Relativism and Meta-Semantics

I’m going to be commenting on Michael Glanzberg’s Context, Content and Relativism (PDF) at Bellingham. The paper is very good, as you’d expect, but I think one of the arguments he is responding to is interestingly different to the argument that I, and some others, have made. (This isn’t to say that some people have also made the argument that Michael makes of course. There are lots of relativists out there!)

What’s under debate is whether a broadly contextualist (what some call an indexicalist) account of epistemic modals and predicates of personal taste are correct. I’ll call such a few contextualism in what follows. Michael suggests that relativists have been moved by the following consideration.

  1. If contextualism is true, then the metasemantic theory of how context sets the value of variables is very unsystematic.
  2. Metasemantics is never that unsystematic.
  1. So, contextualism is false.

Michael argues, correctly in my view, that premise 2 fails. Ironically, that I’m moved by goes quite the opposite way. It’s more like the following.

  1. If contextualism is true, then either the metasemantic theory of how context sets the value of variables is very systematic, or belief reports involving epistemic modals or predicates of personal taste are non-compositional, or epistemic modals and predicates of personal taste are monsters.
  2. Metasemantics is never that systematic.
  3. Those belief reports are compositional.
  4. Epistemic modals and predicates of personal taste are not monsters.
  1. So, contextualism is false.

I’ll defend premise 2, then premise 1. I’ll sort of take premises 3 and 4 for granted.

Consider the pronouns ‘we’ and ‘they’. Here’s one theory about their meaning. The two words are synonyms in the sense that they have exactly the same semantics. Each of them is a function from context to groups of individuals. The semantics have nothing more to say about their meaning. It so happens that the meta-semantics requires that the value of ‘we’ is a group that includes the speaker, and the value of ‘they’ is a group that (at least usually) doesn’t. But that isn’t something that follows from the meaning of the two words.

Now I think this is intuitively crazy. It’s not a mere metasemantic regularity that ‘we’ denotes a group that includes the speaker; it is part of the meaning of the word ‘we’ that it is a first-person plural pronoun.

But how could we argue against the crazy view? I think we could appeal to the following principle. It’s a general rule that there are not any interesting meta-semantic generalisations about the function that maps contexts to values of variables. If there are any such interesting generalisations, then they are parts of the meaning of the word in question. The problem with the crazy view is that it allows for an interesting meta-semantic generalisation concerning ‘we’. By some decent principle of reasoning (affirming the consequent, induction on single case, etc) I infer this is generally impermissible. Hence premise 2.

Now for premise 1. Consider the following example sentences.

(4) Roller coasters are fun.
(5) Andy might be in Prague.

A contextualist theory of ‘fun’ will assign a semantic value to an utterance of (4), in context, something like the following. That sentence is true iff roller coasters are above a certain standard s of funness for experiencers e. Perhaps the context will assign s directly, or perhaps it will assign some comparison class c and s is computed from that. This won’t matter to us, because we’ll mostly be interested in e.

A serious infelicity arises when (4) is uttered by someone who doesn’t in fact enjoy roller coasters. It’s clearly misleading for such a person to utter (4). I’ll assume, as seems plausible, that such an utterance would in fact be false. The natural conclusion is that e must include the speaker. There’s nothing anti-contextualist about that; compare the true contextualist story about the plural pronoun ‘we’.

A similar thing arises with (5). The contextualist story is something like this. (5) is true iff Andy’s being in Prague is not ruled out by a body of information that stands in some epistemic relation R to some agents a. Perhaps R is a constant, perhaps not. R is never stronger than knows, but perhaps it is (always or sometimes) weaker. Again, we’ll focus on the agents a.

If the speaker knows Andy is not in Prague, (5) is misleading. A natural inference is that it is in fact, false. Again, the natural conclusion is that a must include the speaker, and again there is nothing anti-contextualist about that.

Now in both cases we could have this generalisation be a semantic fact or a meta-semantic fact. Let’s assume for now it is part of the semantics. If so, consider what happens in (6) and (7).

(6) Tyler thinks that roller coasters are fun.
(7) Ishani thinks that Andy might be in Prague.

Imagine that I utter both sentences, and that the contextualist account of the embedded claims is correct. Now intuitively for (6) and (7) to be true, neither Tyler nor Ishani have to be thinking about me. Tyler just has to be thinking about roller coasters, and Ishani about Andy’s possible locations. So in these cases, the variables e and a don’t have to include the speaker, though they do intuitively have to include the subject of the belief report.

There are two ways this could, I think, happen.

First, it could be the case that although it is part of the meaning of ‘fun’ and ‘might’ as they occur unembedded that the variables e and a include the speaker, this is not part of their meaning when they are in embedded contexts. That is a way compositionality could fail.

Second, it could be that these terms are monsters. When we have a bound usage of the terms, the variables get their values not from the way things are in reality, but from the higher context of the sentence. So just as we can imagine a language where “Fred thinks that we are idiots” is true iff Fred thinks that he and the people salient to him are idiots, because ‘we’ becomes a bound variable under embedding, we can imagine that the values for e and a take values salient to the thinker in propositional attitude ascriptions. This is conceptually possible, but it would make ‘fun’ and ‘might’ into, in David Kaplan’s sense, monsters.

So I think we’ve now convered the field of ways in which we can reconcile the data about (4) through (7). Either we have to say that the facts about (4) and (5) are generated by a systematic meta-semantic fact: the relevant group always includes the speaker. Or we can say it is a semantic fact, but a fact that goes away under embedding, or the fact stays, but it is a fact about a monster. So premise 1 is true.

This is, to say the least, a rather complicated argument. And I’m sure there are many many ways to respond. Most obviously, one can come up with alternative explanations of the data, explanations that show premise one is not true. But it is interestingly different from the argument Michael detects in other relativist work.

17 Replies to “Relativism and Meta-Semantics”

  1. Hi Brian. I just read Glanzberg’s paper (and I guess I heard a version of it at the Pacific APA). It’s great stuff.

    I have a couple of questions/comments.

    1. By ‘semantic value absolutism’ Glanzberg means the view that the semantic values are sets of worlds. By ‘semantic value relativism’ he means the view that semantic values are sets of worlds and other parameters. In one sense, this is what the disagreement is about. However, there is a sense in which relativists and Kaplan-style semanticists do not disagree about semantic values. On Kaplan’s view, semantic values (propositions/contents, in Kaplan’s sense) are not absolutely true or false but true or false only relative to circumstances of evaluation (so the relativists and the standard semanticists will disagree about what goes into the circumstances but not about whether proposition truth is relative). The real disagreement, in my opinion, is about whether utterances or sentences-in-context are true or false absolutely.

    2. Glanzberg remarks that the relativists cannot account for faultless disagreement. I take it that this is because, to account for faultless disagreement, one must show that there is real disagreement (i.e., a proposition and its negation contradict with respect to one and the same circumstance of evaluation). Relativists, I assume, cannot do that, because the disputants will be constituents/judges of different circumstances of evaluation (since they are “in disagreement”: they are, per definition, different judges). However, it seems to me that there is still a sense in which the parties disagree. Their claims express a proposition and its negation, and relative to each of their respective circumstances of evaluation, the proposition and its negation contradict.

    3. I am not sure I see why old-fashioned index-theory and relativism are more or less the same thing. Index-theory fails to distinguish circumstances of evaluation from contexts of use, but relativism does not fail to do that.

    4. Glanzberg says that to account for modals we need: coordinates for sets of worlds and an accessibility relation in the circumstances of evaluation. But if that’s right, doesn’t everyone need them? Take “it is possible that”. It’s a sentential operator. Its content is the same in all contexts, isn’t it? Or does the operator have different semantic values in different contexts? So, one sem. value if we are talking about physical poss, another value if we are talking epist. poss., yet another if we talking about close possibilities etc. Or (and this seems initially more plausible) we add stuff to the circumstances of evaluation.

    I enjoyed the paper a lot!

  2. Hi Brian,

    You wrote:

    (6) Tyler thinks that roller coasters are fun.

    (7) Ishani thinks that Andy might be in Prague.

    Imagine that I utter both sentences, and that the contextualist account of the embedded claims is correct. Now intuitively for (6) and (7) to be true, neither Tyler nor Ishani have to be thinking about me. Tyler just has to be thinking about roller coasters, and Ishani about Andy’s possible locations. So in these cases, the variables e and a don’t have to include the speaker, though they do intuitively have to include the subject of the belief report.

    I don’t understand your reasoning here. As I read you you’re assuming that any features of the speaker’s context that help determine what proposition is expressed by “roller coasters are fun” in (6) are part of the content attributed to Tyler. If you are making this assumption, I wonder why you think it’s warranted. I would have thought that the speaker’s context can help fix the attributed content without being part of that content. For example, (a) can be true even if Tyler has no beliefs about the speaker’s location.

    (a) Tyler thinks he’s eight miles from here.

    I’m sure I’m just missing something. (And I do see, by the way, that your case might be stronger for ‘might.’)

  3. You wrote:

    …it could be that these terms are monsters. When we have a bound usage of the terms, the variables get their values not from the way things are in reality, but from the higher context of the sentence. So just as we can imagine a language where “Fred thinks that we are idiots” is true iff Fred thinks that he and the people salient to him are idiots, because “we” becomes a bound variable under embedding, we can imagine that the values for e and a take values salient to the thinker in propositional attitude ascriptions. This is conceptually possible, but it would make “fun” and “might” into, in David Kaplan’s sense, monsters.

    Here’s a proposal along these lines, and a suggestion for why the resulting system needn’t be monstrous (nor relativist).

    Suppose semantic values are given relative to a context and an index, and both context c and index i are world-time-person triples ((wc, tc, xc) and (wi, ti, xi), respectively). Then we might make the semantic value of “fun” at c and i equivalent to the semantic value that “fun for xi” would have at c and i. Unembedded, xi would default to xc, since the value of an index coordinate is initially set to the corresponding value of of the context coordinate. This makes the prediction you want that “Rollercoasters are fun” is false if uttered by a speaker xc and rollercoasters are not fun for xc.

    If we assume that the attitude verb in (6) takes functions from centered worlds to truth values (i.e. centered worlds propositions), then “Tyler thinks” in (6) will shift the the person coordinate of the index. So we’ll get (6) coming out true at a context c and index i just in case in all the centered worlds (w,t,x) compatible with what Tyler thinks at tc in wc, rollercoasters are fun for x at t in w. On this account, then, Tyler’s belief wouldn’t in any sense be “about” any of the context coordinates.

    The system isn’t monstrous because nothing ever shifts the context parameter. The system isn’t relativist either because there is no context of assessment parameter, nor does the definition of “truth of an utterance” say anything about contexts of assessment; instead, that definition will say something like this: an utterance of a sentence S by a speaker xc at a time tc in a world wc is true (simpliciter) just in case the Kaplanian character of S maps ((wc,tc,xc),(wc,tc,xc)) to truth.

    What’s slightly non-standard about the analysis is that it has a person coordinate in the index, so it is a “de se”-ified context-index system. I’m not advocating this as an account of predicates of personal taste, just pointing out that there is a way of accounting for the relevant data that is neither monstrous nor relativist.

  4. Brit,

    I’ll try and get Michael to show up and answer these, because they seem to be more questions about the paper than the reply – and good questions too!


    You’re right – I shouldn’t have used ‘about’. What I meant was that the truth of utterances of (6) and (7) doesn’t co-vary with the knowledge or tastes of the utterers. That’s the false prediction of the contextualist, not what I suggested.


    That’s very interesting, and I need to think more about it. Two quick points of interest.

    1) Is this position similar to what John MacFarlane has been calling non-indexical contextualism?
    2) Can we always use this trick to show that a language is monster-free? So consider those languages where the word we’d translate as ‘I’ behaves as a bound variable. Could we offer a context-index account of the meaning of ‘I’ so that no contextual parameter gets bound, just an index? If so, this seems to trivialise the prohibition on monsters.

  5. I’m not sure about the question related to MacFarlane—-I haven’t heard/read about what he means by “non-indexical contextualism”.

    Whether this trivializes Kaplan’s prohibition is a good question. I don’t think it does, since the proposal you’re suggesting would make different predictions from a proposal that posited monsters. The two main arguments that Anand and Nevins give in their paper speak in favor of a monstrous analysis for certain operators in certain languages and against the non-monstrous one you mention.

    For example, the proposal you suggest would allow indexicals in the scope of an attitude verb to shift independently of each other. That is, the proposal wouldn’t predict that either all indexicals under the scope of an attitude verb would shift or none would—-some could get their value from the index, others from the context, so that they could vary independently of each other. A monstrous analysis would say this is not possible, since all the indexicals get their value from the context. Thus, if the attitude verb is a monster, then all of the indexicals in its scope will have the shifted reading; but if the verb isn’t a monster, then all of the indexicals will get their values from the actual context, and so will have their “normal” reading. Anand and Nevins report that the languages they study all satisfy this “Shift Together” generalization, thus offering some evidence in favor of the monstrous analysis for those languages.

    Another piece of evidence Anand and Nevins offer concerns multiple embedding. Take a structure that has multiple attitude verbs. The non-monstrous analysis will in principle allow expressions to access the context no matter how low in the structure you go. So you might get the logophoric (“montrous”) readings after the first attitude verb, but then after the next attitude verb you could have a “genuine” indexical that takes its value from the actual utterance context. So you could have “Sam said that I said that you said…” where “I” gets its value from the index and “you” gets its value from the original utterance context. The resulting reading would be roughly: “in all the centered worlds compatible with what Sam said, the center said that you said…” where “you” picks out the addressee of the actual utterance context, just like it does in English.

    But this is not possible on the monstrous analysis. For example, if in our above sentence fragment, “I” doesn’t pick up the speaker of the actual context, “you” can’t have as its value the addressee of the actual context either. This is so because, on the monstrous analysis, if “I” doesn’t pick up the speaker of the actual context, then that means that the initial “said” has shifted the context, which in turn means that “you” can’t have as its value the addressee of the original utterance context, since that context is “lost” after the monster does its work. Again, for the languages Anand and Nevins look at, the monstrous prediction is borne out—-sometimes you can’t recover the original utterance context in these languages, i.e. after one shifting, the original context is lost and nothing lower in the structure can access it.

    I’m not sure I have all the details exactly right, but you can see the how the proposals might make different empirical predictions. But your point is a good one, since it sheds light on what is at issue here, and what sort of evidence can be used to decide these questions. One might wonder if the non-monstrous analysis would be able to “account” for the Anand-Nevins data, if one added the right stipulations to it. If one did this, that might indeed trivialize the prohibition on monsters; more importantly, the monstrous analysis will still look much more explanatory than the non-monstrous one.

  6. Hi Dilip,

    “Some could get their value from the index, others from the context, so that they could vary independently of each other”

    I would add: and some could get their value from both (for instance, when the attitude verb is factive). And that is one way of replying to one of Jason Stanley’s objections to relativism/non-indexical contextualism. I think, however, that non-indexical contextualism, as it stands, is inadequate (because it cannot easily account for retraction data). Personally, I prefer a closely related alternative that allows for certain harmless monsters.

  7. Hi all
    I think I’ll try to break up some comments into pieces. Here’s a start.

    I think the crucial issue is Brian’s premise (1), though I also want to say a little about premise (2) in a moment.

    The main argument is the behavior of things like epistemic modals and predicates of personal taste in embedded contexts, particularly intensional contexts, but I am not entirely sure if that matters or not. Brian points out that in
    (6) Tyler thinks that roller coasters are fun.
    the relevant experiencer class does not need to contain the speaker (but presumably should contain Tyler).

    I want to quibble about the general rule that the speaker, but lets take it for granted for now. Granting it, there are two sorts of responses. One is that I am not so sure the judgments here go the way Brian says they do. Well, of course, I would have to think that, but at least, I am not so sure the judgments here are all that clear. Suppose we are talking a lot about roller coasters, what makes them fun, what we like about them, etc. According to my view, this conversation will build up a scale for measuring fun, according to which we will asses our claims in the context. (For instance, roller coasters are fun in a way that includes feeling a pit in your stomach, and maybe feeling a rush of fear. This does not always count as fun. If you feel those things at a dinner party, it is usually not fun.) Now, you tell me (6). I think the most natural reading is that you are reporting what Tyler thinks, in terms of the scale we are using in our conversation. Now, it may be natural to assume that Tyler would work with a similar scale, but it still sounds to me like we are saying something the bears on our questions and answers about what counts as fun, for purposes of our conversation.

    I think the intuitions here are weak, and I am not sure if this is the only reading or not. It at least seems to me to be the more salient one. But let us suppose, for arguments sake, that Brian’s reading is also available. Brian concludes we would have a compositionality problem. I am not convinced of this. We have, on my view, a parameter in the logical form of the sentence, which can be set by context. But this does not preclude that in embedded environments, it can be affected by other linguistic material, without affecting compositionality.

    There are broadly two ways one might account for this. One is to note that linguistic material can affect context, and so it might just be that uttering the proper name and ‘thinks’ is a clue to updating the context. This is one theory of how anaphora on referential NPs works. For instance, in Heim and Kratzer’s example:
    I do not think anyone here is interested in Smith’s work. He should not be invited.
    We simply see uttering ‘Smith’ as making Smith highly salient. We might do the same with ‘Tyler’. Now, this might predict that (absent other modifications to context), we see Tyler’s addition to e as remaining. I do not have any clear intuitions on whether this consequence is right (given that my intuitions went the other way already).

    The second way does not work so heavily with context, but notes that something which in one environment functions as a free variable, set by context, can in other environments be affected by other parts of a sentence. They prime example of this, of course, is variable binding, which does not implicate any problems for compositionality. We do not simply have a bound variable here, as e will not be set to be exactly Tyler (and I doubt that would be variable finding anyway). Actually, we are already in difficult territory, without worry about personal taste or modals. It it an old observation that many adjectives are sensitive to subjects in similar ways. For instance:
    Billy made a big pile of sand.
    If Billy is a small boy, ‘big’ gets one standard, if Billy is a construction worker with a bulldozer, it gets a different one. (I think this example is from McConnell-Ginet. Szabo and Reimer have discussed issues like this.) So, I am not going to try to solve this problem completely. But here is a stab. Brian is supposing that it is a rule that e contains the speaker. We might implement this as a presupposition on the value of e. That presupposition might be put in place in way that itself has a free variable, which is normally set by context to be the speaker, but can be marked as co-referential with the subject in some environments (or bound, if you like). This is a pretty quick suggestion, but I think it is enough to show why I think there is maneuvering room. (Szabo makes similar moves, if I remember the paper right.)

    I should mention, though, that I am not convinced that the rule that the speaker is always in e is so hard and fast. It is certainly a strong presumption, but I think it might be a defeasible rule, which indicates it is part of the rules for setting e, not part of the semantics. Here is one sort of case. Suppose you go to a meeting where some very particular, unusual aspect of taste is under discussion. Make it something about acquired taste. Maybe it is, for instance, short experimental films from the 70s, which you have seen almost none of, simply did not get what you did see, and generally feel totally cut off from. Now, if we sit you down in front of one, you will have a pretty direct negative experience. But, at the conference, there is fairly specific scale in play, which has nothing to do with you. What happens if you say ‘Dog Star Man is horrible.’. I confess to having no clear judgment here, and I find no way to hear this without it sounding somehow odd. But I do not think we simply re-set the scale by adding you to the experiencer class. I am inclined to say that you simply express something using the scale generally in play at the conference, whether you experience things that way or not. A more careful assessment might be that we have what Stalnaker calls a defective context, as you and the rest of the group have some irreconcilable presuppositions. It is hard to simply report what is said in such cases—-we often have to resort to glossing individuals’ presuppositions to do so.

  8. Here\‘s a bit more, on Brian\‘s premise (2).
    I think I am basically in agreement about this, but I do think there are a few differences between Brian\‘s way of putting it and mine. Part of the issue is what really counts as \‘systematic\’, and what counts as \‘part of semantics\’.

    I am inclined to think that it is an over-generalization that there are no rules about what maps context to values of variables. Rather, I think the more likely result is that the ones that are not coded up in the language (and so semantic, or grammatical) are defeasible, or at least, work in conjunction with other rules, so it may be hard to spot which are in effect. There are some general reasons to expect something like this. I don’t think we work with metasemantics the way we do with the language faculty. We vary in degrees of ability in meta-semantic tasks, and we get it wrong a lot. But, even so, we do seem to have a pretty good ability to do some pretty hard meta-semantic tasks. It is hard to see how that would happen without some rules. A bit less theoretically, there are some areas where rules seem to be important. For instance, there seem to be some generalizations governing pronoun resolution (presumably part of pragmatics), though they seem to be defeasible, or something like it. This does not directly address the setting of contextual parameters, but I something like it can. Here is one example I have tried out. Suppose you think domain restriction is done by setting a contextual parameter. How is it set? When possible, by an anaphora-like process which looks for predicative material to provide a value. I think the rule that predicative material is required seems to be in place. (I point this out in a forthcoming paper on contextualism about quantifier domains, \“Context and Unrestricted Quantification\”). Take the case

    John came to the party and Sarah came to the party.
    They had fun
    Everyone had fun.

    There is a strong presumption to read the quantifier domain, but not the plural anaphor, as fixed by \‘came to the party\’ and not by \‘John\’ and \‘Sarah\’.

    I actually think this is a bit of a quibble. For one reason, we might insist that anything which is a rule like this is simply a presupposition attached to the parameter. I take such presuppositions to be perfectly good \‘parts of meanings\’, and so we might just have Brian\‘s point back. (Presuppositions show the sorts of defeasible behavior I am opting for, through cancellation, or accommodation.)

    But, to quibble more, I am not sure how good the example of \‘we\’ and \‘they\’ is, as \‘we\’ has agreement features that are clearly present and need to be interpreted. Here is a case that might go Brian\‘s way, though. Consider \‘this\’ and \‘that\’. Traditional grammar separates these as \‘proximal\’ and \‘distal\’, roughly, you cannot say \‘this\’ when the object demonstrated is too far away. Is this part of their meaning? Or a kind of pragmatic rule about their values? Actually, I am not sure which idea is more plausible. It is tempting to say it is a somehow defeasible presupposition, which is at least my take on Brian\‘s point.

  9. Hi Brit
    A couple of thoughts on your points. The one I am most intrigued by is (1), but I’ll say a few quick things about the other points first.

    On your (2). I am not sure that is quite what I had in mind. I do at one point ask, off-hand, why what the relativist calls ‘faultless disagreement’ is disagreement, but I do not pursue it. The reason is that I am arguing that there is no grounds provided by the case of personal taste to appeal to any such thing. The contextualist, I claim, handles everythying necessary in terms of old-fashioned agreement or disagreement (the ‘faulty kind’). I do think, and say briefly, that I take the old-fashioned line that ‘faultless disagreement’ is not a coherent notion (analytically). I do not argue that case, though. I just say that the particular case at hand gives us no reason to to stick to the old-fashioned view.

    On (3): I think the old index theory is a form of semantic-value relativism. It is certainly not the same as the current one, for the reasons you give, among others. (It was a bit naive on matters of context-dependence.)

    On 4: I had in mind that if you think to handle that kind of parameter you have to put it in a circumstance, then you would have to do the same for modals. But, I take it the standard view here is that the set of worlds and accessibility relation are contextual parameters. This is to account for the ways that modal expressions like ‘must’ and ‘might’ change their content from context to context. (Kratzer argued this initially, but it is by now pretty standard.) So regardless of what we think of sentential operators in logic (which are not usually represented as context-dependent), I take it that natural-language modal expressions are quite heavily so.

    But, the big issue, as I see it, is your (1). I think the kind of argument you are offering, that we already have semantic value relativitity to worlds, and this means it is just not an issue if we have relativity to other things. This issue came up in discussing temporalism/eternalism, and remains a question. I do not have a knock-down response, but I am inclined to reject it. I tend to think about semantic values as sets of worlds, but that is, of course, a technical notion, which is supposed to help with compositional semantics, and with the represetations of content. The ‘operators’ response focuses on compositional semantics, arguing that we need to enrich semantic values to make sense of certain operators on contents. That was an argument for temporalism, but does not seem to be relevant for the current relativism (many relativists reject the whole set-up of that style argument, Brian does, I think). But when we think about representing content, worlds are not just another parameter. Rather, as for instance Stalnaker puts it, a set of worlds is supposed to give us a picture of the information encoded in a sentences content (in context). As such, the worlds are not exactly ‘parameters’, but a device for capturing content. The question then becomes whether we can make sense of this picture with relativization (to additional ‘parameters’). I think the old Evans arguement from “Does Tense Logic Rest on a Mistake” lifts here, and basically argued we cannot. Some relativists (reference? I think MacFarlane) argue we can, but I confess I am still convinced by Evans.

    Far from conclusive, I know, but I just don’t see how worlds give us ‘just more relativity of truth’. This certainly gets complicated by things like modals. But even there, the assumption that they are simply operators on content is not so clear. At least, they are context-dependent operators, and what role their being operators on full sentential contents play is not so clear. (In the tense case, that has been argued by Enc to be a bad analysis of natural language tenses. There is a corresponding treatment of modals, though it is far more controversial.

  10. Hi Michael,

    Of course, if you treat modal operators as quantifiers rather than sentential operators, then proposition truth will not be relative. But I suspect that modal operators in ordinary language will be sentential operators. If I am right, then proposition truth will be relative (unless of course you insist that modal operators do not operate on propositions).

  11. Hi Brit
    Yes. I mostly agree. When I was insisting on the context-dependence of natural-language modals, I did not mean to be insisting they are not operators on ‘propositions’ (here meaning, I think, the semantic values of sentences or clauses). I was mostly pointing out that we do have context-dependence to cope with as well, and that mattered to the particular argument I was pressing in my paper. I think the ‘textbook standard’ analysis of natural-language modals these days makes them context-dependent operators on propositions, just as you suggest. I did raise the possibility there are other options, but I do not think they are anything like standard, or even widely accepted. (I have in mind the treatment which makes heavy use of bindable world variables. This is popular for tenses, but I suspect it is a minority view for modals).

    But, I think the cautious claim to make, on the ‘textbook standard’ line, is that modals operate on the semantic values of clauses, and that requires these values to look like sets of worlds. (Well, we could do Davidsonian tricks here, so maye ‘indicate’ rather than ‘require’.)

    I think one of the really interesting, and hard question, is whether this amounts to a genuine form of relativism. I took one of your comments above to be pointing to a similar question. In response to it, I was probing a line which holds it might not, in effect, by asking what the relation of such semantic values is to assertoric content.

  12. Hi Michael,
    Very interesting! Suppose the tenses are quantifiers (or variables). In that case I do not see any reason to distinguish between assertoric content and compositional semantic values (which is not to say that I do see a reason for doing it if the tenses are sent. operators). If there are modal operators in the language, then content will be relative to worlds (or a function of worlds). Of course, you might call content that is tied to an utterance ‘assertoric content’, and say that assertoric content is not genuinely relative. But that is not quite right, I think. For assertoric content is really no different from compositional semantic values, which is to say, relativists and non-relativists are disagreeing about utterance truth and not about content truth. That’s my opinion anyway.

  13. From my perspective as a linguist, it seems to me that Brian’s argument shows that epistemic modals are to be understood in terms of the same theory as logophors and shiftable indexicals, i.e. those pronouns in many languages which get their meaning when embedded relative to a derived context, a perspective Dilip alludes to by citing Anand/Nevins’ paper. In this light, a worthwhile prequel to Dilip’s discussion is Schlenker’s “Plea for Monsters” (link below) and von Stechow’s argument that his analysis is not in fact monstrous (also). von Stechow looks to be right to me on this.

  14. Paul: maybe you’re right that Brian’s argument shows that epistemic modals are akin to shiftable indexicals. But aren’t there important differences between genuinely relative expressions and shiftable indexicals? Shiftable indexicals have different semantic values at different contexts of use/derived contexts; genuinely relative expressions have different extensions at different contexts of assessment.

    Michael: I want to add to my previous post that I think that by introducing the term “assertoric content” you’re admitting that content is genuinely relative (in non-relativistic semantics). For assertoric content may be defined as the content that is tied to assertions. A relativist could invent a term that could do the same trick, say, “r-assertoric content”. Let r-assertoric content be content that is tied to an assertion and an evaluation.

  15. Brit: Yes, I agree that there are differences between genuinely relative expressions and shiftable indexicals. But it’s not clear to me that the argument shows that modals are the former and not the latter.

    Deeper down I think two things: As Dilip suggests, the assumption that there are no monsters may not be sufficient for Brian’s argument, since there may be a non-monstrous alternative that solves the problem (Brian?). Moreover, if, as Shlenker and others think, the correct treatment of shiftable indexicals makes them monsters, there’s no point in making assumption 4 that modals can’t be monsters too.

  16. Paul: I think you\‘re right that we cannot just assume that there are no monsters (pace Kaplan). But Recanati has argued that Schlenker\‘s examples can be explained on the assumption that non-strict indexicals, like \‘two days ago\’, are just indexicals that allow for what he calls \“free circumstance shift\”. So even if Schlenker is right about Amharic, he may not be right about English.

    Here is the link to Recanati\‘s paper:

    Indexicality and Context Shift

  17. This has been a really interesting conversation. I’m sorry I haven’t been able to contribute more.

    I don’t think Michael’s suggestion that we can explain the embedded uses purely by appeal to salience can work. The reason is that in other cases salience doesn’t seem to matter. (I’m indebted here to some examples from Andy Egan.) Consider cases like the following.

    See that guy over there. He’s a really famous chef. He’s a genius. His cakes are delicious.

    It seems I’ve made the guy as salient as possible, but it still seems forced to interpret the last sentence as His cakes are delicious by his standards.

    On the other hand, I agree with Paul that this doesn’t rule out the possibility that the predicates of personal taste are like logophors. So I was wrong to say that a contextualist interpretation would have to be monstrous.

    Indeed, it seems a priori plausible that the unarticulated indexical in predicates of personal taste should behave like PRO. So it seems not only possible, but plausible, that some kind of contextualist theory could explain this data.

Leave a Reply