Beruflich Dokumente
Kultur Dokumente
Semantics Author(s): Barbara Hall Partee Reviewed work(s): Source: Foundations of Language, Vol. 6, No. 2 (May, 1970), pp. 153-165 Published by: Springer Stable URL: http://www.jstor.org/stable/25000447 . Accessed: 24/12/2012 02:18
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Springer is collaborating with JSTOR to digitize, preserve and extend access to Foundations of Language.
http://www.jstor.org
This content downloaded on Mon, 24 Dec 2012 02:18:06 AM All use subject to JSTOR Terms and Conditions
BARBARA
HALL PARTEE
NEGATION,
CONJUNCTION, SYNTAX
AND
QUANTIFIERS:
VS. SEMANTICS
The interesting claim that the syntacticallymotivated 'deep structure'of a sentence is simultaneously the basis of its semantic interpretationwas first presented and defended in (KP). Their hypothesis has gained increasing acceptance, to the point where transformational rules are now widely held to be necessarily meaning-preserving. Early transformational descriptions did not always conform to this requirement, as evidenced, for example, by the optional reflexivization rule of (LK) and the optional some-any sup pletion rule of (Ki), both of which affectmeaning. More recently, explicit objections to the claim have appeared in (Ku), (J-l), (J-2).This paper presents evidence which would appear to disprove conclusively the hypothesis in question. The evidence concerns the interaction of negation, conjunction, and quantifiers in such sentences as the following: (1) (2) (3) (4) (5) Few rules are both explicit and easy to read. Few rules are explicit and few rules are easy to read. No barber givesmany customers both a shave and a haircut. Few parents support four sons throughboth college and graduate school. Few people hate many people.
It will be argued that no syntactically plausible deep structurefor sentences like those above can provide the basis for their semantic interpretation. Sentences (1) and (2), for example,would appear to be syntactically closely related to (6) and (7): (6) (7) The few rules are both explicit and easy to read. The few rules are explicit and the few rules are easy to read.
But whereas (6) and (7) are synonymous, (1) and (2) are certainly not. Furthermore, there is no pair of conjoined sentences synonymouswith (1). But any attempt to derive (1) from a structuresemantically more appropriate
than a pair of conjoined sentences will at the very least force a denial of any
close syntactic relation between (1) and (6), and will furthermore run into serious difficultieswith sentences such as (3) and (4). A battery of arguments along these lines are presented below, along with
This content downloaded on Mon, 24 Dec 2012 02:18:06 AM All use subject to JSTOR Terms and Conditions
154
some specific counterarguments to some proposals made for quantifiers in (L-l). The results of this study are largely negative, and although some implications for the theory of semantic interpretation are indicated, no theory capable of accounting for the semantics of (1)-(5) is proposed here. The results are positive, however, insofar as they vindicate some venerable but recentlymuch criticized transformational rules. We can dismiss the possibility of treating (1) Consider again sentence (1). as an instanceof phrasal conjunction' on two grounds. First, both is charac teristic of derived sentential conjunction and is generally excluded from phrasal conjunction:2 Both John andMary are linguists. (8) *Both John andMary are a happy couple. (9) Second, phrasal conjunction connects structures which are single deep structure categories of the same type; but easy to read is not a constituent at all in the deep structure3and afortiori not a constituent of the same type as explicit.
1 By phrasal conjunction, we mean conjunction of constituents other than S derived in the base component by rules roughly of the form: A-+A and A. Such rules, not universally accepted, are typically postulated to account for sentences like the following: John and Mary are a happy couple. (1-1) Mary's dress is black and white. (1-2) John andMary are alike. (1-3) See, for example, (LP). 2While many speakers totally exclude both with phrasal conjunction, there is a dialect which allows: John and Mary are both similar. Both John and Mary have the same last name. (2-2) For the author these are ungrammatical in the intended sense, and both is an unequivocal indicator of derived sentential conjunction. 3 Paul Kiparsky has kindly reminded me (personal communication) that several people (including Perlmutter, Lakoff, and himself) have recently argued that Chomsky's analysis of easy to read is incorrect and that its superficial subject must after all occur as a deep structure subject. The arguments will not be presented here, since their outcome is not crucial for this paper. It suffices to observe that the following examples share all the relevant properties of (1), and therefore if for any of them it is agreed that the italicized part is not a con stituent in the deep structure, those examples can replace (1) in the argument of this paper. Few students are both popular and likely to succeed. (3-1) Few companies are both run by amateurs and able to attract good pro (3-2) fessionals. Few students had both begun to understand the problem and retained their (3-3) objectivity as informants. Few performers both play brilliantly in their youth and turnout to be great (3-4) artists in adulthood. Itmight be suggested (but tomy knowledge has not been yet) that the superficial subject (2-1)
This content downloaded on Mon, 24 Dec 2012 02:18:06 AM All use subject to JSTOR Terms and Conditions
NEGATION,
CONJUNCTION,
AND QUANTIFIERS
155
Nor can the difference between (1) and (2) be attributed to referential identityvs. non-identity of two deep-structure occurrences offew rules, since (2) contrasts in this respectwith (10): Few rules are explicit and few rules are easy to read. (2) Few rules are explicit and they are easy to read. (10) and (1) is not synonymouswith either of these.
It seems quite clear that the semantic structure of (1) differs from that of
(2) in somehow having the conjunction contained within the scope of a single quantifier. In logic, this kind of distinction is made explicit in the formalism, i.e. the syntax. Thus sentences (11) and (13) (clearly not syn onymous) can be formalized as (12) and (14) respectively: Some men are both married and happy. (11) (12)
(13)
(14)
If some were
the two sentences would be replaced by all in these examples, but within the logical system this fact would be treated as the synonymous, result of certain rules of logical inference, not in any direct way as the result the structure of the formulae. (I.e. it would be a semantic but not a
of
syntactic fact.) Looking at sentences (1) and (2) in the light of formulae (12) and (14),
one might want to postulate that the deep structure of (1) is something like
S
VP
rules S
S and S
are few
rules
are
explicit
to
read
rules
is easy
is in fact the deep subject in all of the above examples and that the conjunction is deep structure phrasal conjunction, e.g. of VP. Such an analysis would, I think, solve the problems of this paper, but would have problems with all the arguments by which deep and surface subjects were originally differentiated. In any case it would certainly have widespread implications formany parts of the grammar, which would have to be investi gated before it could be seriously proposed.
This content downloaded on Mon, 24 Dec 2012 02:18:06 AM All use subject to JSTOR Terms and Conditions
156
(16)
S
S
and S
NP
VP
NP
VP
rules
are few
rules
are few
Something verymuch like this is in fact proposed in (L-l).We returnbelow to specific comments on his proposal. Here we make some comments appli cable to any proposal which would account for the semantic distinction between (1) and (2) on the basis of a deep-structure syntactic difference in relative domination of conjunction and quantifier. (a) Compare the following: (17) (a) Three rules on this page are both explicit and easy to read. (b) The three rules on thispage are both explicit and easy to read. (c) Only the three rules on this page are both explicit and easy to read. (a) Three rules on this page are explicit and three rules on this page are easy to read. The three rules on this page are explicit and the three rules on (b) this page are easy to read. (c) Only the three rules on this page are explicit and only the three rules on this page are easy to read.
(18)
Sentences (17b) and (18b) are synonymous, but the (a) and (c) pairs show a semantic distinction exactly like that of sentences (1) and (2). To account for this phenomenon along the lines of (15) and (16)would require something like the following claims: (i) Quantifiers occur as predicates only with indefinite noun phrases as subjects; quantifiers have some other sourcewith definite noun phrases. (ii) Only occurs as a predicate with both definite and indefinite noun phrases as subjects. (iii) Conjoined sentences containing formally identicalnoun phrasesmay
if that noun phrase contains a quantifier unless not be collapsed contains a definite article, and not even then if there is an only. it also
This content downloaded on Mon, 24 Dec 2012 02:18:06 AM All use subject to JSTOR Terms and Conditions
NEGATION,
CONJUNCTION,
AND QUANTIFIERS
157
While not impossible, (18b) but neither (17a) nor (17c) from (18a) or (18c). such a proposal clearly has little to recommend it.
(b) Semantically, (2) implies (1) but not vice versa. However, if few is
Yet replacedby many in (1) and (2), the direction of implication is reversed. the substitution of many for few in the trees (15) and (16) would give no basis for predicting such a change. It might be argued that deriving few from notmany could account for this phenomenon. However, a few, which sharesnone of the negative properties offew, behaves just like itwith respect to the direction of implication between (1) and (2). (c) The structures of (15) and (16) treatfew as a predicate (presumably as an adjective or verb). This has at least some plausibility for such quanti fiers as many, few, several, and the cardinal numbers (i.e. for those quanti fierswhich can follow the definite article inside a noun phrase), whose pre dicative use, as Lakoff points out, soundsmore archaic than ungrammatical. But there are a number of quantifierswhich cannot even 'archaically'occur in predicate position; they happen to be just the quantifierswhich cannot follow the definite article. Compare (19) and (20): (19) (a) *? the arguments aremany / themany arguments (b) *? the arguments are five / the five arguments (c) *? the arguments are few / the few arguments (a) (b) (c) (d) *the arguments are some / *the some arguments *the argument(s) is/are every / *the every argument *the arguments are all4 / *the all arguments *the arguments are none / *the no arguments.
(20)
proportion of a given set and not its size.5 But this distinction does not coincide with the synonymy or non-syno
4 Sentences such as 'the potatoes are all' do occur in the Pennsylvania German dialect of English, but with a totally unrelated sense (all = all gone). 5 Few, many, and the numerals, etc., can also appear before the definite article with an inserted of, and they then receive a 'proportionate' sense like some, all, etc. Compare (6-1) (6-2) his three sons three of his sons,
where the former implies that the set of his sons has threemembers, while the lattermakes no implication about the total size (except that there be a subset with three members). The point of concern here is simply that all, every, etc., must be regarded as distinct from many, three, etc., in not having the predicative source which could account for one sense of the latter.
This content downloaded on Mon, 24 Dec 2012 02:18:06 AM All use subject to JSTOR Terms and Conditions
158
nymy of pairs like (1) and (2). If for few in (1) and (2)we substitutemany, five, some, or no, we still have non-synonymous sentences; but all or every yield synonymy. Thus the independent syntactic grounds for calling some quantifierspredicates do not lead to the right class of quantifierswith respect to the semantic behavior of quantifierswith conjunction. It would therefore be quite misleading to try to claim independent syntactic justification for structures like (15) and (16) on the evidence of (19). (d) Itwas suggested above that a semantically consistent approachwould require that only also be treated as a predicate. In this case, the counter arguments are even stronger, since not only is only not permitted inpredicate position in ordinary sentences (see (21)), but it can modify structures that are by no stretch of the imagination noun phrases, as in (22). (21) (22) *The three rules on this page are only. The three rules on this page are only explicit and easy to read (i.e. they are not, for instance, interestingor revealing).
as its subject, for then the deep structureof (22)would be identical to that of (18c), and the two are clearly not synonymous. To provide the proper semantic interpretation, the deep structureof (22) would have to contain only as a predicatewhose subject is explicit and easy to read;but as we remarkedearlier, easy to read cannot be a deep structure constituent. It thus appears particularly clear in this case that the semantic must depend in part on derived structure,where explicit and interpretation easy to read is indeed a single constituent in construction with only. To summarize the argument so far, we have suggested that the require ment of semantic appropriateness leads to structures like (15) and (16) as themost likely candidates for deep structures of sentences (1) and (2), but that such an approach leads to several serious problems: (a) itwould appear to require some extremely complex rules for the inter action of the quantifiers, definite articles, and only, and for syntacticallyun necessary constraints on conjunction reduction;
(b) it would not even capture all of the clear semantic facts, such as the vs. many; difference in direction of implication between (2) and (1) withfew
(c) the independent syntactic arguments for treating certain quantifiers as predicates lead to a semantic distinction between quantifiers of set size and quantifiers of proportion, but do not lead to the class of quantifierswhich
behave semantically like few in (1) and (2); and
This content downloaded on Mon, 24 Dec 2012 02:18:06 AM All use subject to JSTOR Terms and Conditions
NEGATION,
CONJUNCTION,
AND QUANTIFIERS
159
predicatewith subjectswhich are not even possible constituents of any type, and certainly not all noun phrases. The suggestion that quantifiers are predicates ismade explicitly in (L-l, Appendix F). Lakoff there claims that sentences containing quantifier predi cates may occur as eithermatrix or constituent with other sentences,with the same surface result but different semantic senses. Thus for the sentence (23) Did many inmates escape?
(24)
Q
S
NP VP
Det
Indef
inmates
escape
many
(25)
Q
S
NP VP
Det
were
many
the
inmates
some
Sentence
(23) is asserted
to be ambiguous
(24) and (25).The ambiguity itself ismarginal, and the structuraldistinction proposed to account for it is called into question by some other evidence.
6 The choice of definite and indefinite articles inmatrix and constituent is curious. It is consistent throughout his examples, and all of the sentences he discusses have indefinite articles (i.e. no articles) in their final shape. Some of the problems of the articles with quantifiers have already been mentioned; theywill not be explored further here since they were not explicitly discussed in (L-1) and are not crucial to our basic thesis.
This content downloaded on Mon, 24 Dec 2012 02:18:06 AM All use subject to JSTOR Terms and Conditions
160
Lakoff claims that any noun phrase can have a quantifier embedded within it, but that only (surface) subject noun phrases can combine with a quantifier from the next higher S. The second part of this claim is false under his assumptions, however, since (26) Does John readmany books?
(27)
Thus itwould appear that this lineof reasoningwould require the possibility of incorporating a matrix-sentence quantifier into at least both the subject and object noun phrases of embedded sentences. But this necessary extension leads to a superabundanceof available deep structures for certain sentences. Consider the following example: (28) Few people readmany books.
Given that both (26) and (27) can derive their quantifiers from higher S's, it follows that both quantifiers of (28) can come from higher S's. Thus one possible underlying structure for (28), and a semantically plausible one, would be:
(29)
NP Det the S
s
VP
N people VP
are few
NP
Det
are many
the
books
This content downloaded on Mon, 24 Dec 2012 02:18:06 AM All use subject to JSTOR Terms and Conditions
NEGATION,
CONJUNCTION,
AND QUANTIFIERS
161
Semantically, (29) is a more reasonable structure7 for (28) than a structure with one quantifier above the kernel sentence and one below it; however, if lower-S quantifiers are deemed necessary to account for the claimed ambiguity of (23), then therewill be five possible deep structures for (28): (i) (29). (ii) A structure like (29)with the quantifiers interchanged. (iii) and (iv) One quantifier in a higher S, the other in an embedded S. (v) Both quantifiers in embedded S's. There may be some dispute as to whether (28) is twoways or threeways ambiguous, but itwill hardly be claimed to be fiveways ambiguous. Itwould be reasonable to claim (i) and (ii) as its deep structures,or (iii), (iv), and (v), but not all of them. The semantic arguments all require the possibility of quantifiers in higher sentences. The suggestion that they also be derivable from embedded sentenceswas motivated primarily by syntactic arguments; the claim that quantifierswere predicates gainedmost of its syntactic plausibility from the apparent similarity of behavior of e.g. numerousand many8: (30) (31) (a) The flowers, which were numerous, were coveredwith dew. (b) The numerous flowerswere covered with dew. (a) (?) The flowers,which were many, were covered with dew. (b) The many flowerswere coveredwith dew.
But counterarguments to this analogy have already been given above. The possibility of deriving quantifiers from lower sentenceswas also used to account for the ambiguity of (23). But note that that ambiguity, at best
7 Since the rule which lowers matrix quantifiers into embedded S's is not stated, it is difficult to be certain whether it could apply to a structure like (29). Certainly normal relativization could not apply: a comparable case with ordinary predicates in place of the quantifiers would yield: (8-1) *People who books which read are best-sellers are extroverts. Sentence (8-1) is blocked by the Complex-NP Constraint described in (R). The downward insertion of quantifiers would also seem to be a 'chopping rule' and should therefore be subject to the same constraint. But itmay be that the product of the rule is not a complex noun phrase and thus that the constraint would not be violated in deriving (28) from (29). (The very fact that the rule is a 'downward insertion' is suspicious, since there are no other such rules as far as I know.) 8Note that the relative clause of (30) must be non-restrictive; it is not obvious that adjectives like numerous can occur in a restrictive relative clause, or that there is any possible relative clause source for the numerous of Numerous animals were driven from the forest. (9-1) Itmay well be true that some quantifiers have essentially the same syntax as quantifica tional adjectives; but it does not appear that those adjectives share the syntax of ordinary adjectives.
This content downloaded on Mon, 24 Dec 2012 02:18:06 AM All use subject to JSTOR Terms and Conditions
162
tenuous, disappears if almost any other quantifier is substituted for many. The arguments for deriving quantifiers from lower S's thus appear to be much weaker than those for deriving them from higher S's, given theKatz Postal hypothesis. Further arguments for nesting of higher S's containing quantifiers appear when we return to examples containing quantifiers and conjunction.We repeat here sentence (3): (3) No barber givesmany customers both a shave and a haircut.
To provide the correct semantic interpretation, both quantifiersmust be outside the conjunction, as shown below: (32)
NEG NP
S
VP
barber NP
S VP
is some
customers
are many
and
shaves
haircut
Since in this case the semantic interpretation can be captured only with quantifiers in stacked higher S's, not with one higher and one embedded, the argument for accounting for (28) in the sameway is strengthened.Thus if quantifiers are to be regardedas predicates at all, theymust apparently come only from higher sentences, and never from embedded ones. But arguments (a)-(d) of the first half of the paper suggest that such an analysis leads to rather serious problems in any case, so this last refinement may be viewed as simply sharpening up the analysis we are basically trying to overthrow. There is an additional, relativelyminor, problem in the use of higher quantifiers. Structures like (32) and (29) have the quantifiers ratherwidely separated from 'kernel'occurrences of the noun phrases inwhich theywill eventually appear; thematching of quantifier to noun appears to rely on the
identity of the nouns in matrix and constituent. But the same noun may
This content downloaded on Mon, 24 Dec 2012 02:18:06 AM All use subject to JSTOR Terms and Conditions
NEGATION,
CONJUNCTION,
AND QUANTIFIERS
163
occur more than once in the lowest sentence, as in the following: (5) (33) Few people hate many people. Many people hate few people.
Sentence (5) may or may not be ambiguous; if not, we will assume (as appears consistent with (L-1)) that (5) plus its passive have readingswith the quantifiers in both of the possible orders or relative scopes. The same can be said for (33) plus its passive. Thus (5) and (33) (with their passives) each require two deep structures.The problem is that they seem to require the same two deep structures, namely (34) and (35). (34)
NP
S
VP
people NP
S VP
are many
people
are
few
(35)
NP
S
VP S NP VP are few
people
people
are many
But (5) and (33) have no readings in common, nor do their passives. Thus
they must not have identical deep structures, and the operation of quanti
fier-loweringmust therefore depend on somethingmore than simple noun identity.The obvious suggestion is indexing, so that the occurrencesof people
This content downloaded on Mon, 24 Dec 2012 02:18:06 AM All use subject to JSTOR Terms and Conditions
164
in (34) and (35) could be matched up in just the appropriateways to differ entiate (5) and (33).But note that this isnot referential indexing in any usual sense, since at least one of the noun phrases in each sentence has a distrib utive sense, i.e. not the same 'many people' for different individuals of the 'few',or vice versa. This is presumably not an insurmountableproblem, but on the one hand it arises as a syntactic problem only if quantifiers are not derived directly within the noun phrase, and on the other hand themecha nisms thatmight be devised to dealwith indexing in this broader sensemight conceivably have to be of such power as to render superfluous the supposed semantic contributions of the quantifiers-as-higher-predicates analysis. Thus these examples appear to furnish an additional small bit of evidence against the treatment of quantifiers as predicates. We can summarize the basic conflict as follows: Semantically, the arguments in (L-1) for deriving quantifiers from higher sentences are very strong, and become even stronger when examples in cluding conjunction are brought in. As in logic, a quantifiermay be inter preted as having a whole complex sentencewithin its scope. Itwould appear that semantic interpretation in fact requires unlimited 'stacking'of quanti fiers, conjunction, negation, only, and perhaps the definite article, although themechanism for the latter is far from obvious. Thus itwould appear fromLakoff's arguments and our extensions of them that if theKatz-Postal hypothesis is to be maintained at all, it is necessary to have structures essentially like (15) and (16) to account for the semantics of (1) and (2). But we showed in the first half of the paper that any such proposal runs into extremely damaging counterarguments. If, on the other hand, theKatz-Postal hypothesis is rejected, it becomes
considerably easier to account for the syntactic and semantic facts. Quanti
fiers can be derived within noun phrases (although the purely syntactic problems involved are non-trivial), negation within 'kernels'after theman ner of (Kl), only as a kernel element freely attachable to any one or more surface constituents (Ku). Conjunction-reduction, insofar as we are con cerned only with generating the set of syntacticallywell-formed sentences, can be viewed as optional and dependent on formal identity alone, with collapsing vs. pluralizing of identical constituents also optional. which have been under consider Those aspects of semantic interpretation
ation here appear and (2), the surface to be explainable structure quite naturally on the basis of surface to that in the
structure, although no specific rules are proposed here. In the case of (1)
shows an overt contrast parallel
logical formulae cited earlier. See also the earlier discussion of only, which strongly supported the claim that some semantic interpretationdepends on surface structure.
This content downloaded on Mon, 24 Dec 2012 02:18:06 AM All use subject to JSTOR Terms and Conditions
NEGATION,
CONJUNCTION,
AND QUANTIFIERS
165
The syntacticallymotivated deep structure is certainly not irrelevant to semantic interpretation; the old arguments about easy toplease vs. eager to please still stand.What we hope to have demonstrated here is that it is simply false to claim that all of the semantic interpretationwhich is de pendent on syntactic structure at all is dependent on the level of deep structure. It is hoped that final abandonment of this claimwill help to over come some of the current objections to otherwise acceptable syntactic rules and simultaneously lead to amore productive formulation of the interesting semantic insights recently gained by such people as Ross, Lakoff, and McCawley. Department of Linguistics, UCLA
BIBLIOGRAPHY (C) Chomsky, Noam, Syntactic Structures, The Hague 1957. (J-l) Jackendoff, Ray S., 'An Interpretive Theory of Pronouns and Reflexives'. Un published, 1968. (J-2) Jackendoff, Ray S., 'An Interpretive Theory of Negation', Foundations of Language 5 (1969) 218-241. (J-3) Jackendoff, Ray S., 'On Some Incorrect Notions about Quantifiers and Negation', in Studies inTransformational Grammar (ed. by S. J. Keyser et al.), Air Force docu ment AFCRL-68-0032, Waltham, Mass., 1968. (KP) Katz, Jerrold J. and Paul M. Postal, An Integrated Theory of Linguistic Descriptions, Cambridge, Mass., 1964. (K1) Klima, E. S., 'Negation in English', in Fodor and Katz (eds.), The Structure of Language, Englewood Cliffs, N.J., 1964. (Ku) Kuroda, S.-Y., 'Attachment Transformations'. Unpublished, 1966. (L-1) Lakoff, George, On the Nature of Syntactic Irregularity, Harvard Computation Laboratory Report NSF-16, Cambridge, Mass., 1965. Note on Negation', inHarvard Computation Laboratory Report (L-2) Lakoff, George, 'A NSF-17, Cambridge, Mass., 1966. (LP) Lakoff, George and Stanley Peters, 'Phrasal Conjunction and Symmetric Predi
cates', inHarvard Computation Laboratory Report NSF-17, Cambridge, Mass., 1966.
(LK) Lees, R. B. and E. S. Klima, 'Rules for English Pronominalization', Language 39 (1963), 17-28. (R) Ross, J. R., Constraints on Variables in Syntax, M.I.T. dissertation, 1967.
This content downloaded on Mon, 24 Dec 2012 02:18:06 AM All use subject to JSTOR Terms and Conditions