to contextually salient standards is simply irrelevant to the claim
that ‘know’ has a similarly context-sensitive semantics.
In fact, we may draw a stronger conclusion from the above discussion. Natural
language expressions that are semantically linked to degrees on scales exploit this link in
a variety of recognizable ways –by allowing for comparisons between degrees on the
scale, and by allowing modifications of the contextually salient degree on the scale. If the
semantic content of ‘know’ were sensitive to contextually salient standards, and hence
linked to a scale of epistemic strength (as “tall” is linked to a scale of height), then we
should expect this link to be exploited in a host of different constructions, such as natural
comparatives. The fact that we do not see such behavior should make as at the very least
suspicious of the claim of such a semantic link. Thus, an investigation into the contextsensitivity of predicates such as “is tall”, “is small”, and their ilk may add to, rather than
remove, the burden of proof on contextualists about ‘know’.
Section IV. Some evidence against contextualism
In this section, I raise three kinds of considerations against contextualism. The
first set of examples suggests that, contra contextualism, shifting standards for
knowledge do not affect the propositional contents of assertions and beliefs. The second
and third sets of examples draws out some differences between the behavior of
uncontroversial context-sensitive expressions and "knows", differences that raise the
"partially known", in virtue of some of their parts being known). Furthermore, I find their
analysis of this particular deverbal adjective not particularly convincing, for reasons I
18
burden of proof on the contextualist's claim that "know" is semantically context-sensitive
in the manner they purport.
Contextualists such as Cohen and DeRose have been admirably clear about the
semantics they advocate for epistemic terms. According to them, the word "know" has, in
Kaplan's terminology, an unstable character. Relative to a context, "know" has different
semantic contents, which are a function of the standards salient in that context. As John
Hawthorne has emphasized in recent work, as a result, contextualism makes some pretty
clear predictions about propositional content. In what follows, I want to expand on some
of the strange predictions here.
Here is a discourse that should, if the contextualist semantics for "know" is
correct, sound perfectly reasonable (certainly, every statement in the discourse is true):
ZOO:
A (looking at a zebra in a normal zoo): I know that is a zebra.
B: But can you rule out its being a cleverly painted mule?
A: I guess I can't rule that out.
B: So you admit that you don't know that's a zebra, and so you were wrong earlier?
A: I’m not a liar! I didn't say I know that is a zebra.
A’s final utterance, according to the contextualist's semantics, is perfectly true. But this
seems a very strange result. It is extremely difficult to make sense of A's denial except as
a lie. But instead the contextualist predicts that it is clearly true.19
One way the contextualist might respond to this kind of case is by arguing that
speech reports sometime have a non-obvious "direct" use. Given this use, A's utterance of
"I didn't say I know that is a zebra" could be read as A denying that she uttered the words
"I know that is a zebra". If so, that could perhaps account for our intuition that A does lie
when she utters these words in this context. However, this reply is a non-sequitor. The
contextualist semantics for "know" straightforwardly predicts a coherent and true reading
of A's statement. The problem for the contextualist semantics is that there is no such
reading of A's statement. This reply appeals to another (alleged) reading of A's
cannot go into here. Thanks to Delia Graff for directing my attention to these papers.
19
When I related this data to Timothy Williamson, he told me that John Hawthorne had
delivered similar examples at an Oxford lecture.
19
statements, one according to which they are not true. But this simply does not address the
problem, which is the non-existence of the reading predicted by the contextualist.
Another contextualist response might be to appeal to a discourse such as the
following, containing the obvious indexical element "here". A and B are speaking to each
other on cell-phones:
CELL-PHONE
A: It's nice here. [said in her office]
B: I don't believe it's nice where you are.
A: Well, it isn't nice here. [said in the corridor]
B: So you were lying?
A: I didn't say it is nice here. But I wasn't lying earlier.
The contextualist might point out that CELL-PHONE is liable to confuse B, one of its
participants, who is unaware of A's movements. Similarly, the contextualist might
maintain that ZOO is confusing for the same reason.
Again, it is difficult to see why this reply is relevant. CELL-PHONE will clearly
be bewildering to B, if she is unaware of A's movements. But there is nothing
bewildering about it to a third party informed of all of the relevant facts. In contrast, ZOO
is thoroughly bewildering to anyone. No ordinary speaker not familiar both with the
details of Kaplan's theory of indexicals, and the proposed contextualist semantics for
"know", can make head or tails of ZOO. In contrast, any ordinary speaker can make
straightforward sense of CELL-PHONE. The problem for the contextualist is that there
should be a reading of ZOO where ordinary speakers can recognize a true reading of A's
final statements, as they clearly can with CELL-PHONE. And this response does not
address this problem.
As John Hawthorne has recently emphasized in a somewhat more theoretical vein
in unpublished work, the problem for contextualist semantics also of course affects
belief-ascription.20 The point can again be made by considering a discourse each
utterance of which the contextualist predicts can be true:
"Contextualism in Epistemology", manuscript delivered to the Rutgers Semantics
Conference.
20
20
ZOO-AGAIN
Bill: That's a zebra.
Hannah (sometime later): Bill knows that that animal is a zebra.
Sarah: Hannah believes that Bill knows that that animal is a zebra.
Frank (sometime later, to Sarah): You know, Bill hasn't ruled out the possibility that that
animal is a pained mule. So Hannah has a false belief.
Sarah: No, I know Hannah quite well. She currently has no false beliefs. For example,
Hannah doesn't have the false belief that Bill knows that that animal is a zebra. And I
never said she did.
According to contextualism, all the utterances made in these discourses can be true. In
particular, Sarah has not contradicted herself. But ordinary speakers, when presented with
this discourse, immediately react with the intuition that Sarah has straightforwardly
contradicted herself.
Contextualism about "knows" also makes some strange predictions about
discourse involving propositional anaphora. Consider the following discourse:
(31) I know that I have hands. But come to think of it, I might be a brain in a vat, in
which case I would believe I have hands, but wouldn’t. Now that I’m considering such a
skeptical possibility seriously, I don't know that I have hands. But what I said earlier is
still true.
According to the contextualist semantics for "know", there should be a clear true reading
of all the sentences in this discourse. But the discourse is very difficult to get one's head
around. In particular, the only interpretation for the expression "what I said" in the final
sentence is the proposition whose truth is being denied in the last sentence. The case is
again markedly different with genuine indexicals:
(32) It's raining here. Had I been inside, that still would have been true. But now that I am
in fact inside, it's not raining here.
When informed of the facts, there is a clear reading of all of the sentences in (32) where
they are true.
In sum, the empirical evidence strongly suggests that shifting epistemic standards
do not affect the propositional content of knowledge-ascriptions. Now, I want to turn to
another argument, albeit one that is admittedly more theoretical in character. The
21
existence of natural language words that express operators on characters, in the sense of
David Kaplan, is controversial. Famously, David Kaplan originally claimed that there
were no such words -- no "monsters", in his vocabulary. But there are certain expressions
that are very good candidates for monsters. That is, there are certain expressions that
clearly give rise to "monster like" readings. Perhaps the clearest example are certain
kinds of modals, such as epistemic modals (see Israel and Perry (1994)). Consider:
(33) a. Why are you opening the door so quickly? For all you know, I am (might be) a
murderer.
b. For all you know, that is Bill Clinton. [pointing at a difficult to discern figure]
c. You have to hurry -- for all you know, it is already noon.
In all of these cases, the effect of "For all you know" is to generate a reading that is
equivalent to the result of evaluating the character of the embedded sentence with respect
to a context. Taking epistemic modals to be quantifiers over contexts, a sentence like
(33b) seems to express the same thing as "There is an epistemically possible context with
respect to which the character of 'That is Bill Clinton' expresses a true proposition."
On this treatment, epistemic modals exploit ignorance about features of the
context. This ignorance is treated as ignorance about what context we now occupy. This
treatment of epistemic modals seems to accord with the contextualist's explanation of the
skeptical problematic, which exploits the thesis that we are often in error about which
context we occupy.21
If "know" is an indexical expression, like "I" or "now", or a demonstrative like
"that", one would expect sentences containing "know", when embedded under apparent
"monster" like operators, to exhibit similar behavior. But consider:
(34) For all you know, Frank knows that p.
There are ways to generate these readings without treating the modal operators as
operators on character. Geoffrey Nunberg (ref.) treats these as cases of "descriptive uses"
of indexicals (and I suppose demonstratives). His idea is that the indexicals and
demonstratives in these sentences have as their semantic contents, the properties that are
their characters. If so, then one could account for the data without treating the modal
operators as operators on characters. The theoretical explanation that one prefers for this
data is irrelevant for the argument to follow.
21
22
If "know" were an indexical, (34) should have a reading where it can be said in a
situation of the following sort. We are perfectly aware that p is true, and that John has
evidence for his belief that p. What I believe you are unsure of whether we are in a
context with loose enough standards to count as one in which John stands to the
proposition that p in the relation picked out relative to that context by the character of
"know". I should be able to utter (34) in such a situation. But such a "monster like"
readings of (34) is, to my ears, not available.
Let's consider a particular case to sharpen intuitions. Suppose that John is a
litigator in a courtroom. Hannah is on the witness stand, testifying in a trial in which Bill
is accused of murder. Hannah is reporting about what she knows about Frank's evidence
for John's guilt. Hannah and John both know that Frank believes that Bill is guilty.
Hannah has already asserted that Frank saw Bill go into the victim's house right before
the time of the murder, and emerge soon afterwards. Hannah herself firmly believes that
Bill is guilty, on the basis of strong evidence. She is aware that John is also firmly
convinced of Bill’s guilt. But she confesses that she is unaware of the epistemic standards
relevant for certainty in the courtroom, so she doesn't feel she can speak for Frank, given
what she knows about his evidence. So John utters:
(35) So, it follows that for all you know, Frank knows that Bill is guilty.
In this context, we should obtain a fairly clear reading of (35), where it means that it is
epistemically possible for Hannah that the standards of knowledge in a courtroom are
sufficiently low, that given what she knows about Frank's evidence, the relation
expressed by "know" relative to that context is one that Frank bears to the proposition
that Bill is guilty.
I think it is very difficult to obtain this reading of (35). In contrast, it is with
complete ease that we obtain the parallel readings of (33). This provides more evidence,
albeit circumstantial, that "know" behaves quite differently than obvious indexicals.22
It is unclear whether there are “monster” like effects on predicative uses of comparative
adjectives. Suppose Hannah and Sarah are driving along for the first time in Ithaca, New
York in mid-winter, an area they both worry is too mountainous in the snow for their
1982 Nissan Pulsar. They are driving into Ithaca on a somewhat undulating road, one
22
23
Here is a third worry with the contextualist semantics for "know". Contextualists
often speak as if there is one contextual standard in a context for all context-sensitive
expressions in a discourse. So, they tend to speak of "how high or low the standards for
knowledge are set" in a context (DeRose (1995, Section 8)), and "how high or low the
standards for tallness are set" in a context (Ibid.). But this is not in fact a good description
of how context-sensitive expressions tend to work. Rather, the context-sensitivity is
usually linked to the term itself, rather than the whole discourse.
For example, suppose John, who is very small for his age, identifies with small
things. He has a picture on the wall in his bedroom of an elephant fighting off a much
larger elephant. He also has a framed tiny butterfly on his wall. When he is asked why he
has both things hung up, he says:
(36) That butterfly is small, and so is that elephant.
John in fact also has a fondness for flat things. On his wall is a picture of a field in
Kansas, and on his desk is a rock. When asked why he has both, he replies:
(37) That field is flat, and so is this rock.
Now imagine picture of a butterfly that’s surrounded by much smaller butterflies; it’s
huge for a butterfly. It’s next to a picture of an elephant that’s surrounded by much larger
elephants. The following is a good description of the situation:
which has a few sporadic small hills. Hannah says, “Hey, this isn’t so bad!”, to which
Sarah replies with a grimace, “For all you know, this is flat.” Perhaps it is possible to hear
Sarah as expressing a proposition with the same informational content as that expressed
by “There is an epistemically possible context with respect to which the character of
‘This is flat’ expresses a true proposition”, where one of the relevant contextual features
is a standard of flatness. But it is certainly not easy to obtain this reading. However, it is
unclear whether this vindicates the contextualist about ‘know’ from the considerations
concerning epistemic modals. For the reasons emphasized in the first section, the
contextualist cannot be maintaining that ‘know’ is similar to a predicative use of a
comparative adjective. She is much better advised to model ‘know’ on indexicals like ‘I’,
‘here’, and ‘now’. But then the lack of a ‘monster’ like reading is worrisome for the
contextualist.
24
(38) That butterfly is large, but that elephant isn’t.23
We can also get the contextual shift with context-sensitive determiners, such as "many",
as in:
(39) In Syracuse, there are many serial killers and many unemployed men.
In this case, the contextual determinants for the denotation of ‘many’ change within a
clause. Famously, of course, the same phenomenon occurs with quantified expressions,
as (40) can express the proposition that every sailor on one ship waved to every sailor on
another:
(40) Every sailor waved to every sailor.
It is no surprise that context-sensitive expressions allow for standard-shifts within
a clause. In each case, the context-sensitivity is linked not to the discourse, but to the
particular context-sensitive term. So what one is speaking about when one speaks of the
"standard of tallness" relevant for evaluating a particular use of "is tall" is simply the
degree of tallness that is associated with the expression "tall" by whatever semantic
mechanism one exploits.
If "know" is a context-sensitive expression, then we would expect precisely
similar phenomena. That is, we would expect it to be smoothly acceptable to associate
different standards of knowledge with different occurrences of "know", just as we
associate different degrees of height with different occurrences of "tall". This should be
unproblematic, as it is with all the other context-sensitive expressions. So, if
contextualism were true, we should expect the following to be fine:
(41) a. Bill knows that he has hands, but Bill does not know that he is not a bodiless brain
in a vat.
b. Bill does not know that he is not a bodiless brain in a vat, but Bill knows he has
Similar examples have in fact been in the semantics literature for a long time. Ewan
Klein, (p. 16 of "A Semantics for Positive and Comparative Adjectives", Linguistics and
Philosophy 4: 1-45) gives the example "This is comfortable, and that is too", pointing
first at a chair, and then at a sofa. Many other such examples occur in Peter Ludlow's
reply to Klein, "Implicit Comparison Classes", Linguistics and Philosophy 1989 12: 519533.
23
25
hands.
Of course, both of these are not fine -- they are, as contextualists themselves have taken
great pains to argue, of dubious acceptability. For example, Keith DeRose rejects Robert
Nozick's account of skepticism on the very grounds that it allows for acceptable
utterances of these sentences, which DeRose calls "Abominable Conjunctions" (DeRose,
Ibid., Section 7). But if contextualism were true, then it would be mysterious why the
sentences in (41) wouldn't be smoothly acceptable, as they are in the case of other
context-sensitive expressions.24
Of course, contextualists have expended a great deal of effort trying to explain
the oddity of the second sentence in (41), (41b). Once a skeptical possibility has been
raised, they say, that has ramifications for the evaluation of future uses of 'know' within
that discourse. But there are two worries for this strategy. First, it leaves the oddity of
(41a) unexplained. Secondly, it proposes a certain pragmatic constraint about the contextsensitivity of "know" that simply has no parallel with pragmatic principles governing the
interpretations of other context-sensitive expression. Thus, these kinds of claims about
how raising skeptical scenarios change the discourse look like stipulations to save the
theory from unattractive consequences such as (41b).25
One response the contextualist may give to this worry is to accept the conclusion
that (41a), say, is acceptable.26 I do not view this as a significant concession. But it is not
clear that the contextualist is in a position to adopt it. For, as DeRose has argued, the
contextualist's case that their closure preserving solution is superior to solutions that deny
closure rests upon their claim that sentences like (41a) are unacceptable. So, conceding
Some of the obvious indexicals don't allow for easy context-shifts across a conjunction.
For example, it's hard to shift the reference of 'I' across a conjunction. But this is due to
the brute metaphysical fact that it's odd for two different people to partake in uttering two
conjuncts of a conjunction. It has nothing to do with the semantics of 'I'.
24
Contextualists starting from David Lewis's original exposition in Example 6 of
"Scorekeeping in a Language Game" (Journal of Philosophical Logic 8 (1979)) have
been addressing this problem. At least Lewis there notes that it is a somewhat mysterious
feature of the account.
26
I am grateful here to discussion with Jim Joyce.
25
26
the acceptability of (41a) robs the contextualist of a powerful advantage they have
claimed their view to possess over some of its rivals.27
Section V. On the Direct Argument for contextualism about “know”
We began by emphasized the large burden of proof on the contextualist about
‘knows’, and in the last section, we added to this by a range of data that suggests that the
content of assertions and beliefs about knowledge are not affected by shifting standards
in a context. The purpose of this section is to consider in detail to what degree the
intuitions with which we began this paper in fact provide support for contextualism about
“know”.
I begin this section by reviewing one of the examples that is supposed to motivate
contextualism about “know”, and how the contextualist semantics is supposed to help
explain the intuitions about the case in question. I will then argue that the contextualist
semantics does not in fact provide a very satisfactory explanation of some of the core
data, and provide two distinct explanations for some of the core data. I conclude by
discussing cases that only a contextualist semantics can charitably capture and argue that
even here, the contextualist victory is somewhat hollow.
Let us begin with the following version of the bank case:
Hannah and her husband are driving home on a Friday afternoon. They plan to
stop at the bank on the way home to deposit our paychecks. But as they drive past
the bank, they notice that the lines inside are very long, as they often are on
Friday afternoons. Thinking that it isn't very important that their paychecks are
deposited right away, Hannah says "I know the bank will be open tomorrow, since
I was there just two weeks ago on Saturday morning. So we can deposit them
Here is another response possibility, again suggested to me by Jim Joyce. In Lewis's
"Elusive Knowledge", he considers two epistemologists on a bush-walk, discussing
skepticism. They still seem to know where they are going, despite their discussion of
skeptical possibilities. According to Lewis, this is because they have
"compartmentalized" their minds. Similarly, one might say that uses of (41a) involve
"compartmentalization". But the point originally was that, as we see from the examples in
(36)-(40), in the case of every other context-sensitive expression, one can have standard
shifts within a clause. It would be absurd to speak of "compartmentalization" in these
cases. Similarly, one expects a smooth, non-contradictory reading of (40a). One cannot
respond to this kind of argument by appealing to compartmentalized thinkers.
27
27
tomorrow morning." But then Hannah's husband reminds her that a very
important bill comes due on Monday, and that we have to have enough money in
our account to cover it. He says, "Banks do change their hours. Are you certain
that's not what is going to happen tomorrow?" Hannah, uttering "I guess I don't
really know that the bank will be open tomorrow."
The contextualist semantics for “know” provides an elegant explanation of this discourse,
consistent with the principle of charity. Both Hannah’s initial utterance of “I know the
bank will be open tomorrow” and her utterance of “I don’t really know that the bank is
open” are true. Once standards have been raised, the relation expressed by “know” in the
final context is not one in which Hannah stands to the proposition that the bank is open.
One salient feature of these kind of cases is that they are most natural when the
denial of knowledge occurs with a (preferably overtly stressed) occurrence of “really”.
The contextualist semantics yields a smooth explanation of the function of "really". In
these cases, “really” functions as a degree modifier. The semantic function of degree
modifiers is to raise the degree on the relevant scale for an object to satisfy the predicate
to which it is appended.28 So, for example, consider:
(42) a. This soup is really hot.
b. Kansas is really flat.
c. Yao Ming is really tall.
In these examples, the terms “hot”, “flat”, and “tall” are associated with contextually
salient degrees of temperature, bumpiness, and height respectively. A predicate such as
“is tall”, in context, is satisfied by those objects that have at least that degree of height.29
The function of “really” in these constructions is to raise (or lower) the contextually
salient degree on the relevant scale. So, an object satisfies “really hot” just in case that
object has at least that degree of temperature to which the degree modifier “really” shifts
the degree of temperature to count as satisfying “hot” upwards.
It is a little difficult to speak of the semantic function of a degree modifier in such
abstract terms – it would be simplest to fix on a particular semantic treatment of degree
and measure constructions, and explain the semantic function of degree modifiers on that
theory. But this clearer route would involve us in too many irrelevant complications.
29
In some degree-theoretic treatments, the adjective itself is taken to express a measure
function, a function from objects to degrees on the relevant scale. Syntactic and semantic
28
28
The contextualist can explain the occurrence of “really” in the kind of cases that
provide evidence for her semantics in an exactly similar manner. These occurrences of
“really” are degree modifiers. The semantic function of “really” in Hannah’s utterance in
the bank case is to raise the contextually salient degree of epistemic strength that
someone must have to count as satisfying the predicate “know that the bank is open” in
that context. Indeed, this explanation is probably forced on the contextualist. For,
according to the contextualist, this is exactly what has happened in the bank case. The
degree of epistemic strength has been raised.
So, this is how the contextualist explains the above kind of case. Initially, the
sentence "Hannah knows that the bank is open" expresses a true proposition. But then the
epistemic standard is raised, either directly by context, or via the use of the degree
modifier "really", the semantic function of which is to raise the standard on the relevant
scale for an object to satisfy the predicate in question (in this case "knows that the bank is
open"). This is an attractive explanation of this data. However, upon closer inspection, it
loses much of its plausibility.
On the contextualist semantics, the function of "really" is as a degree modifier
when it occurs in the above kind of cases. However, there is a good case against “really”
being a degree modifier in these examples. When “really” has a genuine degree modifier
use, it is perfectly acceptable to conjoin it with the unmodified form, as in:
(43) a. Michigan is flat, but not really flat.
b. John is tall, but not really tall.
In contrast, the following is certainly deviant:
(44) * I know that the bank is open, but I don’t really know that the bank is open.
If “really” were a genuine degree modifier when it occurs with “know” in the discourses
we are considering, sentences such as (44) should be perfectly acceptable. The fact that
such sentences are not acceptable is therefore good evidence that such uses of “really” are
not degree modifiers.
machinations then are required to allow the predicative use of the adjective to have the
interpretation described in the text. Again, pursuing this point involves irrelevant detail.
29
Of course, the contextualist about “know” could mount a pragmatic account of the
apparent semantic deviance of (44). But any such defense would have to be consistent
with the smooth acceptability of the sentences in (43).
I now want to describe two different non-contextualist explanations for the bankcase data, ones that seem more consistent with the function of "really" in these examples.
Both accounts of the above bank-type cases allow Hannah to be speaking truly both when
she originally utters "I know the bank is open", and when she utters "I guess don't really
know that the bank is open". But they do so in very different ways.
Here is the first explanation. "Really" seems, in these constructions, to be very
similar to an expression like "strictly speaking". Evidence for this is borne out by the
following similarities:
(45) a.* It's hot outside, but strictly speaking, it's not hot outside.
b. * It's flat in Michigan, but strictly speaking, it's not flat.
c. * I know the bank is open, but strictly speaking, I don't know the bank is open.
If "really", when it occurs with "know", is in the same category with "strictly speaking",
then the fact that sentences such as (44) are infelicitous is explicable.
In the cases that motivate contextualism about "know", one typically finds not just
"really", but occurrences of negation together with "really", as in:
(46) a. I don't really know that the bank is open.
b. Smith doesn't really know that the plane is going to Chicago.
We have already seen that "really" is not a degree modifier in these cases. The question
now is what the function of the negation together with "really" is in these cases. One this
first explanation of this kind of bank-case, the hypothesis about the function of these
expressions as they occur in the kind of cases that motivate contextualism is that they are
instances of "metalinguistic negation", similar to those found in:
(47) a. John is not really happy, he's ecstatic.
b. Yao Ming isn't really tall, he's huge.
c. Strictly speaking, she's not smart, she's brilliant.
In these cases, "not really" functions as a kind of metalinguistic operator, one that
comments on the appropriateness of the speech act. Similarly, the negation together with
30
"really" in (44) comments, not on the truth of the knowledge claim, but the felicity of its
assertion in that context.30
Note that these 'metalinguistic' uses of negation (with 'really') do not entail the
falsity of the content of the sentences in which they occur. That is, (47b) does not entail
that Yao Ming isn't tall, and (47a) doesn't entail that John isn't happy. Similarly, the
corresponding uses of "not really" in the cases that motivate contextualism do not entail
that the knowledge-ascription is false. They merely state that the knowledge ascription is
infelicitous in the context at issue. So, one can explain these sorts of examples that
motivate contextualism, without endorsing contextualism.
This first explanation predicts that Hannah speaks truly both in her original
utterance of "I know that the bank is open", and in her later utterance "I guess I don't
really know that the bank is open." For her second utterance is true just in case an
utterance of "I know that the bank is open" is for some reason infelicitous in the second
context. Thus, this explanation yields a truth-conditional account of the data, albeit via
the hypothesis that these cases involve non-truth-conditional felicity operators.
However, the case for this first explanation is incomplete. That is, it is not clear
that the denials of knowledge in this class of cases that motivate contextualism are
instances of metalinguistic negation, as in (47). In the sentences in (46), the reason one
asserts the infelicity of uttering the sentence in question is because to do so would violate
the maxim of quantity in some egregious manner. No precisely parallel explanation for
the infelicity of asserting the knowledge-claim in the higher standards context is
available. To complete the case, one would need to explain, in Gricean terms (preferably
in terms of the maxim of quantity), the nature of the infelicity of the act of asserting the
knowledge-claim.
The second explanation of these cases exploits a non-contextualist version of
Relevant Alternatives Theory. Consider, for example, the following such account of the
semantic value of "know" ('x', 'w', 't', 'p' are schematic letters replaceable by names of
persons, worlds, times, and propositions, respectively):
Paul Kay ("Linguistic Competence and Folk Theories of Language: Two English
Hedges") has argued that expressions such as "Loosely speaking" and "technically" (and
presumably also "strictly speaking") are what he calls hedges, metalinguistic expressions
operating on some feature of the speech act, like metalinguistic negation.
30
31
[knows] = that relation R such that R(<x, w, t, p>) if and only (1) p is true at w and t, (2)
x believes that p at w and t, and (3) x has ruled out the p-alternatives that are relevant for
x at w and t.
This theory of course needs to be augmented with an account of how the subject of the
knowledge attribution and the facts at the world and time of evaluation determine the
relevant alternatives.31 But let us ignore that incompleteness for now, and focus merely
on how one could exploit the semantic framework to explain some of the data that
motivates contextualism.32
It is crucial, in understanding the distinction between contextualist and noncontextualist versions of Relevant Alternatives Theory, to keep clear about the distinction
between a sentence relative to a context, and a proposition relative to circumstances of
evaluation. In contextualist Relevant Alternatives Theory, the context of use plays a
semantic role, in helping to determine the semantic content of the "know" relative to a
context. In non-contextualist Relevant Alternatives Theories, in contrast, the word
"know" has the same semantic content with respect to every context, namely the univocal
knowledge relation. Propositions containing the univocal knowledge relation may have
different truth-values with respect to different circumstances of evaluation, because the
facts about which alternatives are relevant for the putative knower differ in these different
circumstances.33
Consider the above description of the bank case. Hannah's first utterance of "I
know the bank is open tomorrow" is made at a certain world-time pair, call it <w,t>. This
For example, Rysiew (2001, p. 488) proposes that "...the relevant alternatives are fixed
by what we (normal) humans take to be the likely counter-possibilities to what the subject
is said to know [in the circumstances in question]." Of course, this is only the beginning
of an analysis of "relevant". This shifts the burden from explaining what facts about a
circumstance and a putative knower determine what alternatives are relevant to
explaining what "normal" humans take to be likely counter-possibilities (presumably,
"normal" is not meant in just a statistical sense!). But at least it's a start.
32
Strangely, Rysiew (2001) seems not to notice that the version of non-contextualist
relevant alternatives theory he defends has the resources to give a semantic account of
some of the data that motivates contextualism.
33
An instructive parallel is to be found in Delia Graff's notion of an "interest relative
property" in her paper on vagueness, "Shifting Sands: An Interest Relative Theory of
Vagueness" (Philosophical Topics (2000), 28.1: 45-81).
31
32
pair is not just the context of use for her utterance, but also the circumstance of
evaluation with respect to which we evaluate the proposition she expresses. Presumably,
the relevant alternatives for Hannah to the proposition that the bank is open at w and t do
not include the possibility that the bank has changed its hours. But as the conversation
proceeds (and time passes), the relevant alternatives change. By the time Hannah utters "I
guess I don't really know that the bank is open tomorrow", it is <w,t'>. And at <w,t'>, the
possibility that the bank has changed its hours has become relevant. So, the noncontextualist version of Relevant Alternatives can give a truth-conditional explanation of
this kind of case, without context-dependency. The quadruple <Hannah, w, t, that the
bank is open> is in the extension of the knowledge-relation, while the quadruple
<Hannah, w, t', that the bank is open> is not in the extension of that relation.
In other words, this version of non-contextualist Relevant Alternatives Theory can
give a semantic explanation of many of the cases that motivate contextualism.
Furthermore, as my colleague Thomas Hofweber has emphasized in unpublished work,
non-contextualist Relevant Alternatives theory can also be presented in a form in which it
preserves closure. For example, relative to the same circumstance of evaluation, it
follows on this semantics for knowledge-ascriptions that if x knows x has hands, then x
knows that x is not a brain in a vat. So the desire to preserve closure does not choose
between contextualist and non-contextualist versions of Relevant Alternatives Theory.
A full cost-benefit comparison of contextualist and non-contextualist versions of
Relevant Alternatives Theory simply requires a separate paper. There are too many issues
of interest to pursue. But I cannot resist one brief remark about the choice.
One cost of the non-contextualist version of Relevant Alternatives Theory is that
it allows for the felicitous (and potentially true) utterance, in Bank-case situations, of
sentences like (48):
(48) I used to know that the bank is open, but now that you've raised that concern, I don't
know that the bank is open.
33
Contextualist versions of Relevant Alternatives Theory do not have this uncomfortable
consequence.34 However, as I emphasized above (Section IV, example (31)) there is a
precisely similar uncomfortable consequence facing contextualism:
(49) A: I know that the bank is open.
B.: But maybe the bank has changed its hours. Have you ruled that out?
A: I haven't. Now that you bring that up, I don't know that the bank is open. But
what I previously said is still true.
(49) is a potentially felicitous and true discourse, according to the contextualist, because
A's last use of 'what I said' denotes the proposition A expressed by A's first utterance.
According to the contextualist, this proposition remains true with respect to the time and
world of evaluation relevant for the last part of A's discourse. In short, examples such as
(48) do not obviously bear on the choice between contextualist and non-contextualist
versions of Relevant Alternatives theory, because there are similarly uncomfortable
consequences of both accounts of "know".35
So, we have canvassed two possible explanations for the data that motivates
contextualism. According to the first, a felicity operator, a "hedge" in the sense of Paul
Kay, contributes to truth-conditions of sentences containing it in such a way that the
apparent denial of an embedded sentence expressing a true proposition in fact is a denial
of the felicity of uttering that sentence in that context. According to the second
explanation, one version of a non-contextualist relevant alternatives theory, the change in
truth-conditions is due, not to a change in propositional content, but to a change in the
circumstances of evaluation, and a concomitant change in the relevant alternatives.
But there are certain cases that motivate contextualism that no alternative
explanation, including the two just canvassed, can account for in a way that respects the
See Keith DeRose, "Now You Know it, Now You Don't", in Proceedings of the
Twentieth World Congress of Philosophy (Bowling Green, Ohio: Philosophy
Documentation Center, 1999, Volume 5).
35
Interestingly, whether non-contextualist relevant alternatives theory of the sort that I
presented in the text entails that (49) could be true depends upon whether propositions are
considered as temporalist or eternalist. If propositions are eternalist, then this semantics
does, like the contextualist version, have the potential truth of (49) as a consequence. The
fact that the non-contextualist version has (48) as a consequence, as well as either
34
34
truth-conditional intuitions of all participants. These are cases in which two agents
unaware of each other, when presented with a certain situation, have opposing intuitions
about the truth of a knowledge claim about a person in that situation, solely because the
two agents have different concerns. The classic case is, of course, discussions of
skepticism in a philosophy classroom. The philosophy students in the class conclude that
Joe --an ordinary person walking by on the street, viewed at from the window-- does not
know he has hands. However, Joe's friend Bill, who is not in the philosophy classroom,
and has the same knowledge of Joe's evidence that he has hands as the members of the
philosophy classroom, concludes that Joe does know he has hands. According to the
contextualist, both the philosophy students are correct, and Bill is correct. Only the
contextualist can give such a charitable resolution of apparently conflicting intuitions.
So the benefit of the contextualist's explanation of the data is that it preserves the
truth of our intuitions in a small set of cases. But how charitable, in the end, is the
contextualist's account? If, to preserve these intuitions in a charitable manner, it must
impute to ordinary speakers many errors in judgement, then the benefit of giving a
charitable explanation of these cases of apparently conflicting intuitions is outweighed by
the cost of the error theory imputed to ordinary speakers.
The contextualist touts as a merit of her view that, in contrast to pragmatic
accounts of the data, it preserves intuitions about the semantic content of sentences
relative to contexts as reliable guides to their actual semantic content. Yet, as Stephen
Schiffer has argued, contextualism itself involves a certain error theory about our
intuitions about semantic content.36 Indeed, for the contextualist, our intuitions about the
semantic content of sentences relative to contexts can be just as misleading as advocates
of pure pragmatic strategies hold. And as Richard Feldman has emphasized, this can lend
itself to abuse in the same way that DeRose correctly emphasizes that pragmatic
strategies can.37
temporalism about propositions or (49), points to a small advantage for the contextualist
version, if one is sympathetic to eternalism. Thanks to Richard Heck for pointing this out.
36
"Contextualist Solutions to Skepticism", in Proceedings of the Aristotelian Society
(1996): 317-333.
37
See in particular p. 103 of his "Contextualism and Skepticism", in Philosophical
Perspectives 13, Epistemology (91-114).
35
There are two senses of the term "pragmatic". In the first sense, most closely
associated with Grice, pragmatic processes are, by definition, post-semantic in nature.
When someone utters a sentence, they may choose to implicate a certain other
proposition by uttering a sentence with a certain semantic content in that context. This
other implicated proposition is the pragmatic content of the speech act. In the second
sense, mostly closely associated with Kaplan and Montague, pragmatic processes are
those that determine the referential content of context-sensitive elements in a sentence
relative to a context. Given a suitably contemporary notion of "semantic", this is a role
pragmatics has in determining semantic content.38
In rejecting purely pragmatic accounts (in the first sense) of the speaker intuitions
with which we began, the contextualist rejects views that maintain that we often confuse
the pragmatic content of a speech act in the first sense of "pragmatic content" with the
semantic content of the sentence uttered relative to that context. But the contextualist
about "know" advocates a powerful error theory concerning our knowledge of features of
the context, that is, an error theory about pragmatics in the second sense. For according to
her, we are often ignorant of content-determining features of context. Thus, in preserving
the correctness of our judgements about a narrow range of cases, the contextualist
strategy raises the specter of a much more general error theory.
In the final section of this paper, I give an example of a "crazed" theory of the
meaning of ethical discourse that is parallel to the contextualist account of "know". If the
contextualist strategy of positing unrecognizable indexicality to account for apparently
intractably conflicting intuitions is correct, then it is hard to see how one could argue
against this crazed theory of ethical discourse. In short, the contextualist methodology
raises the very same specter of undermining the core data for the semantic project as the
pragmatic methodology they rightly reject.
So while the contextualist has an account of a limited range of cases that is in
accord with the principle of charity, the contextualist’s account may ultimately be less in
accord with the principle of charity than an account that rejects the semantic significance
of the small class of intuitions that contextualism explains.
38
For discussion of this, see King and Stanley (op. cit.).
36
Section VI. An Unethical Contextualism
Above, I claimed that contextualism about "know" involves an error theory about
speaker intuitions about semantic content, one that poses the same threat of abuse as the
pragmatic strategies roundly criticized by its advocates. In particular, even when we're
explicitly given the details of a context, we still make errors about the content of a use of
'know' relative to that context. One can describe a context in full detail, and ordinary
speakers will still make incorrect judgements about what is expressed. The particular
nature of this error involves confusing (a) the truth of the proposition expressed by a
sentence relative to a given context with (b) the proposition that sentence would have
expressed, relative to another context (see Schiffer (1996, p. 325)).
So, for example, consider a context in which we are in a philosophy classroom
discussing skepticism. We conclude that Joe Bloggs, who is currently waiting for Bill in
front of the local movie theater, does not know he's not a brain in a vat, and so does not
know he's standing in front of the movie theater waiting for Bill. Bill, however, is
thinking that Joe Bloggs knows that he's standing in front of the movie theater waiting for
Bill, and so rushes to meet him there. We think Bill is wrong; Bloggs does not know that
he is standing in front of the movie theater. But according to the contextualist, this
intuition is a thorough confusion. We are here confusing the proposition expressed in our
high-standard context by "Joe Bloggs knows he's in front of the movie theater" with the
proposition that that sentence expresses relative to Bill's context. According to the
contextualist, our intuitions about the propositions expressed, relative to even fully
described contexts, by sentences containing "know", are untrustworthy. For these
intuitions are affected by our intuitions about the truth-value of the propositions
expressed by those sentences in other contexts (usually the one we are in).
Consider the crazed ethical theorist Schmeter Punger.39 Punger is disturbed by the
fact that in ethics classes in philosophy departments, ordinary speakers are inclined to
The view that follows is similar to that of Peter Unger, in his "Contextual Analysis in
Ethics" (Philosophy and Phenomenological Research, March 1995: 1-26), a paper I
discovered unfortunately after I wrote this section. Unger clearly recognizes the basic
point I am trying to make, namely that the two kinds of contextualism are susceptible to
the same kinds of support and objections. The view below is distinct from Unger's,
39
37
reject many ethical assertions made in non-philosophical contexts. For example,
ordinarily, most Americans think that any policy that benefits Americans is morally good,
no matter what its impact is on non-Americans. However, in ethics seminars in
philosophy departments, we are inclined to reject these ordinary ethical intuitions, despite
their robustness and prevalence in ordinary life. Our crazed ethicist is led by these facts to
endorse the following view about the meaning of "good" (and other ethical terms).
Relative to a context, "good" expresses a property true of an action just in case that action
would maximize the happiness of all contextually salient people in that context, and
violates none of the rights of all contextually salient people in that context. Of course,
relative to certain 'high standard' contexts, like in an ethics class in a philosophy
department, all human beings (and perhaps also non-human animals) are contextually
salient. But relative to most contexts, in fact most of the 'ordinary' ones, only a small
group of people or animals are contextually salient. Relative to these contexts, an action
has the property expressed by "good" just in case it maximizes the happiness of the
members of that group, without violating any of their rights.40
This theory of the good is particularly useful for resolving ethical disputes.
Suppose Joe is a surgeon who have five patients, one of whom needs a heart, two of
whom need kidneys, and two of whom need lungs. Joe decides to kidnap Sally, a
perfectly healthy young woman, and use all of our organs to save his five patients. Call
Joe's action of saving his five patients "Harry the action". Hannah and Mary are unaware
of how Joe obtain his organs in performing Harry the action. In particular, they are
completely unaware that Sally has ever existed. In the conversation they are having, only
Joe's five patients are salient. Hannah utters the sentence "Harry the action is a good
action for Joe to have done." According to our crazed ethicist, Hannah's utterance
however, in that it employs a different hypothesis as to the 'character' (in Kaplan's sense)
of "good". In particular, Unger's paper contains a less absurd suggestion as to what
determines the property expressed by "good" relative to a context. My purpose in this
section is to argue that the ridiculous contextualism about "good" discussed below is in
fact what is parallel to contextualism about "know".
40
Of course "good", like other comparative adjectives, is contextually sensitive (a good
dancer is not necessarily a good violinist). But what is at issue here is a context-sensitive
semantics for "good" on one of its readings, namely moral (ethical) goodness. The crazed
ethicist's claim is that "good", when it means moral goodness, has a further dimension of
context-sensitivity, namely the one described in the text.
38
expresses a true proposition. Of course, intuitively, we think her sentence, even relative
to the envisaged context, expresses a false proposition. But we are simply confusing here
the proposition we would have expressed by using that sentence in our higher standard
context, with the proposition that Hannah expressed relative to her context.
Our crazed ethical theorist may cite, as a virtue of her theory, that it resolves
many of the tensions between ordinary every-day ethical discussion, and the sort of
irrelevant "high standards" ethics that occurs in philosophy departments. For instance,
consider again the fact that many Americans do seem to think that actions are good that
benefit other Americans, even if they terribly harm conversationally irrelevant people,
like citizens of third-world countries. Surely, we do not want to undermine the truth of
such ordinary intuitions. Our crazed ethical theorist can explain the tension between the
sort of 'high-standard' ethical intuitions found in a philosophy class, and the 'lowstandard' ethical intuitions that one finds in everyday life, without impugning these latter
intuitions. Relative to a context in which only Americans are salient, the potential
suffering of non-Americans is simply irrelevant to the question of whether a suggested
course of action has the property expressed by "good", relative to that context. Such an
ethical theory goes a long way towards rescuing ordinary intuitions about the truth of
various ethical claims.
There are even skeptical paradoxes about ethical judgements that this ethical
theory allows us to evade. Every one of my colleagues recognizes the obvious truth of:
(50) Doubling the salaries of all the faculty in the philosophy department at the
University of Michigan is good.
But it turns out that doubling the salaries of all the faculty will necessitate firing some
janitors. So it looks like:
(51) If doubling the salaries of all the faculty in the philosophy department at the
University of Michigan is good, then firing all of the janitors is morally acceptable.
So, the skeptical consequence is:
39
(52) Firing all of the janitors is morally acceptable.
(52) seems obviously false. But it follows from the obviously true premise (50), together
with premise (51), which seems hard to dispute. Our crazed ethicist has a solution for
such ethical paradoxes. The moral skeptic is raising the semantically relevant standards
for "good", by shifting the semantically relevant set of people that determine the semantic
content of "good" relative to a context. So, the argument equivocates on the semantic
contents (relative to a context) of the moral expressions used within it.
Of course, the first premise of this ethical skeptical "paradox" may seem to some
less compelling than the epistemological version. But our crazed ethicist would maintain
that this just exemplifies the degree to which we confuse the content-determining features
of one context with the content-determining features of another. Those who find the first
premise of this skeptical argument less than compelling do so because they systematically
confuse "high standards" contexts with "low standards" ones. Perhaps some among us are
more likely to make this confusion in the ethical case than in the epistemological case (no
surprise, given what is often at stake in ethical disagreement). The difference between the
skeptical ethical paradox and the skeptical epistemological paradox is one of degree,
rather than of kind.
One might raise certain objections against this ethical theory. For example, one
might argue that the ethical theory suggests that as soon as a group of people become
conversationally salient, a whole set of previously true ethical propositions becomes
false. But this objection is confused.41 The theory in question claims that "good" is
indexical. Once a new group of people becomes contextually salient, "good" expresses a
different property. The propositions expressed by the earlier utterances remain true; the
relevant actions still have the property expressed by "good" relative to those looser
ethical contexts. It's just that the relevant actions do not have the (different) property
expressed by "good" relative to the new 'higher standards' contexts.
Of course, some of us have the clear intuition that, even relative to contexts in
which people in the third world aren't conversationally salient, certain policies that harm
them tremendously but benefit those of us who are conversationally salient still aren't
41
For an excellent discussion of why this objection is confused, see DeRose (1999).
40
good. But of course our crazed ethicist has an explanation of our intuition here. We are
confusing the proposition that we would express in our 'higher standards' context by
uttering "That policy is a good one" with the proposition that would be expressed by this
sentence relative to the lower standard context.
Again, one might object to this theory that "good" does not behave as if it has this
sort of context-dependency. For example, certainly we don't notice context-shifts with
"good" in evaluating the skeptical argument considered above (see Schiffer (1996, p.
326)). For example, the argument in (a) seems perfectly sound; the argument in (b), with
the obvious context-sensitive expression "here", seems nothing short of batty:
(a) i. It's not morally acceptable to fire all the janitors.
ii. If it's good to double all of the philosophers' salaries at the University of Michigan,
then it's morally acceptable to fire all the janitors.
iii. So, it's not good to double all of the philosophers' salaries at Michigan.
(b) i. It is not wet here. [said standing inside]
ii. If it is raining here, then it is wet here.
iii. So it is not raining here.[said standing outside in the rain]
But our crazed ethical theorist may respond by denying that all context-dependency is to
be modeled on such obvious context-dependency as we see with "here" and other obvious
context-sensitive expressions.42
A final objection to the crazed theory involves some its counterintuitive
consequences that can be brought out by considerations of propositional anaphora.
Consider:
(53) Slashing taxes for Americans is good. Had I considered people in the third world
before making that judgement, what I said still would have been true. But now that we
are in fact considering the effects of our policies on people in the third world, slashing
taxes for American isn't good.
This discourse is nothing short of bizarre. But unfortunately, this sort of example cannot
be used to drive a wedge between this crazed theory of "good" and contextualism about
"know". For, as we have seen in Section IV, the contextualist about "know" predicts the
41
felicity of similarly odd discourses. So the contextualist about "know" is in any case
committed to rejecting this kind of data as evidence for or against a view.
Our discussion of this crazed theory of the meaning of "good" illustrates a
methodological problem with contextualist claims, the very same methodological
problem that DeRose raises against those who seek to give a merely pragmatic account of
the intuitions that motivate contextualism. By rendering impotent a certain class of
speaker intuitions, contextualists seek to render their claims impervious to standard
objections to the kind of proposal it is. Taken to its extreme, this has the consequence of
undermining the kind of data that is relevant for evaluating it. In short, the contextualist
strategy for dissolving a philosophical problem faces the same methodological objection
to the one DeRose raises against merely pragmatic accounts of the intuitions motivating
contextualism about "know". It is particularly important, then, for the contextualist to
defend her thesis that "know" is context-sensitive by appeal to arguments and evidence
that couldn't be used to defend, for example, such crazed theories of meaning like the one
just canvassed. If this cannot be done, then the contextualist methodology of positing
non-obvious indexicality to resolve a philosophical paradox should be rejected.
See Thomas Hofweber's Contextualism and the Meaning-Intention Problem.'' In
Cognition, Agency and Rationality, edited by K. Korta, E. Sosa and J. Arrozola, Kluwer,
1999, p. 93-104.
42
42