You are on page 1of 55

Linguistic Society of America

Metalinguistic Negation and Pragmatic Ambiguity


Author(s): Laurence R. Horn
Source: Language, Vol. 61, No. 1 (Mar., 1985), pp. 121-174
Published by: Linguistic Society of America
Stable URL: http://www.jstor.org/stable/413423
Accessed: 25/01/2010 09:46

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=lsa.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.

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.

Linguistic Society of America is collaborating with JSTOR to digitize, preserve and extend access to Language.

http://www.jstor.org
METALINGUISTIC NEGATION AND PRAGMATIC AMBIGUITY
LAURENCE R. HORN
Yale University
When 'marked' or 'external' negation has not been treated as an additional semantic
operator alongside the straightforward truth-functional, presupposition-preserving or-
dinary ('internal') negation, it has been collapsed with internal negation into a unified
general logical operator on propositions. Neither of these approaches does justice to the
differences and kinships between and within the two principal varieties of negation in
natural language. Marked negation is not reducible to a truth-functional one-place con-
nective with the familiar truth-table for negation, nor is it definable as a separate logical
operator; it represents, rather, a metalinguistic device for registering objection to a pre-
vious utterance (not proposition) on any grounds whatever, including the way it was
pronounced.*
INTRODUCTION
So-called 'external' or 'marked' negation is often exemplified by the reading
of The King of France is not bald which is forced by the continuation ... because
there is no King of France and which is true if France is a republic; by contrast,
the 'internal' reading is either false or lacks truth value. The traditional account
of external negation involves the recognition of a semantic ambiguity; this
position is adopted, in somewhat different ways, by Frege 1892, by Russell
1905, by Karttunen & Peters 1979, and by proponents of three-valued logics.
However, the major recent trend among philosophers and linguists represented
by Atlas 1974, 1977, 1981, by Kempson 1975, and by Gazdar 1979a,b, has been
to reject this putative ambiguity (along with the existence of truth-value gaps
and semantic presuppositions) and to assimilate all instances of natural lan-
guage negation to a single truth-functional operator.
Both views contain much insight and some truth, yet both are seriously
flawed. While two distinct uses of natural language negation must indeed be
admitted, the marked use must be treated not as a truth-functional or semantic
operator on propositions, but rather as a device for objecting to a previous
utterance on any grounds whatever-including its conventional or conversa-
tional implicata, its morphology, its style or register, or its phonetic realization.
In this paper, conceived as a more explicit formulation of some ideas inherent
in Ducrot 1972, 1973, Grice 1967, Wilson 1975, and others, and buttressed by

* Parts of this
paper were presented in a different form at the 1982 summer and winter meetings
of the Linguistic Society of America, and at talks at Columbia and Cornell earlier in the same year.
The seeds for the major thesis germinated at the July 1979 Colloquium on the Possibilities and
Limitations of Pragmatics at Urbino, Italy, and subsequently developed through courses and sem-
inars at the University of Wisconsin-Madison and Yale. The members of these audiences, along
with those who read and commented on earlier versions, are hereby collectively thanked, and of
course absolved of all blame for the ways I may have used (and misused) their suggestions. I would
like to single out for special thanks Barbara Abbott, Jay Atlas, Samuel Bayer, Seungja Choi, Benoit
de Cornulier, Robin Cooper, David Dowty, Georgia Green, Anne Malcolm, Ewan Klein, William
Lycan, David Odden, Jerry Sadock, Dennis Stampe, and Deirdre Wilson. Their contributions were
not important-they were invaluable.
121
122 LANGUAGE, VOLUME 61, NUMBER 1 (1985)

a variety of linguistic data correlatingwith the two uses of negation, I seek to


focus on the descriptive/metalinguisticdichotomy as an instance of pragmatic
ambiguity-a built-in duality of use wherein the descriptive meaning of ne-
gation motivates its extended function. I will discuss the relation of metalin-
guistic negation to truth-conditionalsemantics, and will argue that the tradi-
tional identificationof external negationwith the formulaIt is not true that ...
is misleadingat best. I also attemptto show that other logical operators(and,
or, if-then, and wH-binding)display extended metalinguisticuses of their own;
and I cast furtherdoubt on theories which treat naturallanguagenegation as
either semanticallyambiguousor invariablytruth-functional.
The outline of the paper is as follows. In ?1, some of the more influential
accounts of negation are summarized and discussed: the AMBIGUIST approach
of Russell, Strawson, and the three-valuedlogicians in ?1.1; the MONOGUIST
approachof Atlas, Kempson, and Gazdarin ?1.2; and the NEO-AMBIGUIST po-
sition of Karttunen& Peters in ?1.3. In ?2, I present evidence to supportthe
position that negationmust be taken as pragmaticallyambiguous,with marked
negation as an extended metalinguisticuse of a basically truth-functionalop-
erator. In ?2.1, I arguethat external, presupposition-cancelingnegationis part
of a wider phenomenoncharacterizedas the use of negationto signalthe speak-
er's unwillingnessto assert a given propositionin a given way-or, more gen-
erally, the speaker'sobjectionto the content or form (phonetic,morphological,
syntactic, semantic, or pragmatic)associated with a given utterance. In ?2.2,
I discuss the two uses of negation (descriptiveand metalinguistic)in terms of
what they generally negate: truth (of a proposition) vs. assertability (of an
utterance). In ?2.3, I focus on a characteristicuse of metalinguisticnegation
as a means of removingthe upper-boundingimplicatureassociated with scalar
predications(e.g. It isn't WARM, it's HOT); in ?2.4, I examine one correlate of
the descriptive/metalinguisticdichotomy for English negation-the ability of
descriptive (but not metalinguistic)negation to incorporate prefixally. The
question of the interplay of truth, negation, and implicatureis addressed in
?2.5, where I try to show that markednegationcannotbe semanticallyanalysed
in terms of truth; attempts in this direction, where external negation is taken
as approximately It is not TRUEthat for some semantic predicate TRUE,founder
on the distinction between the semantic notion of truth and the distributional
behavior of the word true in ordinarylanguage.
In ?3, I briefly discuss the extent to which other standardlogical connectives
and variable-bindingoperators may be seen as pragmaticallyambiguous(like
negation) between descriptive and metalinguisticuses. In ?4, I return to an
examinationof other recent approachesto the unity or duality of naturallan-
guage negation which are closer in spirit to the view defended in ?2: in ?4.1,
the monoguist positions of Allwood 1972, of Kempson, and of Atlas, and the
neo-monoguist(or neo-ambiguist)theories of Bergmann, 1977, 1981, of Kart-
tunen & Peters, and of Lehrer & Lehrer 1982;in ?4.2, the analysis of French
negation in terms of a distinctionbetween 'negationdescriptive'and 'negation
metalinguistique'(or 'polemique')in work by Ducrot and his colleagues; and
in ?4.3, the treatmentof negation advanced by Wilson, in many ways akin to
METALINGUISTIC NEGATION AND PRAGMATIC AMBIGUITY 123

that urged here. A summaryof my findings, with a brief cross-linguisticex-


cursus, is given in ?5.
The Appendixpresents an examinationof two types of but clauses in English
(with parallels in Spanish, German, and French) which interact significantly
with the two uses of negationdistinguishedin the body of the paper. Following
Anscombre & Ducrot 1977, I attempt to show that metalinguisticnegation is
compatiblewith only one of these subspecies of but clauses.
EXTERNAL NEGATION: SOME TRADITIONAL ACCOUNTS
1.1. THE AMBIGUISTLINE. For most recent proponents-and opponents-of
the claim that naturallanguagenegationis semanticallyambiguous,the causal
chain stretches back at least to Frege and Russell. Aristotleand the Stoics may
have held analogous views (cf. Bergmann 1977:65,Atlas 1981:125),but the
evidence is unclear.Russell (485)formulatedthe essential puzzle with his char-
acteristic style:
'By the law of the excluded middle, either "A is B" or "A is not B" must be true. Hence
either "the present king of Franceis bald" or "the presentkingof Franceis not bald" must
be true. Yet if we enumeratedthe things that are bald and the things that are not bald, we
shouldnot find the kingof Franceon eitherlist. Hegelians,who love a synthesis,will probably
conclude that he wears a wig.'
This puzzle, Russell maintained,should be unraveledby the eliminationof
definite descriptions (e.g. the present king of France) from logical form, with
the result that sentences like 1-2 are not of subject-predicateform, theirsyntax
notwithstanding:
(1) The present king of France is bald.
(1') 3x (Kx & Vy[Ky-,y=x] & Bx)
(2) The present king of France is not bald.
(2') 3x (Kx & Vy[Ky-y=x] & -Bx)
(2")-3x (Kx & Vy[Ky->y=x] & Bx)
If ex. 1 is unpacked into an existentially quantifiedconjunction, as in 1', its
normal (or 'primary')negation consists in attachingthe negative operator to
the third conjunct, as shown in 2'. This INTERNALLY negated sentence was for
Russell 'simply false' if there was no French king (or more than one of them).
However, he acknowledged,there is a readingof 2 with 'secondary' negation
which is true, given that France is a republic. It is this readingwhich requires
attachingthe negation sign EXTERNALLYto the entire logical form 1', resulting
in 2"above.
Despite the importanceof Russell's analysis, the treatmentof negation as
semantically ambiguous came to be linked historicallywith the treatmentof
logical (or semantic) presuppositionin anti-Russelliantheories of definite de-
scriptions.While the relationof presuppositionobtainingbetween propositions
can be tracedback to Peter of Spain, whose 12thcenturyTreatiseon exponibles
employed the verbpraesupponerein approximatelythe modernsense (cf. Mul-
lally 1945), it was Frege who introducedmodernphilosophersto the problems
of presuppositionand referencefailurefor names and descriptions.In his clas-
sic paper on sense and reference, Frege (1892:40)arguedthat both 3a and its
(ordinary) negation 3b PRESUPPOSE
(voraussetzen) that the name Kepler des-
124 LANGUAGE,VOLUME61, NUMBER 1 (1985)

ignates something:
(3) a. Kepler died in misery.
b. Kepler did not die in misery.
c. Keplerdid not die in misery, or the nameKeplerhas no reference.
To detach this presupposition,Frege claimed, the negation of 3a would have
had to be not 3b, but 3c. While he did not pursue this possibility, it should be
noted that Frege seems to have prefiguredthe later emergence of a presup-
position-cancelingexternal negation whose truth conditions are equivalentto
those of disjunctionslike 3c.
It was, however, with Strawson's re-introductionof neo-Fregean presup-
positions, and the truth-valuegaps resultingfrom their non-satisfaction,that
the semantic ambiguityof negation came into its heyday. For Strawson 1950,
someone who uttered ex. 1 did commit himself to the existence of a (unique)
French king, but did not ASSERT(nor does his statementENTAIL) the existential
propositionthat there is a king of France. In case this existential proposition
failed, Strawson maintained,ex. 1 would be judged neither true nor false. A
statement was indeed made under such circumstances(pace Frege 1892), but
the question of its truth value 'fails to arise'.
While Strawson himself was skeptical of using any version of formal logic
to express his intuitionsabout truthand meaningin naturallanguage,'logicians
before him like Lukasiewicz 1930, Kleene 1938, and Bochvar 1939-and phi-
losophers and linguists since, including Smiley 1960, van Fraassen 1966,
Keenan 1971, Herzberger1973, Katz 1977, and Martin1979, 1981-have pro-
posed varieties of three-valuedlogics in which truth-valuegaps arise, i.e. in
which meaningfuldeclarativestatements can be made which in at least some
contexts are assigned neither of the classical values T or F.
It is generallyaccepted withinthese models that ordinarynegationpreserves
(non-)bivalence:if andonly if a statementis non-bivalent(neithertruenorfalse)
in a given context, its ordinarynegationwill likewise be non-bivalent.This is
shown in the second column of Table 1, the standardtruthtable for negation,
where N stands for the neuter, non-bivalent,or non-designatedvalue. But if
the three-valuedlogician agrees with Strawsonthat ex. 1 and its ordinaryne-
gation 2-read as in 2', as assertinghirsutenessof the French monarch-share
non-bivalencewhen their shared existential presuppositionfails,2 what of the
1
The last sentence of Strawson'smanifestoreads: 'Neither Aristoteliannor Russellianrules
give the exact logic of any expression of ordinarylanguage;for ordinarylanguagehas no exact
logic.' Kempson(1975:86)has pointedout the irony of this conclusion, which containsa definite
descriptionwithoutinducingan existentialpresupposition.In Strawson'slaterwork, only descrip-
tions in subjectpositioninducepresuppositions(andincurtruth-valuegaps when referencefails).
2 Horn 1972
(Chap. 1) argues that the failureof the uniquenesspresuppositionis a somewhat
differentmatter;as noted there, the externalnegationoperatorwhich can be used to cancel ex-
istence (as we have seen) cannot easily cancel uniqueness:
(a) ?*Theking of France isn't bald-there are two kings of France.
(b) ?*It is not the case that the Californiasenatoris bald-there are two Californiasenators.
This probleminvolves the pragmaticsof definitereference(cf. Hawkins 1978);such matterswill
not be addressedhere.
METALINGUISTICNEGATIONAND PRAGMATICAMBIGUITY 125

, _ internal, choice, primary negation


P -P P-external, exclusion, secondarynegation
T F F
F T T
N N T
TABLE1.

reading specified by Russell's 2", which is apparently true in such a context?


This is the sense in which we can say the following:

(4) The king of France isn't bald-there ISN'T any king of France.
One possibility is to assume a second negation operator which is assigned the
truth values shown in the third column of Table 1. The traditional labels for
the two negations thus distinguished are INTERNALand EXTERNAL,although
other terms (as shown) have been employed.
Alternatively, some presuppositionalists have held that negation is not in
effect LEXICALLY ambiguous between the two senses depicted in Table 1, but
SCOPALLY ambiguous as to its position in logical syntax (as it was within Rus-
sell's non-presuppositional two-valued logic). It has often been observed that
the external reading of negation is more natural or accessible when the form
of the negative statement is not as in 2-even when fleshed out in the manner
of 4-but rather as in the following:
(5) It is not true that the king of France is bald.
(6) It is not the case that the king of France is bald.
Arguing from this observation, Smiley, Herzberger, and others have suggested
that we first introduce a one-place connective t, on the model of the Bochvar-
Frege horizontal, to be interpreted as 'It is true that ...' Such a connective
will always yield a bivalent truth value for t(P), given any meaningful statement
(bivalent or non-bivalent) P. Now, while it may or may not be FALSE that the
king of France is bald when France has no king, it is certainly NOT TRUE that
the (non-existent) king is bald. Thus a negation outside the scope of the truth
operator will always yield the opposite bivalent value of that assigned to t(P),
as shown in Table 2.

P t(P) -t(P)
T T F
F F T
N F T
TABLE 2.

We thus obtain the assignments we had for the external negation operator
defined above; note the identity between the rightmost columns in Tables 1
and 2. Indeed, as Keenan and others have observed, we could now define a
derived external negation operator as in 7, rather than taking P to be a primitive
of the logic:
(7) P =df ~t(P)
126 LANGUAGE, VOLUME 61, NUMBER 1 (1985)

However the differences between these approachesare to be resolved, the


existence of markednegative statementswhich are true when their affirmative
counterpartsare neithertrue nor false has led proponentsof semantic presup-
position to conclude that naturallanguage negation must be represented as
ambiguous,either by allowingdual interpretationsof a single surface operator
or by providingdual scope possibilitiesfor negationin logical form. A system-
atic ambiguityfor negation figures crucially in ALL theories which admit se-
mantic presuppositions-but (as Russell illustrated) not ONLY in these. As
Thomason 1973 has observed, unless negation is treated as semanticallyam-
biguous, 'it's not clear that the semantic notion of presuppositioncan be de-
fended at all'.3
The ambiguistposition on negation has been supportedrecently by Martin
1979, 1981-who acknowledges, in the course of his defense of semantic pre-
supposition (1979:43),that know and other factives may have to be taken as
semanticallyambiguous:
'Methodologically,of course, the multiplicationof senses beyond necessity is undesirable...
but a few such ambiguities,especially that of negation,seem perfectlyreasonable.'
1.2. THEMONOGUIST
LINE. The chief difficulty for the ambiguist view of ne-
gation sketched above is that it is by no means obvious, nor is it easily de-
monstrable, that negative sentences like 2 AREsemanticallyambiguous.Fur-
thermore, as pointed out by Atlas 1974, the periphrasticconstructionsof 5-6
do not clearly disambiguate2 in favor of the presupposition-free'external'
reading,except within a readilydefinablesubset of philosophersand linguists.
Withinthe last decade, a countervailingconsensus in opposition to the am-
biguistline has emergedfromthe work of Allwood, of Atlas, of Kempson 1975,
of Boer & Lycan 1976, and of Gazdar.WhereRussell struggledwith diligence
and care to untie the Gordianknot constructedby the king of France, these
monoguists-wielding Occam's razor like a samuraisword-seek to sever it
with one blow; for them, negationis simply not ambiguous,in either meaning
or scope. The burdenof proof is clearlyon the ambiguist(as the defensive tone
of Martin'spassage seems to concede); like other abstractentities, senses must
not be multipliedbeyond necessity. Grice(1978:118-19)calls this the 'Modified
Occam's Razor principle', and Ziff (1960:44)advances the same doctrine as
'Occam's eraser'.
It is, moreover, exceptionallydifficultto PROVEthat the presupposition-car-
rying internalunderstandingand the presupposition-freeexternal understand-
ing of 2 are semanticallydistinct, given that the formerunilaterallyentails the
latter:if the existent king of France is non-bald,it is certainlynot the case that
the present king of France is bald. What the ambiguistmust demonstratebe-
tween internal and external readings is a PRIVATIVE
ambiguity, of the sort

3 The ambiguist/monoguist oppositioncomes from Wertheimer1972, who applies the terms to


two rival theories of the semanticsof modals (cf. ?2.1 below). The statementin the text is a bit
oversimplified,since van Fraassen (1966 and subsequentwork) has elaborateda theory of 'su-
pervaluations'in which logical presuppositionis not defined in terms of the internalnegation
operator,and does not cruciallydependon the ambiguityof negation.Nevertheless, van Fraassen
does figurein the ranksof the ambiguists(cf. Martin1981for relateddiscussion).
METALINGUISTICNEGATIONAND PRAGMATICAMBIGUITY 127

claimed to hold for such examples as these:


(8) a. I just bought a new dog.occ-canis familiaris
canis familaris,
canis familiaris, male
of them is married.
b.and Lee are married .Each
Kim
They are marriedto each other.
Yet, as Zwicky & Sadock 1975have shown, it is just such ambiguitieswhich
are the hardest to substantiateby linguistictests.4
For the linguist,a particularlytellingargumentagainstthe ambiguistposition
is the fact (noted by Gazdar 1979a:65-6) that no naturallanguage seems to
employ two distinct negative operators which function like the two logical
operatorsin Table 1.5This is especially strikingwhen one considers the many
instances in which a languagedoes contain two or more negative markers.In
particular,as exemplified in Table 3, many languagesdraw an opposition be-
UNMARKED MARKED (EMPHATIC)
NEGATION NEGATION
Ancient Greek ou me
Modern Greek 6en (< ou6en) mi
Latin non ne
Modem Irish nach gan
Estonian ei mitte
Tagalog hindi huwag
TABLE3.

4 I do not meanto implythatcases like 8a-b are not trulyambiguous;it is just thattheirambiguity
(if any) is difficultto prove via the standardelliptical 'identity of sense' transformationswhich
block 'crossed readings'in the case of non-privativeambiguities:
(a) Tracy left a deposit at the bank, and so did Lee.
(b) Ralph saw her duck, {andI did too / but I didn't}.
As Zwicky & Sadock note, the existence of crossed readingsin cases of privative ambiguities
cannot in principlebe determined,since the moreinclusiveunderstandingwill always be available
(Atlas 1977to the contrarynotwithstanding):
(c) Ron and Nancy are married,and so are Dick and Jimmy.
(d) Fido is a dog, and so is Queenie.
Thus, contraryto the claims of Atlas 1977and Kempson (1975:99-100),the acceptabilityof (e)
has no bearingon the purportedambiguityof negation:
(e) The king of Franceis not bald, and neitheris the queen of England.
Here both conjunctspermitthe more inclusive 'external'understanding(cf. Horn 1984for related
discussion).
Martin(1981:23-6)notes that the semanticambiguitytest classicallyemployedby philosophers
is based on whethera given sentence can be judged simultaneouslytrue and false relative to the
same possible world, context, or state of affairs(cf. Quine 1960:27on the ambiguityof light; cf.
also Kempson 1982). This criterionis rejected by Zwicky & Sadock for reasons Martinfinds
insufficient;in any event, such privativeambiguitiesas 8b do come out simultaneouslytrue and
false with respect to the state of affairsdescribedin the recent country song title, 'When you're
married,but not to each other'. And in the world of 1905 or today, Russell's classic negative
sentence 2 may well be judged simultaneouslyTRUEand-depending on one's semanticpersua-
sions-either FALSE or non-bivalent(i.e. not true, in any case). The force of the objectionby Atlas
and others to an ambiguousnegation operatoris thus weakened, coming to rest finally on the
generalmetatheoreticaldesideratumof parsimony,ratherthanon any specificempiricalclaimabout
naturallanguageambiguity.
5 Some
apparentcounter-examplesto this claim will be discussed below.
128 LANGUAGE,VOLUME61, NUMBER 1 (1985)

tween an unmarkeddeclarative negation and a marked form (often labeled


'emphatic') which is restricted to embedded non-finiteand/or non-indicative
contexts, frequentlyco-occurringwith subjunctivemood.6
In French, negation can be markedin at least half a dozen morphologically
distinctways, dependingon the syntacticenvironmentandthe semanticcontext
(cf. Gaatone 1971, Heldner 1981):
(9) (ne) ... paslpointlaucunlpersonnelrienljamais
non (pas) + ADV
Swahili similarlycontains several instances of suppletiveand redundantmark-
ing for negation:
(10) ni-na-ku-pend-a vs. si-ku-pend-i
I-PRES-yOUsg-love-INDIC I- NEG I
'I love you' 'I don't love you'
tu-na-ku-pend-a vs. ha-tu-ku-pend-i
NEG -
'we love you' 'we don't love you'
Languages, then, may utilize morphologically differentiated negative forms
for syntactic, semantic, or even arbitrary reasons, but-significantly-never
to mark the one distinction which Russell and the three-valued ambiguists
would lead us to expect.
It might be held that the external reading of negation need not display a
separate morphological coding, since it will be associated with the occurrence
of true within its scope, as suggested in the earlierdiscussion (cf. ex. 7). Thus
Karttunen& Peters (1979:47)propose that 'the "external" negation of ( ...
might be renderedinto English as "It is not true that (."' But this approach
is on shaky ground, given that the occurrenceof the EnglishformulaIt is not
true that (or It is not the case that) is neither a necessary nor a sufficient
conditionfor the emergence of a non-presuppositionalunderstandingof a neg-
ative sentence-as Atlas, Kempson, and others have pointed out. A semantic
theory which invokes an abstracttruthpredicate(TRUE, as in Linebarger1981)
at either the object-languageor metalanguagelevel, as a kind of 'animus ex
machina' for just those sentences where negation seems to be behaving ex-
ternally, is as unconvincingas a syntactictheory which invokes phonologically
and semantically null inaudibiliawithout providing solid motivation for the
existence of such constructs.
Another equally fundamentalproblemfor the utilizationof a truthpredicate
in the representationof external negation is the fact (to be explored in more
detail in ?2.5 below) that the function of TRUE as a metalinguisticoperatorin
a truth-conditionalsemantic theory cannot be directly assimilatedto the be-
havior of true in ordinaryEnglish (or its cross-linguisticcounterparts),apart
from the treatmentof negation.
1.3. THE KARTTUNEN
& PETERSANALYSIS.As mentioned above, the mon-
oguist thesis is incompatiblewith the defense of semantic presupposition.In-
6
One reason for the existence of this particular dichotomy is discussed in Horn 1978a (?5).
METALINGUISTICNEGATIONAND PRAGMATICAMBIGUITY 129

deed, monoguistslike Atlas, Kempson, Boer & Lycan, and Gazdarhave been
more eager than reluctantto jettison any remnantof semanticpresupposition.
But at least one majorrecent approachhas attemptedto combine an ambiguist
line on negationwith a non-semantic-or at least non-truth-conditional-anal-
ysis of presuppositionalphenomena. Lauri Karttunenand Stanley Peters, in a
series of individualand joint publicationsculminatingin Karttunen& Peters
1979, have marriedGrice's notion of conventionalimplicatureto Montague's
truth-conditionalformal semantics. As in Montague'swork, the logic is inten-
sional but classically two-valued. Withinthis framework,a sentence like 1la
not only entails 11', but is truth-conditionallyidenticalto it, since the entailment
is mutual:
(11) a. John managedto solve the problem.
b. John didn't manageto solve the problem.
c. It was difficultfor John to solve the problem.
(11') John solved the problem.
However, 1la differs from 11' NON-truth-conditionally, in that manage to con-
tributes a CONVENTIONAL to 1la-and
IMPLICATURE to its ordinary negation,
1lb-both of which will thereby suggest somethinglike 1lc. This conventional
implicatureborne by lla-b is part of the meaningof these sentences, and is
thus distinct from what Grice calls CONVERSATIONAL
implicatures; some of the
essential differences between these two notions are spelled out in Table 4 (cf.
Grice 1975, 1978;Sadock 1978;Karttunen& Peters 1979).

CONVENTIONAL IMPLICATURES CONVERSATIONAL IMPLICATURES

(a) Make no contribution to TRUTHconditions, but constrain APPROPRIATENESS


of expressions
with which they are associated.
(b) UNPREDICTABLE,arbitrary part of meaning; NATURAL concomitant of what is said or
must be learnedad hoc. how it is said; NON-CONVENTIONAL by
definition.
(c) NON-CANCELABLE;apply in all contexts. CANCELABLE either explicitly (by linguistic
context) or implicitly(by extralinguistic
context).
(d) DETACHABLE:two synonyms may have NON-DETACHABLEif arising via one of the
differentconventionalimplicatures. content maxims(Quality,Quantity,
Relation).
(e) NOTCALCULABLE
through any procedure; CALCULABLEthrough Cooperative
must be stipulated. Principleand maxims.
(f) Akin to pragmaticpresuppositions(non- Possibly relatedto Mill's 'sous-entenduof
controversialpropositionsspeakertakes commonconversation'(1867:501)or
as part of commonground);cf. Ducrot's 'sous-entendu'as discourseor
Stalnaker1974. rhetoricalnotion.
(g) Exhibit a well-defined set of PROJECTION Projectionproperties,if any, are poorly
PROPERTIESenabling implicata of larger understood;conversationalimplicatures
expressionto be computedfrom 'may be indeterminate'(Grice).
implicata of its subparts.
TABLE 4.

Crucially, the falsity of 1lc does not affect the truth conditions for lla-b:
the former is true if and only if 11' is true, the latter iff 11' is false. Rather,
130 LANGUAGE, VOLUME 61, NUMBER 1 (1985)

11c representsan appropriatenessconditionon the normal,felicitous utterance


of both la and llb.7
But now the old Russellian ambiguityof negation rears its hoary head: as
Karttunen& Peters concede, 1lb can be uttered in contexts where 1lc is not
only NOTtaken for granted,as part of the common groundof discourse, but is
in fact the very component of meaningbeing negated or denied. This reading
of 1Ib, which K&P call CONTRADICTION NEGATION, andlink up with the familiar
external negative operator of three-valuedlogics discussed in ?1.1, emerges
more clearly with the right intonationcontour (cf. Liberman& Sag 1974)and
an appropriatecontinuation:

(12) John didn't MANAGEto solve the problem-


it was quite easy for him to solve.
he was given the answer.
K&P point out (46-7) that contradictionnegation-unlike ordinary, con-
ventional-implicature-preserving negation-is incapableof triggeringnegative
polarity items. Thus the ordinarynegation of 13a is 13b, where the existential
shows up as any in the scope of negation:
(13) a. John managedto solve some problems.
b. John didn't manageto solve any problems.
c. John didn't MANAGEto solve {some/*any}problems-
they were quite easy for him to do.
he was given the answers.
d. Bill hasn't already forgotten that today is Friday, because today
is Thursday. (= K&P's 77b)
But with contradictionnegation, as in 13c, no some/any suppletionis possible.
Similarly,we find already ratherthan negative polarityyet in 13d, where con-
tradictionnegation removes the conventionalimplicatumassociated with fac-
tive forget.
Linebargeralso explores the failure of external negation to triggerpolarity
items; following Kroch 1974, she attributesthis failure to the interventionof
an abstract TRUEpredicate immediatelywithin the scope of negation, so that
the first clause of 13c is assigned essentially the following logical form:
(14) NOTTRUE (13a)
We shall see in ?2.5 that this identificationof externalnegationwith the abstract
predicate TRUEfails to generalize successfully.
K&P (47-8) account for their markedcontradictionnegationby assigningit
wide scope with respect to materialwhich is conventionallyimplicated, as in
15b. Such implicata are always outside the scope of ORDINARY negation, as

7 Neither the originalGriceannotion of conventionalimplicature,nor its (re)working-outby


K&P, is uncontroversial.In particular,the propertiesreferredto in (c) and (g) of Table 4 may be
mutuallyincompatible:cf., interalia, Wilson1975,Gazdar1979a,b,Soames 1979,Horn 1979,1981.
Since the relevant issue is the interactionof conventionalimplicaturewith negation, the more
global questions can be safely deferred.
METALINGUISTICNEGATIONAND PRAGMATICAMBIGUITY 131

seen in 15a:
(15) a. ORDINARY NEGATION OF 4): (1 )e; (i)
b. CONTRADICTIONNEGATION OF (): (-] [?eA4)i]; [4iV] ()i])
(Here {e represents the truth-conditional meaning of <(, and b'
its conventional implicata;the members of each ordered pair
denote respectively the 'extension expression' and the 'impli-
cature expression' for the negative form specified.)
In the case under consideration, the ordinarynegation of 1la amounts to
conveying the conjunction 16a, althoughthe first conjunct is implicated and
the second entailed; the contradictionnegation (e.g. in 12) amounts to the
negated conjunction 16b:
(16) a. It was difficultfor John to qj& -(John Jd).
b. -(It was difficultfor John to ' & John pd).
As K&P note (47), the contradictionnegation represented in 15b and 16b is
'by itself non-specific (in the absence of contrastive intonation)in regard to
what it is that the speaker is objectingto'.
In the languageof Karttunen'searlierwork, ordinarynegationis a HOLEto
conventional implicata(a.k.a. presuppositions),and contradictionnegation is
a PLUG.8 This approachechoes Russell's scopal analysis of the two negations
(note that just one selected conjunct is negated in 2' and 16a, but the entire
conjunctionin 2" and 16b)-as well as a similardevice, employing a kind of
translucentbrackets, that was semi-seriously put forwardby Grice 1967 for
derivingthe two readingsof Russell's 2. (Grice 1967finally rejectedthis brack-
eting approach, in favor of a somewhat vague pragmaticanalysis of presup-
position and negation;but he has revived it for reconsiderationin Grice 1981.)
We have arrivedby now at the situationdepicted in Table 5. (Note that, for
Strawson, negation was unambiguouslyinternal-by default, as it were, since
he never acknowledged the existence of sentences like 4 or 12. His position
was probablyuntenable, and will henceforthbe disregardedhere.)
Do truth- Do semantic Is negation
value gaps presuppositions semantically
exist? exist? ambiguous?
Strawson: yes yes no
Russell: no no yes
Lukasiewicz,
Smiley,
Herzberger, ? 'ambiguists'
Katz: yes yes yes
Karttunen&
Peters: no yes (as conven. implics.) yes
Atlas, Boer &
Lycan,
Kempson,
Gazdar: no no no 'monoguists'
TABLE5.
8 Cf. Karttunen 1974. Note that the
'plug' nature of contradiction negation is represented in 15b
by assigninga tautologicalimplicatum.An analysis similarto K&P's is that in Ducrot 1972,to be
discussed in ?4.2 below.
132 LANGUAGE, VOLUME 61, NUMBER 1 (1985)

On the one hand, we have the original ambiguist thesis on negation in both
its classical (Russellian) and revisionist (three-valued) versions-as well as the
K&P compromise position, which some have suggested has all the vices of
each withoutthe virtuesof the other. On the otherhand,we have the monoguist
antithesis which Occamistically denies any ambiguities in natural language ne-
gation, but offers no ready explanation for the intuition shared by ambiguists
of all camps that negative sentences like 2 and 1lb may be used in two radically
differentways, with (as K&P note) distinct linguisticcorrelates in each case.
What we need here is evidently a Hegelian synthesis-one, we must hope,
more explanatorythan that the king of France wears a wig.
METALINGUISTIC VS. DESCRIPTIVE NEGATION
2. In the synthesis I shall advocate here, negation is indeed ambiguous,
contraAtlas, Kempson, Gazdar,et al. But contraRussell, Karttunen& Peters,
ambiguous. Rather, we
and the three-valued logicians, it is not SEMANTICALLY
are dealingwith a PRAGMATIC ambiguity,a built-indualityof use. If I am correct,
we must reject the classical view-cited by Prior(1967:459)and subscribedto
in varying ways by virtuallyall previous analysts-that 'all forms of negation
are reducibleto a suitably placed "It is not the case that".'
That we must reckon with a special or markeduse of negation,9irreducible
to the ordinaryinternaltruth-functionaloperator,is best seen not in examples
like Russell's 2'/2"or K&P's 1lb/12, but in environmentslike 17, where what
is negated is a CONVERSATIONAL
implicatum:
(17) a. SOMEmen aren't chauvinists-ALL men are chauvinists.
b. John didn't manageto solve SOME
of the problems-he managed
to solve ALL of them.
Such examples cannot be collapsed with 12 under K&P's approachwithout
incorporatingconversationalimplicata(like the conventionalimplicataof 15b
and 16b)into the logical form for these sentences; yet conversationalimplicata
by definition are not part of logical form (cf. Grice 1975, 1978, Karttunen&
Peters 1979).
The cases below are even more devastating to any generalized semantic
account of markednegation, which would presumablybe driven to importing
phonetic representationand inflectionalmorphologyinto logical form, within
the scope of the negation:
(18) a. (So, you [miYonijd]to solve the problem.)
No, I didn't [mYonij']to solve the problem-I [maenijd]to solve
the problem.
b. I didn't manage to trap two monGEEsE-I managed to trap two
mOnGOOSES.
A related use of negation is found in the French example below, where the
grammatical gender assignment and the woeful English accent are somehow

9 For expository purposes, the label 'marked (use of) negation' will continue to be employed as
a pre-theoretical descriptor for natural language negative morphemes which do not correspond to
truth-functional internal negation (i.e. to what I shall later term 'descriptive negation').
METALINGUISTICNEGATIONAND PRAGMATICAMBIGUITY 133

broughtwithin the scope of negation:


(19) (Esker too ah coo-pay luh vee-and?)
Non, je n'ai pas 'coo-pay luh vee-and'-j'ai coupe la viande.
Analogously, we see in 20 that one speaker may employ negation to reject
the pragmaticsassociated with the registeror stylistic level chosen by another
speaker in the discourse context (because of insufficientor oversufficientdel-
icacy):
(20) a. Now, Cindy, dear, Grandmawould like you to speak a bit more
like a lady:Phydeauxdidn't 'shit the rug', he {defecated/ pooped
/ had a BM} on the carpet.
b. Grandmaisn't 'feeling lousy', Johnny, she's indisposed.
c. We didn't {'have intercourse'/ 'make love'}-we fucked.
In 21, one descriptionis jettisoned in favor of anotherwhose contributionsto
truth-conditionalmeaning are virtually identical to it in the relevant context,
but which differs from it in focus or connotation:
(21) a. Ben Wardis not a black Police Commissionerbut a Police Com-
missioner who is black. (N.Y. Times editorial, 11/8/83)
b. I'm not his daughter-he's my father. (cf. Wilson, 152)
c. She isn't Lizzy, if you please-she's Her ImperialMajesty.
d. For a pessimist like him, the glass isn't half full-it's half empty.
e. I'm not a TrotskylTE-I'm a TrotskyisT.
Such uses of negation may be marked,but they are by no means marginal
or inconsequential in communication.'( Indeed, an instance like 21d has
'o Closely related to the examples in 21 is the use of negation to focus on and register objection
to a previous speaker's racist or sexist vocabulary. Consider the truth conditions of sentences like
this:
(a) {Niggers/Broads} will benefit from improvements in medicine.
Such sentences have received a good deal of recent attention (cf. Grim 1981, Stenner 1981, Taylor
1981) as philosophers have debated whether an objection to the world view which attaches to the
use of loaded words like nigger and broad is sufficient to render the statements made via these
offensive descriptions automatically false or devoid of truth value. For Grim, if (a) is bivalent
(whether it is true or false), it commits us to a disjunction:
(b) It is either true that {niggers/broads} will benefit from improvements in medicine or false
that {niggers/broads} will benefit from improvements in medicine.
Yet a commitment to (b) seems to entail a commitment to {racism/sexism}. However we may deal
with Grim's problem (cf. fn. 14 below), it is relevant that metalinguistic negation can be employed
by a speaker who wishes to reject the bigoted or chauvinistic point of view embodied in an earlier
statement within the discourse context:
(c) I beg your pardon: Lee isn't an 'uppity {nigger/broad/kike/wop/...}'-(s)he's a strong,
vibrant {black/woman/Jew/Italian/...}.
For someone who utters (c), as with 20-21, the denotative meaning of the statement under attack
(what was SAID) may well have corresponded exactly to that of the rectified statement; the con-
notative meaning (what was IMPLICATED)cannot be allowed to stand unchallenged. R. Lakoff
(1975:19-27) points out that we seem to need euphemisms in exactly those referential contexts
where we also have slurs. It is thus significant that we find both euphemisms and their metalinguistic
rejections in the Grim contexts above:
(d) I'm not 'colored'-I'm black!
(e) I'm not 'a gentleman of the Israelite persuasion'-I'm a Jew!
(f) I'm not a 'lady'-I'm a woman!
134 LANGUAGE, VOLUME 61, NUMBER 1 (1985)

achieved the status of a cliche-second only perhapsto that in 22, where the
play between ordinaryand markeduses of negation has entered immortality
throughvaudeville:
(22) (Who was that lady I saw you with last night?)
That was no lady, that was my wife!
Note that the second speakerin this routinedoes not intendto suggest that his
wife is not a lady; rather, the negation attaches to the implicatureassociated
with the first speaker's utterance. The implicaturalmechanismhere is akin to
that in an example of Grice:
(22') X is meeting a woman this evening.
Use of this 'would normallyimplicatethat the person to be met was someone
otherthanX's wife, mother,sister, or perhapseven close platonicfriend'(Grice
1975:56).Whilenot all speakersmay find this implicatureas 'non-controversial'
as Grice maintains,it appearsto me that-to the extent it is felt to be present
in 22'-it can be removed throughnegation:
(22") No, he's not (meetinga womanthis evening)-he's meetinghis wife.
While the relevant implicatabeing denied or forestalledin 22 and 22"result
from the exploitation of the content maxim of Quantity('Make your contri-
bution as informativeas is required'-Grice 1975:45),manner-generatedim-
plicata may also be rejected; e.g.,
(23) Miss X didn't 'producea series of sounds that correspondedclosely
with the score of "Home Sweet Home" ,' dammit,she SANG'Home
Sweet Home'.
Here what is denied is the reviewer's implicatum,devolving from an exploi-
tation of the Brevity maxim, viz. that 'Miss X's performancesuffered from
some hideous defect' (cf. Grice 1975:55-6).
But we have clearly come a long way from either the well-behavedordinary
internalnegationoperatoror the semanticexternalnegationoperatorof three-
valued logics of the K&P analysis. What we are dealing with in the negative
examples of 17-23 are reflexes of what Ducrot 1972 has aptly termed META-
LINGUISTICnegation-a means for objecting to a previous utterance on any
grounds whatever, including(as in 18-19) the way it was pronounced."
It remains to be shown that these examples involve the SAMEbasic use of
negationas that found in K&P's examplesof canceled or rejectedconventional
implicata, as in 12. To this end, note first that, in the negative sentences of
17-23 (as in 12), felicitous use involves contrastiveintonationwith a final rise
within the negated clause (the 'contradictioncontour' of Liberman& Sag),
followed by a continuationin whichthe offendingitem is replacedby the correct
item in the appropriatelexical, morphological,and phonetic garb-a RECTI-
FICATION,to borrow a term from Anscombre & Ducrot. But it is not only the
intonation and rectification which point to a kinship with 'external' or 'con-
" Ducrot 1973,in
place of his 'negationm6talinguistique'of 1972,uses the new label 'negation
polemique'. I consider the earlier term more felicitous, especially in the light of examples not
discussed by Ducrotand his colleagues,e.g. 18-19. Negationdoes constitutein these cases a way
of rejectingthe languageused by an earlierspeaker,and is thereforeindeed METALINGUISTIC;but
it seems stretchingthe notionof polemicsor argumentation to labelthis varietyof negationPOLEMIC.
METALINGUISTICNEGATIONAND PRAGMATICAMBIGUITY 135

tradiction'negation:in the cases just discussed, like those of K&P, no negative


polarity items are triggered. Thus, while 17b is possible with metalinguistic
negation, its polarity counterpartis not:
(24) *John didn't manage to solve ANYof the problems-he managedto
solve ALLof them.
And parallel to 13c, in which some/any suppletionis ruled out, we also find
this:
(25) I didn't [miYonij] to solve {some/*any} of the problems-I
[maenijd]to solve some of the problems.
Of course, the principalresemblancebetween the instances of markedne-
gation introducedin this section and the classical examples of presupposition-
canceling negation discussed earlieris that both types occur naturallyonly as
responses to utterancesby other speakersearlierin the same discoursecontexts
(or as mid-coursecorrections, after earlierutterances by the same speakers).
It is for this reason that I seek to encompassall these examplesunderthe rubric
of metalinguisticnegation:they all involve the same extended use of negation
as a way for speakers to announce their unwillingnessto assert somethingin
a given way, or to accept another's assertion of it in that way. Given the
behavioralresemblancesjust cited, as well as the prevailingOccamist consid-
erations, there is no obvious reason NOTto collapse the presupposition-can-
celing negation of 4 and 12 with the negation attachingto conversationalim-
plicaturein 17, 22, 22",and 23, to pronunciationin 18a and 19, to morphology
or syntax in 18b and 19, to register or speech level in 20, and to perspective
or point of view in 21.
2.1. PRAGMATIC
AMBIGUITY.
The notion to which I am appealing here has a
rich, if brief, history. Donnellan 1966 coined the term to describe the two
understandingshe found for a sentence like the following:
(26) Smith's murdereris insane.
a. ATTRIBUTIVE:Whoever it may have been who murdered Smith is
insane.
b. REFERENTIAL:That individual [to whom I refer via the phrase
Smith's murderer] is insane.
In 26a, the description Smith's murdereris used essentially; but in 26b, it is
employed as a tool for picking out a specific individualand predicatingsome-
thing of him (or her), regardlessof whether that individualdid in fact murder
Smith.
Similarly,Wertheimerarguespersuasivelythat sentences containingmodals,
e.g. 27, are not semanticallyambiguous,but have either of two uses, as para-
phrased in 27a-b, dependingon the system of rules which is being implicitly
invoked:
(27) Lee {should / ought to} be in Chicago today.
a. EPISTEMIC:Accordingto my calculations,Lee is (probably)in Chi-
cago today.
b. ROOT or DEONTIC:It would be {desirable/ a good idea}for Lee to
be in Chicago today.
136 LANGUAGE, VOLUME 61, NUMBER 1 (1985)

(Kratzer 1977arrivesindependentlyat an analysis which also treats modals as


pragmaticallyambiguous;Palmer 1979, Leech & Coates 1980 offer complex
mixed treatments, incorporatingboth polysemy and semantic indeterminacy
in their analyses of modal 'ambiguities'.)
A thirdexample of pragmaticambiguityhas alreadybeen touched upon. As
suggested by Grice 1967, 1975-and elaboratedby Horn 1972, 1973, Gazdar
1979a,b-scalar predicationslike those below can be consideredpragmatically
ambiguousas between the implicature-bearing version (the 'two-sided' under-
standingof Aristotle, with lower and upperbounds, which resultsin conveying
28a and 29a respectively) and the implicature-free'one-sided' version (with
lower bound only, as in 28b and 29b):
(28) Some men are chauvinists.
a. Some but (for all I know) not all men are chauvinists.
(S
Some e if n
not allll . .
b. men are chauvinists.
LAtleast some J
(29) It is possible that it will rain tomorrow.
a. It is possible but (for all I know) not {necessary/certain}that it
will rain tomorrow.
b. It is at least possible that it will rain tomorrow.
In these examples, as in those of 26-27, the context (either linguistic or ex-
tralinguistic)may explicitly or implicitlyselect one of the two possible under-
standingsas preferred.
None of these proposed analyses is uncontroversial.Thus Donnellan 1978
argues for a SEMANTIC treatmentof the ambiguityin 26 (but cf. Kaplan 1978,
Kripke 1977 for defenses of Donnellan 1966);the standardlinguistic line on
modals treats them as SEMANTICALLY(and possibly syntactically) ambiguous
between epistemic and root (deontic) readings(cf. Hofmann 1966, Newmeyer
1969, Horn 1972, Jackendoff 1972);and there are at least two recent accounts
of weak scalar predicationsof the type illustratedin 28-29 which treat them
as SEMANTICALLY-or at any rate, truth-conditionally-ambiguous (Cormack
1980, Burton-Roberts1984;also cf. Kempson 1982).
Nevertheless, I believe the pragmaticversion of these ambiguitiesis largely
correct(cf. Horn 1984),andthatthe line takenon such constructionsis naturally
extendible to negation. What I am claiming for negation, then, is a use dis-
tinction: it can be a descriptive truth-functionaloperator,takinga proposition
p into a proposition not-p, or a metalinguisticoperatorwhich can be glossed
'I object to u', where u is cruciallya linguisticutteranceratherthanan abstract
proposition. 12
In claiming that negation is pragmatically,ratherthan semantically,ambig-
12
As Barbara Abbott has pointed out to me, u need not even be a specifically linguistic utterance,
as seen by the function of metalinguistic negation in the following musical scenario:
Piano student plays passage in manner L.
Teacher: 'It's not [plays passage in manner JL]-It's [plays same passage in manner '].'
The teacher's use of not is clearly not assimilable to anything remotely resembling truth-functional
propositional negation.
METALINGUISTIC NEGATION AND PRAGMATIC AMBIGUITY 137

uous, I am partly in accord with the classic monoguist position summarized


by Gazdar (1979a:92): 'There are no grounds for thinking that natural language
negation is semantically distinct from the bivalent operator found in the prop-
ositional calculus.' But the spirit (if not the letter) of this position is violated
by my approach, which takes a wide array of uses of natural language negation
to be NON-truth-functional, and indeed entirely non-semantic. (Note that the
pragmatic ambiguity of negation, as conceived here, is not entirely on a par
with the instances of pragmatic ambiguity just cited, where NEITHER under-
standing directly affects logical form.)
2.2. TRUTHVS. ASSERTABILITY. If we temporarily set aside the more extreme
cases of metalinguistic negation (e.g. those affecting phonetic representation,
as in 18a), then the distinction drawn above recalls a distinction made else-
where, the import of which has been insufficiently appreciated: that of the truth
of a proposition vs. the assertability of a statement or sentence. As Grice 1967
has forcefully pointed out, either truth or assertibility can be affected by ne-
gation; it is up to the addressee to determine just what the speaker intended
to object to or deny in the use of a negative form at a given point in the
conversation.
Grice defends the position that ordinary-language or exhibits the truth-con-
ditional semantics associated with the familiar truth table for inclusive dis-
junction, represented in the third column in Table 6. He deals with a potential

p q pV q Vq p p -q (p--q)
T T T F T F
T F T T F T
F T T T T F
F F F F T F
TABLE6.

objection to this claim as follows (V:9):


'If you say "X or Y will be elected", I may reply "That's not so: X or Y ORZ will be elected."
Here ... I am rejecting "X or Y will be elected" not as false but as unassertable.'

Grice puts this distinction to work in defense of his truth-conditional analysis


of conditionals. He begins by conceding that 30 does not have the truth con-
ditions which we should expect of a negated material conditional (cf. the last
two columns of Table 6):
(30) It is not the case that, if X is given penicillin, he will get better.
After all, this does not normally commit the speaker to an assertion of the
conjunction 'X will be given penicillin and won't get better', i.e. the second
line of Table 6. In the same way, I can deny Nietzsche's notorious conditional
31a without committing myself to the conjunction in 31b:
(31) a. If God is dead, everything is permitted.
b. God is dead and something is forbidden.
Grice points out, however, that a speaker uttering 30-or, even more clearly,
30'-is not, in fact, truly NEGATING the contained conditional proposition, but
-

138 LANGUAGE,VOLUME61, NUMBER 1 (1985)

is rather ASSERTINGHIS UNWILLINGNESS TO ASSERTthat proposition(V:5).


(30') It is not the case that, if X is given penicillin, he will get better; it
might very well have no effect on him at all.
Here as elsewhere, 32a is to be interpreted as a refusal to assert If p then q:
rather than as a (descriptive) negation of a conditional whose truth value is
determined in accordance with the material equivalence in 32b:'3
(32) a. It is not the case that if-p-then-q.
b. -(p -- q) (p & ~q)
I do not want to insist here on a wholesale defense of Grice's analysis. His
advocacy of material implication as an adequate representationfor the se-
mantics of naturallanguageconditionalsis especially moot; indeed, the truth
conditions for if-then statements have been passionately but inconclusively
arguedever since the 3rd centuryB.C., when Callimachusobserved that 'even
the crows on the rooftop are cawing about which conditionalsare true' (Mates
1949:234). But the distinction drawn by Grice (and Dummett, cf. fn. 13,
above)-between rejecting a claim as false, and rejecting it as (perhapstrue
but) unassertable-suggests the proper approachfor characterizingthe two
general uses of negation.14

13The same pointis madeby Dummett(1973:328-30),who drawsa distinctionbetween


negation
outside the scope of a Fregeanassertionoperator,'Not (-A)', as opposed to the normalassertion
of a negative proposition,'-(not A)'. The formerinterpretation,he suggests, 'mightbe taken to
be a means of expressingan unwillingnessto assert "A"'; the clearestcandidatesfor this species
of negationare those where 'A' is a conditional.He cites exchangeslike this:
(a) X: If it rains, the matchwill be canceled.
Y: That's not so. (OR,I don't thinkthat's the case.)
Here Y's contributionis not actuallya negationof X's content; rather,we can paraphraseY as
having conveyed (b) or (c):
(b) If it rains, the matchwon't necessarilybe canceled.
(c) It may [epistemic]happenthat it rains and yet the matchis not canceled.
Dummett,in fact, goes beyond Grice, concluding(330) that apparently'we have no negationof
the conditionalof naturallanguage,that is, no negation of its sense: we have only a form for
expressingrefusalto assent to its assertion.'(WhileDummettoffersno explanationfor this curious
state of affairs,Grice 1967:Vdoes give a pragmaticstory for the failureof conditionalsto undergo
ordinarydescriptivenegation.)It shouldbe acknowledgedin passingthatthe notionASSERTABLE-
as employedby Grice,by Dummett,andby me-should properlybe takenas ellipticalfor something
like 'felicitously'or 'appropriately'assertable,wherethe adverbialhedgeis broadenoughto cover
the wide range of examples under considerationhere. I take 'assertability'to be an instance of
linguisticshorthandon a parwith 'Ca'nyou say X?' or 'You can't say Y' forjudgmentsof syntactic
(un)acceptability.
14 The truth
vs. assertabilitydistinctionappearsto vitiateGrim'sargumentfor the non-bivalence
of (a) in fn. 10, above: while (a) does indeed commitus to the TRUTH of (b), it does not commit
us to its ASSERTABILITY.
Grim is in fact aware of this argument (293-5), but he dismisses it on the
groundsthat the existence of true 'unassertables'is 'somethingof an intellectualembarrassment';
an alternativetheory which does not posit them is to be preferred.Yet any such alternativemust
countenancetruth-valuegaps-a majordeparturefromclassicaltwo-valuedsemanticswhich may
be neithera necessary nor a sufficientexpense for dealingwith presuppositionalphenomena,and
is in any case hardlya cause for intellectualpride. If true but unassertablestatementsmust be
countenancedin any case (as the discussionin the text and below suggest), it is a simple matter
METALINGUISTIC NEGATION AND PRAGMATIC AMBIGUITY 139

2.3. THE SCALARCASES. We have noted that languages tend not to distinguish
internal from external negation morphologically. It is thus especially significant
that natural languages seem (almost) always to allow a descriptive negation
operator to double for metalinguistic use as a comment on the utterance, chal-
lenging what is presupposed or implicated as well as what is asserted. One
frequent use of metalinguistic negation-indeed, virtually universal (but cf. ?5
below)-is as a way of disconnecting the implicated upper bound of weak scalar
predicates, as in 17 above, or the following:
(33) Around here we don't LIKE coffee-we LOVEit.
(Lauren Bacall, in TV commercial for High Point decaf)
Again, let us focus on the contrast between 34a-b, or more precisely on their
mutual consistency:
(34) a. Max has three children-indeed, he has four.
b. Max doesn't have three children-(*but) he has four.
c. Max doesn't have three children, (but) he has two.
It seems peculiar at first glance that the same state of affairs can be alternatively
described in terms of Max's HAVING three children and of his NOT having three
children.
Following Mill 1867, DeMorgan 1847, and Grice, I have argued elsewhere
(Horn 1972, 1973; cf. Gazdar 1979a,b for formalizations) that scalar operators
like some (in 28), possible (in 29), like (in 33), and three (in 34) are lower-
bounded by their truth-conditional semantics; and that they may be upper-
bounded (context permitting) by conversational implicature, triggered by
Grice's maxim of Quantity. Given that all men are mortal (or chauvinists-cf.
17a), it's inappropriate, although true, to assert that SOMEmen are. Similarly,
if I know that Max has four children, and this fact is relevant to you, it's
misleading for me to inform you that he has three (although it's true that he
does). In each case, I have said something true but implicated something false.
Within this account, the negation in 34b does not negate the PROPOSITION that
Max has three children; rather, it operates on a metalinguistic level to reject
the IMPLICATUM that may be associated with the assertion of that proposition
(viz. that he has only three). By uttering 34b, the speaker signifies unwillingness

to extend this treatment to cover Grim's cases. (Stenner, in his reply to Grim, advocates essentially
this approach.)
I have suggested elsewhere (Horn 1981, fn. 8) that the problems encountered by Aristotle and
Lukasiewicz in their analyses of future contingent statements hang on the failure to appreciate this
distinction. Pace Aristotle and Lukasiewicz, (exactly) one of the following two statements is indeed
TRUE today:

(a) There will be a sea battle tomorrow.


(b) There will not be a sea battle tomorrow.
But it may well be that (barring precognition) neither of them may be ASSERTABLE.Thus, the very
birth-throes of three-valued logic (in Lukasiewicz 1930) may have been attended by an insufficient
recognition of the truth/assertability distinction: an epistemic problem was misdiagnosed as a
metaphysical one, and an improper treatment consequently prescribed. (Geach 1972:81 argues that
this mistaken line on future contingents was not in fact taken by Aristotle, despite the standard
interpretation; in any case, if the Aristotelian man was straw, Lukasiewicz fleshed him out.)
140 LANGUAGE, VOLUME 61, NUMBER 1 (1985)

to assert a sentence which would induce a misleading implicature. The negation


in 34c, by contrast, is naturally taken descriptively as attaching to the prop-
osition that Max has three children. (The distribution of but in 34b-c is dis-
cussed in the Appendix.)
2.4. INtC(RPORAATEl)NEGAI (IONAS A DIAGNOSTIC. One important correlate of
the negative dichotomy resides in the inability of metalinguistic negation to
incorporate prefixally:
(35) a. The king of France is {not happy/*unhappy}-there isn't any
king of France.
The queen of England is {not happy/*unhappy}-she's ecstatic.
It isn't possible
b. It's not possible for you to leave now-it's necessary.
*lt's impossible
f not probable .
ainnot probable l ,posbe
otimprobable t improbable s
f not
not likely .
likely} e butt certain (vs. fnot likely
not likely} if not impossible)
*unlikely unlikely
{ not interesting but f
*uninteresting J
In each case where the negative operator is used metalinguistically to deny the
appropriateness of using a predicate which would yield a true but misleadingly
weak assertion (cf. the preceding section), the negation cannot incorporate
morphologically as un- or iN--perhaps because it is operating, in effect, on
another level. (The failure of metalinguistic negation to trigger polarity items,
as discussed above, should probably be pegged to the same factor.) The ac-
ceptable incorporated negatives in 35 all involve ordinary, truth-functional uses
of the operator.15
Failure to recognize this diagnostic for metalinguistic negation mars an oth-
erwise cogent defense by Gazdar 1977, 1979a of a monoguist analysis of natural
language disjunctions. On the view that or is semantically ambiguous, a sen-
tence like 36a will be assigned two distinct logical forms, corresponding to the
inclusive (p V q) and exclusive (p V q) interpretations of the connective (cf.

'5 This account of the incorporation diagnostic must draw a sharp distinction between the lex-
icalized prefixal negation of the examples in 35 and the 'contracted' n't in examples like 34b, 35b,
and numerous additional sentences scattered throughout the text. As these examples show, nothing
constrains metalinguistic negation from contracting as an enclitic onto the copula. If (as has been
traditionally assumed) the n't forms are produced by post-lexical syntactic and/or phonological
rules rather than in the lexicon, the distinction is made automatically. However, as Jerry Sadock
has pointed out to me, the adoption of Zwicky & Pullum's 1983 analysis of Xn't as an inflected
form of the auxiliary element X, generated by the morphology, would require a different account
here. I shall simply assume that the grammar has some way of distinguishing the lexical prefixes
un-, in-, and non- (which are incompatible with metalinguistic negation) from the -n't forms (which
are not).
METALINGUISTIC NEGATION AND PRAGMATIC AMBIGUITY 141

Table 6 above):
(36) a. John is either patrioticor quixotic.
b. John isn't either patrioticor quixotic.
c. John is neither patrioticnor quixotic.
This ambiguistthesis is rejected by Gazdar(1979a:81-2) on the grounds that
it makes a 'bizarrelyfalse prediction' when the disjunctionis broughtwithin
the scope of negation. Thus 36b-and 36c, which Gazdartakes to be its para-
phrase-will be assignedtwo readingsby the ambiguistanalysis, with negation
outside the scope of inclusive and exclusive disjunction, respectively. Given
the standardequivalence
(36') ~(p V q) (-p & -q) V (p & q),
the exclusive readingof 36b-c will be true if Johnis both patrioticand quixotic.
Gazdarconcludes thatthe ambiguistthesis on disjunctionis primafacie absurd,
since 36b-c are patently false in this state of affairs.
Yet Gazdar'sargumentcontains a fatal flaw: 36b does allow a readingwhich
is NOT patently false, i.e. the readingwhich involves the metalinguisticuse of
negation. Considerthe following:
(37) a. Maggie isn't EITHERpatriotic OR quixotic-she's both!
b. *Maggieis neither patrioticnor quixotic-she's both!
Not only is 37a a possible discourse utterance, it's also one with which most
British subjects would happily agree (for the most prominentreferent of the
proper name). However, this readingdisappears, as it should, when the ne-
gation is incorporatedin 37b. Since such incorporatednegation can only be
descriptive, 37b is indeeda logicalcontradiction,as Gazdar'sanalysis predicts.
A recently attested instance of just such a metalinguisticuse of negationas
that contained in 37a is the following:
(38) The Constitutiondoesn't say providefor the common defense OR the
general welfare; it says both.
(WalterMondale, at the DemocraticNational Party Conference in
Philadelphia,1982, quoted in the N.Y. Times)
As furtherconfirmationof the claim that the negation in 37a or 38 is metalin-
guistic in the strong sense intendedhere (as a commenton an earlierutterance
ratherthan a propositionalnegation), consider the apparent(graphemic)con-
tradictionin 39a, whichcan neverthelessbe resolved in the appropriatecontext,
illustratedin 39b:
(39) a. Maggie isn't either patrioticor quixotic-she's either patrioticor
quixotic.
b. -Say, {Clive/Fiona}, you have to admit your Maggie is [fYSr
peYtriaDik or kwlksaDik].
-No, I haven't. Maggie isn't [iY.r peYtriaDik or kwtksaDik]-
she's [aySo petriotik o' kwiksotik].
Curiously, Gazdar's argument-flaw and all-was prefiguredin a parallel
attackby Grice (1978:116-18)on a differentambiguisttreatmentof disjunction.
142 LANGUAGE, VOLUME 61, NUMBER 1 (1985)

In Grice's case, the relevant 'strengthening'propertyis not the truth-functional


one of exclusivity, but a non-truth-functional,epistemically based condition
on the appropriateuse of disjunctive statements which Strawson and others
have attributedto the meaningof ordinary-language or. On that view (cf. Grice
1967:I), speakers who utter 'p or q' license the inference that they do not
know for a fact that p is the case (or, of course, that q is). Thus I cannot
felicitously utter 40, on the 'strong' readingof or, if I know for a fact that my
wife is in Oxford (and if this fact is relevant to you when I utter 40):
(40) My wife is either in Oxford or in London.
Grice (1978:116-8)notes that the 'strong'readingfor this disjunction(the read-
ing which builds in the strengtheningproperty)seems to disappearunder ne-
gation, as in 41a; but he does NOTmentionthatthe relevantnon-truth-functional
aspect of the interpretationof such disjunctionsmay indeed be canceled by
metalinguisticuse of negation, as in 41b-d:
(41) a. Your wife isn't (either) in Oxford or in London.
b. Your wife isn't (EITHER) in Oxford ORin London, dammit, she's
in Oxford, as you bloody well know!
c. I didn't do it once or twice-I did it once and once only!
d. It won't be Noam's Pride ORResnic's Choice standing laurel-
crowned in the winner's circle-it'll be Noam's Pride, you can
bet on it!
In these examples, a disjunctionis again disowned-not because it is false,
but because the utteranceexpressingit would be too weak and hence regarded
as unassertable. In 37a, 38, and 41b-d (as in 17, 33, and 34b in the earlier
discussion) negationis used metalinguisticallyto focus on the implicatumtrig-
gered by the maxim of Quantity. In such cases, the negation often seems to
build in a covert just or only which can in fact be expressed directly without
changing the import.16Comparethe versions of 42 with and without the pa-
16
Analogouscases based on clefts are adducedin Horn 1981for an argumentthat the 'exhaus-
tiveness' premiseis not part of the meaningof cleft sentences, i.e. that (a) does not (pace Atlas
& Levinson 1981) ENTAILor (pace Halvorsen 1978) CONVENTIONALLY
IMPLICATE(b):
(a) It was a pizza that Maryate.
(b) Maryate nothingother than a pizza.
It may seem that (c) below, fromAtlas & Levinson, or the analogous(d), suggeststhat the failure
of exhaustivenessto hold DOESconstitutesufficientgroundsfor denyingthe truthof a cleft:
(c) It wasn't John that Marykissed, it was John and Bill.
(d) It wasn't a pizza that Maryate-it was a pizza, a calzone, and a side orderof ziti.
In fact, if such sentences are acceptable(as Atlas & Levinsonmaintain),they are acceptableonly
with metalinguisticnegation,cancelingthe upper-bounding (exhaustiveness)implicatum.Note first
thatjust may be inserted, salva veritate (et sensu), immediatelyafter the negationin (c) and (d)
as easily as in the cases of 42. Furthermore,the same constructionoccurs withoutcleft syntax:
(e) Mary didn't eat (just) A PIZZA-she ate a pizza, a calzone, ANDa side order of ziti.
If Mary ate a pizza along with other items, it is undeniablyTRUEthat she ate a pizza; negation
here, as in (c) and (d), is used to deny not truthbut assertability.(Thatjustor only can be inserted
in these cases shouldnot be taken as evidence that metalinguisticnot can be analysedessentially
as ellipticalfor not only: cf. ?4.1 below. For additionaldiscussion of the relationbetween clefts
and exhaustiveness, cf. Horn 1981.)
METALINGUISTIC NEGATION AND PRAGMATIC AMBIGUITY 143

renthesized element following metalinguistic negation:


(42) a. Max doesn't have (just) THREEchildren-he has FOUR.
b. You didn't eat (just) SOMEof the cookies-you ate ALLof them.
c. Around here, we don't (just) LIKEcoffee-we LOVEit.
d. I don't (just) BELIEVEit-I KNOWit.
Another morpho-syntactic correlate of the metalinguistic/descriptive di-
chotomy exhibited by natural language negation can be found in the distribution
of concessive and contrastive but constructions, discussed in the Appendix.
2.5. TRUTH,TRUE, ANDNEGATION. As noted in ?1.3, the Kroch-Linebarger
line on negative statements treats external negation as an ordinary truth-func-
tional negative operator applied to a semantic TRUEpredicated directly within
its scope. Consider sentences like these:
(43) a. *She DIDNOTlift a finger to help.
b. *We DIDNOT get up until 12:00.
Seeking to explain the unacceptability of negative polarity items when they are
read as 'denials', with rising intonation, Linebarger (35) cites Kroch's definition
of 'external negation' as
'a "metalinguistic" usage in which the negative sentence NOT S does not directly comment
on the state of affairs but instead denies the truth of the statement S previously uttered or
implied. Sentence-external negation can be paraphrased as "The sentence S is not true".'

Linebarger formalizes this account of external negation by representing the


logical form of the 'denial' readings of 43a-b as follows:
(43') a. NOTTRUE(she lifted a finger to help).
b. NOTTRUE (we got up until 12:00).
What rules these out as possible well-formed formulas is that the negative
polarity items (lift a finger and until) are no longer within the immediate scope
of negation; thus they fail to meet what for Linebarger is a necessary (although
not sufficient) condition for the acceptability of the relevant type of polarity
trigger. In the same fashion, Linebarger (36 ff.) notes, the ill-formedness of 44
is correctly predicted by assigning it the 'external' representation 44':
(44) *The King of France didn't contribute one red cent because there is
no King of France.
(44') NOTTRUE(the King of France contributed one red cent) ...
One possible objection to this characterization of 'external' negation is that-
as investigators as diverse as Frege (cf. Dummett 1973), Wittgenstein 1953,
?447, Ducrot 1973:119, and Givon 1978 have observed-ALL instances of ne-
gation (including the ordinary, garden-variety propositional negation which lo-
gicians try to capture) may be viewed as constituting, in Kroch's words, a way
to deny 'the truth of the statement S previously uttered or implied'.
That is, negative statements, regardless of the specific function of negation
in question, are marked with respect to the corresponding affirmative; and they
often pragmatically presuppose a context in which the affirmative proposition
has been asserted or at least entertained. Thus Givon (109) concludes, from
his extensive cross-linguistic survey, that negative statements tend to be uttered
144 LANGUAGE, VOLUME 61, NUMBER 1 (1985)

'in a context where the corresponding affirmatives have already been dis-
cussed, or else where the speaker assumes the hearer's belief in-and thus
familiarity with-the corresponding affirmative'; he shows that the reflexes of
these 'discourse-pragmatic presuppositions' associated with negation include
the restricted distribution, diachronic conservatism, and psychological com-
plexity of negatives vis-a-vis affirmatives.
Whether or not this characterization of negation is correct (cf. ?4.1), a more
serious problem remains for the Kroch-Linebarger position. We have en-
countered many cases which pose insurmountable difficulties for any theory
in which the special metalinguistic negation exemplified in 43-44 is directly
associated with denial of truth: exx. 17-23, 30', 34b, 37a, 42 etc. It hardly
seems plausible to analyse 17a (SOMEmen aren't chauvinists-ALL men are
chauvinists) in terms of a Linebargerian representation like the following:
(45) NOTTRUE(some men are chauvinists) ...
Even in the more semantically-based examples considered by Karttunen &
Peters 1979, such as 12 (John didn't MANAGE to solve the problem-it was quite
easy for him to solve), we encounter the same sort of problem. An analysis of
this 'external' or 'contradiction' negation along the line of the Linebarger model
yields this:
(46) NOTTRUE(John managed to solve the problem) ...
Yet, as K&P correctly observe, the simplest truth-conditional account of sen-
tences like 12 is one in which the proposition corresponding to the parenthe-
sized material in 46 is indeed true in any state of affairs in which John solved
the problem.
The difference between the manage case of 12 and the classic king of France
example in 44 is that propositions containing definite descriptions ENTAIL(as
well as presuppose or conventionally implicate) the corresponding existential
expressions; but X managed to c does NOTentail (though it may presuppose
or conventionally implicate) that it was difficult for X to k (cf. Karttunen &
Peters 1979, Gazdar 1979a).
Metalinguistic negation, as we have seen, is used to deny or object to any
aspect of a previous utterance-from the conventional or conversational im-
plicata that may be associated with it, to its syntactic, morphological, or pho-
netic form. There can be no justification for inserting an operator TRUE into
the logical form for a certain subclass of marked negative sentences, in order
for 'external' negation to be able to focus on it, if metalinguistic negation does
not in general directly affect truth conditions.
Perhaps in these cases of non-truth-functional negation, we could try placing
the negative operator outside the scope of a semantic operator like APPROPRIATE
or CORRECT, rather than TRUE.But this 'solution' merely shifts the problem
back one level, given that metalinguistic negation-unlike ordinary descriptive
negation, or the so-called 'external' semantic negation of Kroch and Linebarger
(which I am arguing does not exist)-is simply not a truth-functional operator
on propositions. Thus representations like 47a are essentially as inadequate as
METALINGUISTIC NEGATION AND PRAGMATIC AMBIGUITY 145

47b for the cases under consideration, given that those aspects of the utterance
which metalinguistic negation is used to focus on may have nothing to do with
the proposition expressed by the utterance:
(47) a. NOT {APPROPRIATE/CORRECT}(p)
b. NOT TRUE (p)
Conventional implicata (or presuppositions) may be analysed as attributes of
propositions (albeit non-truth-conditional attributes); but conversational im-
plicata-and, a-fortiori, morphological and phonetic form, register etc.-can-
not be.
This essential difference between descriptive and metalinguistic negation
provides the most serious problem for the over-Occamistic view of the strong
monoguists, that all uses of negation can be assimilated to the same truth-
functional analysis. It must not be overlooked that marked negation differs
from descriptive negation not only phonologically, morphologically, and syn-
tactically, but also in semantic function. In particular, metalinguistic negation,
as an extra-logical operator, plays no straightforward role with respect to such
central inference rules as double negation and modus tollendo ponens (M.T.P.);
these laws would thus be unstatable if all uses of negation were to be treated
identically. If we chose to tar descriptive negation with the same brush as
metalinguistic negation, we could no longer draw such basic inferences as these:
(48) a. I didn't manage to solve the problem.
.. I didn't solve the problem. (cf. 18a)
b. Maggie isn't either patriotic or quixotic.
.'. Maggie isn't patriotic. (cf. 37a)
In the same vein, Wilson (149), citing disjunctive denials of the type first
noted by Grice (cf. ?2.2 above), observes that the two clauses of 49 seem to
constitute premises in a disjunctive syllogism, viz. 49':
(49) The next Prime Minister won't be Heath: it will be Heath or Wilson.
(49') -p
pVq
.'. q (via M.T.P.)
Yet we don't actually infer q-i.e., The next P.M. will be Wilson-from an
assertion of 49. But instances of DESCRIPTIVE negation DOlicense M.T.P.: if I
know that Heath or Wilson has been elected, and I hear Heath's concession
speech, I do have the right to conclude that Wilson (Harold, not Deirdre) was
the winner. In short, forcing all instances of negation into a single Procrustean
bed-however skillfully the bed may be designed-accomplishes little beyond
playing Pandar to some rather odd bedfellows.
But if metalinguistic uses of negation involve denial of assertability, rather
than of truth, why is it that the syntax used to express this use of negation
often seems to bring in some explicit reference to what is true? Recall that, in
34b (Max doesn't have THREE children-he has FOUR),I claimed that negation
attaches metalinguistically to the conversational implicatum associated with
the utterance of Max has three children, rather than descriptively to the prop-
146 LANGUAGE, VOLUME 61, NUMBER 1 (1985)

osition expressed by that utterance. But at least some speakers can also get
50a, and sentences like 50b are also heard and interpretedwithout difficulty:
(50) a. It's not true that Max has THREEchildren-he has FOUR.
b. It's not {true / the case} that SOMEmen are chauvinists-ALL men
are chauvinists. (cf. 17a)
Parallel syntax can also be found in Grice's disjunctive and conditional ex-
amples, discussed in ?2.2 above.
Does this mean that we're on the wrong track? Do these examples involve
a semanticexternalnegationafterall-so that (Occam'srazornotwithstanding)
conditionals, disjunctions, and weak scalar predications are all semantically
ambiguous?No. Rather, what these sentences show is that the distributionof
the English expressions It is true that, It is the case that, It is so that etc.-
and their correspondentscross-linguistically-is a poor guide at best as to
where the LOGICAL
predicate TRUEis to be applied in the simplest, most elegant
semantic/pragmatictheory of naturallanguagemeaningand communication.
We often say that somethingisn't true, meaningthat it isn't assertable. This
is not ALWAYSpossible: thus it strikes me as odd to insert true into those
metalinguisticexamples hingingon grammar,speech level, or phonetics:
(51) a. ?*It's not true that I [miYonijd] to solve the problem-I
[menijd] to solve the problem.
b. ?*It's not the case that the dog SHATon the carpet-he DEFECATED
on it.
c. ?*Ce n'est pas vrai que j'ai 'coo-pay luh vee-and'-j'ai coupe la
viande.
It is true that the implicature-cancelingexamples of 50a-b remainproblemat-
ical. But it is no less true that, in ordinarylanguage,we often deny or ascribe
truthto a given propositionin manyinstanceswhere the simplesttheory would
representus as in actualitydoing somethingelse entirely. One case in point is
inspired by an example from Wilson (151):
(52) It's not true that they had a baby and got married-they got married
and had a baby.
Here the self-proclaimed'truth negation' focuses on an aspect of the use of
conjunctionwhich Grice 1967, 1975has convincinglyarguedis not partof truth
or meaning proper at all: the interpretationof and in certain contexts as and
then.17

17 Evidence for the view that


conjunctions like
(a) They had a baby and got married
are not semantically ambiguous between 'and also' and 'and then' readings includes the following
facts:
(i) On the ambiguist theory, conjunction in virtually every language would be ambiguous in just
the same way.
(ii) No natural language contains a single conjunction which is ambiguous between 'and also'
vs. 'and earlier' readings; i.e., no language could be just like English except that it would contain
a conjunction SHMANDsuch that
METALINGUISTICNEGATIONAND PRAGMATICAMBIGUITY 147

For an even more clearcutexample, I shall turn to some personallyattested


evidence involving the extended use of true in a non-negativecontext. Several
years ago, I was awakenedfor a pragmaticsclass by the sound of my electric
clock-radio cheerfullydispensingreveille:
(53)

After the familiartune ended, the announcercommented:'Yes, it's true; it IS


time to wake up.' Now, whathas been assertedto be truehere?The proposition
(abbreviated)in 53? Hardly: there is no proposition there, just a bunch of
measures in search of a bugle. (As Georgia Green has remindedme, reveille
does have words-indeed several alternatesets of words; but we cannot infer
from the announcer's comment that he is transderivationallyalludingto any
particularset of words, or in fact that he even KNOWSany set of words.) Rather,
the playing of reveille, given certain non-linguisticconventions in our culture,
can be performedwith the intention of indirectly conveying the proposition
that it is time for the reluctanthearerto awaken. It is this conveyed proposition
which is being called true; the prior indirect assertion of this proposition is
further illustratedby the anaphoricde-stressing in the radio announcer's ut-
terance.18

(b) They had a baby SHMAND


got married
would be interpretableeither atemporallyor-on its asymmetricreading-as 'They had a baby
and, before that, they had gotten married'.
(iii) The same 'ambiguity'exhibitedby and seems to arise in paratacticconjunction,when two
clauses describingrelatedevents arejuxtaposedwithoutany overt connective:
(c) They had a baby. They got married.
Grice'salternativeposition,whichI take to be correct,is that(a) is in fact semanticallyunivocal,
but may CONVERSATIONALLY
IMPLICATE(through an exploitation of the maxim 'Be orderly') that
the events occurredin the orderdescribed.The non-existenceof a conjunctionlike SHMAND could
then be ascribedto the non-existenceof any maximof the form 'Be disorderly.'Conjunctionis
thus potentiallyasymmetricthroughimplicature,in the same way that weak scalar predications
are potentiallyupper-bounded-in both cases, we can cancel or suspendthe implicatumif we don't
want to set it off:
(d) Some, {if not all I and possibly all}, men are chauvinists.
(e) They had a baby and got married,but not necessarilyin that order.
We have, then, one more instanceof pragmaticambiguityto add to the list begunin ?2.1. (Wilson
1975:96-9, Schmerling1975, Gazdar1979aprovideadditionalsupportfor the Griceanpragmatic
line on asymmetricconjunction;McCawley 1981:6-10, Bar-Lev & Palacas 1980offer ambiguist
counterproposals.)
18 Thereare othercases in the literaturein whicha
propositionthatis pragmaticallypresupposed
as part of the 'commonground'can serve as the basis for anaphoric-typede-stressingof material
actuallyutteredfor the first time. These include(a)-cf. Morgan1969-and (b):
(a) How does it feel to be a beautifulgirl?
[no pragmaticpresuppositionnecessary]
148 LANGUAGE,VOLUME61, NUMBER 1 (1985)

Parallelto the conjunctionand reveille examplesjust discussed is the phe-


nomenon induced by rhetoricalquestions of the type which Sadock 1975 has
called QUECLARATIVES:
(54) A: Who the hell buys that cockamamie line about pragmatic
ambiguity?
B: (a) Yes, that's true. (= Nobody does.)
(b) No, that's not necessarily so; there mightbe somethingto it.
Unlike the conjunctionin 52, neitherthe reveille nor the queclarativecase can
involve embedding:
(55) a. I guess I'll have to settle for polyester, because where (the hell)
can you find a 100%cotton jumpsuit anymore?
b. *It's not true that where (the hell) can you find a 100%cotton
jumpsuit anymore.
But this is presumablycaused by syntacticfactors: neithermelodies nor ques-
tions may occur embedded. In effect, the readingsof the type described here
constitute a 'root' or 'main clause' phenomenon;hence the contrast in 55 be-
tween because (which acts like a coordinatorwith respect to other 'root' phe-
nomena)and the subordinatingcomplementizerthat. Crucially,however, what
is being negatedor affirmed,agreedor disagreedwith, in 54B is not the question
in 54A-which, like the tune in 53, has no obvious truthvalue as such (but cf.
Karttunen& Peters 1976)-but ratherthe propositionwhich A is taken to have
pragmaticallyconveyed.
This suggests the line which I urge for 50, for the conditionalcases instan-
tiating 32a (e.g. 30'), and for the disjunctiveexamples 37a and 38: it is not the
propositionactuallybeing asserted which is denied, but the assertabilityof the
proposition (along with any associated implicata)conveyed in the context of
utterance. Following Kripke, what we must deal with here is a divergence of
SPEAKER'S meaning from SENTENCE
meaning.'9

(a') How does it FEELto be a beautifulgirl?


[pragmaticpresupposition:Addresseeis a beautifulgirl.]
(b) I thought you'd make it (... but you didn't.)
[no pragmaticpresuppositionnecessary.]
(b') I THOUGHTyou'd make it (... and sure enough you did.)
[pragmaticpresupposition:Addresseedid make it.]
As in the example 53, the materialfollowing the stressed element in (a') and (b') is treated as
thoughit hadbeen assertedearlierin the discoursecontext, andhence is de-stressed.A frequently
encounteredexampleof pragmatically-triggered de-stressingis in sportscasters'updatesof scores:
(c) will receive majorstress on 6 if and only if the Red Sox have just scored one or more runs:
(c) [And {thatmakes it / the score is now} ...] Red Sox 6, Yankees 3.
The Yankeescore, beingold information,countsas materialalreadyasserted,andso is de-stressed.
19CompareKripke'sgloss (256)on an exampleoriginallydiscussed by Linsky and Donnellan.
The situationis this:
'Someone sees a woman with a man. Takingthe man to be her husband,and observinghis
attitude towards her, he says, "Her husbandis kind to her", and someone else may nod,
METALINGUISTIC NEGATION AND PRAGMATIC AMBIGUITY 149

We should note one additionalway in which the use of the It is true that
preface in ordinarydiscourse differs from the semantic value of truth predi-
cates. Often, the only felicitous discourse-initiatinguse of the affirmativefor-
mula It is true that is a concessive one. If we begin by affirmingIt is true that
snow is white, rather than merely stating that Snow is white, we normally
continueby appendinga clause beginningwith but ... (An instanceof this usage
can be found in the text above, immediatelyfollowing 51c.) I shall not dwell
on this phenomenon here, except to suggest that it seems susceptible to a
naturalconversationalexplanation(a la Grice 1967), and to note that it gives
us one more reason to dissociate the definitionof the semantictruthpredicate
from the behaviorof ordinary-languagetrue (cf. G. Lakoff 1975:259for related
discussion).
OTHER METALINGUISTIC OPERATORS
3. If the approachsuggested here is correct for negation, it is plausiblethat
the naturallanguagereflexes of other logical operatorsshould come in similar

"Yes, he seems to be." Suppose the man in questionis not her husband.Suppose he is her
lover, to whom she has been drivenpreciselyby her husband'scruelty.'
On both Fregeanand Russelliantheories of truth, the statement
(a) Her husbandis kind to her
comes out false (since the individualactuallydenotedby the phraseher husbandis not in fact kind
to his wife)-a resultwhichDonnellan1978finds uncongenial.Yet even on the referentialreading,
Donnellan is not totally confident in assessing (a) as true. Kripke points out that (a) seems to
function ambivalentlyin dialogs like these:
(b) A: Her husbandis kind to her.
B: No, he isn't. The man you're referringto isn't her husband.
(c) A: Her husbandis kind to her.
B: He is kind to her but he isn't her husband.
As Kripkenotes, 'in the first dialog,the respondent(B) uses "he" to referto the semanticreferent
of "her husband"as used by the first speaker(A); in the second dialog,the respondentuses "he"
to refer to the speaker'sreferent.'Since definitepronominalization can 'pick up either a previous
semanticreferenceor a previousspeaker'sreference',each dialogis equallyproper(Kripke,270).
For our purposes,it shouldalso be noted that, as a free alternantof his reply in (c), B could have
responded,
(d) Yes, it's true, {thatfellow / he} is kind to her. But he's not her husband.
Or again, adaptingan even more familiarexamplefrom Donnellan1966,we obtain this dialog:
(e) A: The man in the cornerdrinkinga martiniis a spy.
B: Yes, it's true, he is indeed a spy. But actually, that's water in his martiniglass.
Sports pages often provideinstancesin which the predicatetrue picks out not the entire prop-
osition literallyexpressedby a previousutterance,but some sub-assertionwithinit. Reportingon
a postgameinterviewwith quarterbackDavid Woodleyof the MiamiDolphins,afterthey lost the
1983Super Bowl game, a journalistwrites,
'It was suggestedto Woodleythat when many people rememberSuperBowl XVII, they will
say the Dolphinslost because David Woodleyfailed to complete his last nine passes.
'"That's probablytrue", Woodley said.
'Woodleywas not sayingit was true that the critics will blamehim. He was sayingthat the
critics will be correctin saying the quarterbacklost the game.' (MalcolmMoran,New York
Times, 2/1/83)
150 LANGUAGE, VOLUME 61, NUMBER 1 (1985)

pairs, exhibitingmetalinguisticuses alongsidedescriptiveones. This is indeed


what we find.
Consider,for example, the extension of logical inclusive disjunctionto these
examples:
(56) a. Kim is bright, or {even / should I say} brilliant.
b. New Haven, or the Elm City, is the pearl of south-centralCon-
necticut.
c. Is the conductorBernst[iY]nor Bernst[ay]n?
(cf. The formulatorof relativity theory wasn't Einst[iY]nbut
Einst[ay]n.)
d. The currentPresidenthas appointedmore colored folk-or should
I say blacks-to prominentpositions ...
e. She deprivedher students of a lecture-or (better) sparedthem a
lecture-on the performativehypothesis. (after Wilson 1975)
f. Did Elizabethhave a baby and get married,or did she get married
and have a baby? (after Wilson 1975, McCawley 1981)
As Du Bois 1974 notes, a principal source of non-logical disjunction is the
phenomenon of intentional mid-sentence correction, as in one reading of
56a,d,e-or in exampleslike these, fromDu Bois (8), where the self-corrections
have 'survived presumablycareful editing':
(56')a. I can only very brieflyset forthmy own views, or rathermy general
attitudes. (Sapir, Language)
b. Let us look at the racial, or rather,racist themes in the argument
for population control. (Pohlman, Population: A clash of
prophets)
Metalinguisticuses of conditionalsinclude these:
(57) a. If you're thirsty, there's some beer in the fridge.
b. If you haven't already heard, PunxsutawnyPhil saw his shadow
and we're in for six more weeks of winter.
c. If I may say so, you're looking particularlylovely tonight.
Here each antecedent clause specifies a sufficient condition for the appropri-
ateness or legitimacyof assertingthe consequent, ratherthan for its truth. As
with metalinguistic negation, we can find morpho-syntacticdiagnostics for
metalinguisticuses of disjunctionsand conditionals:note that the disjunctions
in 56-56' cannot be paraphrased by either ... or ..., and that the consequent
clauses in 57 exclude initial then.
Ducrot (1972:175-8)adds to the familiarcases of 57 anothervariety of meta-
linguistic conditional statement, exemplified in sentences which translate as
follows:
(58) a. If the Cite is the heart of Paris, the Latin Quarteris its soul.
b. If the Bois de Boulogne is the lungs of Paris, the neighborhood
squareis its pores.
As Ducrot notes, the antecedentin these cases is understoodas proposingto
justify the metaphorin the main clause (by virtue of accepting the metaphor
METALINGUISTICNEGATIONAND PRAGMATICAMBIGUITY 151

in the antecedent). The sense is 'If you're willing to grant p, you must
grant q.'20
Perhaps the closest pragmaticdoublet for negation, however, is offered by
questions. Whatare generallylabelledECHO questions(or, followingPerlmutter
& Soames 1979:589-90, INCREDULITY questions) might, in the present context,
be renamed METALINGUISTIC questions. As with the most natural occurrences
of metalinguisticnegation, echo questions often seem to require a linguistic
context in which the originalutterance (be it a declarative, an imperative, or
itself a question) has been previously uttered within the discourse. Consider
the circumstanceswhich might evoke these echo questions:
(59) a. You did WHATwith Sally and Bill?
b. Take out the WHAT?
c. Do I WHAT?
The distributionof echo questions is determinedin accordance with the sen-
tence-type they are used to echo. Echoes of declarativesoccur in declarative
contexts, echoes of questions in question environments,and so on:
Mary is dating{Fred/wHo?}l
(59') a. John thinks
*who Mary is dating.
who Mary is dating.
b. John wonders *Maryis dating{Fred/wHo?}
where WHOwent?
Andjust as metalinguisticnegationis impotentto triggernegativepolarityitems
or to incorporateprefixallyas descriptivenegationsdo, echo questions-as is
well known-fail to exhibit normalinterrogativesyntax; they neither exhibit
wH-frontingnor triggersubject-auxiliaryinversion.
There is, then, reason to believe that the existence of parallelmetalinguistic/
descriptive splits for other logical operators,ratherthan supportingthe strong
monoguistline on negation(as Kempson 1975:184suggests), in fact reinforces
the line on negationurgedhere. If we are unwilling,in constructingthe simplest
semanticand syntactictheory, to collapse the or clauses in 56-56' with ordinary
inclusive disjunctions,the if clauses in 57-58 with ordinaryconditionals(what-
ever THEYare), and the echo questions of 59-59' with normalwH-questions,
we must be equally unwillingto claim that all negations are one.

OTHER APPROACHES TO METALINGUISTIC NEGATION


4. The analysis presented here, on which markednegation is taken to rep-
resent a metalinguisticuse of the negative operator rather than (as with de-
scriptive negation) a semantic operator which is part of logical form, bears
varyingdegrees of kinshipto other accounts of negationwhich have been pre-
sented or defended over the last few years. In this section, some of the more

20 For readers who lack the appropriate Parisian frame of reference, the closest domestic coun-
terpart I could devise is one based on my own hometown: If the docks are the burly forearms of
New York, the subways are the pits.
152 LANGUAGE,VOLUME61, NUMBER 1 (1985)

importantof these accounts will be summarizedand comparedwith the view


outlined in ?2 above.
4.1. FALSITY AND NEGATION: THE MONOGUIST THESIS REVISITED. We have
briefly noted the traditionembodiedin the work of linguistslike Jespersenand
Giv6n, and of psychologists like Herbert and Eve Clark, in which negative
statements are taken as generally markedor complex relative to their corre-
sponding affirmatives. There is also a longstandingphilosophical tendency,
manifested in somewhat different ways by Kant, Wittgenstein, Searle, and
many others, of taking negative statements as constitutinga kind of special
speech act of denial, on a differentlevel from the correspondingaffirmative
statement. This tendency, however, has not gone unchallenged. Thus Gale
(1970:201)observes that
'Many philosophershave claimed that negation signifies a person's mental act of denying,
rejecting,or rebuttinga statementthatis actuallymadeor envisionedas beingmadeby some-
one.

But, Gale goes on to point out, this account is not wholly satisfactory. It is
simply not true that the statement addressed by a negative must have been
either made or envisioned as being made. Furthermore,since positive state-
ments can also be used to deny another's assertion, we can have no general
equivalence of the form
(60) It is not the case that S I deny that S.
As Gale notes, if the above equivalence held, then (given the principle of
the excluded middle, S V -S) 60'b should be just as necessarily true as 60'a:
(60')a. Either it is not the case that S or it is [not the case that it is not]
the case that S.
b. Either I deny that S or I deny that deny that S.
Yet while 60'a is indeed valid, 60'b is not. Furthermore,it does not follow from
the right-handside of 60, as it does from the left, that S is false. Negation,
Gale concludes, is part of the propositionalcontent of the statementin which
it occurs, ratherthan markingthe pragmaticfunction of expressingthe speak-
er's propositionalattitudetowardsome affirmativestatementthatwas (or might
have been) made.
Along the same lines, Geach (76) warns againstthe 'widespreadmistake' of
assuming that
'the negationof a statementis a statementthat that statementis false, and thus is a statement
ABOUT the originalstatementand logically secondaryto it'.
Geach uses the behavior of non-declarativesto show that this approach is
mistaken:
'"Do not open the door!" is a command on the same level as "Open the door!" and does
not mean (say) "Let the statementthat you open the door be false!"'
While the negative predicationsmay be linguisticallymore complex than their
correspondingaffirmatives,they are on the same level logically: 'we must ...
METALINGUISTICNEGATIONAND PRAGMATICAMBIGUITY 153

reject the view that a negative predicationneeds to be backedby an affirmative


one' (78-9).
Both Gale's and Geach's caveats are well-taken;however, they seem to offer
no explanationfor why luminarieslike Kant and Wittgensteinmighthave been
deceived into drawingtheir distinctionsbetween the characterof negative and
affirmativestatements.21Nor is any connection drawnbetween the treatment
of negationas a propositionaloperatorand the propertiesof morpho-syntactic
markednessso characteristicof negation in naturallanguage.
The solution to this stand-off, I suggest, is the recognitionthat truth-con-
ditional semantics does indeed (as arguedby Gale and Geach) contain a prop-
ositional negative operator,correspondingto descriptivenegationin the object
language-but that not all occurrences of naturallanguage negation can be
representedin this way. As we have seen, a need clearlyexists to accommodate
a 'denial' use of negation;but once we have weaned ourselves from the strong
monoguistthesis, there is no reason to expect the equivalence in 60 to hold.
In any case, 'I deny that S'-as in 60-is too restrictive a gloss for the
metalanguage-leveluse of negation;we have observeda numberof cases where
a speakeruses metalinguisticnegationnot strictlyto DENYS (or to call S false)
but rather,morebroadly,to REJECT S, or its implicata,or the way it was uttered.
As remarkedin fn. 13, Dummett(328-30) is on the righttrackin characterizing
this use of negation as 'a means of expressingan unwillingnessto assert "A" '
without necessarily constituting a willingness to deny 'A'. However, Dum-
mett's neo-Fregeanrepresentations,utilizingscopal distinctionsto account for
the differencebetween the two ways of takingnegation,may not be sufficiently
general; '-(not A)' is unobjectionablefor descriptive (propositional)negation,
but it is not clear that a representationlike 'Not(-A)' can be interpretedco-
herently for all the cases of 17-23.
Some of the recent radicallymonoguisttheories of negation suffer from the
flaw noted by Galeandby Geach:the failureto distinguishnegationfromfalsity,
and to recognize that to call a statementfalse is to say something(on a meta-
linguistic level) ABOUT that statement, but to apply (descriptive) negation to
a proposition is simply to form another proposition which may itself be true
or false. Thus Allwood, in his seminalunivocal analysis of negation, remarks
(43-5),
'We have in all cases taken negationto be the same basic semanticoperation,indicatingthat
a certainstate of affairsis not a fact. We have taken negationto have exactly the properties
of logical negation:always giving the predicationit operates on an opposite truth value ...
Negationhas the same basic functionas falsity. To negatea certainstatementor to say of the
same statementthat it is false is logicallyto do the same thing, namelyto claim that the state
of affairsdescribedin the statementdoes not obtain.'
Allwood's identificationof negation and falsity is precisely the target of the
warningsissued by Gale and by Geach.
21
It shouldbe noted thatone luminary,Frege, consistentlytreatednegativesentences as simple
assertions of negative propositionsand explicitly warned against confusing a lexical form, i.e.
negation,witha speech-actfunction,i.e. denial(Frege 1919,as calledto my attentionby Jay Atlas).
154 LANGUAGE,VOLUME61, NUMBER 1 (1985)

Kempson's more sophisticated monoguist treatmentof negation also iden-


tifies descriptivenegationin naturallanguagewith 'the falsity operatorof logic'
(1975:95);but from the context of this identification, she may be referring
ellipticallyto the propositionaloperatorwhose semantics correspondsto fals-
ity, i.e. a negative expression OFthe languageratherthan a negative comment
ABOUTit. She goes on to present and challengea variety of presuppositionalist
views of ambiguousnegation in which external or denial negation is taken as
a semantic operator. Since I agree with Kempson that her 'denial' negation
cannot be a semanticoperator,and is indeed 'one of the uses to which negative
sentences could be put' (99), I do not wish to rebut the gist of her account.
But she takes this correct observation as a license to ignore those cases of
'denial' negation whose behavior does not naturallyfall within the proper
bounds for logical negation-or to subsume them within the general category
of propositionalnegation, as Allwood does. Yet as we have seen, no single
logical notion of negation as a truth-functionaloperatorcan collect all natural
languageuses of negation.
While Kempson concedes that 'marked(contrastive stress) interpretations
of negative sentences' may generallyfunction as denials, she argues that 'this
correspondence ... does not carry over to compound sentences' of the type
illustratedby the citationfrom Strawsonmentionedin fn. 1 and repeatedhere:
(61) Neither Aristoteliannor Russellianrules give the exact logic of any
expression of ordinarylanguage;for ordinarylanguagehas no exact
logic.
But, in general, Kempson's citations of marked, 'presupposition'-canceling
negation(cf. Kempson 1975:68,78, 86-7)-as Kiefer(1977:252-3)points out-
'can only be conceived of as answers to a previous utterance.' An example
which Kiefer cites is this:
(62) Edward didn't regret that Margarethad failed because he knew it
wasn't true.
Kiefer's formulation, needless to add, is totally consistent with the metalin-
guistic approachto markednegation.
The most sophisticated,and probablythe most radical,of the contemporary
monoguistson negationis Atlas. Whilehis positionhas shiftedperceptiblyover
the years (from 1974 through 1977, 1979 [cf. also Atlas & Levinson] to 1980,
1981), as he has considered a progressively wider range of data, he has con-
tinued to maintainthat negationis ambiguousneither in scope nor in meaning
(cf. ?1.2 above), even when that position has pushed him into the somber
conclusion that no set-theoreticalsemantictheory can do justice to negation-
and hence to naturallanguage in general. Atlas concludes (1981:127),on the
basis of the kind of data presented in ?2 above, that
'the rangeof interpretationincludesstatementsthatare internalnegations,externalnegations,
and metalinguisticpredications.Not-sentences are semanticallyless specified, and theoreti-
cally more complex, than the traditionin logical theory has heretoforerecognized.'
It should be clear that I share Atlas' misgivingsabout logical theories that
either ignore metalinguisticuses of negation, or take them as a subcase of a
METALINGUISTICNEGATIONAND PRAGMATICAMBIGUITY 155

special semantic external negation operator;but I cannot agree that the ap-
propriatesolution lies in placing all our negative eggs into one 'radicallyun-
derspecified'basket. To put it anotherway, the evidence in ?2 does not support
Atlas' radicalmove of throwingout the model-theoreticbabywith the ambiguist
bathwater.
Ambiguisttreatmentsof negationare not entirelyabsent from modernlogic.
To capture the behavior of external negation, Bergmann1977, 1981 differen-
tiates the unmarkedauxiliarynegation 'not A' from a formalnegative operator
which affects, not truthvalue per se, but 'anomaly'value. WithinBergmann's
'two-dimensionallogic', the truth/falsityaxis intersectsthe anomaly/non-anom-
aly axis, producingfour distinct possible assignments. The ordinaryinternal
negation of A will be true just in case the correspondingexternal negation is
true and A is non-anomalous.
As Atlas notes, however, Bergmann'ssystem inherits empiricaland theo-
retical problems from previous ambiguisttheories, in additionto some which
are created by the innovationsin her account. Double negationno longerholds
for internal negation; furthermore,on Bergmann'sprojection rules, a condi-
tional like the following comes out true but anomalous:
(63) If there's a king of France, then he's bald.
Yet 'intuitively there is no linguistic anomaly in this sentence at all' (Atlas
1981:126-7).
For our purposes, an even more fundamentalflaw exists in Bergmann'sac-
count of negation:there is no obvious way for the 'anomaly'treatmentto extend
from negative statements involving sortal incorrectness (Bergmann1977) or
presuppositionfailure-as in the classic king of France cases-to those in-
volving conversationalimplicata,grammar,style or register, phonetics etc. It
is these cases which most clearly demand a metalinguistictreatmentoutside
the bounds of logical semantics (one- or two-dimensional).22
As we have already seen, similarproblems arise in an account which is in
some ways rathercongenial to Bergmann's.Karttunen& Peters (1979:47)cor-
rectly describe their 'contradictionnegation' as having 'a special function in

22
It is worthnotingthatthe sortalincorrectness(a.k.a. selectionalviolation)examplesdiscussed
in Bergmann1977constitute much strongercandidatesfor presuppositionalityand semanticex-
ternal negation than the referentialcases (e.g. the non-existentking of France) on which most
philosophersand linguistshave concentratedtheirfirepower.It is more compelling-although, as
Bergmannand others have pointedout, still not necessary-to diagnosesentences like (a) and (b)
as sufferingfrom a terminaltruth-valuegap than it is to performthe equivalentdiagnosisfor the
king of France examples:
(a) The theory of relativityis (is not) blue.
(b) Socrates is (is not) a primenumber.
The king of Franceis the kind of thing that can be bald (exx. 1-2 are each true in some possible
worlds), but the theory of relativityis simplynot the kind of thingthat can ever-in any possible
world-be blue. Significantly,as Bergmann(1977:65)notes, the diagnostictest of ?2.4 works as
expectedin the sortalincorrectnesscases, rulingout the external(i.e. metalinguistic)interpretation:
(c) The theory of relativityis {ninterested in classical music.
[#uninterestedJ
156 LANGUAGE,VOLUME61, NUMBER 1 (1985)

discourse' of contradicting'somethingthat the addresseehasjust said, implied,


or implicitly accepted'. This may be a necessary condition for a negation to
be functioningmetalinguistically,but it is not sufficient:many descriptive ne-
gations could be characterizedin the same terms (cf. Atlas 1980).Furthermore,
K&P stipulatethat
'contradictionnegationdiffers semanticallyfrom ordinarynegationonly by virtue of having
a broadertarget ... [it] pertainsto the total meaningof its target sentence, ignoringthe dis-
tinctionbetween truthconditionsand conventionalimplicatures.
This fatally overlooks just how broad the target of marked (metalinguistic)
negation can be.
One additionalcontemporaryaccount of negation,more neo-monoguistthan
neo-ambiguist,is worth mentioninghere. Lehrer& Lehrerdistinguishtwo rival
interpretationsof the relationbetween scalaroperatorslike good and excellent:
the 'hyponymy' interpretation,on which good is a superordinateterm for the
category containingexcellent, and the 'incompatible'interpretation,on which
the predicates good and excellent are mutuallyinconsistent. They point out
that 64a seems to supportthe former analysis, and 64b the latter (cf. the dis-
cussion of 34a-b in ?2.3 above):
(64) a. This wine is good-it's even excellent. (L&L 15)
b. This wine is not good, it's excellent. (L&L 14)
They opt for the hyponymy interpretation,based largely on the acceptability
of 65-a constructionwhich excludes 'true incompatibles',as seen in 66:
(65) That wine is not only good; it's excellent. (L&L 16)
(66) a. *That's not only a cat, it's a dog. (L&L 17)
b. That's not only a car, it's a Cadillac. (L&L 18)
I agree with the Lehrers' conclusion that excellent is a hyponym of good (cf.
Horn 1972, 1973, Gazdar 1979a,b);but I cannot accept their implicationthat
the negative predicationnot good in the first clause of 64b is to be regarded
as somehow ellipticalfor not only good as in 65. Given the scalar natureof the
relation between good and excellent-i.e., that X is excellent unilaterally entails
X is good-64b and 65 will in fact convey the same information;the same point
was made in ?2.4 in connection with the examples of 42. But ONLYthose in-
stances of metalinguisticnegation with an upper-boundingconversationalim-
plicatumarisingfrom the maxim of Quantitywill share this characteristic.
Thus there is no way to extend L & L's ellipticalanalysis of 64b to conven-
tional implicaturecases like 12, phonetic cases like 18a, morphologicalcases
like 18b, or stylistic or connotative cases like those in 20-21:
(67) a. ??I didn'tjust manageto trap two mongeese-I managedto trap
two mongooses. (#: 18b)
b. ??For a pessimist like him, the glass isn't only half full-it's half
empty. ($- 21d)
c. He's not only meeting a woman this evening-he's meeting his
wife. (OK, but 4 22")
Even among those cases which do involve cancellation of a quantity-based
implicatum,the syntax may rendera L & L-style paraphraseawkwardor im-
METALINGUISTIC NEGATION AND PRAGMATIC AMBIGUITY 157

possible:
(67') ?*Maggieisn't just either patrioticor quixotic-she's both. (cf. 37a)
L & L correctly characterizethe 'more than good' readingof the negationin
64b as requiringthat 'the intonationcontour ... remainhighinsteadof dropping,
signaling a clarification to follow'-but this same characterizationapplies
across the board to ALLinstances of metalinguisticnegation, whether or not
they are paraphrasablein the mannerof 65. In short, takingmetalinguisticnot
to stand for not only is as inadequateas taking it to representnot true.
4.2. METALINGUISTIC NEGATION AND 'NIGATION MITALINGUISTIQUE'. AS
noted above, both the concept and labelof metalinguisticnegationare borrowed
from Ducrot 1972, 1973.For Ducrot 1972:37ff.,descriptivenegationconstitutes
a comment on facts, and preserves presuppositions.23Metalinguistic(or po-
lemic) negation comments on utterances and challenges presuppositions.
In Ducrot's system, presuppositions(presupposes) are distinguishedon the
one hand from assertions (poses) and on the other from rhetoricalimplicata
(sous-entendus). An intermediateformal language (which I shall call LD) is
introduced(Ducrot 1972, ?5) for representingstatementsof ordinarylanguage
in such a way as to allow presuppositionsand assertions to be distinguished
in the predicate calculus translationsof LD formulas. The notation XIY rep-
resents a 'predicativepair', where X and Y can be filled by atomic or complex
predicates. Any LD expression of the form XIY(ai, ..., an) will then correspond
to two predicate calculus expressions: one (the translation of X(al, ..., a,)) for
the presupposition, the other (the translation of Y(al, ..., an)) for the assertion.
Natural language operators (only, some) and negation are represented in LD
by 'copulative operations' which convert one predicative pair into another
(Ducrot 1972:147).Two such copulative operations and NEG (presupposition-
preserving descriptive negation) are REF (refutational,i.e. metalinguistic,ne-
gation).Theireffect is indicatedas follows (wherenon-bold-faceNEGeventually
translates into predicate calculus '-' and ET into '&' or 'A'):
(68) a. NEG(XIY) = X NEG Y
b. REF(XIY)= - I NEG (ET (X,Y))
It will be immediatelynoted that the distinctionbetween 68a-b directly(mu-
tatis mutandis)prefiguresthat between ordinaryand 'contradiction'negation
in the work of Karttunen& Peters, discussed in ?1.3. The markednegationof

23
It is often difficult to determine just how a given expression may be (descriptively) negated;
e.g.,
(a) John, too, is coming to the party.
(b) Even John passed the exam.
However, Ducrot suggests that we have an intuitive sense of what the expression presupposes,
and this guides us to discover what constitutes its descriptive, presupposition-preserving negation.
Thus the descriptive negations of (c)-(d) can be given in (c')-(d') respectively (Ducrot 1972:105):
(c) We have finally arrived.
(c') We haven't arrived yet.
(d) For a Frenchman, he knows a lot of logic.
(d') Even for a Frenchman, he doesn't know much logic.
158 LANGUAGE, VOLUME 61, NUMBER 1 (1985)

68b, as in 15b, brings the presupposed (conventionally implicated) material


within the logical scope of negation;but ordinarydescriptive negation, in 68a
as in 15a, respects presupposed (conventionally implicated) materialby ac-
cordingit a kind of logical transparency.(Note that the presuppositionalcom-
ponent in the output of 68b is empty, just as the conventional implicature
component of 15b is vacuous.)
But we have already seen that this scopal distinctiondoes not generalizeto
the entire range of possible applicationsof metalinguisticnegations discussed
in ?2; in particular, such foci of negation as morphological, syntactic, and
phonetic form, conversationalimplicature,register, and connotative meaning
are not part of logical form, and cannot be readily plugged into the format of
68b. Ducrot does acknowledge a 'rhetorical'function of markednegation, to
deny the 'sous-entendus' associated with a given utterance(correspondingto
Grice's conversational implicata);however, his representationsand account
of 'la negation metalinguistique'do not do justice to the protean character
of metalinguisticnegation in French (cf. 19) or English.
Nevertheless, the account of metalinguistic(a.k.a. polemic)negationoffered
in various works by Ducrot and his colleagues (Ducrot 1972:37ff., 1973:124-
5, Anscombre & Ducrot 1977)is certainly helpful and quite suggestive. Thus
Ducrot correctly observes that (as noted above, followingGrice and Dummett;
cf. ?2.2 and fn. 13)the negationattachedto conditionalstends to be interpreted
only as a metalinguisticdevice indicatingthe speaker's unwillingnessto assert
the conditional. Elsewhere, Ducrot points out that metalinguisticor polemic
negation corresponds to a special negative speech act-a way of rebuttinga
previously uttered affirmative.
In an empiricalstudy, Heldnerexpandson the role of Ducrot's metalinguistic
negation and its interactionwith scalar predications(cf. Ducrot & Barbault's
essay in Ducrot 1973). A sample citation involving metalinguisticnegation is
(69) Jules ne chante pas bien, il chante comme un dieu.
'Jules doesn't sing well, he sings like a god.'
Here 'the speakermakes it clear that bien must be replacedby a more adequate
term'-one not necessarily (as with descriptive negation) below bien on the
relevant scale, but possibly higher or on a different scale entirely (Heldner,
92; cf. the Appendix, below, for related discussion).
As Heldner (65) points out, Ducrot and his colleagues originallytook the
descriptive/metalinguisticnegation distinction to be morphologicallyneutral-
ized in French; but more recent work has suggested a candidate for an un-
ambiguous sign of the latter. For Gross 1977, the use of non (or non pas),
immediately preceding the negated item, can only be interpreted 'contras-
tively'-where Gross's 'contrastive'negationcorrespondsto the metalinguistic
or polemic negation of Ducrot and his colleagues. (Anscombre& Ducrot in-
dependentlycite non as an unambiguouslypolemicnegation.)Thusthe negation
in 70a MAY be interpretedcontrastively, but that in 70b MUST be (Gross, 47):
(70) a. Max n'a pas abattu un if, mais (il a abattu) ce pin.
'Max didn't fell a yew, but (he felled) this pine.'
METALINGUISTICNEGATIONAND PRAGMATICAMBIGUITY 159

b. Max a abattu non pas un if, mais (*il a abattu) ce pin.


'Max felled not a yew, but (*he felled) this pine.'
(Note that reduction in the mais clause is obligatory in 70b; this is discussed
in the Appendix.)
Gross (51 ff.) constructs an argument for distinguishing contrastive from
ordinary negation, based on the distribution of partitive de + article vs. simple
de. He judges 71a to be necessarily contrastive, understood with a continuation
(... il boit autre chose); but 71b is understood non-contrastively:
(71) a. Max ne boit pas du vin ''Max doesn't drink
'Max doesn't drinkwine
wine.'
b. Max ne boit pas de vin
Gross finds that non (pas), as expected, occurs only with de + article:
(72) a. Max a bu du vin, non (pas) {de l'eau / *d'eau}.
'Max drank wine, not water.'
b. Max a bu non (pas) {dul*de} vin, mais de l'eau.
'Max drank not wine, but water.'
Clefts, too, force the contrastive reading on negation, and hence demand the
article:
(73) Ce n'est pas {dul*de} vin qu'il boit, mais de l'eau.
'It isn't wine that he drinks, but water.'
If, as is reasonable, we take the use of de without the article to constitute
a classic negative polarity item in French (cf. Gaatone 1971, Horn 1978a,b),
then Gross's correlation of de + article with contrastive (i.e. metalinguistic or
polemic) negation will define a diagnostic for French parallel to the observation
for English (Karttunen & Peters 1979, Linebarger 1981; cf. ?1.3 above) that
external or contradiction negation-and, by extension, the generalized meta-
linguistic operator (cf. ?2.1)-fails to trigger negative polarity items. But the
evidence is a bit murkier than Gross intimates. For Heldner (77), both 74a and
74b are acceptable in isolation:
(74) a. Je ne bois pas du vin, (*mais) je bois de la grenadine.
b. Je ne bois pas de vin, mais je bois de la grenadine.
The former is interpreted as specific in time and space (= 'I am not drinking
wine, I'm drinking grenadine'); the latter is taken as habitual (= 'I don't drink
wine, but I drink grenadine'). Crucially, however, Heldner agrees with Gross's
findings as to the unacceptability of the polarity item in the unambiguously
metalinguistic negation of 72b.24
4.3. WILSONON NEGATION.The English-language account of negation bear-
ing the greatest kinship to the approach taken in ?2 is probably that of Wilson.
She takes as her primary data a wide variety of uses of negation, many derived

24 As carefullydocumentedby Danell (1974;425),much more is going on here than meets the


eye. The interplayof factors determiningthe distributionof pas du X (or pas un X) vs. pas de X
is extremely complex, hingingon such variablesas the scope of negation, the modality of the
sentence, the meaningof the verb, the natureof the complement(s),and the 'degreeof existence'
of the object focused by negation.
160 LANGUAGE,VOLUME61, NUMBER 1 (1985)

from Grice, which are not reducible to garden-varietydescriptive negation,


includingthe following (cf. Wilson, 149 ff.):
(75) a. I'm not happy: I'm ecstatic.
b. The next Prime Ministerwon't be Heath: it will be Heath or Wil-
son. (= 49 above)
c. I don't love Johnny: I love Johnny or Billy.
Althoughthe passage that follows (150), inspiredby examples like those in 75,
does exhibit the errorof identifyingordinarydescriptivenegationwith falsity,
as decried by Geach (cf. ?4.1), we are given a clear descriptionof why natural
language negation cannot always be reduced to the familiarone-place logical
connective:
'To assert that not-p (or to deny that p) cannot be the same thingas to assert thatp is false.
It may also be to assert that p is inadequateto the facts without necessarilybeing false: it
may be too weak, or too strong,or misleading... Once negationandfalsity are distinguished,
semanticstatementsof entailmentand contradictioncould be made in terms of falsity, while
the treatmentof negationcould include, but go beyond, relationsof falsity alone.'
Given the existence of examples like 75, there must be non-truth-functional
aspects to the interpretationof (at least some uses of) negation-instances in
which the value of not-p cannot be simply a functionof the value of p. In these
examples, we see that the falsity of p is a sufficientbut not a necessary reason
for asserting not-p: given that utteringp might suggest q, and that one does
not wish to suggest q, one might say 'not-p' (Wilson, 151).25

25
Kempson and her students, in recent publishedand unpublishedwork (cf. Kempson 1975,
1982, Cormack 1980, Burton-Roberts1984)have drawn a rather differentconclusion from the
existence of examplesof the Grice-Wilsonvariety. Considernegationslike 75a, or the following:
(a) Justindidn'tpaint three squares,he paintedfour.
Cormack(p. 6) pointsout thatthese appearparadoxical:'if Justinpaintedfoursquares,he certainly
paintedthree;if someoneis ecstatic they are certainlyhappy,and so on.' Furthermore,as Burton-
Roberts observes, (b) is apparentlyparadoxical(relevantto standardmodal systems) and yet is
acceptable:
(b) It's not possible that mammalssuckle their young, you ignoramus,it's downrightnec-
essary.
Note that incorporationis impossiblehere, as the diagnostictest in ?2.4 predicts:
(b') *It's impossiblethat mammalssuckle their young, you ignoramus,it's downrightnec-
essary.
We have alreadyconsideredandrejectedLehrer& Lehrer'sellipticalapproachto Cormackand
Burton-Roberts's'paradoxicalnegations'(cf. ?4.1). Burton-Robertsopts for a differentanalysis,
one in which the weak scalar element (possible, three, happy etc.) is taken as semanticallyam-
biguousbetweenthe 'one-sided'reading,whichis lower-boundedonly, andthe 'two-sided'reading,
which is lower- and upper-bounded:
'As Cormackpoints out, the alternativeto this is to invoke a special (denial, quotational)
negationto handlethe phenomenon(an alternativethat she rejects in favor of treatingimpli-
catures semantically).'
But as arguedby Horn 1984, this move by the membersof what I call the London School of
Parsimony-in which a strictlymonoguistline on negationis offset by a radicallyambiguistline
on scalarpredications-is not compelling.Note that the London School frameworkgeneratesan
infinitudeof logical ambiguities:one for each weak or intermediatescalar value, includingevery
METALINGUISTIC NEGATION AND PRAGMATIC AMBIGUITY 161

What Wilson does not make clear is just how the fact that some instances
of not-p count as 'refusals to assert' p is to be related to the fact that OTHER
instances of not-p do contain negationas an object-languageconnective, trans-
lating into logical form as '-'. What is lacking here is precisely a full char-
acterizationof the distinctionbetween negationas a one-place truth-functional
connective (NOT equivalent to falsity, for the reasons noted by Geach) and
negation as a metalinguisticobjection to some aspect of a previous utterance.
In particular,just as not all uses of metalinguisticnegationcan be analysed
as semantic external negation-or as negationoutside the scope of a semantic
operatorTRUE-it is also the case (althoughon a subtlerlevel) that not all the
cases explored in ?2 can be taken as 'refusals to assert' a given proposition
(or sentence; Wilson is not entirely clear on just what p stands for in the pas-
sages cited above). Her characterizationcollects those instances where ne-
gation attaches to conversational implicata, along with those involving con-
ventional implicata or presuppositions (notions whose utility Wilson
challenges); but it does not directly generalize to examples like 18a-b or 19,
where the objection is not to the ASSERTIONof a given proposition(much less
to the truthof that proposition),but ratherto the way that the propositionwas
reified into a sentence, or the way that the sentence was uttered. The use of
negation to signal that a speaker finds a given proposition unassertable(cf.
Grice 1967, Dummett 1973, Ducrot 1973, Grim 1981, as well as Wilson 1975)
is appropriatelymore inclusive than the external negation operators of the
logical ambiguists(Karttunen& Peters 1979, Bergmann1981, and Linebarger
1981), but is itself a proper subcase of the generalized use of negation as a
metalinguisticoperator.
Ironically, it is Wilson herself who cites and attacks two alternativeviews
of marked negative statements-which, while not fully fleshed out, more
closely anticipate the notion of metalinguisticnegation than anything in her
own work or in the dissections of negative sentences at the hands of other
logicians, philosophers, and linguists. The relevant excerpts, from Fillmore
1969 and Kiparsky & Kiparsky 1970, emanate from the heady period imme-
diately following the discovery by generative linguists of those great presup-
positional vistas and swamps; as is typical of that period, they combine keen

cardinal number. Having argued against just such ambiguist analyses (those of Aristotle's De In-
terpretatione, of Hamilton 1860, and of Smith 1969), I remain reluctant to abandon the view set out
in Horn 1972, 1973 (cf. also Grice 1967, Gazdar 1979a,b), according to which scalar operators are
semantically unambiguous, but build in a potential pragmatic ambiguity, based on whether the
context induces a generalized quantity-based implicature (cf. ?2.3 above). My reluctance is rein-
forced on the one hand by the demonstration (Horn 1984) that privative ambiguity cannot simply
be argued away-a step which represents a cornerstone in the London School's approach-and
on the other hand by the arguments in the present paper. I have tried to show that a pragmatic
ambiguity can be motivated for negation, not only in the scalar cases under discussion here, but
in a wide range of examples for which the considerations specified by Cormack, Burton-Roberts,
and Kempson are irrelevant. The 'alternative' rejected by Cormack in the passage cited above in
fact offers the most general and most elegant account of 'paradoxical' and related uses of negation,
while preserving the Gricean line on scalar 'ambiguities'.
162 LANGUAGE,VOLUME61, NUMBER 1 (1985)

insight with a certain lack of rigor and precision:


'Uses of the verb chase presupposethat the entity identifiedas the direct object is moving
fast. Uses of the verb escape presupposethat the entity identifiedby the subjectnoun-phrase
was containedsomewhereby force previousto the time of focus. These presuppositions,as
expected, are unaffectedby sentence negation:
rchased
(58) The dog didn't chase the cat.

(59) He escaped from the tower.


tdidn'tt escapeJ
It seems to me that sentences like 60 and 61 are partlycommentson the appropriatenessof
the words chase and escape for the situationsbeingdescribed.These are sentencesthatwould
most naturallybe used in contexts in which the word chase or escape hadjust been uttered:
(60) I didn't "chase" the thief; as it happened,he couldn't get his car started.
(61) I didn't "escape" from the prison;they released me.' (Fillmore,381-2)
'If you want to deny a presupposition,you must do it explicitly:
Mary didn't CLEAN the room; it wasn't dirty.
Abe didn't REGRET that he had forgotten;he had remembered.
The secondclausecasts the negativeof the firstintoa differentlevel;it's not the straightforward
denial of an event or situation, but ratherthe denial of the appropriatenessof the word in
question [in small capitals above]. Such negations sound best with the inappropriateword
stressed.' (Kiparsky& Kiparsky,351)

These passages are quoted by Wilson (84) in the course of her blistering
attack on all extant presuppositionalisttheories, includingthose of Fillmore
and the Kiparskys. Her objections to the views illustratedhere have more to
do, I think, with her skepticismabout the viabilityof semantic(andpragmatic)
notions of presuppositionthan with the metalinguisticline on so-called 'ex-
ternal'negation;she also (quiteproperly)attackstwo of the weakercandidates
for presuppositionalstatus, Fillmore's bachelor and the Kiparsky'sclean.
In assumingthat markednegationcan only be used to deny 'presuppositions',
Wilson may or may not be faithfulto what Fillmoreand the Kiparskyshad in
mind. In any case, I have arguedfor a differentaccount of the metalinguistic
use of negation-one which strikesme as entirelycompatiblewith more recent
theories of presuppositionalphenomena,includingthe context-cancelablepre-
suppositionsof Gazdar 1979a,band the orderedentailmentsof Wilson & Sper-
ber 1979.
Note, however, that both the excerpts above, from Fillmore and from Ki-
parsky & Kiparsky, specifically allude not only to the fact that metalinguistic
negationis used to object to an earlierutteranceas inappropriate-ratherthan
to judge a proposition previously expressed as false-but also that it occurs
(as does any metalinguisticoperator)on 'a differentlevel', i.e. as a predication
ABOUTthe object language rather than WITHINit. Moreover, while Wilson cor-
rectly recognizes that we cannot define all instances of external or presup-
position-cancelingnegation as 'denials of appropriateness',as the Kiparskys
seem to believe (Wilson, 84-5), their notion does provide a closer approxi-
mation to the general phenomenon of metalinguisticnegation than Wilson's
own view of markednegation as a refusal to assert a given proposition.
METALINGUISTIC NEGATION AND PRAGMATIC AMBIGUITY 163

CONCLUDING REMARKS
5. I have argued here that marked negation is a reflex of an extended meta-
linguistic use of the negative operator in English and other languages. Negative
morphemes generally allow (in principle) both descriptive and metalinguistic
functions; and the context often serves-as is usual with pragmatic ambiguity-
to select one of these uses as the more plausible or salient.
In some cases, a particular morphological realization of negation may in fact
force or exclude a particular understanding. (Again, this is a frequent occur-
rence in the realm of pragmatic ambiguity: cf. Zwicky & Sadock 1975, Horn
& Bayer 1984.) Thus, as we saw in ?4.2, Fr. non (pas), placed immediately
before the item in the focus of negation, must be interpreted metalinguistically.
However, Korean may offer an instance of one morphological negation which
is unambiguously descriptive, as against another which may be interpreted
either descriptively or metalinguistically. The two constructions in question
are the 'short form' an(i), placed before the verb, and the 'long form' an(i)
hada (literally 'not do'), placed after the verb stem suffixed by the nominalizer
-cil-ji. Thus, corresponding to a basic affirmative sentence like 76a, we have
the short-form negative 76b and the long-form negative 76c:
(76) a. Mica ka canta 'Mica sleeps.'
b. Mica ka an(i) canta 'Mica does not sleep.'
c. Mica ka ca-ci ani hantaj
The issue is whether 76b-c, and other frames in which these constructions
function, differ in meaning or use-and, if so, how. Kuno (1980:162-3) finds
that the two constructions are either interchangeable or differ only in emphasis;
others find a subtle difference, in that the former is a 'verb negation' and the
latter a 'sentence negation'. This distinction, as Kuno explicates it, is remi-
niscent of (but not identical to) the internal/external dichotomy we have dis-
cussed.
Other researchers have taken different, often conflicting (if not internally
inconsistent) positions. Choi 1983 considers several possibilities raised in these
studies, and concludes that the closest match for the two Korean constructions
within the Western literature on negation is Aristotle's contrary vs. contradic-
tory negation (cf. Horn 1972, 1978a). In any case, Choi's data indicate that the
preverbal short form is always used descriptively, while the long form is not
restricted to metalinguistic uses-and indeed often 'fills in' for the distribu-
tionally defective short form when the syntax demands. If the choice to use
long-form negation is often interpreted metalinguistically in those contexts
which would have permitted the short form, this interpretive tendency may
well be grounded in the pragmatic 'least effort' factors explored insightfully
by McCawley 1978.
An additional factor relevant to the Korean case is the restricted scope often
associated with the unmarked negative form in verb-final languages. Kuno
notes that the scope of the Japanese negative -na-i is generally limited to the
immediately preceding verb (although quantifiers can 'escape' this restriction,
164 LANGUAGE,VOLUME61, NUMBER 1 (1985)

and come within the scope of a non-contiguousnegation).The normalTurkish


negation-mA-is similarlyrestricted,with the suppletiveperiphrasticformdegil
showing up in contrastive and other contexts.26
Whatever the details of the behavior of negation in specific verb-finallan-
guages (or recalcitrantlanguagesof othertypologies),the over-allpatternseems
confirmed:no languagecontains two negativeoperatorscorrespondingexactly
to descriptive and markednegation, whether the latter is characterizedas an
external semantic operator or (as urged here) a metalinguisticuse of basic
negation. At the same time, every languagecontains at least one negativemor-
pheme which can be used either descriptively(to form a negative proposition)
or metalinguistically(to object to a previous utterance).
One issue remaining is the directionality of the relationshipbetween de-
scriptive and metalinguisticuses of negation:which use is primaryand which
derivative? Or do both uses branchoff separatelyfrom some more basic, un-
differentiatednotion? I have little to contributeon this point, except to note
that the connection is explainablein either direction. Some evidence from ac-
quisitiontends to indicatethat, at least in ontogenetic development, the meta-
linguisticuse is prior;the truth-functionaluse is a later specialization.Fraiberg
(1959:62-6) has written eloquently of the power and autonomy which young
children associate with their first uses of the magic No; negative utterances

26
Cf. McGloin 1982for a discussion of other considerationsrelevant to the interpretationof
Japanese negative sentences, in particularthe interactionof contrastivenegation and the topic
markerwa. She cites this three-waydistinctionin English(Horn1978a:137,adaptedfromJespersen
1924):
(a) She isn't pretty. (= less than pretty)
(a') She isn't (ust) pretty, she is beautiful.(= more than pretty)
(a") She isn't pretty, but she is intelligent.(= other than pretty)
McGloinnotes that the unmarked(descriptive)'less than' readingcan occur whetheror not the
scalar element is suffixedby wa. Thus both (b) and (c) may be read as conveying that it is 'less
than', i.e. cooler than, hot:
(b) Atsuku na-i
hot NEG-PRES
(c) Atsuku wa na-i 'It isn't hot.'
hot TOP NEG-PRES

But only (b) can be given the non-scalar'otherthan'interpretation(e.g. 'It's not hot but it is dirty.')
By contrast, McGloinreportsthat neither(b) nor (c) can be read in the mannerof Englishmeta-
linguistic scalar negations(e.g. (a') above, 17, 34b, 64b, 75), where the negationfocuses on the
upper-bounding implicatureassociatedwithweakscalarpredications.In orderto get sucha reading,
a periphrasticform must be employed:
(d) Atsui dokoroka nietagit-te i-ru yo.
hot far.from boiling be-PRES
'It's far from being hot: it's boiling.'
(e) Atsui nante yuu mon ja na-i. Nietagit-te i-ru yo.
say
'It's not somethingyou can call hot. It's boiling.' (McGloin,57-8)
(Cf. also Davison 1978for a studyof some pragmaticfactorsbearingon the issue of negativescope
in verb-finallanguages.)
METALINGUISTIC NEGATION AND PRAGMATIC AMBIGUITY 165

may well accompanythe very behaviorbeing abjuredor denied, since negation


constitutes a 'politicalgesture', indeed a 'declarationof independence'for the
toddler. Such uses of negation, however we may analyse them in terms of the
adult language, are clearly not truthfunctions. However, for what it is worth,
Rumbaugh& Gill (1977:169-70)reportthat the chimpLana, havingbeen taught
the propositional,truth-functional(descriptive)use of negation as part of her
computer-based symbolic repertoire of 'Yerkish', spontaneously innovated
what can only be viewed as metalinguisticuses of the same negative operator.
Of course, even if we conclude that the generalizedmetalinguisticuse of ne-
gation as a sign of objection or refusal is learnedearlierthan its logical, truth-
functional use, it does not follow that this order of development should be
associated with any logical asymmetryin the account we give for negation(or,
analogously, for the other operators)in an idealized competence model of the
adult language.
I have maintainedin this paperthat conditionson truthmust be kept distinct
from conditions on assertability,and that more explanatoryburdenshould be
shifted from the former onto the latter. I have also arguedthat, while there is
indeed only one semantic negation operator in English and other languages,
the ordinarytruth-functionalinterpretationof this operatormotivates it for an
extended use as a general metalinguisticsign of rejectionor objection, leveled
against the choice of a particularobject-languageexpression or the mannerin
which that expression was overtly realized. I have tried to pinpoint some of
the linguisticcorrelatesassociatedwith metalinguisticanddescriptivenegation,
in supportof the view that all negative tokens can be assigned to one of these
two basic types.
In reply to the query posed by the title of Atlas 1981, 'Is not logical?', some
have answered 'yes' and others (includingAtlas himself) 'no'. I conclude that
the only full and complete answer must be 'sometimes', i.e. when it is func-
tioning descriptively ratherthan metalinguistically.Neither the monoguistnor
the ambiguistapproachto the data we have considered can deal successfully
with the unity and diversity of the phenomenonof metalinguisticnegation.
How then are we to representthe effect of the metalinguisticuse of negation
within a formaltheory of naturallanguagediscourse?27This is a good question,

27
One consequence of the proposal to eliminate so-called (semantic) 'external' negation is that
we are free to adopt whichever theory of descriptions (or of factive predicates) best fits the facts,
ignoring the role of negation as a 'presupposition-canceler'. The approach embodying a pragmatic
distinction between descriptive and metalinguistic uses of negation is neutral with respect to those
issues which divided Strawson from Russell, or Gazdar from Karttunen & Peters. We should not
be surprised, however, if the formal semantics of the resultant theory turns out fairly conservative.
When the function of negation as a plug for presuppositions and implicatures is removed from
logical semantics, then the motivation for semantic presuppositions, for truth-value gaps, and for
supernumerary non-bivalent truth values diminishes, if it doesn't disappear entirely. In contrast,
the conversationalist line (favored at times by Grice, Wilson, and Kempson) on explaining away
presuppositions-in particular, the definite description and factive cases-may prove ultimately
inadequate, as argued by Soames 1976 and by Kiefer (1977, ?4). The jury, despite the new evidence
it may have received, is evidently still out.
166 LANGUAGE,VOLUME61, NUMBER 1 (1985)

but one which cannot be directlyaddressedhere. The formulationof its answer


would, and I hope will, lead to a future paper, perhaps one which takes the
present work as its departurepoint.
APPENDIX
Anothercorrelateof the metalinguistic/descriptive split manifestedby naturallanguagenegation
is found in the distributionalcharacteristicsof concessive and contrastivebut conjunctions.We
have seen that metalinguisticuses of negationtend to occur in contrastiveenvironments,and the
Englishrepresentationpar excellence is but. One typical occurrenceof metalinguisticnegationis
in phrasesof the form not X but Y, functioningfor many purposesas a single constituent. Y here
is proposedmetalinguisticallyas the appropriatesubstitution(or, followingAnscombre& Ducrot,
RECTIFICATION) for X, on any groundswhatever;it is, as usual, irrelevantthatthe rejectedutterance
containingX may have expressed a true proposition.
We often find that not X but Ymay be acceptablein frameswhere not X is not, as in Ala, or-
even more clearly-in Alb:
f?not three children.
(Al) a. We have not three but four children.
not three childrenbut four.
*Not Lee
b
[ Not Lee but Kim wonthe ace.
A particularlystrikingexampleof a metalinguisticnegationwhich absolutelyrequiresits but rec-
tificationappearsin ElizabethStone's explanationof why it is that, for Jessie-the protagonistof
MarshaNorman'sPulitzer-winning play 'night,Mother-suicide countsas a positiveact expressing
not despairbut autonomy:
(A2) Not she chooses to DIE, but she CHOOSES
to die. (Ms., July 1983, 56)
Note that, like 39a in the text, A2 representsa graphemiccontradictionif the negationis taken
truth-functionally.
Rectificationof metalinguisticnegationcan be expressed in a variety of ways, as seen in the
alternateforms of A3a-b:
(A3) a. It isn't hot, but scalding.
b. It isn't hot-it's scalding.
c. #It isn't hot, but it's scalding.
As in examples discussed in the text, hot is objected to on the groundsthat the predicationit
yields, though true, is too weak. But the syntax of A3c forces an interpretationon which but
functionsas a true sententialconnective (ratherthan a rectification),and negationfunctionsonly
as an ordinarydescriptiveoperator.
The reason A3c is pragmaticallydeviant(as signaledby the crosshatch)is that-given the fact
that anythingscaldingis also (at least) hot-it is inconsistentto assertof anythingthatit is scalding
yet not hot (cf. Cormackon 'paradoxicalnegation',discussedin fn. 25). Similarly,the metalinguistic
interpretationof A4a (cf. ?2.3) disappearswith the unreducedsyntax of the but clause in A4b:
(A4) a.We don't have three children{b-w have four}
but four.
b. #We don't have three children,but we (do) have four.
When such sententialbut conjunctionsare acceptable(as descriptivenegations),they tend to be
interpretedas concessions, and assigneda characteristicintonationcontour:

(A5) a. We don't have three children,but we do have two.

b. It isn't hot, but it is warm(#scalding).

c. Negation isn't ambiguoussemantically,but it is pragmatically.


The acceptabilitycontrastin A5b, or that between A5a and A4b, intuitivelyhinges on just what
METALINGUISTIC NEGATION AND PRAGMATIC AMBIGUITY 167

can count as a concession. The appearance of supportive do in what must be taken as an emphatic
environment, and the heavy stress on the auxiliary, are additional linguistic correlates of the con-
cessive but clause.
My discussion of these two but constructions will lean heavily on the extremely insightful analysis
of the cross-linguistic counterparts of these sentences given by Anscombre & Ducrot. They begin
by pointing out that Spanish differentiates pero from sino, and German aber from sondern; but
French contains just one surface connective mais. However, it enters into two distinguishable
distributional patterns, corresponding to the PA (pero/aber) type and the SN (sino/sondern) type.
Then, in the construction
(A6) NEG-p SN q
the negative must be syntactically overt and unincorporated, and the entire sequence must come
from one speaker:
no es consciente, sino totalmente automdtico.
(A7) Sp.: Eso
*es inconsciente,
nicht bewusst,
Ger.: Das ist sondern ganz automatisch.
t *unbewusst,
'It is {not conscious / *unconscious} but (rather) totally automatic.'
Here q is presented as the motivation for denying p and, crucially, SN-but is compatible with
polemic (i.e. metalinguistic) negation.
While the typical use of A6 directly follows a previous speaker's assertion of p, this is not a
necessary condition on SN-but. Thus we can get the following, in both Spanish and German ver-
sions;
(A8) X: Pierre is nice.
Y: He's not just nice, SN quite generous.
Here the object of Y's denial (he's just nice) hasn't actually been asserted, but is inferable via the
'loi d'exhaustivite'-Ducrot's version (1972), independently arrived at, of Grice's maxim of Quan-
tity.
However, the construction
(A9) NEG-p PA q
necessarily involves descriptive use of negation (i.e. when a negative is present; unlike SN, PA is
not restricted to negative contexts). Here p and q must have the same 'argumentative orientation'
on a given scale, and p must be 'argumentatively superior' to q. (Anscombre & Ducrot's argu-
mentative scales, also expounded in Ducrot 1973, are similar to-but not identical with-the prag-
matic scales of Horn 1972, 1978b, and of Fauconnier 1975.) Thus we get A10 but not All:
(A10) Sp.: No es cierto, pero es probable.
Ger.: Das ist nicht sicher, aber das ist wahrscheinlich.
'It's not certain, PA it is probable.'
(All) Sp.: #No es probable, pero es cierto.
Ger.: #Das ist nicht wahrscheinlich, aber das ist sicher.
'It's not probable, PA it is certain.'
Though the SN vs. PA distinction is morphologically neutralized in French, Anscombre & Ducrot
point out that certain diagnostics can be used to distinguish the two corresponding forms of mais.
When mais = PA, we can add cependant, neanmoins, pourtant, en revanche, or par contre; when
mais = SN, we can add au contraire or (in familiar style) meme que, or we can use paratactic
syntax with no overt conjunction:
(A12) Ce n'est pas certain, maispA c'est pourtant probable. (= A10)
(A13) II n'est pa grand -il est tres grand.
maissN tres grand.
'He isn't tall, SN very tall'
When the SN-mais does occur overtly, however, its clause requires deletion (cf. 70 from Gross,
and the English A4):
(A 14) #11 n'est pas grand, mais il est tres grand.
168 LANGUAGE,VOLUME61, NUMBER 1 (1985)

As we noted for the parallelEnglishexamplesA3c and A4b, A14 is pragmaticallydevianton the


concessive (PA) readingof mais which forces the negationto be taken descriptively.
Given that mais clauses with seulement(like but clauses with only orjust) force the SN inter-
pretation,they are incompatiblewith lexicalized(incorporated)negation(cf. A7):
(A15) a. II n'est pas intelligent, mais seulement bucheur.
b. #I1 est inintelligent, mais seulement bucheur.
'He is ab.unintelligent j butjust a grind.'
[b. unintelligent'J
The failureof SN-butto co-occurwith incorporatednegationis reminiscentof the observation(cf.
Horn 1972and ?2.4 above) that metalinguisticnegationdoes not incorporateprefixallyin English;
cf. the examples in 35b above.
Since non (pas) can be read only as metalinguisticnegation(cf. Gross and ?4.2), it can occur
only in environmentswhich otherwisepermitSN (ratherthan forcingPA) readingsof mais:
(A16) a. II n'est pas francais mais il est beige.
'He isn't French, butpAhe is Belgian.'
b. II est non pas francais mais (*il est) belge.
'He is not FrenchbutsNBelgian.'
(A17) a. C'est non seulement vraisemblable, mais certain.
'It's not just likely butsNcertain.'
b. *C'est non pas certain mais reste possible.
'It's not certainbutsNremainspossible.'
The ungrammaticalsentences are ruledout becausethe metalinguisticnonpas negationforces the
SN reading,which the context excludes.
Finally, Anscombre& Ducrotshow that -p SN q constitutesa single speech act, while -p PA
q representstwo speech acts which may be associatedwith two separateinterlocutors.The meta-
linguisticnegationnon (pas) requiresthe rectificationof the offendingitem p by the affirmative
statementq withinthe same speech act, and is thus incompatiblewith PA-mais.
In general,Anscombre& Ducrot'sanalysiscarriesover remarkablywell to concessive (PA)and
contrastive(SN) but clauses in English. Exx. Ala-b, A2, A3a-b, and A4a all illustrateSN-but in
English, while A5 must be read as PA-but. Like the unacceptableSpanish,German,and French
sentences, the crosshatchedcases of A3c, A4b, andA5b are simultaneouslydisambiguatedin both
directions (by their syntax and/orcontext of utterance),and hence can be neither PA nor SN.
English,of course, is on the Frenchside of the isogloss, wherethe overt morphologicaldistinction
between PA and SN forms found in Spanishand Germanis neutralized.
The concessive PA examples are worth exploringa bit more closely. Ex. A5 exemplifiesthe
usual pattern:two scalartermsarejuxtaposedin the constructionNEG-p PA q, withp taken to be
a strongerelement than q on a given scale. In the clear cases, such scales can be defined by
unilateralentailment:four is strongerthan three because any simplepositive propositionwith the
scalarelementfour entailsthe correspondingpropositionwith three, but not vice versa. The scale
containingscalding, hot, and warm which is implicitlyevoked in A3 and A5b can be similarly
constructed.
In each case, the use of the weaker scalar term conversationallyimplicatesthat (for all the
speakerknows) no strongerterm on the same scale could be substitutedsalva veritate(cf. ?2.3).
Additionallinguisticcorrelatescan be found for the existence of these scales, as illustratedby
these constructions:
(A18) not only X but Y
not even X, {let alone / much less} Y
X, (or) indeed Y
at least X and possibly even Y
X if not (downright)Y
(where Y exceeds X on some relevantscale)
But the notion of scale which is relevant for these constructionsis wider than entailmentcan
accommodate.The entailmentcases are special instances of what is more broadlya pragmatic
METALINGUISTICNEGATIONAND PRAGMATICAMBIGUITY 169

relation, defined as much by knowledgeand beliefs about the world sharedby the speech partic-
ipants as it is by the languageitself (cf. Horn 1972,Ducrot 1972, 1973, Fauconnier1975).
Considerthese additionalexamplesof well-formedand ill-formedconcessive (PA) but conjunc-
tions:
(A19) a. I don't have my master'sdegree, but I DOhave my {bachelor's/#doctorate}.
I DID spend a few years there.
b. I wasn't born in L.A., but #I was born in New York.
(rather)in New York. [SN but]
c. Of course it isn't cotton-but it is cottony soft. [Cottonelle]
rich/presentable. 1
(A20) He isn't handsome,but he IS a Catholic/ a linguist.
#ugly/?#mean. J
In the well-formedconcessive examplesin A19, it is straightforward to constructa scale on which
the negated element outranksthe item being affirmed.In A20, however, the concessive pattern
expands to admita case in which the two elementsdo not standin an obvious scalarrelation,but
where they do occur as fellow membersof an implicitlyinvoked set of attributes.The examples
in A20 mightbe paraphrasedas, e.g.,
(A20') He isn't handsomeAND rich, but (at least) he is rich.
(In the same way, (a")in fn. 26 may be read concessively as 'She isn't pretty and intelligent,but
at least she's intelligent.')
The one case which stronglyresists acceptabilityis that where the affirmationof the latteritem,
q, is judged incompatiblewith the negationof the former,p, either (as in A4b or A5b) because it
is a STRONGERratherthan WEAKERitem on the same scale, or (as in A20) because it is just too
mind-bogglingto constructthe set of which the two items in questionfunctionas fellow members
(e.g. the set of attributescontaininghandsome and mean). Considerthis unlikely, but actually
attested, instance:
(A21) Tippingis not so common in Nepal. Tippingis not compulsorybut it is obligatory.
('Nepal travelcompanion',by S. D. Bista & Y. R. Satyal, cited in the New Yorker,
7/19/82)
Even here, we infer that the writers(if they were not totally confused)were assuminga scale on
which compulsoryoutranksobligatory-i.e. where anythingcompulsoryis automaticallyobliga-
tory, but not vice versa.
In fact, however, even the apparentlydeviant #-markedexamplescan be renderedacceptable
when ingenuitypermitsconstructionof the relevantpragmaticscale. Suppose that you have an-
nouncedthatyou are lookingfor peoplewiththreechildren(to fill out a questionnaire,for example,
or to offer aid and solace); then, if I assume that havingfour childrenqualifiesme almost as well
(or even better), I can nominatemyself by utteringthe suddenlyredeemedA4b.
Abbott 1972, citing some unpublishedobservationsof CharlesFillmore, discusses this set of
examples:
(A22) a. John was born not in Boston, but in Philadelphia.
b. #John was born in Philadelphia,but not in Boston.
c. (#)John wasn't born in Boston, but he was born in Philadelphia.
While A22a is good on what we've been callingthe SN reading,the syntax of A22b-c forces the
PA interpretation:the formerbecause its first clause lacks negation, and the latter because its
second clause is unreducedand containsan overt conjunction.As both Fillmoreand Abbottnote,
A22b suggests the (unsatisfiable)expectationthatJohncould have been bornboth in Philadelphia
and in Boston, while A22c seems to have 'an associatedassumptionthatthereis a scale connected
with places to be born in, and that Boston representsa more extreme point on that scale than
Philadelphia'(Abbott, 19). For me, one context which rendersA22c acceptableby commissioning
the constructionof just such a scale is the following:a castingdirectorfor a school play in a small
town in Iowa or Mississippi,needinga fifth-graderto portrayJFK in a forthcomingproduction,
is being convinced to settle for John.
If eitherthe non-focusedmaterialhe was bornor the conjunctionbutitself is deletedfromA22c,
170 LANGUAGE,VOLUME61, NUMBER 1 (1985)

we obtain the SN readingwhich, as in A22a, is acceptablewithoutany special context:


(A22') John wasn't born in Boston, {but/ he was born}in Philadelphia.
Crucially,auxiliarynegationallows both the PA reading(as in A22c) and the SN reading(as in
A22'), dependingon the syntax of the second clause. However, post-auxiliary('constituent')ne-
gation can only be taken metalinguistically(like non (pas) in French, which also immediately
precedesits focus or target),andis thus incompatiblewith the concessive PA reading.Thus, while
A22c is the PA version of A22', the latter's post-auxiliary-negated paraphraseA22a has no ac-
ceptable PA counterpart:
(A22")*Johnwas born not in Boston, but he was born in Philadelphia.
These Englishfacts parallelthe FrenchexamplesfromAnscombre& Ducrot, viz. A16-17 above.
As seen in the discussion of A4, the requisitepragmaticscale may force an inversion of the
ordinarysemantically-based(entailment-generated) scale involvingthe same elements. Thus too,
the #-marked version of A19a becomes acceptable if the speaker feels that the interlocutoris
looking, essentially,for someonewith a graduatedegree, ratherthan particularlya master's. Sim-
ilarly, A23 in isolation seems implausible,since it alludes to a scale on which being a private
outranksbeing a corporal.
(A23) #He isn't a private,but he is a corporal.
Yet just such a scale CANbe constructed,if the context is fleshed out in the rightway: the Colonel
has orderedthe Lieutenantto find a privateto blamethat last missionon. The Lieutenantreports
back to the Colonel:
(A23') I've found a soldier we can volunteerfor that mission, sir. He isn't a private, but he
is a corporal.Will he do?
Note that in this same context, the scalartermsalmost, barely,not even etc. reverse theirnormal
distribution,helpingto confirmthis ad-hoc inversionof the standardscale:
(A23")He's almost a private.
Inspection seems to indicatethat similarunusual(if not outlandish)contexts can be constructed
to redeem the unacceptablePA examplesfrom Anscombre& Ducrot, e.g. A1la-b above.
The English examples discussed in this Appendixare consistent with Anscombre& Ducrot's
thesis that the negation(optionally)figuringin the concessive PA-butconstructionsis necessarily
descriptive, while the negative requiredby the SN contexts may be either descriptiveor (more
frequently)metalinguistic.Thus the contrastbetween the SN and PA types of but constructions,
in languageslike Frenchand English-as well as (if more subtly than) in languageslike Spanish
and German-constitutes anotherdiagnosticfor metalinguisticvs. descriptiveuses of negation.28

REFERENCES
ABBOTT,BARBARA. 1972. The conjunction but. University of California, Berkeley, MS.
ALLWOOD, JENS.1977. Negation and the strength of presuppositions. Logic, pragmatics,
and grammar, ed. by Osten Dahl et al., 11-57. Goteborg: University of Goteborg,
Dept. of Linguistics.
ANSCOMBRE,JEAN-CLAUDE, and OSWALD DUCROT.1977. Deux mais en francais? Lingua
43.23-40.
ATLAS,JAY DAVID. 1974. Presupposition, ambiguity, and generality: A coda to the
Russell-Strawson debate on referring. Pomona, MS.
28
After this paper had been typeset, I discovered that the essential elements in Anscombre &
Ducrot's analysis of mais are given in substantially the same form (but without the supporting
distributional evidence) by Adolf Tobler (Vermischte Beitrage zur franz6sischen Grammatik III,
Leipzig: S. Hirzel, 1908, 93-4) and by J. Melander (Etude sur magis et les expressions adversatives
dans les langues romanes, Upsala: Almqvist & Wiksell, 1916, 1-4). Tobler and Melander distinguish
a modification or restriction (= PA) sense of mais (Tobler's 'einschrankender mais'), corresponding
to aber, from an exclusion (= SN) sense (Tobler's 'ersetzender mais'), corresponding to sondern,
with the latter reading restricted to contexts following negated clauses and triggering syntactic
reduction in the mais clause.
METALINGUISTIC
NEGATION AMBIGUITY
ANDPRAGMATIC 171

. 1977. Negation, ambiguity and presupposition. Linguistics & Philosophy


1.321-36.
-. 1979.How linguisticsmattersto philosophy:Presupposition,truth, and meaning.
In Oh & Dinneen, 265-81.
-. 1980. A note on a confusionof pragmaticand semanticaspects of negation. Lin-
guistics & Philosophy3.411-14.
-- . 1981.Is not logical?Proceedingsof the 11thInternationalSymposiumon Multiple-
Valued Logic, 124-8. New York: IEEE.
-- C. LEVINSON.1981. It-clefts, informativeness, and logical form. In
, and STEPHEN
Cole 1981:1-61.
BAR-LEV, ZEV,and ARTHURPALACAS.1980. Semantic command over pragmatic priority.
Lingua51.137-46.
BERGMANN, MERRIE. 1977. Logic and sortal incorrectness. Review of Metaphysics
31.61-79.
. 1981. Presuppositionand two-dimensionallogic. J. Phil. Logic 10.27-53.
BOCHVAR, D. A. 1939. On a three-valuedlogical calculus and its applicationto the
analysis of contradictions.MatematiceskijSbornik4.287-308.
G. LYCAN.1976. The myth of semantic presupposition.
BOER,STEVENE., and WILLIAM
Bloomington:IndianaUniversity LinguisticsClub.
BURTON-ROBERTS, NOEL. 1984. Modality and implicature. To appear in Linguistics &
Philosophy.
CHOI, SEUNGJA. 1983. Some aspects of negation in Korean. Yale University, MS.
COLE, PETER. 1978 (ed.) Syntax and semantics 9: Pragmatics. New York: Academic
Press.
. 1981(ed.) Radicalpragmatics.New York: Academic Press.
, and JERRYL. MORGAN.1975 (eds.) Syntax and semantics 3: Speech acts. New
York: Academic Press.
CORMACK,ANNABEL. 1980. Negation, ambiguity, and logical form. University of Lon-
don, MS.
DANELL, KARL JOHAN. 1974. La concurrence pas de vin-pas du vin. Studia Neophilo-
logica 46.409-25.
DAVISON, ALICE. 1978. Negative scope and rules of conversation: Evidence from an OV
language.In Cole 1978:23-45.
AUGUSTUS.1847. Formal logic. London: Taylor & Walton.
DEMORGAN,
DONNELLAN, KEITH. 1966. Reference and definite descriptions. PhilosophicalReview
75.281-304.
-- . 1978. Speakerreference, descriptionsand anaphora.In Cole 1978:47-68.
Du Bois, JOHNW. 1974. Syntax in mid-sentence. Berkeley studies in syntax and se-
mantics,Vol. I, pp. III:1-25. Berkeley:Dept. of LinguisticsandInstituteof Human
Learning,University of California.
DUCROT, OSWALD. 1972. Dire et ne pas dire. Paris: Hermann.
-. 1973. La preuve et le dire. Paris:Maison Mame.
DUMMETT,MICHAEL. 1973. Frege: Philosophy of language. London: Duckworth.
FAUCONNIER,GILLES. 1975. Polarity and the scale principle. CLS 11.188-99.
FILLMORE, CHARLESJ. 1969. Types of lexical information.Studies in syntax and se-
mantics, ed. by Ferenc Kiefer, 109-37. New York: HumanitiesPress. [Reprinted
in Steinberg& Jakobovits, 370-92.]
FRAIBERG, SONIA. 1959. The magic years. New York: Scribner.
FREGE, GOTTLOB. 1892. Uber Sinn und Bedeutung. Zeitschr. f. Philos. und philos. Kritik
NF 100.25-50. [Translatedas On sense and reference, in Translationsfrom the
philosophical writings of Gottlob Frege, ed. by Peter Geach & Max Black, 56-78.
Oxford: Blackwell, 1952.]
. 1919. Negation. Beitrage zur Philosophie des Deutschen Idealismus 1.143-57.
[Reprinted in Translations from the philosophical writings of Gottlob Frege, ed.
by Peter Geach & Max Black, 117-35. Oxford: Blackwell, 1952.]
172 LANGUAGE,VOLUME61, NUMBER 1 (1985)

GAATONE, DAVID.1971. Etude descriptive du systeme de la negation en francais con-


temporain. Geneva: Droz.
GALE,R. M. 1970. Negative statements. Am. Phil. Quart. 7.206-17.
GAZDAR, GERALD. 1977. Univocal 'or'. CLS book of squibs, ed. by Samuel E. Fox et
al., 44-5. Chicago: CLS.
. 1979a.Pragmatics.New York: Academic Press.
. 1979b.A solution to the projectionproblem.In Oh & Dinneen, 57-89.
T. 1972. Logic matters. Berkeley & Los Angeles: University of California
GEACH, PETER
Press.
GIv6N, TALMY. 1978. Negation in language:Pragmatics,function, ontology. In Cole
1978:69-112.
GRICE,H. P. 1967. Logic and conversation:The WilliamJames lectures. HarvardUni-
versity, MS.
. 1975. Logic and conversation. In Cole & Morgan, 41-58.
. 1978. Further notes on logic and conversation. In Cole 1978:113-27.
. 1981. Presupposition and conversational implicature. In Cole 1981:183-98.
GRIM,PATRICK.1981. A note on the ethics of theories of truth. In Vetterling-Braggin,
290-98.
GROSS,MAURICE.1977. Une analyse non presuppositionelle de l'effet contrastif. Ling-
visticae Investigationes1.39-62.
HALVORSEN, PER-KRISTIAN. 1978. The syntax and semantics of cleft constructions.
(Texas linguistic forum, 11.) Austin: Dept. of Linguistics, University of Texas.
HAMILTON,SIRWILLIAM OFEDINBURGH. 1860. Lectures on logic. Edinburgh: Blackwood.
HAWKINS, JOHNA. 1978. Definiteness and indefiniteness. London: Croom Helm.
HELDNER, CHRISTINA. 1981. La portee de la negation. Stockholm: Norstedts Tryckeri.
HERZBERGER,HANS. 1973. Dimensions of truth. J. Phil. Logic 2.535-56.
HOFMANN, T. R. 1966. Past tense replacement and the modal system. (Harvard Uni-
versity Computation Lab, Report 17.) Cambridge, MA.
HORN, LAURENCE R. 1972. On the semantic properties of logical operators in English.
Los Angeles: University of California dissertation. [Reprinted, Bloomington: In-
diana University Linguistics Club, 1976.]
- . 1973. Greek Grice: A brief survey of proto-conversational rules in the history of
logic. CLS 9.205-14.
-. 1978a. Some aspects of negation. Universals of human language, 4: Syntax, ed.
by Joseph Greenberg et al., 127-210. Stanford: University Press.
. 1978b. Remarks on neg-raising. In Cole 1978:129-220.
-- . 1979. Even, only, and conventional implicature. Paper presented at LSA annual
meeting, Los Angeles.
. 1981. Exhaustiveness and the semantics of clefts. Proceedings of the 1 th Annual
Meeting of the North Eastern Linguistic Society, 125-42. Amherst: University of
Massachusetts, Dept. of Linguistics.
-. 1984. Ambiguity, negation, and the London School of Parsimony. To appear in
Proceedings of the 14th Annual Meeting of the North Eastern Linguistic Society.
--, and SAMUELBAYER.1984. Short-circuited implicature: A negative contribution.
To appear in Linguistics & Philosophy.
JACKENDOFF,RAY S. 1972. Semantic interpretation in generative grammar. Cambridge,
MA: MIT Press.
JESPERSEN,OTTO. 1924. The philosophy of grammar. London: Allen & Unwin.
KAPLAN,DAVID.1978. DTHAT. In Cole 1978:221-43.
KARTTUNEN,LAURI. 1974. Presupposition and linguistic context. Theoretical Linguistics
1.181-94.
- , and STANLEY PETERS.1976. What indirect questions conventionally implicate. CLS
12.351-68.
, 1979. Conventional implicature. In Oh & Dinneen, 1-56.
KATZ, JERROLDJ. 1977. Propositional structure and illocutionary force. Cambridge, MA:
MIT Press.
METALINGUISTICNEGATIONAND PRAGMATICAMBIGUITY 173

KEENAN,EDWARD L. 1971. Two kinds of presupposition in natural language. Studies in


linguistic semantics, ed. by Charles J. Fillmore & D. Terence Langendoen, 45-52.
New York: Holt.
KEMPSON,RUTH M. 1975. Presupposition and the delimitation of semantics. Cambridge:
University Press.
- . 1982. Negation, ambiguity, and the semantics-pragmatics distinction. Paper pre-
sented at LSA annual meeting, San Diego.
KIEFER, FERENC. 1977. Two studies on presupposition. Lingua 43.247-71. [Review ar-
ticle on Wilson 1975 and Kempson 1975.]
KIPARSKY, PAUL, and CAROL KIPARSKY. 1970. Fact. Progress in linguistics, ed. by
Manfred Bierwisch & K. E. Heidolph, 143-73. Mouton: The Hague. [Reprinted
in Steinberg & Jakobovits, 345-69.]
KLEENE, S. C. 1938. On a notation for ordinal numbers. J. Symb. Logic 3.150-55.
KRATZER,ANGELIKA. 1977. What 'must' and 'can' must and can mean. Linguistics &
Philosophy 1.337-55.
KRIPKE, SAUL A. 1977. Speaker's reference and semantic reference. Midwest Studies
in Philosophy 2.255-76.
KROCH, ANTHONY. 1974. The semantics of scope in English. New York: Garland.
KUNO, SUSUMU. 1980. The scope of the question and negation in some verb-final lan-
guages. CLS 16.155-69.
LAKOFF, GEORGE.1975. Pragmatics in natural logic. Formal semantics of natural lan-
guage, ed. by Edward L. Keenan, 253-86. Cambridge: University Press.
LAKOFF, ROBIN. 1975. Language and woman's place. New York: Harper & Row.
LEECH, GEOFFREY, and JENNIFERCOATES.1980. Semantic indeterminacy and the modals.
Studies in English linguistics, ed. by Sidney Greenbaum et al., 79-90. London:
Longman.
LEHRER, ADRIENNE, and KEITH LEHRER. 1982. Antonymy. Linguistics & Philosophy
5.483-501.
LIBERMAN, MARK, and IVAN SAG. 1974. Prosodic form and discourse function. CLS
10.402-15.
LINEBARGER,MARCIA. 1981. The grammar of negative polarity. MIT dissertation. [Dis-
tributed by Indiana University Linguistic Club.]
LUKASIEWICZ,JAN. 1930. Philosophische Bemerkungen zu mehrwertigen Systemen des
Aussagenkalkuls. [Translated as Philosophical remarks on many-valued systems
of propositional logic, in Polish logic 1920-1939, ed. by Storrs McCall, 40-65.
Oxford: Clarendon Press, 1967.]
MCCAWLEY,JAMESD. 1978. Conversational implicature and the lexicon. In Cole
1978:245-59.
-- . 1981. Everything that linguists have always wanted to know about logic, but were
ashamed to ask. Chicago: University of Chicago Press.
MCGLOIN,NAOMIHANAOKA. 1982. Negation in a verb-final language: A case study from
Japanese. University of Wisconsin, MS.
MARTIN, JOHNN. 1979. Some misconceptions in the critique of semantic presupposition.
Bloomington: Indiana University Linguistics Club.
. 1981. Negation, ambiguity, and the identity test. University of Cincinnati, MS.
MATES, BENSON. 1949. Diodorean implication. Phil. Rev. 58.234-42.
MILL, JOHN STUART. 1867. An examination of Sir William Hamilton's philosophy. 3rd
edn. London: Longmans.
MORGAN, JERRYL. 1969. On arguing about semantics. Papers in Linguistics 1.49-70.
MULLALLY, J. P. 1945. The summulae logicales of Peter of Spain ('Treatise on expo-
nibles'). Notre Dame, IN: University of Notre Dame Press.
NEWMEYER,FREDERICKJ. 1969. English aspectual verbs. (Studies in linguistics and lan-
guage learning, 6.) Seattle: University of Washington, Dept. of Linguistics.
OH, CHOON-KYU, and DAVID A. DINNEEN. 1979 (eds.) Syntax and semantics 11: Pre-
supposition. New York: Academic Press.
PALMER, FRANK R. 1979. Modality and the English modals. London: Longman.
174 LANGUAGE, VOLUME61, NUMBER 1 (1985)

PERLMUTTER,DAVIDM., and SCOTTSOAMES.1979. Syntactic argumentation and the


structure of English. Berkeley & Los Angeles: University of California Press.
PRIOR, A. N. 1967. Negation. Encyclopedia of philosophy 5.458-63. New York: Mac-
millan.
QUINE, WILLARDV. 0. 1960. Word and object. Cambridge, MA: MIT Press.
RUMBAUGH,DUANE M., and TIMOTHYV. GILL. 1977. Lana's acquisition of language
skills. Language learning by a chimpanzee: The Lana project, ed. by Duane M.
Rumbaugh, 165-92. New York: Academic Press.
RUSSELL, BERTRAND.1905. On denoting. Mind 14.479-93.
SADOCK, JERROLDM. 1975. Toward a linguistic theory of speech acts. New York: Ac-
ademic Press.
-- . 1978. On testing for conversational implicature. In Cole 1978:281-97.
SCHMERLING, SUSAN F. 1975. Asymmetric conjunction and rules of conversation. In
Cole & Morgan, 211-31.
SMILEY, TIMOTHY.1960. Sense without denotation. Analysis 20.125-35.
SMITH, STEVEN. 1969. Meaning and negation. Los Angeles: University of California
dissertation. [Published, The Hague: Mouton, 1975.]
SOAMES, SCOTT. 1976. A critical examination of Frege's theory of presupposition and
contemporary alternatives. MIT dissertation.
- . 1979. A projection problem for speaker presuppositions. LI 10.623-66.
STALNAKER, ROBERTC. 1974. Pragmatic presuppositions. Semantics and philosophy,
ed. by M. K. Munitz & P. K. Unger, 197-214. New York: New York University
Press.
STEINBERG,DANNY D., and LEON A. JAKOBOVITS.1971 (eds.) Semantics: An interdis-
ciplinary reader. Cambridge: University Press.
STENNER, A. J. 1981. A note on logical truth and non-sexist semantics. In Vetterling-
Braggin, 299-306.
STRAWSON, P. F. 1950. On referring. Mind 59.320-44.
TAYLOR, KRISTE. 1981. Reference and truth: The case of sexist and racist utterances.
In Vetterling-Braggin, 307-18.
THOMASON, RICHMOND. 1973. Semantics, pragmatics, conversation, presupposition.
University of Pittsburgh, MS.
VAN FRAASSEN, BAS C. 1966. Singular terms, truth-value gaps, and free logic. J. Phil.
63.481-95.
VETTERLING-BRAGGIN, MARY. 1981(ed.) Sexist language:A modernphilosophicalanal-
ysis. Totowa, NJ: Littlefield, Adams.
WERTHEIMER, ROGER. 1972. The significance of sense: Meaning, modality, and morality.
Ithaca: Cornell University Press.
WILSON, DEIRDRE. 1975. Presupposition and non-truth-conditional semantics. New
York: Academic Press.
-- , and DAN SPERBER. 1979. Ordered entailments: An alternative to presuppositional
theories. In Oh & Dinneen, 299-323.
WITTGENSTEIN, LUDWIG. 1953. Philosophical investigations. Trans. by G. E. M. An-
scombe. New York: Macmillan.
ZIFF, PAUL. 1960. Semantic analysis. Ithaca: Cornell University Press.
ZWICKY,ARNOLDM., and GEOFFREYK. PULLUM. 1983. Cliticization vs. inflection: Eng-
lish n't. Lg. 59.502-13.
ZWICKY, ARNOLD M., and JERROLDM. SADOCK. 1975. Ambiguitytests and how to fail
them. Syntax and semantics 4, ed. by John Kimball, 1-36. New York: Academic
Press.
[Received 20 July 1983;
revision received 24 January 1984;
accepted 29 May 1984.]

You might also like