Meaning As An Inferential Role PDF

Meaning as an Inferential Role Author(s): Jaroslav Peregrin Source: Erkenntnis (1975-), Vol. 64, No. 1 (Jan., 2006), pp.
1-35 Published by: Springer Stable URL: http://www.jstor.org/stable/20013377 . Accessed: 04/10/2013 10:20
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Springer is collaborating with JSTOR to digitize, preserve and extend access to Erkenntnis (1975-).
http://www.jstor.org
This content downloaded from 128.250.144.144 on Fri, 4 Oct 2013 10:20:40 AM All use subject to JSTOR Terms and Conditions
Erkenntnis (2006) 64:1-35 DOI 10.1007/sl0670-005-4698-l
? Springer 2006
JAROSLAV PEREGRIN
MEANING
AS AN
INFERENTIAL
ROLE
ABSTRACT.
inferential meanings, paper we "semantics" reasonably
While
according
to the inferentialists, meaning

to semantics
is always a kind of
doubt that actual role, proponents approaches as they see them, can be generally to inferential In this roles. reduced a formal framework for considering the hypothesis of the "general propose
of other
often
inferentializability
and seen as
of meaning".
"inference" engendered and
We
provide
very general
which restrict kinds
definitions
of semantics to
of both
can be con?
the question study We inferences. by
ourselves
logical
stants; and especially to the question of the fesaibility of seeing themeanings of those of classical logic in an inferential way. The answer we reach is positive (although with
some provisos).
1. THE INFERENTIALIST TRADITION can be divided, with a degree theoreticians of meaning Contemporary as of oversimplification, into those seeing the meaning of an expression a matter or stands for, and of what the expression denotes principally those seeing it as a matter of how the expression is used. A prominent the latter is assumed seek the basis of among by those who an in the in the of usage meaning expression, "language games" we play a and with them is assumed it; prominent place among by those who claim that meaning is a matter of the role of the expression w.r.t. the place as Brandom the inferential are, rules, which especially our crucial for out, (1994) points game of all-important From this viewpoint, the meaning of "giving and asking for reasons". an expression its inferential role. is, principally, games, rules of the language
(1985, p. 31) characterizes the inferentialist tradition (which, according to him, can be traced back to Leibniz) in the fol?
lowing way: The philosophical tradition can be portrayed as providing two different models for the significances which are proximal objects of explicit understanding, representational
and that inferential. inference We call "representationalism" may is to be explained in favor away the of more semantically primitive reductive representational view
Brandom
relations.
2
... "inferentialism", reductive subsentential them. on order
JAROSLAV PEREGRIN
the of other one which of the would would inferential mean the
By
hand,
complementary
semantically of features containing
explanation in terms expressions
define
representational of sentences relations
Various
degrees Thus,
of commitment first
to inferentialism for
can be found
also
within
losophy.
the writing of some of the founding fathers of analytic phi?

Frege's account the concept of "conceptual
content", which he presents in his Begriffsschrift distinctively inferentialist:
(1879, pp. 2-3),
is
The contents of two judgments can differ in two ways: first, itmay be the case that [all] the consequences which may be derived from the first judgment combined with
certain the which same others can always be derived this may also not be from the the case second ... I call others; secondly is the same in both judgment the part combined of the with content
the conceptual
content.
Similarly, Tractarian meaning Mathematics

The rules
Wittgenstein
assumed
a distinctively to the more In his Remarks
inferentialist inclusive on use
stand? of of
point in a particular stage of the development of his thought from the

representationalism of the Investigations. (1956,
of of the
theory the Foundation
pp.
24,
398) we
can
read:
or right. inference for determine the They as to say I want of these signs.
meaning giving
cannot be either wrong logical ...We can conceive the rules of the signs their meaning, because they are rules signs inference
the use
Recently,
the philosophical
foundations
of
inferentialism
have
been
elaborated especially by Brandom

It is only insofar and as judgments that something as a semantic inferences
(1994, p. 144):
to in explaining the circumstances under which it is appealed so are properly of doing made and the proper consequences or states the with theorist associated qualifies expressions interpreted by a theoretical of a content. or deserves to be called concept interpr?tant,
Hence, semantic
beginning of both sions
according explanations
with go on as proprieties to explain singular
to
Brandom
(2000,
p.
30),
the
inferentialist
of the terms
inference
...
explain
propositional expressed by
content,
and
in terms expres?
such
content conceptual and predicates.1
subsentential
All
often
of this indicates that the idea of identifying meanings

roles is worth However,
with
investigating. 1960/1961, (see, e.g. Prior challenged to claiming amount that challenges straightforward 1993). The most seem to carry that some of our expressions the meanings clearly as creatures of inference. This invites the general cannot be envisaged which I would like to address in this paper:
inferential
has been its viability or Fodor and LePore
question,
MEANING AS AN INFERENTIAL ROLE

(*) Which kinds can be conferred on words by means
3
rules?
of meanings
of
inferential
I will for
neither
inferentialism
sophical framework precise relevant may not I will
nor argue a specific version of inferentialism, nor as a philosophical discuss its philo? position, - see so elsewhere have done foundations esp. Peregrin (I endorse which would though allow some to make the question points (*) reasonably I think are which significance
2001, Chapter VIII, 2004b). Instead, I will focus on establishing a

and I will make rather technical
for the debate be obvious.
their exact
philosophical
struc? that languages inferential possess simply presuppose name are sen? some in that worth its there i.e. tures, any language can tences which be (correctly) inferred from other sentences. (Can we have a language without such a structure? We can have something, but I do not think it would I will also of the word.) from the truth-valuations then, (How, commitmentfrom X does in a non-metaphoric be a language structure that is not this presuppose of the sentences or from into being? sense derived
truth-conditions.
sees it as a kind of Brandom or entitlement-preservation to say that A is inferable to X is com? is committed is to say that whoever [entitled]
it come
mitted [entitled] to A. This may lead to a finer inferential apparatus of the kind ofthat presented by Lance 1995 - but Iwill not go into these
details here.) In addition which incompatibility, ferentialist explanations. to the very related point that X\one has A amounts asserted to inference, we will sometimes also plays consider a vital the relation role within of in?
to another is connected (This important nature It is often of inference. assumed
an obligation to a prescription: to assert A when over to from the that X to the X, pass thought that A, etc. But this is obviously not the case, for such pre? thought never be it could to assert scriptions simply obeyed, being impossible or think all consequences all consequences of one's assertion, of one's more X h A a con? Hence is much as construed thought. reasonably straint: As of the exclusion it does not seem of the possibility reasonable to deny A when one has as?
serted X, i.e. the incompatibility of X with the negation of A.)

there must those be a meaning-conferring we do not of other words, of a word of other is specifiable words i.e. to presuppose inferential exclude that for each word pattern independent the possibility that the in a mutual with dependence
meanings
only that the pattern constitutive of the a of word involves other words. From this viewpoint it meaning to talk, more be better about semantics might generally, furnishing on a single word. for a language than about conferring meaning
meaning
JAROSLAV PEREGRIN is conferring on one of on other on all meanings its words may be
for a language semantics (Establishing whereas its words; conferring meaning inextricable from conferring meanings Hence it may be better if we
words.) question as
reformulate
the above
(**) Which kinds of semantics are determined by inferential rules? However, must first the theme In what "inferential". which to logical vocabulary, I will restrict my attention follows, stratum in of the most language (especially perspicuous we come to to the which have the formal regiment employ languages of the possibility natural ones). This should not be read as rejecting offers the inferential treatment of other contrary, words usual versa2). logical I believe semantics parts of our vocabulary. can of any kind of expression of inference On the be con? to be able to deal with a rigorous manner, we this question in it; and this is all the terms occurring Let us start from the term sections.
clarify or explicate for the upcoming
strued as a kind of "inferential role" (though in the case of empirical

the concept this presupposes extending to "inferences" from the world limits, However, words. here I want to restrict myself its beyond to language and vice to the simplest case of
2. A
INFERENCE AND
INCOMPATIBILITY
(S, hs), where S is a a relation between is \-$ S. If the sequence S elements of and of of elements finite sequences A then is in the relation hs to the statement An) of statements (A\,..., {strong) set whose structure inferential are called elements is an ordered pair statements and
we will write simply

A\,...,An \~sA
We
the letters X, will use the letters A, Au A2,..., B, C for statements, statements. sets If X Kfor of and Z finite for U9 thereof, sequences Y, all a set of its will be the then of is X* statements, consisting sequence statements. constituent
We define Cn(U) = {A\ there is a sequence Xsuch zndX\-sA} We will say that U is closed if Cn(U) = U. that X* ? U
We will say that(S, hs) is standard iff for every X, Y, Z, A, B, C:
(REF)A hs A
(EXT) ifX, Y hs A, then X, B,Y\-SA (CON) ifX, ?, ?, Y hs 5, then X, A,Y\-SB (PERM) ifX, A, B, Y hs C, then X, 5, ?, 7 hs C (CUT) ifX, ?, Y hs 5 and Z hs A, then X, Z, Y hs 5
The properties of hs spelled out by these sch?mas will be also and called
reflexivity, extendability, are also (The sch?mas mutation An and cut.) incompatibility set of statements and the sequence
?sA\,...,An
contractibility, permutability as identity, known thinning, structure
transitivity. per?
contraction,
A\,...,An
is an ordered pair (S, J_s)> where S is a a set of finite sequences of elements of S. If _L_sis we to will write _Ls, belongs
We
is consistent say that a set U of statements sequence X such that X* ? ?/ and _L_sX. We will say that (S, _L_s) is standard iff for every X, will
if there
is no
F, Z,
^4, B, C
(EXT) if Xs X, 7, then Xs X, A, Y (CON) if J_sX, A, A, Y, then _LS ^ ^, ^ (PERM) if ?s X, ?, 5, Y, then J_s X, 5, A, Y

Let (S, hs) be an inferential structure. A. Let us define J_s as follows:
_L_sX=Def. X
hs A forevery
The resulting incompatibility structure (S, J_s) will t>e called induced ^<S,hs>.
Let conversely (S, i_s) -U Y, X, Z be an incompatibility Y and Z structure. such that Let _LS Y, A, Z. X hs A =Def. for every
The resulting inferential structure (S, hs) will be called induced by <S,?s>.
THEOREM incompatibility is standard. 1. If an inferential induced by structure is structure it is standard. standard, If an then the
incompati?
bility structure is standard, then the inferential structure induced by it
6 PROOF. patibility tive. Hence Most we of have
JAROSLAV PEREGRIN so let us prove only that if an incom? inference is transi? then the induced that if
structure
it is trivial, is standard, to prove
(i)
and
?SW,X,A,Y,
W^forevery W and W
suchthat ?SW,B, W,
(ii)
then
_LS W, Z, W
for every W and W
such that _LS W, A, W,
(iii)
1^W,X,Z,Y,W
for every W and W
suchthat ?SW, B, W.
It is clear that (ii) is equivalent to (ii') i_s W, X, Z, F, W

?SW,X,A,Y,W and hence to
for every W, X, Y and usuell
that
(ii") for every Xand

and W such
Fit is the case that _LS W, ^ Z, F, H^ for every fF

F, W.
that UW,X,A,
And
A
it is clear that (iii) is a consequence of (i) and (ii"). D

an ordered are triple ful?
structure is (gis) inferential generalized iff the following (S, hs, J-s). It is called standard
conditions
filled: (1) (2) (3) (4) (S,hs) is standard; (S, _Ls) is standard; if -LsX then X hs A for every A; ifX hs A then _LS F, X, Z for every Y and Z such that ?SY,A, Z. fulfills the following:
A standard gis is called perfect, iff itmoreover
(5) ifX hs A for every A, then _Ls^ (i.e. J_s is induced by hs) Z for every F and Z such that ?sY,A, Z, then X\-SA F, JJT, (6) if _LS
(i.e. hs Thus, is induced byJ_s) incompatibility is reducible to inference in a perfect structure,
(X is incompatible iff everything is inferable from it) and vice versa (A is inferable form X iff everything which is incompatible with A is
also Let incompatible us now prove with X). general result concerning standard gis's.
one more
THEOREM
inconsistent PROOF. Let
2. Let (S,hs,?s)
only if ?$X. be that
be a standard gis. Then Cn(X*) is
Cn(X*) sequence Y=A\,...,An This further means
inconsistent. there
This means such
that
there
exists
of statements
exist Xl5...,X?
that 7* COz(X*) and _L_s^ so that X* ? X* and
Xi hs At. But due to the extendability and permutability
of hs, it
is incompatible follows that X hs Ai. Thus, whatever with Ax must be as _L_s is with X; hence L$X,A2,...,An, and hence, incompatible case as it is the that is whatever Then, -L_s^2, ...,An,X. permutable,
incompatible with A2 is incompatible with X,

? X, X, and so on. Ultimately, _L_s^b,... Am it is the case that J_sX. D contractible, _LsX,...,
it is the case that

X, and, as _L_sis
3. Suppose we have subset of {0,1}5.
INFERENCE AND TRUTH-PRESERVATION a set V of truth valuations of elements of S. i.e. a can be identified with subsets of a semantic system. Then we can and the property's of incompat?
(Thus, S.) The pair (S, V) will be called the relation define Ns of entailment
valuations
ibility as follows: X l=sA iff v(A) = 1 for every ve V such that v(B) = 1 for every 5gX* JJ-sX iff for no veV it is the case that v(B) = 1 for every Bex Then (S, |=s, -"-s) is a gis; and we will say that it is the gis of (S, V).
It is easily checked Let us call a gis that this gis is standard. if there is a F such that (S, hs, J_s) truth-preserving is the structure seen that standardness of We have (S, hs, J_s) (S, V). is a necessary now we will condition of truth-preservingness; show that it is also a sufficient condition hence that a gis is truth-pre? serving iff it is standard.
THEOREM
PROOF. Let
3. A gis is truth-preserving if it is standard.

V be the class of all closed and consistent subsets of S.
We will first prove that then Xhs A iffX Ns A. The direct implica? tion is straightforward: if X hs A and X* ? U for some Ue V, then
and hence, as Uis So we closed, AeU. AeCn(U) the inverse implication. Hence let X hs A. This means that whenever AeU; i.e. that AeU for every U such that only UeV have to prove
and X* C U,
JAROSLAV PEREGRIN
(i) X* cu (ii) U is consistent (i.e.Y* C U for no Y such that _LS F) = (iii) U is closed (i.e. Cn(U) U). is reflexive, JT C O? (JT). As it is transitive, Cn(Cn(X*)) = AsJrs Cn(X*). This means that Cn(X*), in the role of U, satisfies (i) and (iii),
and hence if it is consistent, then AeCn(X*). As a consequence we
have: either AeCn (X*% orfC Cn(X*) for some Fsuch that J_sF. In both cases itmust be the case that Z hs A for some sequence Z all of
whose to jf. Due to the extendability and con belong means some F A of this F that for with the hs, hs sequence tractibility same elements as Xand of hs, X hs A. hence, due to the permutability Now we will prove that J-sîff-^^; and as the direct implication to prove is again obvious, it is enough that -&$X entails _Ls^- So let it be the case that -U-sX. This means that X* ? U for no Ue V; and as members
X* C Cn(X*\
and bound V is the to be
that Cn(X*) ? U for no UeV. But as Cn(X*) is closed

set of all closed inconsistent. and consistent according subsets Hence, to Theorem of S, Cn(X*) 2, J_s^. is
Thereby the proof is finished.

This means standard; structures: and that a structure therefore we have
D
if and is truth-preserving a reason to be interested only if it is in standard
for is not truth-preservation what logic is about? True, for an inferentialist, not is but even truth-preservation prior to inference, as want to have inference she would probably explicable truth-pres? at the end of the day, if not at the beginning. ervation She might want to inverse the order of explanation and claim that 'truth is that which is preserved to standard attention There may appear the notion framework logic. This might inference'. by structures? to be reasons of inference is one Hence not to do should we pay special
so. Thus,
for example,
Lance
(1995) and Lance and Kremer
(1994, 1996) have forcefully

to the Brandomian
that argued inferentialist than classical natural an
language
to relevant, rather structure of that the true inferential - we not be standard should not expect that it indicates
appropriate that does justice
will comply to (EXT). However,

distinction. important a collection R Given of rules,
here I think we must be mindful of

we can ask what can be inferred
it seems clear ...) in their terms. And justified, (proved, sense A that in the intuitive of "inference", can, for example, always A is always of the nature of R be inferred from A, independently substantiated
justified given A, and hence (REF) appears to be vindicated. Also
ifA
can be inferred - in this sense - from X (A is justified given X), it can be inferred from anything more than X (A is justified given any
is the same, simply the justification of X ignoring superset X we should Hence from this viewpoint extra elements of X). rules. structural all the other accept (EXT), and similarly This indicates that given a "substandard" that A is the in terms inferential is inferable any also the iff
relation, from X
question whether A is inferable from X in terms of the relation is

that response of the given relation sense of the the obvious properties (in the intuitive of inference plus we in the in the and inference Hence inference have narrow, word). sense sense: whereas in terms of R in the narrow wider inferability sense to R, inferability in terms of R in the wide simply amounts amounts standard inference to the standard closure of R. Hence inferential in the wider it seems structures sense that might does inevitably lead to them. to be precisely this closure which interconnects (i.e. consequence). as truth-preserving to which according We not it claimed on that the only an is merely also on the in special attention be vindicated by the fact for answer the obvious Besides ambiguous. to it is in the given relation it, there from X if it can be obtained inferable it is so
Moreover, inference with inference
truth-preservation should be construable of our construal, account for
non-inferentialist expedient ferentialist tion. This
account, indicates
"sub-standard": structure.
why it should
according the basic whereas for
but truth-preservation, to which it underlies inferential
truth-preserva? structure be should
Furthermore, is simply the case that taken (and it is usually construal
be a part of, and extendable to, a standard on the non-inferentialist it construal the closer it is to truth-preservation the better granted that the results of Tarski and
G?del block the possibility of covering the whole of it), on the in

ferentialist be a unique vation must Hence account inference fact, on not be the case. However, there should to it for the way truth-preservation: truth-preser? have arisen from inference. even if we accept in favor of the relevantist the arguments this need from there is a sense inference, standard and thus amounting construal, + relevance] basic ? than inference, we cannot transform this into in which we can still see In to truth-preservation.
for as
the relevantist =
inference but take as we the
truth-preservation take truth as more
do not
'underlying'
at face value = inference
and we must relevance
truth-preservation
10
JAROSLAV PEREGRIN
And
the "subtraction" closure. it seems
of relevance has been
seems
to amount
to forming
the and and
standard Also inferential
that what
structures
delimit of incompatible incompatibility a set being sense of in the narrow sets, then besides incompatible sense in a is of this collection there the wider member again being
applies structures: If we
mutatis
inference just said about to incompatibility mutandis a collection
which
nition But we are more
it is incompatible iff its incompatibility follows from the defi?

of the collection should note the obvious properties of incompatibility. plus that generalized inferential structures, especially
the rules (3) and (4) interconnecting

deeper non-relevantists:3 she admitted
inference and incompatibility,
a In fact, they may mark from this viewpoint. problematic the relevantists and issue between than just terminological it seems that there is a sense to be possible that of "inference" the sense in which inference the relevantist, eved in which inference she would prefer), if is
standard always there is a sense ways these
this is not (thought on "incompatibility" still deny that
and is al? in of
standard, may senses are tied are also
There inferential
at face value, then language we can take truth-preservation itself as we might whether wonder ac? sentences of natural is that many The point standard. language a context: thus though within embedded only when quire truth-values it He is bald does not have a truth value the statement by itself, we is wise. Hence The king of France following is He entails is wise bald France followed The may say that by king of is surely not entailed is bald, but this statement The king of France by us a This is wise. France The is bald followed He gives king of by we can see truth-preservation itself as reason to wonder whether acquires one when to this, of course, is to solution to (PERM). The standard complying are not context to those sentences which restrict logical investigations it at face of the also remains but there taking possibility dependent; to semantics from the different lead to an approach value which may one investigated in all, we All the in press-a, for a sketch). (see Peregrin and that both from the non-inferentialist conclude as can see we inference inferentialist "approxi? viewpoint here can be can we ta? get for example, the alleviating
by together different arguments structures. If we take natural
the ex falso against
incompatibility and incompatibility rule (3). quodlibet the standardness
from
the "approximating" (though truth-preservation mating" in the former ken literally generally, case). How, only we saw that, out inference? of Well, truth-preservation the relation of truth-preservation arising from
MEANING AS AN INFERENTIAL ROLE relevance
11
minimal
amount to the from relevantist inference might requirement standard relation the How? inferential relation. containing to the topic, which will take a more ever, we will general approach our explication from the of the terms "semantics" and emerge "determines". But aspect beforehand, of the situation. let us consider one more non-infer
entialist
4. from of
EXPRESSIVE RESOURCES OF SEMANTIC SYSTEMS the inferentialist from
Though means
something to "capture" uations
we establish semantics viewpoint, by a more common semantics is inferences, viewpoint, is here prior to inferences which and we use inferences only it. From the latter viewpoint, delimit what the acceptable truth-val? is possible, i.e. represent can be semantically characterized classes and of possible languages statements;
system and a statement worlds", "possible in which it is true. Hence by the class of worlds are potential values semantic of differ as to their "expressive power". statements all possible in the order in which and
of a semantic
worlds may two
Take the system ({A, 5}, {{A, 5}, {A}, {?}, 0}), i.e. a system with
truth-valuations. are If we number the valuations statement we whereas they A belongs to 1 and 2, whereas switch to the possible-world-perspective, B we expresses can, these for {1,3}. There add more is no a listed, we can see that the to 1 and 3; hence B belongs if then A statement statement expresses expressing C expressing {1,2}, {1}, or {1}:
{1,4}, or, say, {2,3,4}. This can be improved by extending the lan?
guage: example,
({A,B,C},{{A,B,C},{A},{B},?})4
rigorous, we need some more a Let F= semantic be terminology. system. For every state? (S,V) ment A from S9 let \A\ denote the set of all and only elements of V which contain A; hence let considerations To make
MI=Def.
{Ue
v\AeU}
A subset V of V is called expressible inF iff there is an A <E S so that F is called iff subset of V is \A\^V. every (fully) expressible
expressible. F is called Boolean set is expressible and expressible is again expressible. is obvious (It expressible, algebra.) then its statements iff the complement of any of any two expressible sets that if a semantic is Boolean system can be seen as constituting a Boolean expressible the union
12 Fully semantic our expressible
JAROSLAV PEREGRIN systems constitute in some a proper subset of the set of we to restrict may want respects is that it seems that if our ultimate lack of for var? 'nature' some the it or
point not then we should take a language, resources too A natural seriously. expressive language may, lack some words ious contingent and consequently reasons, target is natural to be a matter of its
systems; however, to them. The attention
but this does not seem sentences, are always flexible natural languages resources new creation of expressive may often seem reasonable to simply
to take in their stride enough whenever Therefore needed.5 presuppose some full expressibility,
at least something close to it (like Boolean expressibility).

Fully semantic theorem: THEOREM 4. The gis of an expressible semantic system is perfect. expressible systems have systems have. An example not all which properties is spelled out by the following
PROOF. Let X Ns A for every A. Let # be an element of S such that |#|= 0. Then X Ns # and hence X cannot be part of any element of V
which neither to no such element, not contain #. But as # belongs now case that -U-sY,X,Z can X, and hence Let it be the -U-sX. Now there is a Ue V for every 7 and Z such that -tt-Y,A,Z. suppose does
= such that X* C U, but A % U. Let BeS be such that \B\ {U}. Then
-U-sA,B9 can see that but not -1LSX, B. D
obviously We
the proof ton. This it can be the empty
of the semantic the only properties system used in set and of every single? of the empty the expressibility and indeed indicates that we do not need full expressibility; or even expressibility of that Boolean shown expressibility, are set plus the expressibility set is enough.6 of the complement of every
expressible
5.
SEMANTIC SYSTEMS AND SEMANTICS
we have a non sense only provided the question (**) makes Clearly i.e. little sense to it makes of "semantics" inferentialist explication of such an explication. an inferentialist the possibility who denies such an independent the fact that we will provide However, expli? cation tialism point should into of our not effort be construed is in checking as building the foundation of our of inferen? the rejection the the On contrary, approach. the prima facie whether indepen
MEANING AS AN INFERENTIAL ROLE dence could
13
and if so, then inferentialism be eliminated perhaps we to which to be vindicated would appear (at least to the extent is quite general). of semantics admit that our delimitation of semantics So how can we delimit a general concept independent
of inferences? Elsewhere
eralization used
(see Peregrin 1995) I argued that the gen?
as the concept is of semantics of the concept interpretation, ver? to "minimal" in logic, comes down mapping compositional I subsequently others. and falsifying argued ifying some sentences still often sense be as amounting to the same semantics (in the I concluded and that what makes the word); are differences the different within substantially seen
(Peregrin 1997) that different semantic interpretations of this kind

can intuitive of
interpretations respective In other
truth valuations which spaces of acceptable they institute. to the conclusion I came that the most words, general is a space of truth-valuations, i.e. that it is of "semantics" explication of semantic system. This conclusion concept by our above provided accords with the approach pioneered paper by van Fraassen for (1971) and I
recently elaborated by Dunn and Hardegree

the scope of the present beyond give here only a digest. seems Semantic interpretation truth-valuation of sentences:
(2000). As it would be
this at length,
to argue
to go hand in hand with a obviously sentences at least some of them), by (or or true this become false. However, semantically being interpreted, mean does not necessarily that semantic fixes the truth interpretation a as sun shines" sentence values of all sentences such "The surely true or false by being made to mean what does not become it does. on to What semantic does is limits interpretation generally impose truth-valuations: that if "The sun shines" e.g., it determines possible is true, then "The sun does not shine" must be false; hence that the sentence to be "The sun shines and the sun does not shine" is bound that semantic should put false, etc. This means always interpretation some constraints on the possible of truth-valuations sentences.7 of Moreover, many philosophers language (most notably to have must that all there is consist Davidson, 1984) argued meaning in truth conditions. Now let us think about the ways truth conditions can be articulated: we must say something of the form
X is true iff F,
where X is replaced of the conditions the truth conditions and F by a description by the name of a sentence i.e. a sentence. Hence we need a language in which - a are expressed then However, metalanguage.
14 our theory will work of
JAROSLAV PEREGRIN only at face value the of long as we take the semantics in fact we will merely have reduced to a sentence considered sentence, X, so the the of
metalanguage truth conditions
the metalanguage, the one replacing 7 namely the semantics of the latter be explicated equally X would recourse sources Hence obviously set an infinite regress instead the like
to require And that as that of rigorously
in motion. to make do with the re?
This indicates that it might be desirable to refrain from having

to a metalanguage and of the object language, suppose that we would language to use a under sentence investigation. of this very
language in place of 7 Which sentence should it be? The truth conditions of X are clearly best captured by X itself; but using X in
truism. But, at least place of 7 would clearly result in an uninteresting in some cases, there is the possibility sentence of using a different of the same language. So let us assume that we use a sentence Z in place of 7 of true if..." or "Xis true only if..." with Z in place "Xis Saying to claiming the "..." amounts that X is entailed by Z and that X a entails Z, is mammal" '"Fido is true if respectively. (Claiming is a dog" to also is claiming that that "Fido is a mammal" is entailed by
"Fido
"Fido
amounts verifies
is a dog'".) And
claiming - or X that
claiming that X
is entailed by Z
in turn
acceptable. valuations which other
Hence, that part of the specification represents can be accounted for without mobilizing
Z truth-valuation which verifies every so not not truth-valuation is any doing the specification of the range of acceptable truth of truth-conditions the resources of an?
language.8 If we accept this, then the question (**) turns on the relationship of sentences) between semantic of truth-valuations systems (spaces structures between finite of sen? and inferential sequence (relations on the way in which tences and sentences), in particular the latter are of "determining" capable tion to this determining. the former. So let us now turn our atten?
6. An inference
THE INFERENTIALIZABILLITY OF SEMANTICS can be seen as a means certain truth-val? of excluding can seen as X h A be stipulating
uations
of the underlying language: which contain X and do not contain A. all truth-valuations excluding a certain structure determines semantic In this sense, every inferential are we in if that truth condi? agree system meanings grounded (and tions, thereby it also confers meanings on the elements of
the
MEANING AS AN INFERENTIAL ROLE underlying meanings language). are conferable And hence
15
which question structures. This
the question which kinds of is with connected the inferentially intimately can be determined semantic systems by inferential leads to the definition: the gis (S, h, _L) determines the
semantic system {S, V), where V is the set of all v which fulfill the
following conditions:
= = (i) if v(B) 1 for every constituent B of X and X\~ A, then v(A) 1; (ii) if _LX, then v(A) * 1 for at least one AeX*.
Now seen the latter that every structure inferential seem trivial: question might prima facie semantic structure; system has an inferential not determine this very inferential system? structure we does have this
answer
semantic system system (though, a system which has the same inferential course, structure). Let S= {A,B} and let F consist of the two "truth-value-swapping" i.e. the valuations all valuations, {^4} and {B}. Let us consider instances possible ations we exclude of by inference for S, and for each of them its adoption:
an is notoriously negative: a determine might different
the However, of a semantic of
the
the valu?
hA hB
A\-A
0, {B} 0,{^}
A -{B}
B\-A
A,B\B\-B
A\-B
A,B\-B
{A}
no combination {A, B}; of the inferences is capable is capable of of
This
means
that
excluding
the valuation
and also no combination
excluding 0 without excluding either {A} or {B}. In other words, no inferential structure determines the system ({A, B},{{A},{B}}).
Now bility, in addition, the possible consider, and the valuations excluded by them: instances of incompati?
10 LA LB
LA,B
0,{A},{B},{A,B} {A},{A,B} {B},{A,B}

{A,B}
With their aid, it becomes possible to exclude {A,B}, by stipulating LA, B. However, it is still not possible to exclude 0 without excluding either {A} or {B}. Hence no gis determines ({A, B},{{A},{B}}). Now
16
JAROSLAV PEREGRIN we have just considered is the negation of A (and vice
this appears for the semantics alarming: to make B into needed is what precisely such an ordinary In fact, this characterize its negation However, statement
versa). This indicates that inferentialism might fall short of conferring

as that of the standard negation. meaning we not be surprising at all. What should a if is the that statement (i) negation stipulation need to
isfalse, and (ii) if a statement isfalse, its negation we can in terms of inferences what is that a stipulate are true. In terms of statements is true if some other can also
is true, is true.
cannot be that some statements stipulate are true, a statement is that if some statements cover (ii). but we still cannot (i) the situation is similar w.r.t. disjunction As can be easily observed, In the former and implication. the valua? case, it is easy to exclude one true and the disjunction false of the disjuncts tions which make
we incompatibility true jointly, hence covers false, which
(by the inferences A h A v B and B h A v B), but we cannot exclude all those which make the disjunction true and both disjuncts false. In
that the implication is true if the easily guarantee is false or is true, and that it is true only if the antecedent consequent ?? ?> B h B), but we cannot B and A, A is true (B h A the consequent true if the it is antecedent is false. that guarantee the Does with of not this mean that of the standard semantics for the not classical the com? it is indeed The axi latter, we can
propositional
calculus is not inferential? And if so, how does it square

the very calculus that the axiomatic (i.e. with not the semantic contradict for does inferential) one? In fact, its completeness. delimitation
the completeness show proof pleteness the calculus
coincides
which does inferential, of the calculus omatization structure does of not
yields us its inferential of the semantics determine
but this structure, a As the calculus.
matter
another it determines which, however, semantics, fact, vindi? set is what with the calculus of tautologies shares the (which cates the completeness proof). 7. THE GENTZENIAN GENERALIZATION different from the one used so far and write
Let us now X\ instead of ?X.
adopt
a notation
17
under one roof In this way, we can get inference and incompatibility a as h finite of statements to treat relation between sequences starting of length not greater than one. The of statements and finite sequences
ordered pair (S, hs) with hs of this kind will be called a weak infer?
ential holds structure. (where G Such a structure will be called standard of length if the following at most one): is a sequence of statements
(REF)A hs A
(EXT) ifX, F hs G, then X, B, Y hs G (CON) ifX, A, A, Y hs G, then X, A, Y hs G (PERM) ifX, A, B, Y hs G, then X, B, A,YhsG (CUT) ifX, A,Y\-SG and Z hs A, then X, Z, F hs G (EXT7) ifXhs,thenXhs,4
inferential If (S, hs) is a weak structure which arises out of structure, then hs to the strong instances inferential with
non? restricting It restriction. will be its is obvi? called sides, strong empty right-hand ous that the strong restriction structure is itself of a standard weak hand, we restrict hs to instances with empty an which will be structure, incompatibility right-hand structure. restriction It is easy of the original called the incompatibility a structure to show that if is standard, then both its strong restriction standard. If, on the other sides, we get and make its incompatibility up a standard condition than restriction are also inferential that we standard. structure. can add statements this on Moreover, they generalized indicates (EXT7)
The
the right-hand side of hs (of course if we thereby do not make

what, then, about 1). However, longer i.e. for finite sequences tion, arbitrary allowing relaxing on the right restric?
it
side of
hs, and letting the right hand side be freely expandable just as the left hand side is? It is clear that what we reach in this way is in
fact Gentzen's of this kind will structure sequent will be be called calculus. called a standard The ordered quasiinferential if the following pair (5, hs) structure. holds: with Such hs a
(REF)A hs A
(EXT) ifX, F hs ?/, F, then X, A, Y\-SU,V and X, Fhs U,A, F(CON) i?X,A,A, Fhs U, then,v4, Fhs U; ifXhs U,A,A, KthenXhs U,A,V (PERM) ifX,A,B, Fhs U, thenX, B,A, Fhs U\ ifXhs U,A,B, FthenXhs U,B,A, V (CUT) ifX, A, Y hs U and Z hs K, A, W,
then X,Z, Fhs V,U,W
18
JAROSLAV PEREGRIN
If (S, hs) is a quasiinferential structure which arises out of
structure, restricting of out this of
then hs to
the weak instances i.e. hs to
inferential with the right
hand
side of length not greater than 1, will be called

The restriction strong structure which arises it is obvious restriction, restricting
its weak
strong instances
restriction. inferential restriction. weak with reason clusion
with right-hand side of length precisely

that
1, will be called its strong
if a qis is standard, then both its Again, are standard.The restriction and its strong restriction problem structure this version is that there seems to be of an inferential a single-conclusion to prefer one. Thus, if we subscribe we submit that as committed inference to are the multiple-con? the Brandomian variety originally to something, to something a matter in particular as bringing to of of
inferentialism,
inferences or entitled
treating people as treating their commitment/entitlement about their commitment/entitlement is easily what it takes
imaginable to one or more committed/entitled more it is much rewards), complicated take to treat her as committed/entitled things. Moreover, based normally Tennant (1997, it seems on the that
to something else. And while it as implicitly treat somebody and things (in terms of sanctions to to imagine what to at least one actual of notion it would of many is as
the form
of our
single-conclusion take it, "in normal puts arguments practice, one from premises to a single conclusion".9 Be it as it may, it is are contrast to in inferential that ones, structures, quasiinferential over a set statements. finite of of any system capable determining 320)
arguments inference:
Hence
we call
if we call a semantic system (S, V) finite

it semifinite 5. For iff V every is finite, semifinite we can claim
iff S is finite, and if
THEOREM finite.
semantic
system
(S, V)
there
is a
qis (S, h) that determines
it;moreover,
if S is finite, then also h is
PROOF.
V=
Let (5, V) be a semantic system and let V be finite. Let
of S that do not and let V* be the set of all valuations {vlv..vw}; we we construct to what will call the critical For ve V. V* every belong v in As for each j= 1,..., the way. following v0K, / v7 quasiinference
m. Hence for each j there is a sentence A{ GS such that v(A{) ^ Vj(A{); i.e. such that either (i) v(A{) = 1 and Vj(A{)= 0, or (ii) v{A{) = 0 and
? 1. Assume, Vj{A{) case (i) are A\,..., for the sake A* and of simplicity, in case the sentences that the sentences A in . (ii) are A*+x,...,
Let

a\ Ak i_ jk+\ aw
19
be
the
v-critical
quasiinference and
quasiinference. does not exclude case v7 is excluded.
= < VjEV, either j k, and then Vj(A{) 0, or j>k,

in neither On
the case that is obviously from valuation V; for given any It the other hand,
this any
and then Vj(A{)= 1;

the v-critical
? 0 = 1 when j< k and v(A{) quasiinference excludes v; for v(A{) when j>k. Now it is clear that (S, h), where h consists of the
v-critical finite quasiinferences number of for every instances. veK*, D determines (S, V). It is also
clear that if S is finite, then also V* is finite and hence h consists of

only
now other follows is: what about which naturally question are not semifinite semantic also deter? systems which systems? Are seem not that It but of them would mined are, many surely by qis's? here. The reason is that but we will not go into these questions are10 The determinedness a system infer? by a qis is not yet what would make sense. Hence we will now try to explicate in the intuitive the ential more intuitive of concept inferentiality adequately.
8. Let us now we we have have return
FINITE BASES
to the enterprise
dealt with so far not
of explication of the question (**): "determined" and "inferential", but "semantics", tackled "rules". The point is that the idea behind furnish expressions, and speakers, who - we treat their inferential the power from another by (perhaps she is committed idea from taking one to to the latter) is that we have a finite a set of other of the rules.11 to inferential of inference rules. first
inferentialism consequently statements be committed as
it is us, with languages, inferable one to the former
is that
whenever
and as incompatible with each other. number of rules and that a statement statements This structures What element if it can be derived that we should of a specific from means
The
is inferable them with our
restrict
the help attention relation
is that of constituent
kind, namely collection inferability by means of a finite of inferential us an is inferential rule? Let call an ordered pair whose is a finite of S (finite
those whose
an of elements of S and the second sequence a over of elements of sequence S) (quasi)inference if P over is a set be ("of parameters"), an (quasi)inference then over a (P-) S in
S. (Hence if (S, hs) is a (quasi)inferential structure, then hs is a set of

(quasi)inferences.) (quasi)inferential Now rule S will
20 which stance some elements
JAROSLAV PEREGRIN of S are replaced by elements be any of P.12 (We will
usually write Xh
7 instead of the less perspicuous

rule over S will
(X, 7)). An
in?
of a(quasi)inferential (quasi)inference over S which can be gained from the rule by a systematic replacement of the elements of P by the elements of S. (REF),
A\-A.
for example,
is an
inferential
rule:
Its quasiinferential rule:
form,
then,
is an
example
of
a quasiinferential
XhX
more rules emerge that However, interesting only when we assume the set of statements is somehow structured. for for If, every example, as A a B, we can two statements A and B there is a statement denoted have the pattern
AaBÂ
AaB\-B A,B\-AaB A a B as the conjunction establishing So let us assume we have fixed relation them. of inference How? We which interests of A and B. some inferential us is the one which
rules, and the 'derives from'
some way need inferences of inferring obviously some metainferences or metainferential from inferences, rules. Hence we introduce over S, which is an the concept of meta(quasi)inference ordered ences pair whose over S and first whose constituent second is a finite is an of infer? sequence over inference S. A of
(P-)meta(quasi)inferential over S with some elements P. We by will a slash separate and we down Y\we will
rule over S will from the
of S in its constituents
be a meta(quasi)inference replaced by those
the antecedent separate
the consequent elements of its antecedent
of such a rule by
semicolons. Thus, the metainferential

be written X,A, Now basic what as follows: U;Zh want V,A,W?X,Z,YV is that the
rule constituted by (CUT) will
V,U,W relation by means derives of basis some from the
inference rules
finitely specified metainferential specified
inferential rules: A
{quasiinferential
finitely is an ordered
21
triple (S,R,M), where S is a set, R is a finite set of (quasi)inferential

rules over is a finite set of meta(quasi)inferential S and M a ante? in M have that all metarules S. (Let us assume non-empty can be treated with the empty antecedent for metarules cedent rules over simply as rules and put into R.) The (quasiinferential structure gen?
erated
structure whose is the (quasi)inferential by (S, R, M) (quasi)inferential relation is the smallest class of (quasi)inferences
contains all instances of elements of R and is closed to
over S which
all instances of elements ofM. A (quasi)inferential structure is called finitely generated iff it is generated by a (quasi)inferential basis. A semantic system is called finitely (quasiinferential iff it is determined
(quasi)inferential by a finitely generated Now it is clear that as far as finite structure. languages are concerned,
(quasi)inferentiality and finite (quasi)inferentiality THEOREM
simply coincide.
6. Every finite (quasi)inferential semantic system is fi? nitely (quasi)inferential. If the number of statements is finite, then there obviously is
number of of (quasi) inferences non-equivalent the determination of the system. D from the
PROOF.
only viewpoint The Take arithmetic
a finite
situation the
in respect to infinite languages. is, of course, different semantic constituted of Peano system by the language a state? and the truth-valuation which maps (PA) single the needed inferential relation and consists of all
ment on truth iff it is true in the standard model. This system is

the empty antecedent true in the standard model in the consequent. true within statements the standard model have merable, the semantic to more system. The than system is surely (trivially) inferences inferential: which a statement of PA
as the class of However, enu? is not recursively not finitely inferential.
But
amounts ential ferential
in fact it seems that inferentiality

delimitation relation inference
in the intuitive sense
infer? by any kind of a finitely of the systems we aim at should
be derived from the basic inferential rules not by just any metain?
If R is a set of inferential rules, but in a quite specific way. to say that A is inferable, of R, from X rules, then we want by means iff there is a sequence of statements A with and such that each ending an or its of element is either element of X is the consequent of an instance occur tial, of a rule earlier rather from R such that all elements of the antecedent inferen? it as a an in the sequence. is, strictly speaking, (REF than a metainferential rule. But we can regard
22 metainferential consisting derivable
JAROSLAV PEREGRIN rule with an empty antecedent.) to M This amounts of the five Gentzenian structural rules. Indeed, A is infer?
able from X by means

from R by
of R if and only if the inference X h A is

rules. D
these
THEOREM
(in the sense
7. A is inferable from X by means of the rules from R

that there exists a 'proof) just in case Xis inferable from
R by means of REF, CON, EXT, PERM and CUT.

As the proof of the inverse implication is straightforward, us one. let Hence let A be inferable from X. This prove only the direct means that there is a sequence of statements such that Ax,...,An = A and an or X is element of is either inferable every A? An by a rule are among Ax,.. n= If which from R from statements 1, then .,Ai_x. PROOF.
there are two possibilities: either AeX* and then Xh A follows from REF by EXT; or A is a consequent of a rule from R with a void antecedent, and then h A and hence X h A due to EXT. If n > 1 and
An Ah,..., is inferable from Aim h A, where a rule from R, then Ai{,.. .,Aim by X h AJ forj= X h A due 1,.. .,m. Then X,..., some
to CUT, and hence X h A due to PERM and CON. D leads us to the following definition: We will call a (quasi)inferential basis standard iff R contains REF and M contains CON, EXT, PERM and CUT (hence if the (quasi)inferential basis is This
standard, standard strictly generated standard, then in the standard the (quasi)inferential sense of the earlier iff, moreover, M structure definition). contains which And no we it generates will call rules. is it A
other
(quasi)inferential
structure will be called strictly standard if it is

basis. inferential (Hence every structure is
(quasi)inferential by a strictly standard and especially every strictly standard
finitely generated.) we implicitly inferences also It seems that in stipulating stipulate are derivable from them by the structural which all the inferences are in structures which hence we should be interested rules only seem or at It that it is least standard. strictly might strictly standard, standard the role inferential of the systems which of are an the most "inferential natural candidate for system"; explicatum inferential is that no finitely the trouble system (and hence the inferential one) is capable of accommodating standardly of classical operators logic. it is possible conjunction to fix the usual truth-functional by means of the obvious of meaning inferential pattern, semantic
however, surely no simplest
Though the classical
MEANING AS AN INFERENTIAL ROLE the same is not for reasons sketched earlier possible, and nor for the classical disjunction is to fix the truth-functional by means of ?/was/inferential
23
classical tion. What classical way:
negation is possible operators
(?6), for the and implica? of all the meanings e.g. in this
patterns,
A,Â\
V A,Â A\-AvB
BhAvB AvBh A,B

B\-A^B A,A-^B\-B
\-A,A->B
Hence
strictly
(as discovered already by Gentzen)

standardly quasiinferential. inferential.
all of classical
it is not
logic is
strictly
Nevertheless,
standardly For an even have
this situation need not be too inferentialist, if she wants to restrict herself to single-conclusion to believe she might reasons, claim, independent
embarrassing inferences. that the only
(primordial) way to furnish an expression with a meaning

not
is to let it
be governed are rules; so if there are 'meanings' which by inferential are in this way, conferable not worth the name. they meanings But things are not this simple. We have seen that many meanings of a
very familiar and seemingly indispensable kind fall into the non
inferential Classical category. negation the standard semantics for arithmetic. these are non-meanings? To be sure, the inferentialist are the may meanings that all the others engineering. tionist ments 'natural' or disjunction; not Is the inferentialist defend the line to mention saying that
that
straightforwardly are late-coming of our artificial products She might claim that the only 'natural' logical and that the classical ones are their and
inferential
the only ones; and language constants
are some which are delimitable

ones), available with
inferentially (presumably the intui?

artificial through adjust? explicit reflections
tampering However, in this way,
only after metalogical the natural meanings. not want to have no
if she does she appears
to let classical choice but
logic go by the board to settle for (strictly
quasiinferential systems. The latter, as we saw, are strong standardly) for the classical but as pointed out above, there are enough operators, reasons to see the multiple-conclusion as less natural than sequents
24 the single-conclusion strictly ones.
JAROSLAV PEREGRIN Fortunately there is a sense can be after in which regarded all enjoy officially every as a some admit?
standardly quasiinferential inferential system, (standardly) of
system so we may systems
the quasiinferential advantages the multi-conclusion sequents. ting
without
9. What we will system class of we need
THE EMULATION THEOREM show now is that for every strictly standardly a standardly inferential system tautologies). (and hence especially definitions:
exactly
quasiinferential the same with First, however,
there
exists
inferences some more
If X h A\,..., An is a will be the metain rule over S, then its emulation quasiinferential = B h h YXZ rule YAXZY- B-...; ferential 0, i.e. if the YAnZ (if n B/ is emulation X of the the antecedent then the is of rule h, shape
R,M) will be the empty). An emulation of a quasiinferential basis (*S, inferential basis (5, Rf,M') such that R' is the set of all those elements
of R which the union proper (i.e. not quasiinferential) of all elements of the set of restrictions inferential the emulations rules. of all elements are is rules, and M' to M inferences of are proper
and
of R which
quasiinferential of a strictly standard that an emulation Now we are going to prove an structure inferential basis generates quasiinferential (5, i?,M) the qis results from taking to the structure which is identical which generated THEOREM structure by (5, i?,M) 8. The is its strong and dropping emulation restriction. all genuine quasiinferences: quasiinferential
of a strictly
standard
PROOF. (S,R,M)
(5, Rf, Mf).
Let (S, hs) be the quasiinferential structure generated by and let (5, h?) be the inferential structure generated by
What we must show is that for every sequence X of ele?
ments
X hg of R, which
of S and every element A of S it is the case that X hs A iff

the inverse first. As R' is a subset implication rule from M' every metainferential
A. Let us consider
to show that it is enough is not an element of M preserves hs, i.e. that for every such rule case if X\ hs A\,..., that is h A the it h An/X ; Xn X\ h A\;... rule each metainferential then also Xhs^4. However, Xn hs An, of M' but not of M must be, due to the definition is an element which of the former, an emulation of a quasiinferential rule from R, i.e.
must be of the form FîZ h B;...;
YAnZ h B / YXZ h B, where
MEANING AS AN INFERENTIAL ROLE X h A\,..., X h A\,..., An belongs to R. Hence to R what we have
25
An belongs (and <mdYAxZ hs B,..., YAnZ hs B, then YXZ hs 5. of hs if n = 0, then lows from the standardness > and if n 0, then in the following way: X hs A\,..., YA\Z\~s YXZhs B BA2,...,An An assumption assumption (CUT) (CUT)
hence
to prove is that if X hs A\,..., An) But this easily fol? simply by (EXT),
Y...YXZ...Z\-SB...B
YXZ hs 5
(PERM)and (CON)
Hence ifX h? A, then X hs A. The proof of the direct implication, i.e. of the fact that ifX hs A,
then X\~l A, h* B,..., 7îZ Xhs ,4 and is trickier. We 7v4?Z h* B, then some that if Xhs prove A\,... ,An and 7XZ B. then From this it follows that h? will X\-*SA; notational and as AY-\A due to (REF),
if X hs ^ and YAZ h? B, then 7XZ h? 5; and in particular that if

A\-\A9 need
Xhs A entails Xh^.

First, we will then we will use conventions. If X=Ai,...,An
xhh; 7
as a shorthand
A\ hs 7;...
for
; ^4M rs 7.
Moreover,
[u]x[V\hh; y
will be the shorthand for
UAxV\~l Hence such that Y;...;UAnV\~l we need Y
now what
X hs A\,...,
to prove is that for every Xand case it is the that TîZ An,
Ax,...,An h^ B,...,
YAnZ h* B entail 7XZ h* B for every 7, Z and B. We will proceed

that X h A\,..., to R. Then if First, assume by induction. An belongs n ? contains its emulation, i.e. the metainferential rule 1, then M' means This that if YAXZ YAiZt-B,...,YAnZhB/YXZ\-B. h* B,..., on the other hand, YAnZ h? B, then indeed 7XZ h? B. If,
26
JAROSLAV PEREGRIN
n= 1, then XV A belongs toK and hence X\-\ A; and the fact that YAZ h* B entails YXZ h* B follows by (CUT).
Now assume a metaquasiinferential will prove only that X h A\,..., that X h A\,..., of An is the result of an application rule from M. As stan? is strictly (S,R,M)
dard, the only possibilities

the
are CON, EXT, PERM and CUT. We
case of CUT. assume less perspicuous Hence can be written in the form X, Z, F hs K, ?/, W so An
that
X,^, and, by Fhsi7, induction and hypothesis, \-sV,A,W; that
1. [M] h* 5, and U[N] hh* B entailsM, X, ?, F, JV hh* 2. [M] B h* B. TV entails K, ?, 0^[7V] M, Z, What we want to prove is that then [M] V, ?/,W[7V] hh? B entails assume h* TV 5. Hence that M, X, Z, F, V, U, W[N] hh* B. This is [M]
to say, we assume
3. [M]V[N]\-hlB, 4. [M]U[N] hh* B, and 5. [M]H^V]hh*5.

3. and 5. yield, via (EXT),
6. [M,X]F[F,7V] hh* ?, and 7. [M,X]^[F,7V]hh*5;

whereas 8. M,X,A, Now 4. and 1. yield B. amount to
F,iVh*
6, 7 and
8 together
9. [M,X]F,^,^,[F,7V]hh*5,
from which M,X,Z, via 2. D we get
F,7Vh*?
COROLLARY.
For every strictly standardly system quasiinferential a the same class of inferential there exists system with standardly a there exists for every Hence such system inferences. especially same class of inferential with the system tautologies. standardly
27
10. AN EXAMPLE: CLASSICAL PROPOSITIONAL LOGIC IN THE LIGHT OF INFERENCES
We
saw that not even the classical propositional

in the sense that there are that inferences the of which by is constituted
logic (CPL) is
can delimit the usual the explicit have just
inferential
very class of truth-valuations semantic of CPL. definition proved not while semantic inferential -i tells us that there the
However, is a standard semantics the same CPL
we corollary structure inferential does
determining
system possessing structure is it? if we
CPL, class of on
which, a determine Which
tautologies.
It is easy to see that the semantics and a, rules:
base
of CPL
the primitive operators is determined by the following
quasiinferential
(2)AaB^ B
(3)A,B\-AaB
(4)A,-iA\
(\)AaB\~A
(5) h A,-*A
What those we must not having do is to replace the genuine rules quasiinferential one statement in the exactly single consequent, (i.e. (4)
and (5)) with their emulations. This is to say that we must replace (4) and (5) by (4')AÂ\-B (5f)X,Ah B;X,Âh Note
possible
B/Xh
that in view of the fact that (1), (2), (3), (4') and (5') constitute a
axiomatization of CPL amounts
of CPL, the fact that they determine the to the result for the tautologies completeness logic. But it is, in a sense, more than the usual one and it throws general some new light on the fact that the axioms of classical logic, despite their completeness, do not pin down the denotations of the operators to the standard are truth-functions. is that the axioms (The point even some with non-standard with compatible interpretations of some falsities some false and with of negations being disjunctions
pairs of falsities being true.What

and
is the case is that if the axioms hold
are truth functions, then they are if the denotations of the operators to be the standard bound are com? truth functions. But the axioms patible with the From indicated non-truth-functional our vantage point we can interpretation see that classical of logic the is
constants.13)
28 complete class of semantics semantics. in the sense tautologies of CPL;
JAROSLAV PEREGRIN that which that they, its axioms is the determine same do as not a semantics that of the determine with the
standard this very
however,
Let us give some illustrations of how proofs within (l)-(5) get emulated by those within (l)-(3), (4') and (5'). Consider the inference
Â h A,
which is valid inCPL. With

1. -,A,Â\-
(4) and (5) it can be proved rather easily:

(4)
2. h A,Â 3. -.-..4 h
This gets emulated
(5) from 1 and 2 by (CUT)
as follows:
-,AhA 1. Â, 2. -.-.?, A h A

3. -.-..4 h A
from(4') by(PERM) from (REF) by (EXT)

from 1, 2 by (5')
Or consider the proof of the theorem

-.(?A-1.4)
1. AaÂ?-A
2. yiA-.i4h-.i4 3. A,Â\-
(1)
(2) (4)
4. A a -vi h
5.
from 1, 2 and 3 by (CUT) and(CON)

(5)
6. h ->{Aa ->A)
The emulation
\-Aa->A,-?(Aa-Â)
from 4 and 5 by (CUT)

looks as follows:
now
1. AaÂ\-A 2. AaÂ\-Â (2) 3. A,Â[-^{AaÂ) 4. A a Â h -.(i? a -.i?) 5. ->(A a -iii) h ^{A a -,A) 6. h -,(A a Â)
This means in a rather very space weak that classical sense: there
(1) (4') from 1, 2 and 3 by (CUT) and (CON) (REF) from 4 and 5 by (5')
delimiting tautologies
of acceptable a space of truth valuations to it w.r.t. which is equivalent more inferences). (and generally single-conclusion
though logic may be seen as inferential, is no inferential its way of delimiting a is there truth-valuations; however, way of
MEANING AS AN INFERENTIAL ROLE 11. EXTREMALITY CONDITIONS Hence intuitive dardly lection what seems of to be a good "inferential candidate semantics" for the explication is the concept of of
29
the
concept inferential of
stan?
semantic
inferential structural seem
i.e. a system generated system, by a col? rules the and metainferential containing is obviously of a piece with the ideas
Gentzenian and
rules. This that
of the natural deduction program (Prawitz 1965; Tennant

so it would the inferentialist
1997; etc.)
a should display agenda large overlap with the agenda of this program. We have also seen that characteriza? there is a direct way from the natural quasiinferential to tion of structural inferential their operators superstandardly Let us consider rule disjunction. A v B is partly characterized by the
characterization. inferential
Ah AvB Bh AvB
but the characterization rule has to be completed by the genuine quasi
inferential
AvBh
This rule gets
A,B
emulated as
AhQBhC/AvBhC
of disjunction characterization yields us the metainferential from of natural well-known the systems deduction. Note that the rule can be looked at as a "maximality metainferential condition". which
Let us say that the statements A, B, C fulfill the condition <&(A, B, Q iS Ah C and BY- C. Then A v B can be characterized in terms of the
following two conditions:14
and (i) ?(A,B,AvB); is the strongest statement such that Q>(A, B, Q; (ii) AvB A v B h C for every C such that 4> (A, B, Q Why
could pattern.
i.e.
is this interesting? Because this kind of maximality

be we this seen state as implicit that Ah AvB as an to the statement and Bh AvB of an
condition
inferential we, of else.
perhaps When
and when and nothing
moreover, A v B, we
put insinuate
forward <&(A, B,
(exhaustive) ...) is fulfilled by AvB
characterization
30 However, weaker it is clear that
JAROSLAV PEREGRIN if AvB fulfills
Q>(A, B, ...), then everything so the "nothing else" can only does, (i.e. entailed by AvB) mean it means that for every such hence non-weaker" "nothing statement C it must be the case that Aw BY- C. I have i.e. to characterize I am asked what children Imagine a son and a answer "I I have and the class of my children a unique I am not giving character? Strictly speaking daughter". a contains I am only that class this of the class ization stating as I a that what But it is and say normally expected boy girl. should taken yield an exhaustive to imply (by way the class that implicature) I state. And characterization, Grice of what statement be would my a conversational called
the condition
seen as insinuated by my of disjunction. also to motivate to the point, this train of thought More appears seman rules which it is that insistence Gentzen's introductory only As Koslow Section the operators. characterize 2.1) (1992, tically shows, ditions: it is natural the to see introduction
one fulfilling is the maximal in question can be a similar maximality implicature characteristic that O is the pattern stating
con? in terms of extremality it precisely rule via the the elimination rule yields rule gives all that there is to the that the introduction assumption it seems that we can, Hence of the connective. behavior' 'inferential an inferential the classical after all, delimit pattern by disjunction that instead This indicates if we assume the maximality implicature. rules metainferential the non-structural for (which allowing to merely over from strictly standard standard to passing amounts an we that admit could inferential stating structures) perhaps of the the involves inferential operator maximality stating pattern by Can we the pattern. see all the other statement which classical holds logical whenever operators B analogously? from A.
of
fixed
Gentzen
the maximal
(1934) himself gives the example of implication: A ?? B is

is derivable
So here ?(A, B, C) would be A B C

Alternatively, we can characterize implication as the minimal opera?
tor fulfilling
A.A^BVB
31
i.e. as an operator A,C\-B/C\-A^B
fulfilling
this and, moreover
of maximality But, does not the replacement by minimality spoil the intuitive picture outlined reflected exhaus above, where maximality as now ?> B A tiveness? Not for is within the antecedent of the really; basic pattern, The is such assuming exhaustiveness basic that pattern together comes now with to yield reads: A it entails B. not maximality, but minimality. A ?> And A B
exhaustivity, else, is such that together with A it entails B.
?> B, and nothing
Now
follows only
it is clear that if something follows from A ?> B, then it

from
so the "nothing sense else" makes stronger, anything as "nothing Hence the exhaustivity boils down to non-stronger". the same, also then it is stronger, conjunction i.e. C h A ?> B. as the maximal operator
if C does We can
characterize
fulfilling
C\-A Ch B.
Negation,
seems that
ifwe want it to be classical, is unfortunately more fishy. It

the only pattern available is
A,C\-B
-.Ch A.
which
the negation itself contains and this sign to be determined; to largely spoil the picture. Is there a remedy? appears We trade the second part of the negation-pattern, could, perhaps, i.e. the law of double for something else e.g. for the negation, are truth-functional. "external" that all our operators It is assumption clear that the only truth-function a more frank which always maps a statement on
itsmaximal
for more out
incompatible is the standard negation (see Peregrin 2003

solution would and negation the intuitionist without a substitute. negation be to simply strike What would be consequently the
But details). the law of double Of course
the result?
32 intuitionist inference. logic. This In this
JAROSLAV PEREGRIN indicates, classical as I have discussed in detail elsewhere
(see Peregrin 2004a), that it is intuitionist logic that is the logic of

from the infer? logic is not natural its unnaturalness from this is (however, viewpoint viewpoint so and its outweighed maybe overridingly by simplicity). sense, entialist
12. CONCLUSION we must first specify what exactly we mean for inferentialism, Arguing In this paper I have tried to by the term: there are several options. can be merged that two of the options indicate into a single one, in its turn, is the hottest candidate for becoming the inferen? which, tialism. The winner On 'emulating' ferentialism". of natural inferentialism" and hence deduction. is the "superstandard of inferentialism", capable as encompassing treatable "standard quasiin to the framework the technical side, it comes down (Its appears then we immediate much to be should as we too stricter weak; "standard neighbor, while the stronger logic as the
is obviously
"quasiinferentialism" If we accept this, most does
less natural.) also see intuitionist
have taken pains to indicate, natural this logic. However, not preclude the way to classical which is natural in surely logic, some other respects utter I be? and whose would be, inaccessibility Inferentialism is a descriptive lieve, a failure of inferentialism. project is meaning?; the question what the natural concerned with whereas
&prescriptive ismore concerned with the program program we How Thus the latter should do while could logic?. perhaps question if it concluded do ban classical that one can make logic simply to extant at former is the face the bound take without it, meanings deduction is an inferential matter, then the question If meaning that are prima facie "non-inferentian there be meanings some I think it may "favor" Hence that inferentialism, though over in any unnatural "semantic does not result others, meanings are I am convinced that the thesis that all meanings asceticism". or less direct more creatures is viable. of inferences value how and face could
ACKNOWLEDGEMENTS Work on this paper was the Research Grant No.
supported
by
401/
04/0117 of the Grant Agency
of the Czech Republic. This final ver
MEANING AS AN INFERENTIAL ROLE sion has been
33
insightful are many earlier versions during
to the extraordinarily thanks improved significantly comments and helpful of the referees of this journal. There to whom I am thankful other people for comments either on the paper or on the I would and discussions. of van Benthem, Ondrej ideas contained Mark like to name in the paper at least Michael Lance and Jar?
lectures Johan
Kremer, oslav Zouhar.
Majer,
NOTES 1 See also Lance (1996, 2001) and Kalderon (2001). 2 See Peregrin (in press-b) for more details. 3 As was pointed out to me by Michael Kremer.
To be precise, of instead we two now have different So truth-valuations, {1}" since should be now read we statements. "expressing about are evaluating as "expressing a
three
truth-valuation which yields {1} when restricted to {A,B}". 5

Consider The the recurring problem discussions substitutional tion. basic would seem to be that we simply
vs.
objectual quantifica? assume cannot that all
entities (including those not known to anybody) within our universe must have names. But this is a red herring (independently of whichever side of the quarrel we
for what the proponent stand), not that every entity is named, to form a name the resources 6 it can be shown Moreover, of but as substitutional that soon each it is nameable as it becomes standard gis quantification in the needed such needs sense that to assume has is language (cf. Lavine 2000). is a function/ that there
that
mapping sentences on sentences so that for every X, A and B, LA, f(A) and if X,A h B and X,f[A) h B, then X h B, is perfect. (This is important for/is the usual
proof-theoretic 7 Obviously the calculus, notion in the case of negation.) of such exclude I owe extensional all valuations this observation languages save a as single to Michael those one. of Kremer. the predicate this
constraints
However,
is
not be the case for any natural clearly language. 8 Of course when with terms and empirical dealing empirical a way to "connect we need them with the world" hence language however, 9 This sion is on capable can be is not
languages, a either
then we trusted
need meta?
inference the side
or else a direct of mediating the connection, connection which, established only practically. to be a knock-down the multiple-conclu? supposed argument against in press, Restall for a defense). naturalness (see, e.g., However, clearly the one.
single-conclusion 10 A simple example was suggested to me by the referee of this paper: Suppose V consists of all valuations which make only a finite number of sentences of the infinite set S true. Then there is obviously no quasiinference which would exclude an unacceptable
of
1* Note
valuation
without
excluding other
also
an
acceptable we do not
one.
that this does not mean

rules we adopt do engender not
that we have to be aware of all the rules binding

rules and have to foresee all the
us:
the
consequences. 12 Note that we hence parameters
require possible
that
it is even
any particular for a rule to contain
sentences
are
no parameters
replaced whatsoever.
by
34
Hence we do not that
JAROSLAV PEREGRIN
rules be and the fact that in what
require restrict our the
purely
"formar';
follows we will deal especially with rules of this kind should be seen as amatter of the
fact of that we will inference attention semantics to the of other semantics expressions of
13 - see This is a fact noted already by Carnap (1943) but rarely reflected upon Koslow (1992, Chapter 19), for a discussion. 14 This form is borrowed from Koslow (1992), whose book offers a thorough dis?
cussion of the technical side of the issues hinted at in this section.
underlying
constants. The rules logical will surely be non-formal.
REFERENCES
R.: 'Varieties Science, of Understanding', Presses University Harvard in N. Rescher Reason pp. and 27-51.
Brandom, Rationality Brandom, Brandom, MA. Carnap, MA. Davidson, ford. Dunn,
1985, inNatural
(ed.), Lanham,
of America,
R.: R.:
1994, Making 2000,
It Explicit,
Articulating Formalization
Reasons,
MA. Press, University Cambridge, Harvard Press, University Cambridge, Harvard
R.:
1943,
of Logic, Into Truth
University
Press,
Cambridge, Ox?
D.:
1984,
Inquiries
and
Interpretation, Methods
Clarendon
Press,
J. M.
and G. M. Oxford.. and LePore,
Hardegree: E.:
2000,
Algebraic
in Philosophical Isn't Conceptual
Logic,
Clarendon, Fodor, J.A.
1993,
'Why Meaning
(Probably)
Role', 15-35.
in E. Villaneuva Fraassen, Frege, Gentzen, G.: G.: B. C: 1879,
and Knowledge, Atascadero, pp. Ridgeview, (ed.), Science New York. and Logic, Semantics Macmillan, 1971, Formal Halle. Nebert, Begriffsschrift, 'Untersuchungen ?ber das logische Schliessen I-IF,
1934, E.:
Mathemat?
ische Zeitschrift
Kalderon, 129-160. Koslow, A.: 1992, M.
39, 176-210.
2001, A 'Reasoning Structuralist and Representing', Philosophical Studies 105,
Theory
of Logic, The
Cambridge Journal
University
Press, Re?
Cambridge. M.: Lance, search!^, Lance, M.:
1995,
'Two
Concepts
of Entailment',
of Philosophical Nous
113-137. 1996, 'Quantification, Substitution and Conceptual Content', 30,
481-507.
Lance, M.: 2001, 'The Logical Structure of Linguistic Commitment III: Brandomian Scorekeeping and Incompatibility', Journal of Philosophical Logic 30, 439-464. Lance, M. and P. Kremer: 1994, 'The Logical Structure of Linguistic Commitment I:
Four Systems M. and of Non-Relevant Commitment Entailment', Journal of Philosophical Commitment Logic
Logic 23, 369-400.

Lance, P. Kremer: of Relevant 1996, 'The Logical Structure of Linguistic II: Systems 25, 425-449. Lavine, J.: Commitment Entailment', Journal of Philosophical
S.: 2000, 1997,
'Quantification 'Language and
and
Peregrin, 2, 1-23.
Ontology', its Models',
1^3. 124, Synthese Nordic Journal of Philosophical
Logic

J.: 2001, J.: 2003, Yearbook and and Aldershot. Childers 193-205. in L. B?hounek
35
Peregrin, Peregrin, L?gica
Meaning 'Meaning 2002,
Structure, Inference',
Ashgate, in T.
& O. Majer
(eds.), The
The
pp. Filosof?a, Prague, as 'Making J.: 2004a, it Explicit', Peregrin, Logic Yearbook Filosof?a, 2003, pp. 209-226. Prague, Peregrin, (eds.), English Peregrin,
(ed.),
L?gica
J.: 2004b, und Semantik', in A. Fuhrmann and E.J. Olsson 'Pragmatismus am M., Frankfurt 89-108 Denken, Ontos, pp. Pragmatisch (in German; version available from http://jarda.peregrin.cz). J.: The in press-a, 'Semantics as Based Kluwer on Inference', in J. van in press. Benthem et al. of Alternative
Dordrecht, Age Logics, (eds.), as a Rule', J.: in press-b, in Sellars' Semantic Peregrin, 'Developing Legacy: Meaning on Wilfrid M. Lance The Self-Correcting and P. Wolf (eds.), Enterprise: Essays Amsterdam. Sellars, Rodopi, Prawitz, Prior, Restall, A. D.: N.: G.: 1965, Natural 1960, in press, of Logic, N.: Deduction, 'Roundabout 'Multiple Methodology The Taming & Wiksell, Almqvist Inference Ticket', Analysis in Proceedings Conclusions', Stockholm. 21, of 38-39. International
the 12th.
Congress Tennant, Wittgenstein, Oxford.
and Philosophy of Science. the Clarendon True, 1997, Press, Oxford. of L.: 1956, Bemerkungen ?ber die Grundlagen der Mathematik,
Blackwell,
Jaroslav
Peregrin
Department of Logic Institute of Philosophy Academy of Sciences, Jilsk? 1, 1 Prague 110 00, Czech Republic
E-mail: jarda@peregrin.cz
Manuscript submitted 28 January 2004 Final version received 29 March 2005

Meaning As An Inferential Role PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Meaning As An Inferential Role PDF

Uploaded by

Copyright:

Available Formats

Meaning as an Inferential Role Author(s): Jaroslav Peregrin Source: Erkenntnis (1975-), Vol. 64, No. 1 (Jan., 2006), pp.

Erkenntnis (2006) 64:1-35 DOI 10.1007/sl0670-005-4698-l

to the inferentialists, meaning

the question study We inferences. by

semantically of features containing

explanation in terms expressions

representational of sentences relations

the writing of some of the founding fathers of analytic phi?

content", which he presents in his Begriffsschrift distinctively inferentialist:

(1879, pp. 2-3),

Similarly, Tractarian meaning Mathematics

a distinctively to the more In his Remarks

inferentialist inclusive on use

point in a particular stage of the development of his thought from the

theory the Foundation

elaborated especially by Brandom

content conceptual and predicates.1

of this indicates that the idea of identifying meanings

has been its viability or Fodor and LePore

MEANING AS AN INFERENTIAL ROLE

sophical framework precise relevant may not I will

2001, Chapter VIII, 2004b). Instead, I will focus on establishing a

for the debate be obvious.

to another is connected (This important nature It is often of inference. assumed

serted X, i.e. the incompatibility of X with the negation of A.)

clarify or explicate for the upcoming

strued as a kind of "inferential role" (though in the case of empirical

we will write simply

MEANING AS AN INFERENTIAL ROLE

We will say that(S, hs) is standard iff for every X, Y, Z, A, B, C:

contractibility, permutability as identity, known thinning, structure

(EXT) if Xs X, 7, then Xs X, A, Y (CON) if J_sX, A, A, Y, then _LS ^ ^, ^ (PERM) if ?s X, ?, 5, Y, then J_s X, 5, A, Y

bility structure is standard, then the inferential structure induced by it

6 PROOF. patibility tive. Hence Most we of have

it is trivial, is standard, to prove

for every W and W

such that _LS W, A, W,

for every W and W

It is clear that (ii) is equivalent to (ii') i_s W, X, Z, F, W

for every W, X, Y and usuell

(ii") for every Xand

Fit is the case that _LS W, ^ Z, F, H^ for every fF

it is clear that (iii) is a consequence of (i) and (ii"). D

A standard gis is called perfect, iff itmoreover

MEANING AS AN INFERENTIAL ROLE

be a standard gis. Then Cn(X*) is

Cn(X*) sequence Y=A\,...,An This further means

This means such

that 7* COz(X*) and _L_s^ so that X* ? X* and

Xi hs At. But due to the extendability and permutability

incompatible with A2 is incompatible with X,

it is the case that

3. Suppose we have subset of {0,1}5.

3. A gis is truth-preserving if it is standard.

that Cn(X*) ? U for no UeV. But as Cn(X*) is closed

Thereby the proof is finished.

(1995) and Lance and Kremer

(1994, 1996) have forcefully

that argued inferentialist than classical natural an

appropriate that does justice

will comply to (EXT). However,

here I think we must be mindful of

justified given A, and hence (REF) appears to be vindicated. Also

that 7* COz(X) and _L_s^ so that X ? X* and

that Cn(X) ? U for no UeV. But as Cn(X) is closed