When I
first met Jonathan Dancy three years ago, I remember him mentioning the early
emotivist papers in Analysis. One night year and a half ago, I was working at
library and things weren’t really going anywhere so I thought I’d check those
out. I recall finding interesting things and thinking about writing something
about them. Didn’t happen. I got the same idea when I was at UNC this year, but
it turned out that they didn’t have the first issues of Analysis at the
library. So, when I got back toEngland,
I thought I’d check those papers again. To my horror, those journals were not
there anymore. Thankfully, the librarians were friendly enough to get them from
the cellars. Here’s some things that got my attention long time ago.

The very
first issue of Analysis from 1933 has W.H.F. Barnes’s paper ‘A Suggestion about
Value’ (pp. 45–46). This paper occasionally referred to is a very clear
statement of emotivism. Here’s the first paragraph of the paper:

‘Value judgments
in their origin are not strictly judgments at all. They are exclamations
expressive of approval. This is to be distinguished from the theory that the
value judgment, “ A is Good, “ states that I approve A. The theory that I am
now putting forward maintains that “ A is good,” is a form of words expressive
of my approval. To take an illustration :
¾When I say “ I have a pain,” that
sentence states the occurrence of a certain feeling in me : when I shout “ Oh !
“ in a certain way that is expressive of the occurrence in me of a certain
feeling. We must seek then for the origin of value judgments in the expressions
of approval, delight, and affection, which children utter when confronted with
certain experiences.’

All this
sounds very modern. A distinction is made between subjectivism and
expressivism. Things then turn slightly for the worse. Barnes adds that value
judgments ‘will only posses meaning in so far as the society in which they are
used is agreed on what things it approves’. Thus, they have ‘meaning by
referring to the actual nature of the thing, not to any non-natural quality it
possesses’. Denial of non-naturalism is of course well and good but that sounds
a lot like relativism. He then says that disagreements about value are settled
by saying ‘I like it and you don’t, and that’s the end of the matter’. He is
right that that goes for food and drink but I wonder about morality.

The last
paragraph of the paper picks up a bit. He says that opposition between the
approval of one man and that of others lies at the bottom of controversies
about value. What happens is that we point out details in the good objects
which are the targets of our approval and hope that others realise that they
approve these features too and come to accept that the object is good. This
sounds better. I cannot find an argument though from the paper for the view. I
wish Analysis would still accept such papers.

There is
another very interesting paper in the same first issue by C.A. Mace entitled
‘Representation and Expression’. Mace argues against Richards’s distinction
between scientific and emotive uses of language. He wants to replace this
distinction by the distinction between representative and expressive functions
of sentences. The latter is to express a state of mind – whether a belief or an
emotion (this reminds me of Mark S’s arguments against such a view). He also
claims that the representative function of the sentence inherited from the way
in which the state of mind, expressed by the sentence, exhibits a reference to facts.
Unfortunately, he doesn’t say anything about ethics.

In the
second issue of Analysis from 1934–1935, G.R.T Ross gives ‘Another Suggestion
about Value’. Ross’s criticism of Barnes is not wholly convincing but in any
case his view is that ‘A is good’ means ‘ I like it ; and you will’. He is
quick to realise that proving such a prediction to be true of a general
audience will be impossible.

early emotivist views had of course already been put forward by Ogden and
Richards in their 1923 The Meaning of
(and perhaps by Mill in his System
of Logic
). Later on Ayer’s 1936 book Language,
Truth and Logic
and Stevenson’s 1937 paper ‘The Emotive Meaning of Ethical
Terms’ became the classic presentations of emotivism.

The first
reference to the so-called Frege-Geach problem is usually to Geach’s 1958
Analysis paper ‘Imperative and Deontic Logic’. On page 54, he has the following

arises here a difficulty for what may be called performatory theories of the
predicates “good” and “true” – that to predicate “good” of an action is to
commend it, and to predicate “true” of a statement is to conform or concede it.
For such predications may occur within “if” clauses; the predicates “good” and
“true” do not then lose their force, any more than other predicates used in
“if” clauses do; but “if S is true” is not an act of conforming S, nor “if X is
good” an act of commending X’.

Geach went
on to develop this argument in his 1960 paper ‘Ascriptivism’ and 1965 paper
‘Assertion’. In addition, John Searle brings up a similar problem in his 1962
paper ‘Meaning and Speech Acts’. Now, I’ve always thought that this would have
been a slow reaction to a view that was presented decades ago. Usually, if
there is something obviously wrong with a view, philosophers are pretty quick
to react. Funnily enough, I’ve think I’ve found the same problem from an
earlier paper in the early issues of Analysis I’ve been reading. This is
probably common knowledge in some circles but I haven’t seen it mentioned

The paper is from the 4th issue of Analysis published in 1936. It’s by H.B. Acton entitled ‘The
Expletive Theory of Morals’ and basically it is just an objection against
Ayer’s emotivism in Language, Truth and
published the same year. Here’s the argument in the first paragraph:

‘Mr. Ayer,
in his account of what might be called the expletive theory of Morals (Language, Truth and Logic ch. vi),
argues that statements containing normative ethical, [sic] terms, such as “
murder is wrong”, are not really propositions at all, and so cannot be true or
false. He also holds that casuistry is possible, and that casuistry is “ a
purely analytical investigation of a given moral system ”. Hence he admits that
casuists deduce propositions from propositions. Either, then, Ayer is
inconsistent, or else no normative statements such as “murder is wrong” can
form part of casuistry as understood by him. For in order to make deductions
you must have propositions from which the deductions are made.’

I find this
argument pretty neat. Ayer accepts practical ethics which is based on
inferential relations between different elements in the ethical theory. Such
relations can only be based on inferential relations between propositions. But,
Ayer has denied that there are moral propositions – moral utterances express
pro-attitudes. Something has to give.

Actoneven gives an example of a valid moral inference:
‘Deliberate taking of human life, except in lawful war, or in self-defence, or
in the carrying out of the legal punishment, is wrong.’

2. ‘”Therefore
it is wrong for a doctor to kill an unborn child in order to preserve the
mother’s life.” (for such a doctor is not acting in self-defence, or lawful
war, or as executioner.)’

He says
that according to Ayer’s view neither the premise or the conclusion is a
proposition and therefore 2 cannot be deduced from 1. And, here comes the

‘Now it
seems to me quite possible that, since 1 and 2 do look like propositions, and
since, further, we can, by making
suitable assumptions, deduce the one from the other, Ayer’s theory of what
constitutes meaningfulness is defective. That is, it seems more likely that 1
and 2 are propositions, than that
Ayer’s theory of meaning is true’. To me that looks like the main idea of some
of the later versions of the Frege-Geach. Are the any other predecessors? 

22 Replies to “The Proto-Frege-Geach Problem

  1. Ross gives a clear statement of the Frege-Geach problem in 1939 in the Foundations of Ethics (his target is Carnap). I suspect that you might find an even earlier statement of this if you look through some of Ross’s papers leading up to the Foundations.

  2. Nice post, Jussi. Well-worth investigating. It doesn’t sound to me like Acton’s argument gets to the heart of what I understand the Frege-Geach problem to be, but I never knew about it, and it’s certainly in the neighborhood of how the problem is often understood (namely, as specifically about inference, rather than as about embedding).
    If you pay attention to what Hare is doing in the early chapters of The Language of Morals when he discusses conjunctive and negative commands, though, he’s clearly responding to the worry that if moral language is imperative-like, moral sentences won’t have the right sorts of logical properties. So I take it that pre-1950 it must have been well-known to be an obstacle to noncognitivist views that they need to explain how moral sentences can have the appropriate logical properties.
    On other sources: Urmson claims to trace the early history of emotivism in the English-speaking tradition in The Emotive Theory of Ethics and I don’t have it with me, but I’m pretty sure Barnes came up. I think he also mentions some things that you didn’t find – for example, I seem to recall that there’s a reference to some indirect discussions of some papers that were never actually published. One gets the idea that emotivism had been quite a hot topic for four years or so before Ayer published L,T&L.
    I also recall that Urmson was careful to hedge that he was only tracing the references in the English-speaking tradition – I have it third or fourth-hand, for example, that there are supposed to be clear statements of emotivism by scandinavians from as early as the late 19th century, a claim I’m not qualified to investigate, though I’m happy to spread rumors…

  3. I agree this is very interesting and just took a look at my Urmson. Interestingly, he does mention the Barnes and says that paper is “apparently a fragment of a paper on Hartmann’s Ethics read to the Jowett society…”
    Urmson also claims that the first full-fledged expression of an emotivist theory is Broad’s 1934 paper to the Aristotelian Society (in which the theory is attributed to A.S. Duncan Jones)

  4. I a similar vein, I ran across this interesting article a while ago (it mentions some of the European background):
    The Theory of Value and the Rise of Ethical Emotivism
    by Stephen A. Satris
    Journal of the History of Ideas, Vol. 43, No. 1. (Jan. – Mar., 1982), pp. 109-128.
    JSTOR link (had to split it to get it to show up):

  5. Hi Jussi. Thanks for the great post. About those rumors…
    The earliest noncognitivist accounts I’ve seen are from Axel Hagerstrom and some of his Swedish contemporaries. Robert Sandin’s 1962 “The Founding of the Uppsala School,” in the *Journal of the History of Ideas*, places the development of Hagerstrom’s views between 1911-1917. And I might be making this up, but I think I remember (it was about eight years ago, now) reading in some of Hagerstrom’s work on legal language about problems of deriving specific legal claims from more general legal principles, if the general legal principles were merely exclamative- or imperative-like. If it wasn’t Hagerstrom, it was one of his contemporaries. At any rate, Hagerstrom and others in the Uppsala School were certainly interested in the logical features of moral and legal language and of the attitudes expressed by utterances of moral and legal sentences.

  6. Brad, thank you for the Satris reference. That article contains a wealth of information.

  7. The Scandinavian Mark S mentions must be my compatriot Axel Hagerstrom (there should be two dots above the ‘a’ in the surname). In his inaugural lecture as professor in Uppsala 1911 he presented a view very similar to what eventually came to be called ’emotivism’. I suppose there is an English translation of this lecture, the title should be something like ‘On the Truth of Moral Beliefs’. Hagerstrom probably took his cue from some German philosophers. In particular disciples of Brentano, such as von Ehrenfels. On another historical note: someone rightly pointed out above that Broad outlined an emotivist view in 1934. Broad mastered Swedish (the no 1 philosophical language), spent some time in Sweden (Uppsala, in fact) and was perhaps influenced by Hagerstrom.
    Another point: From what Jussi reports it seems to me that Acton did not identify the Frege-Geach problem. An emotivist could say that someone who accepts (1) in Jussi’s post but does not accept (2) has inconsistent attitudes. In contemporary jargon, such a person violates supervenience, which, the emotivist might say, is a conceptual feature of moral judgements. To explain what goes on when moral sentences occur in embedded contexts is trickier.
    Finally, someone mentioned an article by Steven Satris. He also has a book on the history of emotivism, which is not bad at all (published by Kluwer, I believe).

  8. Mark K says Ross was responding to Carnap. The Carnap that I’ve found where he advocates a version of noncognitivism is Philosophy and Logical Syntax, from 1937 at around page 28.

  9. Thanks for the nice comments everyone! And, for the further info. Very interesting.
    I want to defend the Acton problem, as I shall call it, a bit. I’m not sure the story about the contradictory attitudes does the trick. One problem is that it’s not clear why having inconsistent attitudes is a *logical* failing in the way that failing to infer (2) from (1) (it’s not even clear what the inconsistent attitudes are in this case). If I boo killing in general (with 3 exceptions) but I don’t want to boo doctors killing, I’m surely odd but hardly quilty of accepting a logical contradiction.

  10. Jussi,
    I grant you that the emotivist cannot say that accepting (1) and not accepting (2) amount to a logical contradiction in the strict sense. I merely meant to say that the emotivist can say that anyone who accepts (1) and does not accept (2) is misusing moral language, in the sense of violating a constraint (supervenience) that is plausibly seen as conceptual.

  11. Mark S is right strictly speaking that the Acton problem isn’t quite the Geach problem as the latter explicitly concerns embedding. But it’s surely at its heart. To make sense of inference we need a story about consistency and inconsistency. Given such a story we can then tell a story about negation. And negation is right at the heart of the F-G problem, as Unwin’s important papers illustrate. (If we want enough logical regimentation to do all of propositional logic after all, in principle, we need just negation and conjunction and no one thinks conjunction is terribly hard.)
    My suspicion is that the heart of the issue is to articulate and motivate an account of exactly what the point is of the introduction and elimination rules for “not” (or perhaps a la Dummett for “absurdity”) for a language that includes things that are not strictly speaking assertoric and not strictly speaking propositions. If we had that, I suspect, everything else would just be bookkeeping. More or less. Ish.
    Apropos the pre-history of the Geach problem, someone once pointed out to me that there is another pre-Geachian statement of more or less it in Reichenbach. Unhelpfully I don’t have the reference though I’m pretty sure it was in Elements of Symbolic Logic.

  12. Hi, Jimmy.
    I disagree with your claim that ‘no one thinks conjunction is terribly hard’. I think it’s just as hard as negation, if you pay attention to the data. The reason it is often thought not to be as hard is because it is thought to be okay to conflate believing a conjunction with believing each conjunct. But belief doesn’t really agglomerate over conjunction. You can believe P and believe Q without believing P&Q. So we can state Unwin’s problem for conjunction just as well as for negation: there are three places to insert an ‘&’ in ‘Jon thinks that murdering is wrong’:
    &1 Jon thinks that murdering is wrong and thinks that stealing is wrong
    &2 Jon thinks that murdering is wrong and stealing is wrong
    &3 Jon thinks that murdering and stealing is wrong
    Assuming without loss of generality that our expressivist thinks ‘wrong’ sentences express disapproval, these would get treated as:
    &1* Jon disapproves of murdering and disapproves of stealing
    &2* ?????
    &3* Jon disapproves of murdering and stealing
    Now, if you’re willing to collapse &1 and &2, then you can collapse &2* to &1*, and claim to get all of the relevant distinctions. Some people are stuck with this anyway, for example, in most kinds of possible world treatments of the objects of the attitudes, so in a sense, it doesn’t seem as bad as making the corresponding conflation between ‘Jon does not think that murdering is wrong’ and ‘Jon thinks that murdering is not wrong’. But it’s still a conflation.
    The moral is that Unwin didn’t really show that negation is ‘at the heart’ of the issue; what he showed is that it is a special case of the issue – and he gave us a particularly perspicuous way of seeing just what the issue really is. Moreover, though you’re right that all you need for propositional logic is negation and conjunction, things don’t stop there; the embedding problem is completely general. For example, there’s exactly the same problem for quantifiers:
    E1 There is something that Jon thinks is wrong
    E2 Jon thinks that something is wrong
    E3 Jon thinks that murdering something is wrong
    And the same problem for tense:
    P1 Jon thought that murdering is wrong
    P2 Jon thinks that murdering was wrong
    P3 Jon thinks that having murdered is wrong
    And also for ‘necessarily’:
    N1 Jon necessarily thinks that murdering is wrong
    N2 Jon thinks that murdering is necessarily wrong
    N3 Jon thinks that (necessarily murdering) is wrong
    In every case, what expressivists need in order to solve the embedding problem, is an account of the thought ascribed to Jon in the ‘2’ sentences. But in every case, this is distinct from what is going on in the ‘1’ and in the ‘3’ sentences, and in every case, the ‘1’ and the ‘3’ sentences are the only ones that fall out of the expressivist treatment. The problem is that as they are standardly understood, with normative predicates corresponding to attitudes toward the subjects of the sentence, the expressivist analogue of ‘Jon thinks that murdering is wrong’ doesn’t have enough structure to generate readings of the ‘2’ sentences.
    I agree, of course, that if you can get an adequate treatment of the case of negation, that will lead to an adequate treatment of the others, at least for a simple normative language. But that’s not because negation is the main problem; it’s because it’s subject to the same problem as everything else, so you can’t have solved the problem with negation unless you’ve solved the more general problem. (I also agree that negation is a good thing to start by thinking about, since it is simpler than the other constructions – which I take it was part of Unwin’s point.)

  13. The more I think about Acton’s paragraph above the more I’m starting to think that even when he doesn’t explicitly state that the problem is to do with embedded contexts this is what he has implicitly in mind as a part of the problem. If casuistry is analytical investigation of a given moral system, then that moral system is in an unasserted, embedded, conditional context. Doing that work doesn’t require that the casuists have accepted the system or that they are expressing the relevant positive and negative claims when they are discussing the validity relations between the claims of the system. As Acton points out, for Ayer, in such context all the sentences are meaningless and in this case they are not even being used as exclamations. But, still the investigation is supposed to reveal inferential connections between the sentences. So, I’m starting to think he was pretty good on this. I might be reading too much into him though.

  14. Hi Mark,
    My sociological claim that no one thinks conjunction is terribly hard I now withdraw in the face of decisive refutation by counterexample.
    The philosophical thought behind it I am not so sure about. Here is the Unwin triptych:
    N1: It is not the case that Jon believes that O(p)
    N2: Jon believes that it is not the case that O(p)
    N3: Jon believes that O(not-p)
    Which Unwin translates:
    N1*: Jon does not hurrah p.
    N2*: ???
    N3*: Jon hurrahs not-p.
    The challenge is to characterize the thought Jon is supposed to have in case 2. With which we are to compare the Schroeder triptych:
    &1: Jon thinks O(p) and Jon thinks O(q)
    &2: Jon thinks O(p) and O(q)
    &3: Jon thinks O(p&q)
    &1*: Jon hurrahs p and Jon hurrahs q
    &2*: ???
    &3*: Jon hurrahs (p & q)
    What expressivists need, you say, to solve the embedding problem, is an account of the thought ascribed to Jon in the 2 sentences. But this problem looks a lot more tractable in the case of &. There is an intimacy between &1 and &2 that is quite absent in the case of N. For what we want to say Jon thinks in the &2 cases straightforwardly entails and is entailed by what Jon thinks in the &1 cases in virtue of the introduction and elimination rules for &. And that, we can then go on to say, is just what the word “and” (at least qua propositional connective) means. So the state Jon is in in &2 is just the state that plays that inferential role in his thinking. Of course that raises various further issues but the key point is that this sort of strategy is just unavailable with the N cases given the gulf that divides what we say about what Jon thinks in N1 and N2.

  15. Hi, Jimmy.
    I think that when you say, ‘There is an intimacy between &1 and &2 that is quite absent in the case of N’, you fit right into my diagnosis, which was that conjunction seems less hard because it seems easier to conflate &1 and &2 than to conflate N1 and N2.
    I do agree with you that if expressivists want their view to work, maybe they should be willing to make some conflations and settle for some not entirely satisfactory results. In fact, when I was first thinking through the negation problem, before I really understood how it worked, I thought about it in just this way – that if my idea about negation worked, then I could put it together with the conflation of &1 and &2 to get propositional logic.
    And if I set lower standards for expressivism, on the grounds that it has extra-semantic motivations, then I could imagine getting sort of comfortable with that. But if I set the same standards for expressivism as I would set for any other semantic theory, the ‘intimacy’ between &1 and &2 does nothing to undermine the point that they are not equivalent, and in particular, that &1 can be true and &2 false at one and the same time.

  16. Hi Mark,
    What I proposed was that we make sense of the thought attributed to Jon in &2 by reference to the inferential role in which it stands to the thoughts attributed in &1. But that is not to deny that they are distinct thoughts, not to deny that you can have the one without the other (we certainly don’t want to insist that Jon’s thoughts are closed under even very obvious entailment) and consequently surely not to conflate them.

  17. Jimmy – I see that that’s what you said, but in that case I don’t see why you can’t do the same thing for N2. It is, after all, what Horgan and Timmons do about it quite explicitly, without going in for interpretive claims about how to understand what is going on in Gibbard or Blackburn.

  18. The thought is promising though not straightforward. Obviously for starters, we can’t exploit the clear intimacy between the straightforward 1 sentence and the problematic 2 sentence to get such an account going. And, my (alas, far from adequate) knowledge of the literature on how best to motivate and characterize an inferential role semantics for the logical constants, quite generally and not just in this context, gives a strong impression that, while conjunction is a bit of a doddle, negation is a proper headache.

  19. Wait a minute. Surely we have two, if not three very different problems here. The Acton problem concerns the possibility of inferential relations between moral thoughts if they are not propositional. This is something that Hare, for example, addresses in LM. The Frege-Geach is something else altogether. The question isn’t whether we can make sense of inferential relations between boos and hurrahs (that much may be granted), but that there is no booing or hurrahing going on at the antecedents of conditionals. Instead, we seem to have equivocation, if we accept the expressivist story for the asserted contexts. A fortiori, solving the problem for negation and conjunction does nothing to solve the Frege-Geach problem. So I disagree, it seems, with Mark and Jimmy.
    And that’s not all – I also disagree with Jussi. The sort of casuistry that Acton has in mind is going from a universal generalization to a particular instance – “Abortion is wrong” is a consequence of “All murder is wrong” and “Abortion is murder”. You could, of course, also take a hypothetical attitude toward a moral system and investigate its consequences – I could think about the implications of sharia for studying philosophy, for example. It would be more natural to phrase that in terms of subjunctive conditionals than indicative ones, though: “If traveling on a Friday were wrong, it would be wrong to go to many philosophy conferences”. This is a different sort of ‘casuistry’.
    Finally, is conjunction really as tough as negation? N1 and N2 are clearly not equivalent, since Jon could have no beliefs about O(p). But &1 and &2? I’d like to see a credible case. It seems to me that if I believe that this computer is new and I believe that Aristotle is dead, I do believe that this computer is new and that Aristotle is dead. I might not assent to the latter if asked, either for pragmatic reasons (are you suggesting a connection between the two by your question?) or because of a false belief about what I believe.

  20. Antti – take the paradox of the preface. I believe that P, where ‘P’ is the first sentence in my book – that’s why I said it in my book. I believe that Q, where ‘Q’ is the second sentence in my book – that’s why I said it in my book. And so on. But I also believe that something I said in my book is false – i.e., that one of P,Q,… is. But I don’t believe that P&Q&…&(one of P,Q,… is false). I’m quite certain that this conjunction is (necessarily) false. So ipso facto belief doesn’t agglomerate across conjunction.
    Incidentally, I don’t believe that I said that solving the problem for negation and conjunction was enough to solve the embedding problem. I believe that what I said was that there is no problem specifically about negation, but rather a structural problem that comes up for any other linguistic construction, as well, including not just conjunction, but quantifiers, modals, and tense.

  21. Antti,
    I agree that there are different problems here but they do seem related. I’m not certain though that it is fair to psychologise Acton’s problem. I’m remember wrong but he doesn’t mention thoughts at all in the paper. Rather, he seems to be interested of the linguistic level – of normative statements. That’s a different target than the attitudes. Also, his target is Ayer for whom on the linguistic level moral terms were meaningless (in the way that it isn’t for Gibbard for instance).
    I’m sure you remember the famous claim from LTaL that goes like this:
    “If I say “You acted wrongly in stealing that money”, I am saying no more than “You stole that money””.
    That seems to suggest that you can remove the moral terms from the sentence without changing its meaning – the proposition that it expresses.
    But, if that’s true then the inference we have is first (I’m changing murder to killing because of certain complications Acton actually discusses):
    1. All killing is wrong
    2. Abortion is killing
    3. Therefore, abortion is wrong.
    On Ayer’s view, what I am saying in the premises and the conclusion would be
    1*. All killing
    2*. Abortion is killing
    3*. Therefore, abortion.
    But, that’s just mere gibberish no matter what feelings I attach. Rather, we would like to say that the inference is valid because of its form that is truth-preserving for any propositions:
    1**. All Fs are Gs.
    2**. Hs are Fs.
    3**. Thefore, Hs are Gs.
    We should be able to detect the validity of 1-3 because it has this form merely at looking at the argument without any emotions we might have towards killing and abortion. But, Ayer’s view doesn’t seem to allow to do this – it doesn’t give propositions that could be involved in the valid argument.

  22. I don’t have my Geach with me, but as I recall, the connection between the problem about inference and the problem about embedding is that one way one might try to solve the embedding problem is by saying that a normative sentence in an unasserted context has a different meaning from a normative sentence in an asserted context. But the reason this won’t work is that then you’re equivocating on the normative sentence, and modus ponens won’t go through.

Comments are closed.