`Reference and Deference` – amended version (2006)

advertisement
Reference and Deference
ANDREW WOODFIELD
Acknowledgements and affiliation as unnumbered footnote :
The first draft was presented on 16th September 1998 at the Karlovy Vary Symposium
on the philosophy of Hilary Putnam. I am grateful to the organisers at the
Philosophical Institute of the Czech Academy of Sciences, and to the participants who
gave helpful feedback. Redrafting was done during my tenure of a Mind Association
Research Fellowship. Jonathan Berg, Jessica Brown, Adam Morton, François
Récanati, and the Mind and Language referee all made helpful comments.
Address for correspondence: Department of Philosophy, 9 Woodland Rd, Bristol
BS8 1TB, UK
Email: A.Woodfield@bristol.ac.uk
Abstract:
According to Putnam, meaning and reference depend on acts of
structured cooperation between language-users. For example, lay-people defer to
experts regarding the conditions under which something may be called ‘gold’. A
modest expert may defer to a greater expert. Question: can deference be neverending? Two theories say no. I expound these, then criticize them. The theories deal
with semantic processes bound by a ‘stopping’ constraint which are not cases of
ordinary deferring. Deferring is normally done for a reason, and a rational person is
always disposed to defer if there is good reason.
1. The Possibility of Endless Deference
When Putnam first wrote about the division of linguistic labour (in ‘The
Meaning of “Meaning”’ 1975), he gave prominence to cases in which ordinary
speakers defer to experts concerning the application of natural kind words. This often
happens because laypeople are poor at recognizing members or instances of natural
kinds. Putnam argued that the recognitional knowledge is present in the linguistic
community taken as a whole. Experts know the recognition criteria associated with a
kind-word, and other people defer to their judgements. The extension of the kindname in the community’s language gets demarcated through a process of structured
cooperation.
1
The idea of semantic deference can be extended in several directions. Firstly,
as Putnam noted, deference is practised not only with natural kind words but with
many other sorts of words as well.
Secondly, expert recognizers are not the only category of experts. For we
sometimes want expert general advice about what the proper criteria are. And
sometimes what we seek is a good verbal definition.
Expert recognizers are members of society who are skilled at applying criteria.
Depending on the word in question, there may exist subgroups of professionals who
know or claim to know more than the average person about which are the right
criteria. Other people - including expert recognizers - may defer to their rulings. Also,
when trying to define a word explicitly, we often defer to a definition provided by an
authoritative linguist or text or academy. Since fixing criteria and verbally defining
are different tasks from recognizing instances, there is room for a subdivision of
linguistic labour. There can be role-switching: it may be that you are better than me at
explicating the sense of some word (so I defer to you on this), while I am better than
you at recognizing instances (so you defer to me on that).
Meaning, being multi-dimensional, generates opportunities for social cooperation on a range of different tasks. The three tasks mentioned (identifying the
right criteria, applying agreed criteria, and defining) are not the only ones where
expertise is recognized. Advertising copywriters are experts at choosing apt forms of
expression for a desired message; legal experts decide which definition is the relevant
one to employ (e.g. when a term has different definitions within different systems of
regulation); and so on.
The third extension of Putnam’s idea is from linguistic semantics to mental
semantics. Putnam showed that the meanings of many words are not fully
internalized. His work inspired various forms of externalism about the contents of
mental states. Externalism about mental contents says that the contents of some
thoughts and concepts are individuated by factors external to the thinker. Externalists
say, for example, that my having the concept gold requires that I be related in some
way to the outside world. There are different externalist theses about which relation
has to hold (see Donnellan 1993). Putnam-inspired externalists say the relation
consists in my having had direct or indirect causal contact with gold. Social
externalists say the crucial relation between me and the concept gold is socio-
2
linguistically mediated. If social externalism is right, other people can play a role in
fixing what my thought-content is. There can be a division of conceptual labour
consequent upon the division of linguistic labour (see Burge 1979).
Linguistic deference and conceptual deference are widespread phenomena.
However, some philosophers say that deferring has to stop at some point, for if
everyone were deferential with respect to a given word or concept no one would ever
succeed in attaching a definite content to it. That thesis is the stimulus for this paper.
Fodor certainly holds the thesis. In his latest book Concepts, Fodor (1998,
p154) says, ‘Adherence to conventions of deference couldn’t be a precondition of
conceptual content in general, if only because deference has to stop somewhere; if my
ELM concept is deferential, that’s because the botanist’s isn’t’. (cf. Fodor 1994, p33).
Récanati is another philosopher who espouses the thesis that deference must
eventually stop. He says that no content is expressed by the utterance of a term if it is
a term ‘used deferentially by everybody, in a mutual or circular manner’ (Récanati
1997, p92).
Although the ‘must stop’ thesis seems initially plausible, further reflection
suggests that matters are not straightforward. Suppose that three people, a customer A,
a jeweller B and a chemist C, are attempting to settle whether a certain brooch
belonging to A is made of an especially pure type of gold which is mined only in
South Africa. A proposes that it is. The jeweller B performs a test. On the basis of the
result, B asserts that the gold is not the special South African type. A defers to B’s
judgement. But B is not confident that his crude diagnostic technique was the right
one, so he consults the chemist C. C performs tests which invoke different criteria
from the jeweller’s. C concludes that the brooch is made of the special type. B defers
to C’s judgement. A happily defers to C as well. And there the matter rests; no further
investigations are performed. However C’s judgement relied on an assumption. C
took it that certain impurities which he found in the gold were sufficient to establish
that it was the special South African type. But it later transpires that this set of
characteristics is not decisive. When C learns this fact, he realizes that he might have
misclassified the gold in the brooch. If another chemist employing better criteria were
to test it and pronounce it not of the South African type, then C would probably defer
to the other chemist’s opinion.
3
Empirical testing ends because the parties have no more time or inclination to
continue. But in principle the testing could continue, and each interim conclusion
could conceivably be undercut. The proponent of a judgement reached at any point
would be prepared to defer to another person’s superior contrary judgement. Real-life
processes of deferring do stop. But it is not necessarily true that the deferring process
had to stop at the point where it did. Nor is it true that the deferring necessarily has to
stop somewhere. It could in theory go on indefinitely. The participants are not stuck
without a resolution just because the latest judgement is defeasible. On the contrary,
they commit themselves to the best judgement available at the time. For example, the
owner puts the brooch up for sale at a price which reflects the current assay.
We may readily imagine circles of mutually deferring agents. Take three
scientists A, B and C who continually strive to improve their methods of detecting
some elusive disease. On Monday, A defers to B’s opinion about whether the disease
is present in a particular blood sample. On Tuesday, B revises his opinion in the light
of a contrary assessment made by C, on the grounds that C used a superior criterion.
Meanwhile, A learns more about the disease and develops a much better diagnostic
technique than any so far. On Wednesday, A issues a new verdict on the original
blood sample. C defers to A’s new assessment, because he recognizes that A, armed
with better criteria, has become more authoritative. Meanwhile, B is working away on
an even better method. And so it could go on.
Objection: ‘These scenarios involve people deferring to one another over
questions of fact, whereas the kind of deferring practices we ought to be considering
are those that hinge on meanings’.
Actually, the above disputes do concern the extensions of words, and
extension is an important aspect of meaning. Every time that a participant defers to
another’s categorization of an item, he defers (implicitly) to the other’s judgement
about whether a certain word applies to the item.
More importantly, non-stopping deferential processes can equally well occur
in disputes about intensions. People sometimes defer to experts’ advice about how an
expression in a language should be construed. This frequently happens in cases where
its sense is popularly misconstrued. Also disputes may arise amongst expert linguists
about the best way to specify the meaning of an expression. With any type of semantic
4
issue, there can agreement giving way to disagreement which is resolved by new
agreement.
Deferring is not passing the buck. Suppose that UK Customs and
Excise decides to employ EU criteria for classifying certain food-products and judges
that toffee-apples count as confectionery rather than fruit by those criteria and hence
are liable to UK Value Added Tax. The UK authority itself arrives at that judgement.
It does not pass the buck. It is not the case that the UK lets the EU decide whether
British toffee apples are to be subject to VAT, nor is it that the UK lets the EU decide
that they count as confectionery. (That the EU plays no part in taking such national
decisions is an aspect of the principle of subsidiarity which governs relations between
the Union and its member states.)
A decision to defer may be entered into responsibly and for good reason. Every
rational speaker should be disposed to accept semantic corrections from those who are
more knowledgeable. Since experts are rational speakers, they too have the same
conditional willingness to defer. There is a sense in which the pressure to defer never
goes away. Let us look more deeply, then, into the reasons why two philosophers hold
that in order for a word or a concept to have a determinate content, deference must
eventually stop. I shall suggest that they hold this because they treat A’s deferring to B
on the model of A’s borrowing B’s meaning. I shall argue that this model is wrong.
2. Fodor’s Model of Semantic Deference
Fodor writes about ‘deferential concepts’ in The Elm and the Expert (1994,
lecture 2), and Concepts (1998, ch 4). (These works will be cited respectively as ‘EE’
and ‘C’.) He says that lots of our concepts are deferential (EE p34).
Fodor holds that a concept is a mental particular, an internal representation
belonging to an individual (C p23). A concept possessed by one person D (the
Deferrer) is deferential just in case it appropriates or co-opts the content of someone
else’s concept. Public language plays a role in mediating the co-opting relation
between two people’s concepts in cases where D gains access to the content of E’s
concept through E’s verbalisations.
The thesis that deferring equals co-opting someone’s content is offered as a
solution to the ‘elm/beech’ problem. Suppose a linguistically competent adult S
knows the words ‘elm’ and ‘beech’ but cannot perceptually discriminate elms from
5
beeches. This person is not interested in trees; her stereotype of an elm is qualitatively
the same as her stereotype of a beech. But she knows that elms and beeches are
different species, so she must be credited with distinct concepts of elm and of beech.
Problem: how is it possible for her to have two concepts c1 and c2 such that c1 has the
content elm and c2 has the content beech? What makes the contents different if she
cannot tell the difference?
This sort of case afflicts practically every theory of content. It is a problem for
internalists who believe that conceptual content is in the head, and for externalists
who say that for a concept to have the content elm is for it to stand in an external
relation to elms. It is a problem for such externalists, because they need to find a
mind-world relation that does hold between c1 and elms but does not hold between c1
and beeches.
Fodor says that D can exploit the concepts of an expert in the following way:
she selects a botanist who calls a tree ‘elm’ when and only when it really is an elm,
and she uses the botanist’s verbal responses as indicators. So D can tell - with the
expert’s assistance - when a tree is an elm. Concept c1 can get hooked on to elms by
an indirect route, a route that goes through the expert’s concept of elm. Similarly, c2
can co-opt the content beech from the botanist’s concept of beech.
Fodor’s position is epitomised in EE on p36: ‘What philosophers call
‘linguistic deference’ is actually the use of experts as intruments; not Marxist division
of labour in semantics but capitalist exploitation in epistemology’.
It is easy to see why Fodor insists that not everyone’s concept of elm can be
deferential. On his model, being deferential is a form of dependency. If D’s concept
has a content which devolved from someone else’s concept, the donor concept must
have already possessed a content. It is possible that the donor concept was deferential.
There can be a chain of content-borrowings. But every chain must trace back
ultimately to a concept that has non-borrowed content.
The thesis about content-borrowing needs to mesh in with a theory of
independent content-fixing. In fact, there are lots of theories in the ring. The one
currently favoured by Fodor is ‘Informational Semantics’, which says, roughly, that a
concept’s having the content elm consists in its standing in a certain relation of
lawlike correlation with elms.
6
Informational Semantics marries neatly with his view of deference. For Fodor
can claim that what constitutively fixes the content of a concept is always the same,
whether the concept is deferential or non-deferential. D’s concept of elm picked up its
content through E. But the theory of content-fixing is required only to explain the fact
that it has the content elm, not how it came to have it. When D’s concept c1 is up and
running, its having the content elm consists in the fact that c1 itself is informationally
locked on to elmhood.
The relationship between deferrer and expert may be summarised as follows.
(i) E’s concept had the content elm independently, before D’s concept did.
(ii) In the circumstances, if E’s concept had not had the content elm, D’s concept
would not have come to have it. The fact that c1 ended up with that content was
causally dependent upon E’s provision of an access-route to elms, and also causally
dependent upon the fact that E’s concept carried elm-information.
(iii) In alternative circumstances, the same information could have got to D through E
even if E had had no concept of elm at all. Suppose that E were not an expert
recognizer but an unwitting indicator. Imagine, for instance, that E is specifically
allergic to elms. In their presence E’s face always flares up in a red rash, just as some
people react whenever they eat peanuts. And suppose D had used E’s allergic reaction
as her measuring-instrument. D’s concept would have ended up carrying the same
information. Thus although in the actual situation E’s concept played a key role, it was
not necessary for D to interpret E’s concept as a concept. The essential role that E’s
concept played was that of being a natural sign.
(iv) But it was necessary for D to engage in some interpreting. For example, D needed
to understand that the tokenings of E’s concept and the vocalisations were nomically
correlated with a certain type of tree, and D had to treat E’s responses as indicators. D
had to know what she was aiming for, otherwise she would not have kept up the
attention and effort. She intentionally made use of E’s responses in order to get her
own concept locked on to a certain kind of tree.
(v) D’s concept obtained its content by tapping the information carried by E’s
concept. Not every concept can get its content by tapping in to another, therefore
Fodor is right that deference, on this model, must eventually stop.
Now for some comments. The first thing to note is that deference as Fodor
conceives it is far removed from the ordinary notion. I shall not list all the
7
dissimilarities; many of them are obvious. But one difference deserves special
attention.
If a novice judges at time t1 that a certain tree is an elm, then enters into
communication with an expert who tells the novice at t2 that the tree is not an elm, and
the novice accepts at t3 that the tree is not an elm, this is a standard case of deferring,
as noted by Putnam. At t1, the novice already possesses a concept which he expresses
through the word ‘elm’, and this concept already has an intentional content and an
extension. At t3, what he accepts is that the tree in question is not in the extension of
that concept, given the content that the concept actually had at t1. He now believes that
his judgement at t1 was false. In order to make sense of the novice’s change of belief,
we need to assume that he took the expert to be correcting a particular application of
the concept, with the content that it had at t1, in a situation where both assume they
agree on what the content was and is. This form of interaction is not one where a
novice’s concept picks up its content from the expert. But in Fodor’s model the
deferential relation holds if and only if D’s concept picks up its content from E’s. It is
quite clear, then, that whenever ordinary ‘Putnamian’ deference occurs, Fodorian
deference does not occur.
The sort of situation where Fodorian deference is supposed to occur is when D
acquires a concept through interaction with E. The whole point of the story is that the
novice gains a concept whose content is fixed by facts about the expert. Prior to the
information-tapping, D has no concept of elm because none of her mental
representations is locked on to elmhood. E, the measuring-instrument, is D’s only
channel of semantic access.
The second comment I wish to make is that Fodor’s story misrepresents what
the participants take themselves to be doing. Such considerations belong to what
Fodor calls the ‘epistemology’ of deference (and not to the semantics). He says that if
we view matters his way, the existence of deferential concepts is ‘seen to be of some
interest for epistemology’ (EE p34); but as I see it, his model distorts the
epistemology so drastically that the most charitable interpretation we can put on it is
that it purports to be a revolutionary alternative to common-sense, a denial of the
participants’ manifest image of the process in which they are engaged.
Actually, Fodor does not say what the beliefs and intentions of the deferrer and
the expert are, other than say that the deferrer has to set things up cleverly and
8
carefully, pick the right expert, choose a test environment that permits the expert to
manifest his recognitional skill, enlist the expert’s co-operation, etc. Above all, the
deferrer has to be ‘able to pursue policies with respect to another person’s mind’ as
well as his own. Let us probe a little into what these policies might be.
A deferrer D1 who is fully in control and in-the-know might plausibly be
credited with the following strategy. D1 thinks: ‘I am no good at recognising elms. So
I shall employ E, who is an expert recogniser. Under optimal detection conditions,
when presented with trees, E will judge that a tree is an elm if and only if it is an elm.
When optimally co-operative, E will reveal his sincere judgements to me by uttering a
certain sound. I will rig these conditions up, and I will judge that a tree is an elm just
in case E utters the sound under these conditions.’
Although D1 stands a good chance of making correct elm-categorisations if the
conditions hold, a person who initially lacked the concept ELM could not possibly
entertain these thoughts. This man has beliefs and intentions about elms qua elms.
For example, D1 believes: E’s noises indicate the presence of elms. D1 obviously
possesses the concept ELM, even before he has started to defer. Prior possession of a
given concept cannot be an empirical precondition for acquiring it. Fodor does not
want to say that deferrers conceptualize their policy in such terms.
So consider a deferrer D2 who has a conceptually more austere set of beliefs
and intentions, but who is still more or less in control. This man thinks to himself: ‘I’ll
use E to help me categorise trees. There is a certain kind of tree which E is very good
at recognising. If I pay him enough, he will always reveal to me his judgement that a
presented tree is of that kind. I’ll take his utterance of a certain sound as a natural sign.
When the optimal conditions of observation and co-operation obtain, I will judge, of
any tree that we both see, that it is that kind of tree, if and only if E makes the sound.’
D2 can think these thoughts without possessing the concept ELM. But he must
already possess a concept that denotes elms. It’s true that this novice simply treats E’s
utterances of the sound as a reliable natural sign of one kind of tree - which is all that
informational semantics requires. Nevertheless, he has a concept of the kind of tree in
question. For example, he thinks: E’s noises are reliable indicators of that kind of
tree. He employs the concept THAT KIND OF TREE. This concept of his refers to
the kind elm. Fodor 1998, in particular, is committed to saying that its content is elm,
9
for he holds that if two concepts are co-referential (as are the concept ELM and D’s
concept THAT KIND OF TREE), they have the same content (see C pp 12-15).
People cannot use measuring-devices profitably unless they know what the
devices measure. Those who exploit an expert as a measuring device need to know
what they are doing; they need some conception of what the expert is responding to.
In the present case, D already has a concept whose content is elm; so it is not true that
D acquires that concept by borrowing the content from E’s concept.
It is very hard to see what the epistemic situation of an acquirer of a deferential
concept could possibly be. To avoid begging any questions, one would need to
attribute to such a person a set of attitudes that was conceptually less rich than D2’s.
A defender of Fodor might say that I am exaggerating the extent to which a
content-borrower has to be conscious of borrowing. If ‘deferring’ is defined as
something that has to be done deliberately and in full awareness, then indeed the
Fodorian content-borrower does not literally qualify as a ‘deferrer’. But let’s not get
hung up on literal construals. Fodor is using the word metaphorically. (This point was
made by Michael Devitt at the Karlovy Vary conference.)
This seems right, but it does not wholly insulate Fodor from the objection. I
concede that deferring to experts can become automatic; it need not be undertaken
consciously. Nevertheless, a deferrer adopts a policy on how to handle the other
person’s expertise. Having an unconscious policy means having unconscious beliefs
and intentions. Such beliefs and intentions must have contents. The objection was that
D would surely have some attitudes whose contents contained the content elm. If D
did not have any such attitudes, D would not understand what was going on.
I think it is time to invite Fodor to bite the bullet. He should admit that if
anyone ever acquires a concept via the deferential route, such a person is not really in
control of the process. The information-tapper does not know which concept he is
going to get, in fact his beliefs about the informational transactions between himself
and the expert are pretty much irrelevant.
Note also that such a person would be unable knowingly to set up the optimal
test-conditions. To achieve such a feat one would have to select an expert whom one
knew to be a reliable measuring-device of the kind of tree in question, and one would
need to know which sound made by the expert was the sign of that kind of tree. After
all, the expert might make several different sounds which are correlated with different
10
kinds of trees. On the whole, I think that if anyone ever did get informationally locked
on to anything as a result of such a set-up, it would either be a matter of luck, or a
consequence of someone else’s design (or Nature’s).
I see signs in Fodor’s latest book that he is moving in this direction. For he
says (in. C p.124) that informational semantics plus conceptual atomism favour the
view that primitive concept-possession is non-cognitive, and atomic conceptacquisition is a not a cognitive process. Information from the world colonises us and
implants contents in our heads, rather as germs colonise us and give us diseases. The
current position harks back to the causal-historical theory of reference: picking up the
content of a general concept from an expert is like borrowing the reference of a proper
name from another user. If these are the models that Fodor 1998 favours, then he
ought to deny that the epistemology of the receiver is of any interest.
My own view, for what it’s worth, is that there are no such mental particulars
as Fodorian deferential concepts. The processes that he hypothesizes do not exist. But
I shan’t try to defend that view here. It is clear enough, anyway, that his story about
deferential concepts in no way supports the thesis that deferring in the normal sense
must stop.
3. Récanati’s Deferential Operator
In order to appreciate Récanati’s theory of deferential operators, we need to
fix our minds upon quasi-quoting. I’ll give examples of speakers who do this; the first
example is mine, the others are adapted from Récanati 1997. It is important that in
each example we understand that the utterer of the sentence was the person
responsible for putting in the quasi-quotes. We may assume that the speaker signalled
them by a gesture or by vocal emphasis. I highlight quasi-quotation marks by writing
them as asterisks - this helps to distinguish them from ordinary quotation-marks.
Récanati’s paper is a response to certain doctrines about ‘quasi-belief’ that
were proposed by Sperber. In my exposition I shall ignore those aspects of the debate
which are inessential for our purposes and I shall not say anything about Sperber’s
position.
Consider utterance (1) made by a non-Czech speaker while looking at a menu
written in Czech:
(1) For lunch I shall have *kachna*.
11
The speaker quotes a Czech word which he does not understand, but at the
same time he uses the word as a syntactically functioning part of an English sentence.
Those in his lunch party who speak English and Czech will be able to work out what
food he will be eating. English monolinguals will not be able to identify the dish,
except perhaps under the description ‘food called “kachna” in Czech’. Note, though,
that the speaker did not explicitly say that the food was called ‘kachna’, nor did he
speak about the Czech language.
Quasi-quoting is one of several functions that scare-quotes can be used to
perform: in general, the speaker who puts an expression in scare-quotes intentionally
signals that he distances himself from the expression in some way and that it is not
functioning normally. Precisely how or why he distances himself from it is to be
pragmatically worked out with the help of contextual information. In this case the
speaker expects the audience to realize that he is quasi-quoting the foreign word
because he does not know how to translate it.
Perhaps (1) is not a well-formed English sentence, in which case it does not
express a determinate proposition. But the speaker intends to communicate some
thought. If we ask ‘Which thought does the speaker intend to communicate?’, there is
a problem. For it is hard to find a satisfactory characterization in indirect speech. He is
thinking of a certain type of food, but under what mode of presentation? Clearly he is
not thinking of it as duck. But it seems equally unwarranted to say that he thinks of it
under the metalinguistic description the referent of the Czech word ‘kachna’. The
thought: I shall be having the referent of the Czech word ‘kachna’ for lunch is more
complex in structure and in content than the thought that he actually entertained.
My second example is a variant of one that Récanati gives. Sally has overheard
conversations between members of a religious sect. The sect-members trot out a
phrase which is used by their guru, taking for granted that the phrase is meaningful.
They trust the guru and believe that everything he says is true. The phrase in question
is ‘alpha-enlightened’. Actually, none of the followers knows what it means and nor
does the guru: the phrase is meaningless. In utterance (2) Sally uses the phrase in
quasi-quotes, intending to allude to those who use it straight.
(2) Jesus Christ and John Lennon are *alpha-enlightened* beings.
12
Arguably, utterance (2) is a token of a syntactically well-formed sentence of English.
There is a problem about identifying which proposition, if any, (2) expresses. Also
there is the problem of characterizing what Sally believed.
The third example is of an unconfident speaker who quasi-quotes a meaningful
expression of English, intending that it be construed as having its normal meaning but
also intending to signal that he does not fully understand it. A schoolboy has heard a
teacher say ‘Cicero’s prose is full of synecdoches’. The boy is not sure what
synecdoches are. He is able to convey this fact while passing on the teacher’s
message. The boy uses quasi-quotes, saying to his friend:
(3) Cicero’s prose is full of *synecdoches*.
People sometimes do this to hedge against being asked to explain what a term means.
The audience will realize that the speaker may not fully understand the term, so they
do not embarrass him by asking. The quasi-quoting speaker can still play a helpful
role in communication. Once again, however, there is a problem about characterizing
the speaker’s belief.
Récanati offers a solution to these characterization problems. I take him to be
offering two proposals. One is that overt quasi-quotes in public language are
deferential operators. The other is that there exist corresponding deferential operators
in thought. At both levels, he draws upon a distinction between character and content.
Kaplan, the originator of the distinction, invoked it in order to describe the semantics
of indexical expressions. The expression-type is asociated with a general meaning-rule
(its character), and the character specifies how contexts of utterance combine with
particular utterances to fix their semantic values (i.e. their contents). For instance, the
first personal pronoun ‘I’ is governed by the rule that each utterance of ‘I’ refers to the
producer of the utterance. That is its character. If we look at a particular utterance, say
Greta Garbo’s saying on a particular occasion, ‘I want to be alone’, the content of that
occurrence of ‘I’ is Greta Garbo.
Here is Récanati’s definition of a deferential operator (pp91-2):
DEFN: The deferential operator Rx( ) applies to a symbol  and yields a complex
expression Rx() whose character is distinct from that of  (if  has one). The
character of Rx() takes us from a context in which the speaker tacitly refers to a
certain cognitive agent x (which can be an individual or a community of users) to
13
a certain content, namely the content which  has for x, given the character
which x attaches to .
Let us see how the quasi-quotes in (1), (2) and (3) satisfy this definition. In (1)
the quasi-quotes apply to the Czech word to yield the complex expression ‘*kachna*’.
This expression has a character. The character is a function from the word ‘kachna’,
plus the context in which the speaker tacitly referred to Czech speakers, to the content
which Czech speakers attach to the word ‘kachna’. In fact, the content is duck. So the
speaker of (1) thereby expressed the proposition that he would eat duck for lunch though he did not know this. (Récanati does not use this sort of case, so I won’t
mention it again.)
In (2), Sally’s utterance of ‘*alpha-enlightened*’ was a complex expression
formed by a deferential operator. Its meaning or character is a function from tokens of
‘alpha-enlightened’ and from contexts in which Sally tacitly referred to utterers of it to
the contents which the expression-tokens have for their respective utterers. What
content does it have for the sect-members? It has none. Therefore the complex
expression-token has no content either. Sally’s utterance fails to state a proposition.
In (3) the boy’s complex deferential term ‘*synecdoches*’ has a character such
that its content is identical with the content that ‘synecdoches’ has for the teacher. So
the boy succeeds in asserting that Cicero’s prose is full of synecdoches. No scarequotes are needed in reporting the proposition expressed.
Récanati’s second proposal is that deferential operators exist inside the mind.
He claims that in the examples the speakers’ thoughts are deferential, and his theory
enables us characterize those thoughts.
To set the theory up Récanati needs a few background assumptions. He
assumes that there is a Language of Thought. When S believes that p, S stands in the
‘believes’ relation to the proposition that p; this may be analysed into two
components: (i) S stands a relation of ‘accepting’ to some sentence of Mentalese, (ii)
the Mentalese sentence means that p.
The words of S’s Mentalese are the vehicles of S’s concepts. Every Mentalese
symbol has a definite character for S: ‘it is hard to think of a symbol being mentally
entertained without being “interpreted” in some fashion or other....if a mental sentence
is well-formed, it must possess a definite meaning - a character - even if it falls short
of expressing a definite content’ (p91).
14
He makes the ‘individualistic’assumption that when S utters a sentence in a
public language, the public character of the sentence is irrelevant to the question
‘Which thought did S express?’. For it may be that S did not understand the public
character. What matters is the character that the sentence has for S. This character
must be one that S himself grasps; it will be the character of the Mentalese sentence
which S pairs with it (p91).
At the same time, he embraces some of the insights of externalism. It is
possible for S to have attitudes involving mental representations whose referential
contents are epistemically indeterminate for S. Sometimes people do not know which
proposition it is that they believe. How can this be squared with the individualistic
assumption? By distinguishing between character and content. Character is in the
head, even if content is not.
Having made this distinction for Mentalese, Récanati now explains how to
characterize the speakers’ beliefs, using English as if it were their language of
thought.
In uttering (2), Sally accepts this mental representation:
(2 -LOT) Jesus Christ and John Lennon are Rx(‘alpha-enlightened’) beings.
It has a character for her. But the Mentalese term Rx(‘alpha-enlightened’) has no
referent, because the sect-members use a term that has no meaning. Sally’s Mental
sentence fails to express any proposition, hence Sally has no propositionallyidentifiable belief.
The boy who utters (3) accepts a mental representation (3-LOT):
(3-LOT) : Cicero’s prose is full of Rx (‘synecdoches’).
Since the boy is quasi-quoting the teacher, the content of his mental symbol
Rx(‘synecdoches’) is the content that ‘synecdoches’ has for the teacher. Assuming the
teacher knew what the word meant, the content was synecdoches, and so the content
of (3-LOT) is that Cicero’s prose is full of them. This, then, is what the boy believes.
The key feature of a complex symbol bound by a deferential operator is that it
is directly referential. The content is fixed not descriptively but indexically, in virtue
of the fact that the content stands in a contextual relation to the thinker. The complex
deferential symbol is like a demonstrative. When S entertains a demonstrative mode
of presentation of an object O, S thinks OF O under the mp That. Similarly, when the
boy entertains the complex symbol Rx(‘synecdoches’), he thinks directly OF
synecdoches.
15
What is it like to think in such a way? When Récanati says (p96), ‘Deferential
representations are meta-representational at the level of character but not at the level
of content’, he does not mean they they present themselves to S as being metarepresentational. Actually, he does not try to describe what it’s like to entertain a
deferential symbol. But he might convey it in the following way. Imagine being in a
situation where you have just heard the teacher say ‘synecdoches’. You are thinking of
the speaker and the sound indexically, under perceptual mps. You then think OF the
sound’s semantic value for the teacher, in thinking: Cicero’s prose is full of those.
I agree that this is a possible psychological state. I think that Récanati has put
his finger on a phenomenon which no one else has previously diagnosed: mental
quasi-quoting. Moreover he is right that if a given symbol is to have a determinate
content, it is not possible that everyone always puts quasi-quotes around it.
Récanati offers an interesting theory that explains a specialized range of
phenomena. Where I want to take issue with him is over his assumption that the
theory has much wider application. He extends mental quasi-quoting far outside the
range illustrated by these examples. For instance, he claims that all of the characters in
Burge’s 1979 thought-experiments employ deferential operators. In fact, he claims
that children, language-learners and other imperfect understanders of picked up words
normally bind such words inside deferential operators. These claims are false. They
systematically generate wrong interpretations of the conversations that take place
between novices and experts, learners and teachers.
The way he reconstructs Burgean cases is instructive. In the best-known
scenario from Burge 1979, the character simply says to the doctor, ‘I have arthritis in
the thigh’ without putting ‘arthritis’ in scare-quotes. It is crucial to Burge’s argument
that the subject should not use quasi-quotes or distance himself from the word, but
should use it as part of his own working vocabulary. As Burge notes, ‘The argument
can get under way in any case where it is intuitively possible to attribute a mental state
or event whose content involves a notion that the subject incompletely understands’
(1979, p79). Suppose we run Burge’s argument using the ‘synecdoche’ example. We
imagine a boy, Alf, who takes over the word from the teacher and uses it straight. Alf
says ‘Cicero’s prose is full of synecdoches’. The meaning is different from that of (3).
Alf has assimilated the term into his working vocabulary, and according to Burge he
expresses the concept synecdoche.
16
But Récanati offers a revisionary interpretation. He assumes that Alf missed
out the scare-quotes. If you ask Alf whether he was quasi-quoting the teacher, it does
not matter if Alf denies it. You may presume he quasi-quoted the teacher unwittingly.
The important hypothesis is that Alf’s Mentalese sentence contained a deferential
symbol.
That is the first step in the radical reinterpretation. The next step is to
introduce the distinction between Kaplanesque content and character. This is a move
which Burge never even considered. Burge worked with the old, undifferentiated
Fregean notion of conceptual content. Once we make the move, however, we separate
two questions:
(a) Is the character of Alf’s thought socio-linguistically determined?
(b) Is its referential content socio-linguistically determined?
Récanati’s answer to (a) is: No. Character is in the head. It cannot be that the
boy entertains the public concept synecdoche, because it is impossible that anyone
should think with a notion whose character they do not understand.
Concerning (b), the answer is: Yes, sort of. The referential content of Alf’s
belief is identical with that of (3-LOT), the quasi-quotational belief. The relevant
intuitions about twin-earth seem to march in step with Burge, because referential
content is outside the head. Suppose that the quasi-quoting boy has a counterpart on
Twin-Earth, where the language is slightly different from English. Suppose that twinboy has a twin-teacher who said ‘synecdoche’, but his word did not mean synecdoche.
Twin-boy defers to twin-teacher’s meaning. It is not the case that twin-boy’s utterance
or belief refers to synecdoches. This seems to fit with the intuitions we have about
Burgean cases.
However, these parallels at the level of referential content cannot conceal the
fundamental differences between Burge’s view and Récanati’s concerning conceptual
content. According to Récanati, the boy who thinks (3-LOT) has a directly referential
belief about synecdoches, which he thinks of, in context, as Those. That is how Alf
thinks of them, if Alf’s mental representation is (3-LOT). But on Burge’s view, this
not how Alf thinks of them. He thinks of them as synecdoches. In the thoughtexperiments, the belief-attributions are supposed to be true when construed opaquely,
as specifying the subject’s conceptual mode of presentation. The attributions are not to
be taken de re. Moreover, on Burge’s view, Alf’s thought about synecdoches does not
17
depend upon his being in a context in which he tacitly refers to the teacher. Alf’s
mode of presentation is in no way indexical or demonstrative.
Récanati is pushing for an individualistic reinterpretation, albeit one that
makes room for mental indexicals whose referents are determined contextually. Burge
was arguing against individualism, for the view that conceptual content was
sociolinguistically individuated. Note that. for Burge, speakers who utter a word
which they do not understand at all are not credited with the concept expressed by the
word, but partial understanders may be said to possess the concept.
Récanati, on the other hand, assimilates Burgean cases to indexical beliefs.
The contents are externally individuated, but social factors (other people and their
utterances) come in only as elements of the subject’s context of utterance. The context
helps to fix a referent, and the referent itself is a constituent of the content. It does not
matter whether the subject partly understands the word or completely fails to
understand it. The mechanism of reference-fixation is the same in either case.
So much for the geography of the issues. Let us turn now to an evaluation of
Récanati’s theory of deference. I said above that I think the apparatus works for
certain special types of utterances and thoughts, but that he over-extends it. He claims
that children and imperfect understanders entertain quasi-quote representations as a
matter of course; that these representations are the appropriate vehicles for quasibeliefs whose characters we know directly, but whose contents we do not know. He
also holds that learners move gradually from a state of quasi-belief to fully
conceptualized belief: ‘ We start by accepting a representation without understanding
it; this attitude of acceptance leads us to use the representation in a certain way; and
by so using the representation we end up understanding it. What makes learning
possible is the use to which the representation is put, and that use itself depends on the
initial attitude of acceptance that motivates it.’ (p89). Moreover: ‘This continuity
suggests that deferentiality is a matter of degree’ (p94).
A great deal more explanation is needed here. It seems impossible that there
could be a gradual process of moving out of quasi-quotes. It’s clearly not a process of
bit-by-bit removal (like taking one’s clothes off), nor is it a process of decay (like
quotation-marks fading away on a page as the ink loses its colour). The learner starts
off using mental symbols like Rx(‘synecdoches’) and Rx (‘kachna’) and ends up using
18
completely distinct symbols like synecdoches and duck. Prima facie, there has to be a
saltation - a switch of symbol-type - at some point.
Communication-based learning typically proceeds as follows: somebody tells
you something, you understand what they are saying, you believe them, you add the
knowledge to your store, and if necessary you delete previously held beliefs that are
incompatible. This is also what goes on in ordinary cases of deferring. For the learning
to work, you must already possess the concepts (in order to understand), and they
must be the same concepts that the teacher or expert is employing (otherwise there is
communication-failure).
I want to end this section by showing how Récanati’s theory misrepresents
typical acts of deferring, which play such an important role in learning. Consider a
dialogue between two characters, Alf and L. Alf is the boy who picked up the word
‘synecdoches’ from his schoolteacher and who does not fully understand it. L is an
expert linguist. L knows the dictionary definition. Their conversation is set against an
important background fact, that the schoolteacher misunderstood the word. (He used
it to refer to metonymies.)
The conversation is as follows:
(i) Alf says: ‘Cicero’s prose is full of synecdoches.’
(ii) L replies: ‘No it is not. It’s true that his prose is full of figures of speech. But
very few of them are synecdoches.’
(iii) Alf replies: ‘I accept what you say. Cicero’s prose is not full of synecdoches.’
If the belief expressed in the second sentence at (iii) is to count as the negation
of the belief Alf had at (i), both beliefs must involve the same predicative concept.
This point was made in section 2. Admittedly, that does not settle the question of
which concept it was. Note, however, that L is definitely employing the concept
synecdoche. If L’s judgement contradicts Alf’s original belief, then Alf’s belief must
have involved the concept synecdoche too. It’s a typical Burge-type situation. Alf
must have possessed the concept synecdoche right from the start, even though he did
not fully understand what synecdoches were.
How do matters look from Récanati’s point of view? Alf was not using the
concept synecdoche, nor was he referring to synecdoches. He was using a deferential
symbol to refer to what the teacher called ‘synecdoches’. Consequently, in this case he
was unwittingly referring to metonymies. If Récanati’s acount is right, the two
19
participants are talking at cross-purposes. At step (ii) the linguist misunderstands what
Alf meant at step (i). He takes himself to be rebutting Alf’s claim. But in fact his
remark is not a denial of the proposition that Cicero’s prose is full of metonymies. L’s
remark is simply irrelevant.
Alf’s response at step (iii) compounds the misunderstanding. For he does not
spot the fact that L’s remark was irrelevant. Récanati can analyse Alf’s response in
two ways, depending on whether or not Alf is still using the same deferential operator
at step (iii). Either way makes the dialogue problematic.
Suppose that Alf at (iii) still defers to the teacher’s meaning. He is referring
unwittingly to metonymies and is now saying that Cicero’s prose is not full of them. It
appears to him that he is deferring to the linguist, but this is an illusion. The linguist
expressed no opinion on whether Cicero’s prose is full of metonymies.
Suppose, on the other hand, Alf at step (iii) is no longer operating with the
teacher’s meaning. Suppose the term now has some other meaning for him. The
question arises whether he is really agreeing with the linguist. He appears to be
agreeing, but that could be an illusion. He cannot really be agreeing unless he
understood what the linguist said at step (ii). If he did understand, he must have
possessed the concept synecdoche, despite the fact that at step (ii) he did not fully
know what synecdoches were. Thus Récanati has to concede that this cognitive
condition is possible, just as Burge said it was. Hence he ought to concede that Alf
could have been in this cognitive condition at step (i). The motivation for introducing
the deferential operator is undercut.
Récanati might object that this argument is unfair because in this conversation
x (the semantic deferee) is not identical with L (the conversational deferee). But why
is this unfair? It is clearly possible for a child to pick up a word from one source and
then use the word when talking to a different person. In the conversation the child
might defer to the new person’s judgement. In later conversations, the child might
defer again to other people’s judgements. This sort of thing happens all the time; it is
the standard process by which a child learns from other people. My criticism is that
the theory misrepresents what is going on in the majority of such conversations.
Normally the child is not quasi-quoting the original source. My argument has the
following form: if incomplete understanders, in converations like this, were normally
quasi-quoters of an original source, then most of the conversations would be riddled
20
with misunderstanding; but it is absurd to suppose that there is so much
misunderstanding; therefore, the quasi-quotational construal of normal cases not right.
4. The Epistemology of Deferring
Does deferring have to stop? The considerations given in this paper do not
prove that deference can be never-ending, either linearly or circularly. The focus has
been on criticizing two theories that say deference must stop. The alleged need for a
‘stopping’ constraint is not established by these theories, because neither of them
gives a correct account of semantic deference. Until the phenomenon is correctly
identified and described, the question will remain unaswered.
Deference is not a relation of content-co-opting, nor is it a relation between
representations. Deferring is an intentional act done by a person for a reason. That to
which the agent defers is a presumed authority’s judgement. In cases where the act of
deferring is part of a social interaction, each party coordinates his actions and attitudes
with the other’s actions and attitudes. A precondition for felicitous deferring is that
both parties understand one another. Through their exchange of utterances the expert
grasps what the issue is (and the deferrer is satisfied that this is so), the deferrer
identifies what the expert’s judgement is, and - optionally - the deferrer lets the expert
know that he agrees. In accepting the expert’s judgement, the deferrer commits
himself to using the expression in accordance with the advice given. Thus the expert’s
judgement has an objective empirical content, but it also treated by D as having
normative force. D accepts that he should speak in the way that E recommends.
D has to have a good reason for entering into such a commitment. In the
standard case, D’s attitudes to E are set against a background of presumptions shared
by D and E. Both parties take for granted that there are norms which determine the
proper meaning of the word, norms to which they both owe allegiance. D defers to E
on a particular issue because D takes E to be a good guide, given the meaning that the
word already has. D does not take E to be the giver of meaning. No fact about E
constitutes the word’s meaning what it does. D knows that experts are fallible. D
regards E’s judgement as good evidence that the word means such and such, but D
does not suppose that E makes it the case that the word means such and such. The
same goes for E. Both parties mutually understand that the word has a proper meaning
which it would have had even if E had never existed.
21
But isn’t it true that when I defer to you, I treat you as my norm? In a sense, it
is true; but we have to be careful about what this amounts to. The word ‘norm’ derives
from the Latin word ‘norma’ meaning ‘carpenter’s square’; the dictionary defines
‘norm’ as standard, pattern, type. If I want to copy your pattern, I will take your
practice as my yardstick. The deferrer does treats the deferee as a yardstick, or as
Fodor says, a measuring instrument.
But that is not the whole story, for no mention has been made of the
background presumptions. I have to want to conform to your pattern for the right
reasons. Semantic deferring is underpinned by the desire to speak in accordance with
the norms that determine objectively correct use. It is to these norms that I ultimately
subscribe. If I defer to you, it is because I trust you as an interpeter of them and I let
their authority devolve to you. Under some circumstances I may revoke this
permission and withdraw my trust. Where no independent standard of right and wrong
exists, there is no deference, only subservience.
Department of Philosophy
University of Bristol
References
Burge, T. 1979: Individualism and the Mental. In P.French, T. Uehling and
H.Wettstein (eds), Midwest Studies in Philosophy Vol IV: Studies in
Metaphysics. Minneapolis: University of Minnesota Press.
Donnellan, K. 1993: There is a Word for That Kind of Thing: an Investigation of Two
Thought Experiments. Philosophical Perspectives 7, 155-71.
Fodor, J. 1994: The Elm and the Expert. Cambridge, MA: MIT Press.
Fodor, J. 1998: Concepts. Oxford: Clarendon Press.
Putnam, H. 1975: The Meaning of “Meaning”. In Putnam, Philosophical Papers Vol
II: Mind, Language and Reality. Cambridge: Cambridge University Press.
Récanati, F. 1997: Can We Believe What We Do Not Understand? Mind and
Language 12, 84-100.
22
Download