0ak.hu About Writing CV

Contextualism

The following is a tutorial essay for Knowledge and Reality with Daniel Kodsi. In effect, it is a quick agenda for an hour-long discussion on a guiding question, which for this week was ‘What is the difference between contextualism and subject-sensitive invariantism? Which is better motivated by shifting patterns of knowledge ascriptions?’. A free-standing and much more readable version of some of the material may eventually go up on my Substack.

Utterances (e.g., knowledge ascriptions) express meanings, which are true or false in various circumstances.

(Speaker-)contextualism about ‘know’ holds that the meaning of knowledge ascriptions depends, more than traditionally recognized, on the context of utterance—e.g., the possibilities salient to the speaker, or the practical interests of the speaker. Invariantism denies this. It’s more common to prepend ‘subject’ to ‘sensitivism’ / ‘insensitivism’ than to prepend ‘speaker’ to ‘contextualism’ / ‘invariantism’; but I think this is helpful for exposition.

(Subject-)sensitivism about knowledge holds that the truth value of knowledge ascriptions Claims about the extension of an utterance are to be understood as claims about the extension of the content of an utterance. After all, I’ve said that contextualism and sensitivism are views that go beyond what’s traditionally recognized. depends, more than traditionally recognized, on the circumstance of evaluation—e.g., the possibilities salient to the subject, or the practical interests of the subject. Insensitivism denies this.

Insensitive contextualism combines subject-insensitivism about knowledge with speaker-contextualism about ‘know’; I’ll use ‘contextualism’ elliptically for this view. Sensitive invariantism combines subject-sensitivism about knowledge with speaker-invariantism about ‘know’; I’ll use ‘sensitivism’ elliptically for this view. As the above characterizations imply, the traditional view is insensitive invariantism. English know is polysemous between knowledge of some fact (Spanish saber, Chinese 知道) and familiarity with some thing (Spanish conocer, Chinese 认识). I take knowledge ascriptions to ascribe knowledge in the first sense. The second sense doesn’t reduce to the first, but perhaps the first sense reduces to the second (though I don’t have space to discuss this further here). Sensitive contextualism is also possible in principle; I’ll put off discussion of this until the end.

§I adds some more detail (i.e., caveats) to this quick introductory sketch, to make the difference between contextualism and sensitivism clearer. §II outlines some particular contextualist and sensitivist views. §III presents some examples of shifting patterns of knowledge ascriptions, comparing how contextualism and sensitivism treat them. I conclude that, at least with respect to shifting patterns of knowledge ascriptions, contextualism is better motivated than sensitivism (although it may have fatal flaws elsewhere).

I. The mechanics of knowledge ascriptions

Above, I discussed knowledge ascriptions as if they were always utterances. I don’t mean to suggest that knowledge ascriptions cannot be in writing, or even in (possibly non-linguistic) thought. But I do require knowledge ascriptions to ascribe knowledge: i.e., to express that some subject knows some proposition.

So, here is a schematic picture of how knowledge ascriptions work. When one makes a knowledge ascription, its character and context determine its content. (So, character is a function from context to content.) The content of a knowledge ascription in some context is something like a proposition—something true or false in various circumstances. (So, propositional content is a function from circumstance to truth-value.) In general, content (or meaning, or intension) maps circumstance to extension. When a sentence is composed of terms, then its content is composed of their content, and its extension is composed of their extensions. (Though I think this gets a bit tricky to spell out more precisely.)

Here’s an example to illustrate. Suppose we both say ‘I know that she’s in Oxford!’; in one sense, we’ve said the same thing: our utterances have the same character. But they occur in different contexts: your utterance is uttered by you, and mine by me. So, my claim was about me, yours about you. In this other sense, we’ve said two different things: our utterances have different content. This stems from the fact that the English word ‘I’ is indexical: its content (a function from circumstances to entities) depends non-trivially on the context of utterance.

This illustrates some context-dependence in knowledge ascriptions that even the invariantist accepts. Some of this might even be in the word ‘know’—for instance, if our present-tense utterances are made at different times, then each claim is contextually tied to the time of the utterance. What’s particular about contextualism is holding that there’s some extra indexicality with the word ‘know’.

Meanwhile, sensitivism doesn’t really interact with these mechanics for knowledge ascriptions. Above, I described it as requiring some extra sensitivity to the circumstance of evaluation. But this really just means taking knowledge to require more than traditionally recognized. Consider my knowledge ascription again: even the insensitivist recognizes that its truth depends on facts about me, the subject. For instance, the truth of my claim depends on whether I believe that she’s in Oxford or not, since knowledge requires belief. What’s particular about sensitivism is holding that there’s some extra subject-sensitivity with knowledge.

Sometimes, the line blurs between what is and is not traditionally recognized. For instance, looking ahead, one form of sensitivism requires that error possibilities are not salient (in the right way) to the subject. But one might hold that the salience of error possibilities (expressed by, e.g., ‘for all I know, it might not be true!’) destroys belief. But this seems like a traditional requirement on knowledge, not extra subject-sensitivity. One way out is to classify this view as sensitivism not only about knowledge but also about belief. But this still illustrates that (especially for sensitivism) bright lines are difficult to draw.

One final note is the distinction between truth conditions and assertibility conditions. Invariantists may hold that the assertibility of knowledge ascriptions depends on the speaker. For instance, perhaps I ought not assert, ‘she knows that she’s in Oxford!’, if I myself don’t even believe that she’s in Oxford. If I nevertheless did produce such an assertion, we might still judge it true (on the grounds that she does know that she’s in Oxford). The particular claim of the contextualist is that such contextual features can make a knowledge ascription not only unassertible but outright false. There is the thought that invariantists can mimick contextualist successes by converting truth-conditions to assertibility-conditions. This might work for converting false utterances to unassertible ones; but I’ll exploit the fact that this doesn’t work for converting true utterances into assertible ones.

II. Salience versus interest

One version of contextualism relies on attention: the subject’s belief must be true in any case noticed by the speaker. The analogous version of sensitivism requires that the subject’s belief be true in any case noticed by the subject. A slightly different form of contextualism relies on salience: the subject’s belief must be true in any case taken seriously by the speaker. The analogous version of sensitivism requires that the subject’s belief be true in any case taken seriously by the subject. Consider the following dialogue:

Salience

A: I know that I’m in Oxford.

B: You might be dreaming at home in London…

A: Don’t be silly — I know that I’m in Oxford.

Clearly, the response from A is acceptable. But neither attention contextualism nor attention sensitivism can vindicate A’s response, because B has brought error possibilities to attention. On the former, the knowledge ascription now expresses something false. On the latter, what the knowledge ascription expresses is now false. Meanwhile, both salience contextualism and salience sensitivism can vindicate A’s response, because the error possibilities brought to attention are not taken seriously by A. On the former, the knowledge ascription still expresses the same truth. On the latter, what the knowledge ascription expresses is still true. So, this (un)shifting pattern of knowledge ascriptions tells against attention contextualism and attention sensitivism.

Another version of contextualism relies on practical interests: the subject’s belief must be an appropriate premise for the speaker’s practical reasoning. The analogous version of sensitivism requires that the subject’s belief be an appropriate premise for the subject’s practical reasoning. Another way to put appropriateness for practical reasoning is that the subject’s belief must be true in any practically relevant case. Consider the following reasoning:

Interest

I’m excited that I’m going to France next summer, so I must know that I’m going to France next summer. This implies that I should start learning a bit of French.

# [It also implies that I’m not going to have a heart attack soon, and so I don’t need to buy life insurance this year.]

Clearly, the reasoning is faulty. On interest contextualism, the knowledge ascription which permits the first bit of reasoning is true, but the knowledge ascription that would permit the second bit of reasoning is false. On interest sensitivism, the knowledge that one is going to France next summer disappears as soon as one attempts the second bit of reasoning. Meanwhile, on salience contextualism, the ‘second’ knowledge ascription But wasn’t there only one knowledge ascription? We might consider using some proposition as a premise in practical reasoning a type of self-knowledge ascription. is false only if the second bit of reasoning counts as taking the possibility of a heart attack seriously. If salience requires more than this, then salience contextualism fails to immediately rule out the faulty reasoning. On salience sensitivism, the knowledge ascription becomes false only if the second bit of reasoning counts as taking the possibility of a heart attack seriously. If salience requires more than this, then salience sensitivism fails to immediately rule out the faulty reasoning. So, the notion of salience is fairly tightly constrained; if it is too weak, it runs into trouble with the Salience cases; if it is too strong, it runs into trouble with the Interest cases.

The salience views also run into trouble in failing to immediately rule out this alternate version of the Salience dialogue:

Salience

A: I know that I’m in Oxford.

B: You might be dreaming at home in London…

A: Sure, I don’t know that I’m in Oxford. ? [But there was nothing wrong with what I said, before you spoke.]

The continuation seems infelicitous (at least at first pass), but on the salience views, it turns out true because the error possibility was not previously salient to A. One way of hearing the continuation might be to consider the concession that one doesn’t know as destroying belief, and thereby preventing knowledge; but while one still believed, one very well could have known. Still, it seems unnatural. There might be other reasons that the continuation is true but unassertible, on the salience views. However, the interest views smoothly predict that the continuation is bad: if the first part of A’s response is correct because the error possibility is (somehow) relevant, then all along either the knowledge ascription meant something false (to put it in contextualist terms), or what it meant was false (to put it in sensitivist terms).

So, we have the interest versions of each view as leading candidates, with the salience versions somewhat behind (and attention versions discarded). Consider a final version of the above dialogue:

Clarification

A: I know that I’m in Oxford.

B: You might be dreaming at home in London…

A: I meant knowing with reasonable certainty.

To my ear, A’s response—which involves a clarification as to what ‘know’ meant—is acceptable. Contextualism vindicates this, while invariantism does not. In the next section, I’ll take a deeper look at how shifting patterns of knowledge ascriptions motivate contextualism and sensitivism. In particular, I’ll present some cases which contextualism seems to handle much better than invariantism, even sensitive invariantism. These patterns thus motivate contextualism better than sensitivism.

III. Shifting Patterns

In Some Sense

(1) Everyone’s coming to the party! Well, not (absolutely) everyone, but still everyone.

(2) My cup is empty. Well, not (completely) empty, but it is empty.

(3) He knows she’ll be here soon. Well, he doesn’t (entirely) know, but he does know.

(4) We ate the snacks. # [Well, we didn’t (totally) eat them, but we did eat them.]

Notice the marked difference between (3) and (4). Just as contextualism about ‘everyone’ and ‘empty’ vindicates (1) and (2), contextualism about ‘know’ vindicates (3). On the stressed usage of each term, a stricter standard is applied. (By contrast, (4) seems bad because there’s no way to get two readings of ‘eat’ such that one is true and the other is false.)

At first pass, it looks like sensitivism might also vindicate (3), if two standards can be relevant at once. For instance, it might be that he knows enough that he should wait for her arrival, but not well enough that he should place a large bet on whether she will get into a road accident on the way. Unfortunately, this diagnosis ends up being a form of contextualism: it requires the two usages of ‘know’ to have different content. If invariantism is correct and they have the exact same meaning, then it’s hard to see how this meaning can be both true and false of the same circumstance.

The most plausible route for sensitivism to take is to give an error theory: on the best reading of (3), it’s not literally true that the subject knows—this is just an instance of loose talk. But in cases of loose talk (‘It was $300. Well, not exactly $300, but it was $300.’), we would expect the speaker to concede the literal falsity (‘No, it wasn’t literally $300.’). But the speaker in (3) seems able to felicitously insist that the subject does literally know.

My judgements about (3) therefore support contextualism and undermine sensitive invariantism. Of course, if one’s judgements differ—perhaps (3) does sound as bad as (4)—then the support will go the other way.

Disagreement

(1) Is she tall? By my standards, yes; by your standards, no.

(2) Does she know how to sing? By my standards, yes; by your standards, no.

(3) Does she know that there are 52 partitions? By my standards, yes; by your standards, no.

(4) Did she come to the tutorial? # [By my standards, yes; by your standards, no.]

Again, notice the marked difference between (2) and (3) on the one hand, and (4) on the other. Just as contextualism about ‘tall’ vindicates (1), contextualism about ‘know’ vindicates (2) and (3). In each case, there is no real disagreement. (By contrast, (4) seems bad because the dispute must be more real.)

However, it does seem like sensitivism might vindicate these judgements. In particular, we might interpret ‘by my standards’ as elliptical for ‘if my standards are the right ones for her’. But this seems like a stretch. Both (2) and (3) sound just like (1), but the sensitivist explanation for them is quite different. The elliptical treatment of (1) sounds different from the natural reading. So, this explanation seems much less smooth than the contextualist one.

One interesting wrinkle is that (2) seems to involve some contextual factors above and beyond those in (3). In particular, we might take knowing how to sing as knowing that this is a way to sing, for some way which meets a given threshold. Then, there is some context-dependence in what threshold is relevant for something to count as singing. (Notice that it’s harder to hear the default reading of (2) with ‘Does she know how to get to my office?’; the default reading for this sounds like it’s about whether her confidence is sufficient, not whether her way is sufficient.) Thus, there may be some context-dependence in knowing how to sing, which results from context dependence in what counts as sufficient for singing.

Bound Readings

(1) Among the adults in the family, only some are tall; but every child in the family is.

(2) Among the mathematicians, only some know that CH is independent of ZFC; but every philosophy student in the room does.

(3) Among professional cyclists, only a few bike in Oxford; but every Magdalen student does.

Notice that there are two readings for (1): on the first, the threshold for tallness is constant, implying that most of the adults are shorter than the children. On the second reading, the threshold for tallness is relative: the children are tall for children, but most adults are not tall for adults.

There seem to be two analogous readings for (2): on the first, the threshold for knowing is constant, implying that most mathematicians haven’t even heard that this result holds (which is presumably the extent to which the philosophy students know the result). On the second reading, the threshold for knowing the result is relative: the students have heard that the result holds, and so know it for philosophy students, but most mathematicians (let’s suppose) don’t know how to prove this result, and so don’t know it for mathematicians.

In the same way that contextualism about ‘tall’ suggests two distinct readings of (1), contextualism about ‘know’ suggests two distinct readings of (2). With the same word, the speaker can mean different things for different people. By contrast, (3) totally lacks the second reading which (1) and (2) have.

At first pass, sensitivism seems to do well in predicting the second reading for (2): a sensitivist might hold that mathematicians only know results if they know how to prove them, but philosophy students know results if they simply hear that they hold. The trouble is that such a sensitivist view has a hard time generating the first reading, on which most mathematicians have not even heard of the result. Holding fixed the circumstance of evaluation, (2) is not ambiguous on sensitive invariantism: the two readings of (2) on sensitivism must coincide, since ‘know’ always picks out the same relation of knowing. Either this relation is relative in the circumstance (e.g., because the interests of the two groups differ), such that we get something which looks like the second reading, or it is not relative in the circumstance (e.g., because the interests of the two groups coincide), such that we get something which looks like the first reading. There only seem to be two distinct readings because the truth value of this single reading depends on subtle differences in the circumstance of evaluation. Meanwhile, (1) remains ambiguous with a constant circumstance of evaluation; it genuinely has two readings. But it seems likely that (1) and (2) should be treated the same way. So, this case also supports contextualism and undermines sensitive invariantism.

For all I’ve shown, contextualism may end up worse off than sensitivism: e.g., perhaps contextualism struggles to provide a solid basis for knowledge norms. Of course, I have not even considered all cases of shifting patterns of knowledge ascriptions; some other cases will surely tell in the other direction. But ideally, I’ve shown that contextualism enjoys a decisive advantage over sensitivism in at least some important cases of shifting patterns of knowledge ascriptions. I was actually planning to defend SSI; but I changed my mind after considering the cases I discuss in the last section. So at least in my (limited, naive) case, these were decisive. Most of these cases are dismissed at some point in the Hawthorne book; this probably means that most of them don’t end up working, and the other few have been discussed elsewhere since.

References