One thing I’ve become much more aware of in Leeds is the distinction between semantics and metasemantics. There is a worry that ignoring this distinction is creating some confusion. Metaethicists have traditionally claimed to be doing moral semantics whereas in fact they have mainly been doing metasemantics. Today, there is a new debate in which people also say that they are doing semantics. This time this claim is actually true. However, as a consequence, the new debate is not at the same metasemantic level as the previous debates. I know this is a bit confusing but let me try to elaborate a bit.


Very roughly, semantics investigates what the semantic value of different terms is. Most moral terms (right, wrong, good, evil, kind, cruel…) are used in predicate expressions. On the traditional Fregean views, the semantic value of predicate is a first-level function from objects (in the moral case usually from actions) to truth-values (there are other options too but let me use this simple view as an illustration). So, crudely put, the semantic value of ‘is wrong’ could be the function, F, which gives  the value True for murder, the value True for breaking promises, the value False for singing, and so on. Of course, this function would have to be much more specific. It would have to take every possible individual act-token as its argument, and give a value of either True or False for each one of them. In this sense, given the disquotational property of truth, it seems as if moral semantics is closer to normative ethics.

The metasemanticists are, in contrast, interested in the question of in virtue of what does a given term have its semantic value. Thus far, much of the metaethical debates has been about how to answer this question. So, the expressivists might argue that the semantic value of ‘is wrong’ is the previous function F because statements which use this predicate conventionally express certain conative attitudes towards actions. Naturalists, in contrast, might argue that the semantic value of ‘is wrong’ is F because this would best explain our core uses of the predicate or because this best fits the platitudes which relate ‘wrong’ to the network of other moral terms. And, finally, the non-naturalists might claim that the semantic value of ‘is wrong’ is F because there is a class of actions which share the non-natural property of wrongness. The older metaethical debates assumed that these views in metasemantics are fairly neutral about the semantics of ‘wrong’ – i.e., about for which arguments the function gives the value True and for which the value False.

The new debate is between views that could be called invariantism, contextualism, and relativism. As I understand them, these are views about the shape or the structure of the function F. Usually these views are about ‘ought’ but I’ll continue to use ‘wrong’ here instead. According to invariantists, functions like F have a very simple structure like the one described above. All we need to describe this function is a name for each action, and the value True or False (with capitals!) for each one of them.

            The contextualists, in contrast, claim that we should add another argument-place to the F for the speakers who are talking about the wrongness of the given actions. This is because whether claims of the form ‘action A is wrong’ is True or False depends not only of the action but also about what information the speaker and/or others in the same context of discourse have. So, the function F* defended by a contextualist, might give the value True for the argument of giving a chocolate bar to a friend if the speaker knows that the bar is poisoned and the value False for the same argument if she lacks any evidence about this being the case. (I should add that there are interesting contextualists debates about who belong to the same context of discourse. Some contextualists also add a third parameter to F to denote the standards that are accepted in the context of the utterance).

            Finally, the relativists agree with the invariantists that there is only one argument place in F for the actions that are the arguments of the function. However, they reject the idea that the value of the function for each of its arguments is either True or False. Rather, the relativists believe that the value of the function F** for is for different actions either true-relative-to-the-context-of-assessment or false-relative-to-the-context-of-assessment. And, whether the value of F** is true or false relative to the context of assessment depends on the informational state of those who are assessing the truth of the given utterance. So, if we consider whether Mark’s earlier claim ‘what Ben did was wrong’ is true or false relative to our context of assessment this might depend on whether we know whether the chocolate bar he gave to Ann was poisoned or not. Thus, the very same proposition might get a different truth-value (with small letters) relative to different contexts of assessment.

            I hope I got those views at least along the right lines. Anyway, the point I wanted to make was that all these views seem to me to be metasemantically neutral. Whether F, F*, or F** is the real semantic value of ‘is wrong’, it still seems that expressivists, naturalists, and non-naturalists can still have the same disagreement about in virtue of what the predicate ‘is wrong’ has that particular semantic value. This is why I believe that the views in the new debate are metaethically neutral. However, it should be emphasised that these views are not normatively or morally neutral. Whichever function of the type F, F*, or F** is the real semantic value of ‘is wrong’ will determine what it is wrong for me to do in any given situation. So, this means that those who have been working in the moral theory should take an interest on how this debate goes.  

21 Replies to “Metaethics, Semantics, and Metasemantics

  1. Jussi, interesting and helpful post. (To me, anyway.) I remain embarrassingly muddled about these issues, but it seems to me that matters aren’t quite as tidy as you suggest. If you can point out to me how I’m wrong, that would be enormously helpful for my own research.
    (1) I’m suspicious of your characterization of the dispute between expressivism, naturalism, nonnaturalism etc. as metasemantic. While these positions might be associated with certain metasemantic claims, it seems to me that their essential claims are semantic (that is, if they make any claims about moral language at all, which many of them do not; e.g. many naturalists and nonnaturalists claim to be characterizing goodness and wrongness, not what we mean by ‘good’ and ‘wrong’, and so they would maintain that their theses are metaphysical rather than semantic or metasemantic.)
    It seems to me that what makes a view about moral language naturalistic, nonnaturalistic, or expressivist is not so much how it says the term gets its semantic value, but what it says that semantic value is (or what kind of thing it says it is). I’m not sure how to do this in terms of functions, so forgive me for translating it into talk of properties. Naturalists (if they make any semantic claims) maintain that the semantic value of a moral term is a natural property. Nonnaturalists maintain that the semantic value of a moral term is a nonnatural property. (For the most part, neither naturalists nor nonnaturalists bother telling us exactly which property–the former because they think this requires difficult empirical research, the latter because they don’t think that anything informative can be said–but this just means they are doing very vague semantics, not that they aren’t doing semantics at all). And expressivists claim either that the semantic value is something other than a property, or that moral terms don’t strictly have semantic values but some other kind of meaning (depending on what you take ‘semantic value’ to mean).
    (2) Let me pick up the function/property worry. I would hope that a semantic theory for a normative term could do more than merely provide a function, understood as a set of ordered pairs, (action, truth value). As you say, that would collapse moral semantics into normative ethics. As I see it, moral semantics aims to give an informative characterization of the function (i.e. identify the property) in general terms. Nonnaturalists and ‘nonreductive naturalists’, of course, deny that this can be done, at least without using other normative terms, whereas reductive naturalists like myself (and I think I can include Judy Thomson, Philippa Foot, and Gil Harman, for example) claim that it can. While this may seem like a mere technical worry, I think it has significant implications for the connection between metaethics and normative ethics. If identifying the semantic value for a moral term consists in identifying the property it picks out in general terms, rather than articulating a list of ordered pairs, then it doesn’t necessarily settle normative questions and doesn’t collapse into normative ethics: it remains to be determined which actions etc. have that property and which don’t. (Of course, given some semantic values, normative consequences follow pretty immediately. But given others–e.g. nonnaturalism–few if any normative conclusions follow without substantive normative enquiry.)
    Anyway, I’m interested to know what you think about this.

  2. Steve,
    I think I am symphatetic to your way of phrasing the debate – I just thought the Fregean way was clearer, served my purposes of illustrating the new debate better, and fitted a general, global semantic framework in a way I can understand. And, I agree that expressivists, naturalists, and non-naturalists do not only disagree about metasemantics but also about metaphysics, psychology and so on.
    Anyway, I think that I can rephrase what I said in a way that fits what you say. It could be that everyone accepts the claim that the semantic value of ‘wrong’ is the property of wrongness. The debate between invariantists, contextualists, and relativists would then be a debate about in what way, if any, this is a relational property – the having of which depends on the speakers’ or assessors’ informational state. And, again, the way in which this property is relational will have consequences for normative theories that try to capture which acts have this property.
    What the metaethicists seem to disagree about is still one level up. They could all accept *that the semantic value of ‘wrong’ is the property of wrongness*. However, they would disagree about how we should understand this claim. This matches the way you say that the metaethicists disagree about what kind of a thing the semantic value is. This I take it is again a metasemantic disagreement. The naturalist and the non-naturalist will understand this claim as a claim that ‘wrong’ refers to either the natural or non-natural property of wrongness, whilst the expressivist will understand it (a semantic claim made in the moralising practice) as an expression of attitudes. In any case, in whichever way we interpret the claim ‘the semantic value of ‘wrong’ is the property of wrongness’, this again will be neutral between invariantism, contextualism and relativism. In the function talk, this would mean that metaethicists not only discuss in virtue of what moral terms have their semantic values but also how we should understand the descriptions of the functions that give us the semantic values.
    About (2), I don’t want to collapse semantics into normative ethics either so I am symphatetic to what you say in the end. My main thought was just that the invariantism, contextualism vs. relativism is not a debate on the same level as the expressivism, naturalism vs. non-naturalism debate however we carve the semantics/metasemantics and normative ethics/metaethics distinctions. The latter debate seems to me to be in any case one level up even if there are difficulties in how to put that thought naturally.

  3. I think you have a good point, Jussi, but there’s one bit here I’m confused about. Does a contextualist really agree that the semantic value of ‘wrong’ is a property? I would have thought contextualism explicitly denies this. The semantic value is a function from contexts to properties, according to contextualism.
    Maybe the confusion arises because an expression may have two semantic values. According to contextualism, there is a primary semantic value (what you have to understand in order to be a competent user), and this is not a property but a character, a function from contexts to properties. Then on any occasion of use, ‘wrong’ will also denote a property, and this is also a semantic value: a content.
    Talk of a ‘relational property’ seems off base to me, though. Imagine someone reading Kaplan and concluding, ‘Oh, I get it, the word “I” has a semantic value, namely an individual person, only it’s a relational person.’ Not the best way to put it, to say the least.

  4. Thanks Jamie. I think you are right. In the original post, I tried to do with mere functions which is more natural to contextualism. In the response to Steve, I tried to quickly translate this to property-talk but you are right that this isn’t as straight-forward as I thought. I was thinking of the secondary semantic value as a property as this I take it is what the term as used in the context contributes to the truth-value of the whole utterance. But, you are right that the contextualist might say that primary semantic value is the mere character of the term.
    I also grant that the talk about relational properties is not the best way to put this. I was thinking more of the claims like ‘He is taller than me’. You might think that the (secondary) semantic value of ‘is taller than me’ is in Steve’s framework a property that is a relational property which makes the given claim true if the person I am talking about has it. And, this is a predicate about which it is natural to be a contextualist. I was thinking that the contextualist could think that moral terms work like this predicate even if this is not explicit from their form.
    But, you might be right that this is a bad way to think about contextualism. I hope the original posting’s way of putting things is slightly better. The main point is that in contextualism only relative to the information of the speakers does the utterance express a thought which can be True or False – and yet this idea is neutral about the metaethics of the resulting thought.

  5. Sorry. Thinking this through more. I think you are right that the contextualist needn’t say at all that the properties that are secondary semantic values of moral terms are relational. Just the fact which these properties are is relational. This makes it all the more easier to say that these properties can be either natural or non-natural or the talk about them expression of conative attitudes.

  6. Hi, Jussi,
    I just wanted to add a couple of small points about your characterization of contextualism. You’ve got two different formulations in your original post. (I’m going to put your point in my terms for reasons that I hope will become clear.) On the one, normative modals have truth-values relative to an informational parameter and on the other, there are two parameters, one selects information, the other a standard. It seems to me that there are two parameters and that we need to allow the first parameter to sometimes take circumstances as a value. (The reason is that if normative modals are always relative to information, we won’t be able to capture their intuitively objective uses.) So at least ONE contextualist isn’t captured by your characterization. (I also *think* that in Kratzer’s original proposal, the first parameter only took circumstances. If so, her proposal wouldn’t be captured either.)
    The second point is that I don’t think contextualism is best characterized as having an argument-place for a name of an action. I don’t know of any contextualist whose view could be characterized that way, but maybe there is one in metaethics somewhere that does that; I’m mostly familiar with contextualism about modals in the philosophy of language and linguistics. There the default contextualist view is that modals are quantifiers over possibilities. The two parameters restrict or restrict and order the possibilities in the modal’s domain and we evaluate the whole modalized proposition for truth by seeing whether the prejacent is true in some, all, the best, etc. of the possibilities in the domain. (The prejacent is roughly, the proposition that is getting modalized, e.g. that water is H20 in “necessarily, water is H2O” and that you don’t lie in “you must not lie”.)

  7. Hi, Jussi.
    Sorry I’m late to the discussion. I agree that the semantic/metasemantic distinction is important, but I also don’t think that ‘traditional’ debates in metaethics are all so cleanly metasemantic as you suggest. True, Brink and Boyd and Jackson are doing metasemantics in the course of defending their various brands of naturalism. But that doesn’t mean that naturalism itself is a metasemantic thesis, or even that you need to do metasemantics to defend it – it just means that in the course of the particular views that these philosophers were defending, they got engaged with metasemantic questions. As Steve notes, the variety of naturalism that he has been defending has led him to engage directly with first-order semantic questions. And for myself, I’ve defended a kind of naturalism without defending a view about either semantics or metasemantics.
    I also don’t think that expressivism is well classified as a metasemantic view. It’s rather a quite different kind of semantic theory – so different from the other kinds of semantic theory that you discuss that your way of saying what makes something a semantic theory doesn’t even apply to it.

  8. Thanks Janice. These are good clarifications. I don’t think either of them challenge the main point of the post so I can be fairly symphatetic to them.
    I worry about adding the standards as another parameter. This seems to me to lead to bad first-order relativism which is objectionable in many ways. And, thus, I’m not sure what need would there be for contextualist parameters in the uses where truth was not relative to information. Why wouldn’t we just give an invariantist account of those uses?
    The second point: well, this is an issue of a big disagreement in the current metaethics. There are people like Ralph Wedgwood who defend the very model you describe. On his view, ‘ought’ functions as an operator that takes propositions in its scope. An ought claim is then true if the proposition within the scope is true in the modally ideal world of the relevant set of worlds that is picked out by the context-sensitive parameters. This, I take it, is the semantic picture of the standard modal logic with a contextualist flavour.
    However, there are people who resist this picture. They think that ought is not an operator but rather a two-place predicate that corresponds to a normative relation between the action and the agent mentioned in the sentence. If this alternative view was right, then it seems like the semantics of ought would match the simple Fregean picture I gave. Ralph does argue well against this view but his view is also problematic. It creates modal propositions we have no way of understanding in ordinary language.
    I think I can make the same point though by using the picture of contextualism you give. If the contextualist parameter about the standards accepted in the context of discourse both restricts and ranks the possibilities so that this determines the truth-value of the proposition, then I admit that this isn’t metaethically neutral. Contextualism collapses into metaethical subjectivism. This is a metaethically problematic view so this would be bad for contextualism.
    To avoid this, the moral contextualist can agree that the information parameter only picks out the relevant worlds but which one of them is the best is independent of the standards the speakers accept but rather dependant on the features of that world (I think Ralph accepts this when he talks about correctness). But this better view is metaethically neutral as there will be an expressivist, naturalist, and non-naturalist reading for the talk about the modally ideal world where the relevant propositions have to be true in order for the modal claims to be true.

  9. Thanks Mark. I think you are right that I might have overstated what I had in mind. I was thinking that once we have given the semantics of moral terms there will be different explanations for how to understand these semantics and how the moral terms came to have the semantics they have. And these explanations can either be naturalist (and as you say there are many naturalist stories), non-naturalist, or expressivist. But, of course people who are naturalists, non-naturalists, and expressivists have also been interested in other questions and not taken part in the metasemantic debates.
    I was also thinking about expressivism after I wrote the entry. I was sort of reporting how some of the expressivists I talk to seem to understand their own view. They seem to accept the claims of the truth-conditional semantics but give an expressivist reading for them. And, they are aware of Dreier’s point that they have to give a further story about how those claims are to be understood as they are complex claims.
    But, I know they might be wrong about how to understand their view best. I was thinking that in the picture of expressivism you give in Being for there would still be distinct semantic and metasemantic element. Maybe we could think of the characterisation of the expressed attitudes you give as the semantics of moral terms (something perhaps corresponding to the semantic values of the terms) whilst the story about expression relation in terms of the norms of assertion would give the metasemantics of this view. What I take it is the message in your book is that metasemantic story of expressivism is not semantically neutral.

  10. Hi, Jussi,
    Just a couple of other quick clarifications about contextualism that I don’t think affect your main point. (I’m not sure that I agree with your main point, but I don’t want to focus on that.) Sorry to go on about this, but I’ve found that there is a lot of confusion about what contextualism is among folks not familiar with the literature in phil language or linguistics on this topic and I think it’s important to get clear on what the view is.
    “Contextualism” is the name of a kind of semantic theory about modals according to which they have a certain formal features that include parameters for contextually determined values. Contextualists can disagree about what parameters there are to get filled. Having said that, if we follow the literature on modals in linguistics, we’ll say that to count as a contextualist view, a semantic proposal must hold that sentences containing modals have the characteristic formal features. Ralph’s view is not an instance of contextualism in the linguist’s/philosopher of language’s sense, since he thinks that “ought”s are sentential operators. The view I’m discussing takes them to be quantifiers. (some familiar examples: “not” is an operator, “every” is a quantifier.) Treating modal expressions as quantifiers is a hallmark feature of contextualism about modals as it is understood in phil language and linguistics and since it is also the canonical semantics for modals in both areas, I think it worth following that usage. (This is also how the term is used in the literature on epistemic modals and since linguists and philosophers of language think that the epistemic and normative modals are best given a unified account, that’s another reason to prefer this usage.)
    A second clarificatory point: there’s a common, incorrect assumption that the ‘ought’s in question are always moral, so that a contextualist about ‘ought’s who accepts a second, standard parameter is required to hold that it is a varying moral standard that gets contextually selected. But there are other sorts of ‘ought’s and the contextualist is best understood as aiming to provide a unified account of all of them. So, the standard needn’t be moral. It may be legal or prudential, for example. Given this, it is open to a contextualist to hold that, when used morally, ‘ought’s always select an absolute, moral standard in a context of utterance–that’s what makes them moral ‘ought’s. So, contextualism about ‘ought’s need not collapse into moral relativism. (I’m not sure what you mean by “metaethical subjectivism”. Is that an instance of metaethical relativism that holds the standards are the speaker’s? If so, then the same move blocks contextualism’s collapse into that too.)
    I’m not sure why you think that a standards parameter would make contextualism collapse into relativism. Is it because you’ve been making that assumption? Also, I’m not following your point about invariantism.
    Anyway, I’m straying badly from the main thread of your post, but I wanted to clear up an confusion about one of the views under discussion.

  11. Of course, you can use the word ‘contextualism’ as the name for the specific semantic theory about modals that take them to be operators. I’m not sure though I see why we should. There is a more general sense of contextualism. According to Stanford Encyclopedia, epistemic contextualism for instance “is the view that the proposition expressed by a given knowledge sentence (‘S knows that p’, ‘S doesn’t know that p’) depends upon the context in which it is uttered”. And, not all, contextualists in this sense think of the term ‘know’ as a quatifier over possibilities. To generalise this definition of contextualism, you can think that one is contextualist about a range of statements, if one thinks that the propositions they express depend upon the context in which it is uttered. I do much prefer this more inclusive definition of contextualism that covers the contextualists in your sense and many others who are contextualists too. One advantage in metaethics is that you can use contextualism to give a unified account of talk about good and bad, right and wrong, excellent and evil and so on (in addition to oughts) even if it’s hard to see how these could be quantifiers.
    About the second point. Very few metaethicists ever talk about the moral oughts whatever they are but rather about the generic overall ought (most of them don’t even think that there are different kind of oughts – just different grounds for what you ought to do). There is a good reason for this and for thinking that this ought does not take a standard as a context sensitive parameter. Take all the standard relative oughts in the situation: the moral oughts, legal oughts, rational oughts, prudential oughts, and so on. There is still a question of what you ought to do, all things considered. What standard is this ought relative to? If you say that it is relative to some standard x, all you get is new x kind of ought that you can add to the previous list. And, then you can ask again, what you ought to do overall. At some point, you’ll have to stop the regress.
    It’s true that contextualist views don’t collapse into relativist views if the standard do not need to be the standards which the speakers accept. But, then I’m not sure I see how the parameter would be context-sensitive.
    The point about invariantism was that you said that there are objective uses of oughts and that contextualism as I phrased it cannot capture them. I was wondering what motivation there is to be contextualist about these uses in the first place.

  12. Oh yeah. I was meant to add that it doesn’t seem inconsistent to think of oughts as both operators and quantifiers as this seem to be on a different level. Think of standard modal operators possibly and necessarily. We call these operators as they take propositions within their scope. I take it that this is their syntactic role. Yet, in semantics, we do take them to be quantifiers over possible worlds. It seems to me that the structure of Ralph’s view is like this and so compatible with the definition of contextualism in linguistics.

  13. Hi, Jussi,
    You’re right that “contextualism” gets used in different ways in different areas of philosophy. My thought is that because it’s generally agreed that ‘ought’, ‘must’, and ‘may’, unlike “knows”, are modals, we should stick with the term as it’s used in the literature on modals generally.
    Another reason is that linguists agree that modals should be given the most unified account compatible with the data, so that what we have is a bare modal expression that is sometimes used epistemically, sometimes alethically, sometimes deontically, etc. It would be weird to not have a single name for that view, but have a name for each of the different uses. (We don’t, for example, have different names for utilitarianism, depending on the area to which we apply its principles.)
    An aside about unity: Why is this important? Because contextualism and its rivals are descriptive semantic theories and as such they are beholden to the empirical data. Unity and simplicity are important methodological criteria of adequacy among linguists doing descriptive semantics for the same reason compositionality is: descriptive semantic theories are to figure in explanations of usage by competent speakers; the more complex and unwieldy the theory is, the less plausible it is that the meanings the theory assigns to the expressions are ones that ordinary speaker’s grasp. Given this, anyone working on descriptive semantics for an expression can’t ignore either usages of what linguists hold to be the same expression or the semantics of expressions linguists hold to be related.
    Of course, if you use “contextualism” just as a thesis about modals when used to make moral claims, then the proposal I offered doesn’t have a shifting standard parameter. But I’m not using ‘contextualism’ in that way. On my contextualist view, ‘ought’ is sometimes used to express a legal requirement, e.g. “visitors must stop at stop signs”. This is the same ‘ought’ as in “Doctor ought to prescribe Z” and “one ought to help others when one can”. By ‘same’ I mean they make the same contribution to determining a proposition on an occasion of use. What makes the claim expressed in the first example a legal one and in the second two moral ones (the first of the two so-called ‘subjective’, the second ‘objective’) is contextual modulation. What distinguishes the first from the second two, is a shifting standard parameter. What distinguishes the second two from each other is a shift in the information/circumstances parameter, with the first taking information, the second circumstances as a value.
    So, what about the reason you offer for thinking that there is a generic overall ‘ought’? Well, I think we need a better reason to think there is one than that, since I myself am not sure there is a more general question of what you ought to do. I’d like to have the content of that question explained to me. Here is one thing you could mean: You could mean to ask what you ought to do, according to the standard that is most important to follow? You’d need a way of determining which standard that is, but that would be a metasemantic and metaphysical issue. Most importantly, the ‘all-in “ought”‘ would still be context sensitive. I myself don’t know how else to understand that question.
    Also, I have a question about what the generic ‘ought’ view applied to the examples above would say. Are they all generic ‘ought’s? That wouldn’t fit with ordinary speakers’ intuitions, I’d expect. It would also predict that certain uses would express contradictions that don’t. Suppose we’re talking about the traditions of some tribe regarding selection of their next chief. The man that would be selected by following the traditions we agree would be a very bad man. It seems that I can felicitously say “Well, they must pick him, but they really shouldn’t”. Saying that is felicitous if we’ve got different values for the standard parameter selected in the first and second conjunct, but not if we’ve got a single, generic ‘ought’ in both.
    Last point: You may well be right about Ralph’s view. I’d taken him to be offering a semantic proposal, but you’re no doubt more familiar with his view than I am.

  14. A question for Jan: Why do you say ‘knows’ is not a modal? One might have thought it was because, for example, in epistemic modal logic, the box symbol is sometimes interpreted as something like ‘It is known that’.

  15. Hi, Campbell,
    I don’t know enough about how ‘it is known that…’ functions in epistemic modal logic to quite follow your line of thought. Could you say a bit more?

  16. Well, to be honest, I’m not sure what it means to say of a word that it is ‘a modal’. But I was thinking it might mean something like: it’s the sort of thing which can be represented by an operator in modal logic. If that’s wrong, perhaps you could just explain what makes a word a modal.

  17. I see. I guess I think there are two different ways you could use ‘a modal expression’. One way would be as you suggested, as a indefinite description for any term that is amenable to treatment in a modal logic. That’s seems fine, but leaves it open whether there is any natural language that contains any particular modal term in that sense. Then I see that “so and so knows that such and such” may turn out to be a modal expression in that sense.
    I was using that term, though, as a description for an expression in a natural language whose semantic features are best understood as functioning as a quantifier over possibilities. I guess I was expressing doubt that this was so for “so and so knows such and such”, but I don’t have a compelling argument for that. I wonder if linguists have any work on that. If I were to try to develop a descriptive semantics for ‘knows’, that’d be where I’d start. I should add, though, that I’ve looked at some of the linguistics literature on modals and I’ve never seen anyone include ‘knows’ on their list of modal expressions. Looking at what the SEP entry has to say on epistemic logic, I see that it reports that Hintikka has proposed a formal treatment for “knows” according to which c knows that a is true iff a is true in all the worlds compatible with what c knows. That would give ‘knows’ the same truth-conditions that solipsistic contextualists give to speaker c’s assertion of “must a” (where the modal is being used epistemically, not deontically). I’m not a solipsistic contextualist, but I do think “must a” has solipsistic uses. But I don’t hear those uses as saying the same thing as “c knows that a”. Do you?
    Just glancing at the Hintikka proposal, it looks to me very like, in some respects, the way Stalnaker in “Inquiry” represents an agent’s belief state. This way of understanding the proposal makes a lot more sense to me. On this understanding of
    “c knows that a iff in all possible worlds compatible with what c knows, it is the case that a”
    the RHS doesn’t give the truth-conditions for a use of the expression on the LHS. Rather, the RHS specifies the conditions under which c counts as knowing a. That would give epistemic modal logic a role in theorizing about mental states and the norms governing them, without taking on any semantic commitments which, to my ear and limited knowledge of the linguistics literature on modals in English (and German), seems implausible.

  18. Hi, Jussi.
    I agree that on an expressivist semantic picture like that discussed in Being For, there is still a semantic/metasemantic distinction – we just need a little bit more abstract characterization of what that distinction is.
    Also, I do think that it’s a mistake to understand expressivism as a metasemantic theory in the sense that you articulated. If I give a semantic theory, and then give a theory about the semantics of the language that I used to give my original semantic theory on which it received an expressivist interpretation, then that’s a kind of ‘meta’-semantics, but it’s not the foundational theory of how words come to have their meanings that Boyd and Jackson (for example among metaethicists) and others are engaged in.

  19. Jan,
    Thanks, that’s helpful. But there’s still something I don’t understand. It may be, as you suggest, controversial whether ‘knows’ is well analysed modally, i.e. in a way which involves quantifying over possibilities, but I would have thought the same goes for ‘ought’. Deontic logic has its detractors (see this discussion on PEA Soup). So I don’t see why it would be uncontroversial that ‘ought’ is a modal, but not ‘knows’.

  20. Hi, Campbell. I’m not sure I see the connection between deontic logic and the claim that in English “must”, etc. are best understood as quantifiers over possibilities.
    But I don’t think your question depends upon there being an interesting connection. You’re thinking: if being a modal requires being a quantifier over possibilties, then shouldn’t we understand *some* of the disputes about how to understand the semantics of “must” etc in English as a dispute about whether they’re modals?
    I hedged this a bit by saying that “modal” is a term for an expression “best understood” as a quantifier, etc., but that’s not quite fair to those who don’t think “must” etc. are quantifiers. So, let me be a bit more careful: I think “modal” gets used among linguists and most philosophers of language as an expression for terms that it agree are canonically held among experts to be quantifiers over possibilities. Then we can raise the question about whether they are really so best understood. (In some of my posts about flexible contextualism, I said a few things about why that’s the canonical view about “must” etc.) I don’t think that’s the canonical view among linguists about ‘…knows…’.
    I guess the punchline is that, since we’re talking about descriptive semantics and so need to have a look at what linguists say about various expressions, it seems to me our discussion will proceed most easily if we adopt their usage for “modal”. I take it that usage treats “modal expression” as a term for expressions that are canonically thought in linguistics to function as quantifiers. That would keep “…knows….” off the list, but still allow for debate about whether there is good linguistic evidence to put it on the list.
    Remember that this issue arose because of another terminological one, i.e. how to use “contextualism”. Since both issues are terminological, one important standard for settling them, I’d have thought, is best fit with most general use (esp among experts). That’s the one I’m relying on. I’m not going to call the cops if someone wants to use the expressions differently, but there’s an increased risk of miscommunication, I think.

Comments are closed.