metaphor and non-metaphor: the semantics of adjective-noun combinations

5

Click here to load reader

Upload: john

Post on 03-Jan-2017

234 views

Category:

Documents


2 download

TRANSCRIPT

Page 1: Metaphor and non-metaphor: the semantics of adjective-noun combinations

Book reuiews ,4~

Jan M.G. Aarts and Joseph P. Calbert. Metaphor and non.metaphor: the semantics of ad.)iecti~e-noun combinations. Ttibingen: Niemeyer. 1979. 240pp. DM54.00.

Polysemy, a phenomenon ranging from subtle extensions of meaning to obscure metaphors, from the lexically integrated to the very unconventional, is, a pervasive problem in linguistics. Aarts and Calbert (henceforth A&C) look at this problem :;pecifically with respect to adjective-noun (Adj + N)combina- tions in an atter~lpt to show how the meaning of the phr~se can be predicted from the meani~lgs of its component lexical items via a process of selecting specific lexically given senses and building extended senses This book repre- sents some wort]~Lwhile descriptive work, t,lou~.h it gets bogged down in a fairly ad hoc model which hides a number of important generalizations.

The basis of A&C's system is a lexicon representin~ one or more senses fo'r each Adj or N, and a process of matching the senses c~f Adj's and N's in combination to derive one or more interpretations f~r the who!e phrase. Matching invo]Lves comparing contextual (CONT) features associated with the Adj senses with semantic features of the N senses for ccmpatiibility in ~ cyclic fashion: the first N sense is matched with each Adj sense in turn on :ih~. first cycle, the second N sense is matched with each Adj sen.,;:e on ~Lhe secont~ cycle, etc. If no compatible match is found on any given cycle, then special/ 'ans]er or metaphoriz~tion rules apply t.o yield an interpretation. Thu.,', CONT L:,~tures do not serve, ~s in some models, to mark some syntactic strings as semantically anomalous.

By means of the particular structure of the matchinl~ process, A & C define an ordering 9f the set of all possible readings of ~ ~,hrase with respect to probability. They profess that the larger linguistic :~r~d non-linguistic context can reduce the probability of a certain reading, or impose a new reading on the Adj + N phr~,se, but they try nevertheless to exploit as much information given in the phr~Lse before looking further. It is hard for n,l~ to see just what the significance of this scheme is suppose, I to be. If A & C a~re trying to specify the cognitive p,rccesses involved in sel,~.'ting an appropriate reading,, they are probably rrfi:~,taken. Current psycholLogical work in langt,age processing indi- cates that ser~ses are arrived at largely -n the opposite w~y, i.e., in a "top.- down" fashion whereby the wider context suggests possible sev, Lses which are only then confirmed or discounted by ma~.ching possible senses of the lexica! iit,~ms.

A&C as s~lme an essentially Katzk~a semantic framework with a few iinnova- tions. Each ~ense of a lexical i~em is a complex of features, which are either binary (marked " + " or " - " ) , or non-binary (marked " E " ) . Each C ONT feature is a,,;sociatcd via lexical~ redundancy rules with a predicatio, gl re!a~,r (PR), which indicates hew ,~he adjective is predicated of the noun when that CONT fca~.ure is involved, in a successful match. Each PR is a ~Iwo..place predicate. The meanings of the PR's are inconsistent!ty formulated: son~efimes

Page 2: Metaphor and non-metaphor: the semantics of adjective-noun combinations

50 Book reviews

the N features appear as the first argument of a PR, and sometimes the Adj features come first, Sometimes a PR relates Adj and N features, and sometimes a PR relates Adj features :,~ a statement involving another PR of which N features are one argument.

For instance, heavy has the CONT feature -[ + CONC]- ( " - . . . - " indicates a CONT feature), which is associated with PRl. PRI(N,A) means that N has the property A. Thus heavy table attributes the property of heaviness to table. Angr¢ has alternative CONT features -[ + HI-, and -[ E comm]-. -[ + HI- is ass<x:iated with, 1~ ~..,+, PR2(N,A) means that N experiences the mental state or c~+~,'aJ condition A. -[ E comm]- is associated with PR3 which embeds PR2; PR3(N, PR2([ + Hi, A)) means that the referent of N manifests that the referent of 1-'~ Hi experiences A. Man is [+ Ft] so +.hat in angry man we get a match ~nvolving PR2 which means that the marl experiences the mental state of anger. Letter, on the c..ther hand, is [ E comm] so that in acgry letter we get a match • l t , ; r * n D D "} . . 1 _ _ ' - - k

.~- t'X,.~' W];|l~ll +nvo,,,,,~ means that the letter manifests that [ + H] experiences the mental state of anger.

A & C claim that when PR3 is used as in this last interpretation, the referent ~ f I+ H] must be explicitly given in the wider linguistic context, e.g., by of Jo,~n in a~+grv letter of John. This claim is supported by a single unconvincing cxampl. +, and in any case is falsified by the sentence 1 received an angry letter today, which does not explicitly identify the referent of [+H]. The referent must. of course, be the writer of the letter, but this follows nowhere from A&("s semantics. Problems like this notwithstanding, A&C's use of PR's ~]Jow~ them to make some revealing generalizations about h~w the application ~f adjectives is extended from objects of one kind to obje,.:ts of another~ as f:~und in happy person, hat~,gy home, happy days, happy evem'., etc.

A ve,-'y interesting aspect of A&C's lexicon is the inclusion of features with ~onnetartve as opposed to denotative value. The only connotative features specified in the lexicon are [E + ] and [ E - ], signifying posidve and negative emotional evaluations. Further connotative features are not given in the iexicon but can be derived for metaphors as described below. Love is [E + ], while ;~ate i~ [ E - ]; patriot is [E + i, while chc~uvinist is [ E - ]. The status of connotative meaning in semantic the,>ry is certainly problematic, but A&C argue :bz~t it must be lexically represemed because it plays o, role in determin- ing the d~notative meaning of larger +anits: If a N has a connotative E-value and i++~ in construction with a gradable Adj requiring an extended reading, the deno~.ation of the phra~ may have an evaluative (Ev) denot~ttive feature whir h depends on tt:-e E-value of the N. Gradable Adj's are marked [I -+-- ] depending on their po!~rity+ following Bier, visch. Thus deep is [I + ], while shallow is [! - ]; big is [1 + ], while little is [ I - ]. The Ev-value of the Adj + N phrase genera!ly is a function of the (denotational) l-value of the Adj and the ¢connotational) E-value of the N. For example, blunder is [ E - ], while love is [E + ]; accordingly b~g love is [Ev + }, while big blunder is [ E v - 1. There are

Page 3: Metaphor and non-metaphor: the semantics of adjective-noun combinations

Book reviews 51

e:tceptions to A &C's generalizations which they naark lexically. Their observa- tions aboat the relevance of connotation to determining denotation are excit- ing and should be 1.ooked at more closely.

The first sense given in a lexical entry is the b,~ic sense. In addition to the basic sens.e there may be a number of extenaed senses as well, which are each iirL some way parasi:ti¢ on the basic sen, e. A&C draw a distinction between metaphorical and non-metaphoric~d ex.ended senses on the basis of an intui- tive difference between the two. A metaphoricaJl sense generally background~, i.e., demotes to the connotative lewd, the features of the basic sense. Extended senses aot represented in the lexicon may be invoked as described below. A & C discuss the persistent problem of determining which extended senses are lexically integrated and which are not.

Non-metaphorical extended senses are accounted for by generative (GEN) features and transfer (TR) features, which apply only to N's. GEN features are specified in the lexicon, while TR features are n,c,t. The function of either is t,a allow an object to b:, ~ referred to in two ways. For example, [+ H, ~ profe:;- sional] N's redundantly take the GEN feature [[ + H]:[ +ACT]]. This means that an associated Adj can optionally apply to the action typically associated with the professional instead of to the human referent directly. Pianist is [+ H, ~- professional], and good applies via the appropriate PR's to humans or to actions; therefore good pianis~ ~ ambiguously predicates goodness of the person himself, or of the action he typically performs as a pianist. Similarly, [~ insl;rument] N's are assigned the GEN featLtre [[ E instrument]:[+ ACT]l, which plays a role in the interpretations of good knife.

TR features do the same thing as GEN featu~,:s, but must be invoked only if the m;atching process is unsuccessful on some cycle. GEN features refer to some factual belief about the referent : f the lexical item which all members of the language community share. Wh,'zeas everybody knows tha~t the action typicaltly associated with a pianist is playing l:he piano, or that the action typica]lly associated with a knife is cutting, there is no consensus about which action is to be associated with ma~ or stick. Therefore this informatiion cannot be represented in lexicon. Nevertaeless, contextual information ,:an be used (A&C do not suggest how) to identify eJ,n actiov with stick or man to which the meaning of the associated Adj c~tw apply. This is the domain of TR features. For iv~stance, good stick in one context might mean that the stick i:~ good for walking with, in another that it i.~, good for bonkir~g someone on the h,ead with.

I question the usefulness of the distinction between GEN and 'FR features. First, the distinction is based on purely non-linguistic knowledge of the world which should not have to be represented in the lexicon. Second, the diistinction invoh,es the assumption that the action to whiieh reference is made; in words with a lexical GEN feature is fixed. This assumption, contrary ~o A&C's arguments, is cleverly false. Th~:y argue that good wrench, for in~aance, must mean that the wrench is good :in the funcI~iLon normally associated with

Page 4: Metaphor and non-metaphor: the semantics of adjective-noun combinations

52 Book reviews

wrenches, and not that it is go~xt in the function of breaking windows, etc. But it ,,s easy to imagi ,e contexts in which good wrench has exactly this last reading. Tleir argument rests on the observed peculiarity of ?good wrench to break a wb~dow. But notice that :good wrench to break a window with is fine,

Furthermore, I question the motivation for adding features like [ + ACT] via Gt-N and TR features at all. First, recourse to such features is entirely ad hoc. Ida,ally, semantic features should reflect referential properties of lexical items. But to allow dennst to assume the feature [+ ACT] does not :make sense in this respect: a dentist cannolt be a,a action. A & C argue that [+ACT] is associated with a change of reference, pointing out that a teat:her, for instance, can be either a professional teacher, or someone who occasionally teaches, as in for a poltcemano he's a good teacher. Here teacher exhibits the feature [+ACT], A & C claim. But if this were the case, good teacher could not refer to a professional teacher to mea~ that he is good as a teacher, or to an occasional teacher to mean that he is good as a person, i.e., the rei'erent of the N would have to be linked to whether the A ~ predicates the person or the action, which it clearly is not.

Second, re~:oarse to the features GEN and TR probably obscures a much more general ~ccount of these ambiguities: every, use of a gradable Adj involves some e~plicit or implicit scale: if I say it's big I leave it to the hearer to infer what it's big relative to, and if a say he's good the hearer must infer what he's good at, b~:~ed on considerations of relevance, expectation, etc. Given the existence of rules of some sort which allow the hearer to infer the correct scale, there is no a priori reason to suppose that the same principles are not at work when an Adj occurs before a N instead of after the copula. If we assume that the occurrence of a N can bring certain objects or actions to mind which then play a role in the determination of an appropriate scale, the observed amb;~gu- ity is predicted. If good dentist refers to a good person who is also a dentist, then the relevant scale has been determined from the wider context; if good dentist refers to a person who is good at dentistry, then the determination of a relevant scale has probably been made under the influence of the immediate context. This proposal needs to be worked out, but I suspect it has some tr~ath to it, and it certainly would be more general than A&C' s account.

Finally, A& C discuss metaphorical extended senses, which differ from the senses defined by GEN and TR features in backgrounding ~dl or part of the b~ic sense. Metaphorical! senses given lexically are marked by secondary (SEC) features, of whicll there are two kinds, experiential and symbolic. Experiential SEC senses aLre based on a (still) clear experiental relation to ~he b~ic sen.,~e. Examples are blind in the sense of unheeding, and bright in the sense of smart. Symbolic SEC senses are based on some seemingly arbitrary retation to the basic sense. Examples axe square in the sense of conservative, and chicken in ~he ~nse of coward. The backgrounding of the basic sen:~e even in the ~ of symbolic metaphors is an important observatiion. A & C don' t

Page 5: Metaphor and non-metaphor: the semantics of adjective-noun combinations

Book reviews 53

mention, however, that the backgrounded basic senses can actually play a role in the interpretation of larger units. Thus, He's ,1 square but he's rounder than Harry might mean that he is less conservative than Harry., and Sarah's a chicken, but she doesn't cluck as much as she used to might mean that sl~,e is no longer .as overtly cowardly.

Metaphors which are not rep:resented in the lexicon arise via memphoriza- tion ru~es, which like TR features come into play when the matching process is unsuccessful on some cycle. The form; of these rules are aot particularly insightful. Essentially, met~tphorization i,:: olves backgrounding ~:he basic sense, copying the feature involved in the unsuccessful matching process ~(~L CONT feature or a feature in conflict witk a CONT femure), and, in the case of Adj metaphorization, adding an appropriate PR and copying I- and Ev..features according to the basic sense.

For instance, sad has the CONT feature -[ + H]-, which conflicts 'with the feature [ - A N ] of tree; therefore, matching is un~successful in sad tree and the .....m'~t~"h'~'q"~ti'~"..r.., ~'-------'-'-- rules .~,,jmo" r,,~,,,. ~,q,,,,o~,~a,, ~e,'-"-'~. A "'"-'~v'"-''"~"~"a''";""l'"a' ,""'~A;,.a~,,,~ . . . . . . ,.a,, ~'~o~ achieved, for example, by backgrounding the [-. AN] of tree and copying the [ + H] f~om sad onto tree. "lhe. resulting interpreta.tion has the denotation that a human being experiences sadness, and the connotation that the speaker's attitude: toward the human referent is the same in some respect as thalt towards a tree.

The basis of any metaphor is generally taken to be some similarity or analog)'. Two unfortunate things can be observed about the role of ,fimilarity in A&C's interpretations ef metaphors: they do not provide an account of the type of similarity involved, and they relegate the similarity to the. connotative level. When a metaphor is actual]iy used there is virtually always ,;ome specific way in which things are said to be simiiat or an~th~goos, and this way ,of being similar ,or analogous is intuitively part of the; ctenotation. In the most im- mediate interpretation for Harry is an elephant not just any simiilarity to elephants is attributed to ttarry, but size. And the attribution of size. is in no way connotative; it is the point of the sentence, and can plausibly be denied !by the hearer.

A&C. admit that the recognition of the relev~Lnt similarilty or :~a21ogy in a metaphor is important and invollves a number of pragmatic parzmeters. But without e~:plicating the nature of these factors, they haven't sai,:l anything significant about metaphor.

In conclPusion, I would recommend A&C's book to th,:,se haterested in polysemy for the scope of its observations, though some of l~hem :~eem to be mistaken. I wo,.dd not recommend it for its theoretical insights or sophistica- tion, thcmg, h it does make some interesting points.

John Dinsmore University of California, Berkeley

USA