DOCUMENT RESUME

ED 363 104 FL 021 546

AUTHOR Cheng, YuminTITLE The Interpretation of Linguistic Signs and the Role

of Inference.PUB DATE Oct 91NOTE 28p.; For the serial publication as a whole, see FL

021 540.PUB TYPE Viewpoints (Opinion/Position Papers, Essays, etc.)

(120) Journal Articles (080)JOURNAL CIT CUHK Papers in Linguistics; n3 p125-151 Oct 1991

EDRS PRICE MF01/PCO2 Plus Postage.DESCRIPTORS *Communication (Thought Transfer); *Grammar;

*Inferences; *Linguistic Theory; Models; Phonology

ABSTRACTThis discussion on the interpretation of language and

the role of inEerence argues that language is a tool in the sensethat it provides the basic operators and rules of operation forspeech just as arithmetic does for calculation. It is suggested thatbecause language is a tool that is restricted in its expressivepotential by the linguistic system, speakers often have to rely oninferences made on the basis of the meaning of sentences to get atthe messages conveyed, and learn to take advantage of this process tocarry on communication. The discussion is divided into the followingsections: (1) language is used to say something; (2) the model ofsaying things; (3) the interpretation of sign meanings; (4) theinterpretation of grammatical signs; (5) lexical systems; (6) theinterpretation of lexical signs; (7) the role of inference in signinterpretation; (8) logical inference at all levels of language use;(9) inference at the phonological level; (10) inference at thelexical level; (11) inference at the grammatical level; (12)

inference at the lexico-grammatical level; and (13) a model forinference in sign and sentence interpretation. (Author/JL)

*********************:!.1. **********************************************

ReproductioLs supplied by EDRS are the best that can be madefrom the original document.

***********************************************************************

"PERMISSION TO REPRODUCE THISMATERIAL HAS BEEN GRANTED BY

TO THE EDUCATIONAL RESOURCESINFORMATION CENTER (ERIC)."

U.S. DEPARTMENT OF EDUCATIONcmpce ot Educabonat Research and Improvement

EDUCATIONAL RESOURCES INFORMATION

1CENTER (ERIC)

his document hes been reproduced as6recmyed from the person or organtzabononcunahno it

r Mmor changes have been made to onprovereProduchon Qua Idy

Points of vtew or opinlons stated ,n th.sdocc .went do not necesSardy represent othc.atOERI posdbn or pOhCy

CUHK PAPERS IN LINGUISTICS NO.3.1991 : 125 - 151.

THE INTERPRETATION OF LINGUISTIC SIGNSAND THE ROLE OF INFERENCE

Cheng YuminFudan University

1. Language Use

1.1 Language is used to say something

Although it mav seem too elementary a thing tosay that language is used to say something, it isapparently not so understood by most linguists andspeakers in general. With regard to everythingelse we are ready to admit that if we usesomething to do something else, then this"something" does not go into the "something else"and must not be included in it as a part or acomponent. We use a camera to take a photo, theart of drawing to produce images of real things,arithmetic to settle accounts, and so on and soforth. In all these cases, the former (camera,art of drawing, and arithmetic) are instrumentalto the creation of the latter (ghotographs,images, and accounts), but do not go into them toform a part. In this sense, using something to dosomething else is different from using somethingto make something. When we use wood to make atable, the wood goes into the table, andconsequently, theoretically speaking, there is somuch wood less, as wood, in this world. When wemake a statue with wax, the same thing happens.But language does not diminish with use, so lan-guage use belongs to the category of using some-thing to do something else.

Besides, there is no aifficulty in seeing thata photo of a person is not the person himself. Adrawing of a table is likewise not the table. Byinference, the thing we say is not the thing wesay about.

Yet with language the differentiation of thetool from the artefact, and the differentiationof the thing we say from the thing we say aboutare difficult to make--because the tool(language) is not anywhere to be found outside ofthe artefacts (utterances); and very often thething we say about (as in the case of ideas) isnot to be found outside of what we say (againutterances). So naturally utterances are easily

125

taken to be language.But utterances are not language, just as a

picture is not the art of drawing and a bankaccount is not arithmetic. The latter are behind,under or in the former, and one can deduce thelatter from the former, but the two are notidentical. So far I have been arguing thatlanguage is a tool in the sense that it providesthe basic operators and rules of operation forspeech as arithmetic does for calculation. Thistool has no natural form of expression except inits products--utterances, but we have long sincededuced it from its products and outlined it invarious grammars and dictionaries. Further,theorizing about the nature of language, linguistssince Saussure have followed the concensus that itrepr-7;sents a system of signs, each of which is acombination of a signal, serving as its form, anda meaning, serving as its content.

The inference linguists have failed to makeis: since meaning is inherent in the linguisticsystem, and this system is, as we have discussedabove, instrumental to the production ofutterances but does not go into them as acomponent part, it follows that system meaningmust be differentiated from utterance meaning. Inusing language, we are taking advantage ofsomething which has its own meaning to indicatethe meaning we want to express.

Since language ;s a tool that is restricted inits expressive potential by the linguistic system,speakers often have to rely on inferences made onthe basis of the meaning of sentences to get atthe messages conveyed, and learn to take advantageof this process to carry on communication. "Whattime is it?" is an explicit and straightforwardquestion. But it seems that people are notentrely comfortable with its straightforwardness.Therefore, what one often hears on a casualoccasion is "Have you got the time?" or simply"Got the time?" The meaning of the sentence canonly be: "Have you got a watL so that you mayknow the time?" But we know by inference that theperson is not interested in knowing if we havethe time, but in knowing the time himself. Thisis not the meaning of the sentence, but it is themessage it carries. Apparently, this oblique formof inquiry is preferred because it sounds lessimposing: one would feel freer to say "No", sincethe question is simply whether we have the time,while the request for help is our own inference.

126

3

Once inference is understood as a regularcomponent in langtlage use--both in encoding anddecoding a message--it follows that conversationalimplicatures and indirect illocutionary forcesbecome special cases of this general process.Further, as will be demonstrated 4..n Section 3 ofthis paper, the inferential processes at thevarious levels of language use all follow thegeneral Gricean pattern of "what is said is R, butwhat is meant is g".

1.2 The Model of Saying Things

What all the above is driving at is that thereare two combinations of form and content at twodifferent levels. Language represents onecombination of form and content, defined bySaussure as the combination of the signifiant andsignifié (for our purpose we will call them signform and sign meaning respectively). Whenlanguage is put to use, this systematiccombination of form and content in turn becomesthe form of a speech event (hence the formulation"language is form"), which is meant by the speakerand understood by the hearer to indicate, to hintat, the content of the speech event (the things wewant to say, which from now on will be referred toas the message), thus forming another form-contentcombination. This dual relationship can be shownas follows:

FIGURE1

F

Form

ORM

Content

CONTENT message

SI H

MIND

( The stralght-lIne arrows Indicate the language use process of the hearer:

the dotted line arrows that of the speaker.)

127

4

The processes indicated in FIGURE I areactivated not only when there is some kind ofindirect meaning or message to be worked out, asone may tend to think, but are constantly at workin language use. The inferential processes haveto be there, if language is thought of as a systemwith jf5 meaning used to carry messages. Even whena sentence is used in its literal meaning, i.e.when meaning and message coincide, we still needthe inferential processes to turn the meaning,which is a potential of the linguistic system,into a message, which is something that is reallyconveyed in communication, and to reveal that thetwo actually agree. A sentence like:

(1) Shakespeare was a great playwright.

is said to be used in its literal meaning becauseevery sign meaning in it is turned directly,w:Ithout any change, into a corresponding signmessage, and when the sentence message, resultingfrom all the sign mensages, enter's the hearer'smind to be processed, the final discourse messageremains the same. However, this is not always thecase. If the same sentence (1) is said in acontext where the interest is concentrated onShakespeare's qualities as a philosopher, then thesentence message of (1) is still "Shakespeare wasa great playwright" because no change is broughtabout by the mind in the sign messages (ineveryday terms: every word is used in its literalmeaning) , but after processing the sentencemessage, the mind may produce as discourse messagesomething like "That Shakespeare was a greatplaywright doesn't necessarily mean he was a greatphilosopher" or "Since everybody acknowledgesShakespeare as a great playwright, that alreadyimplies that he was not a great philosopher". Evena factual sentence like

(2) Shakespeare was born in '564.

may in a suitable context carry the message of "Hewas not to be blamed for not knowing something",which is also produced by the mind afterprocessing the sentence message.

From this angle, the seemingly unexpectedmessage of (3), said in reference to WilliamGladstone (Allan 1986):

(3) The prime minister is an old woman.

128

5

If we divide the whole process of understanding anutterance into 3 stages of sign message assignment(I in FIGURE 2), sentence message assignment (II)and discourse message assignment (III) , thensentences (1) and (2) in their literal meaning areinterpreted already at I, passing through II andIII unchanged. When (1) is used to carry themessage that Shakespeare was not a great philoso-pher and (2) that Shakespeare may be excused fornot knowing something, they are interpreted intheir literal meanings at I, then pass II withoutany change, because as individual sentences theymake good sense. But at III, (1) and (2) in thisinterpretation would seem irrelevant and thereforethey are reinterpreted in accordance with theCooperative Principle, and discourse messages areinferred for them. A literal interpretation of(3), however, is blocked at II by the clash be-tween the male subject and female predicate. It issent back therefore and goes through I once againwith old woman treated as one sign, the resultantinterpretation goes through II and III unchanged.

From the above analysis we can see that theunderstanding of an utterance, generally speaking,has to go through the 3 stages of I, II and III,and that II serves as a check on the results of I.In the following I will discuss the interpretativeprocess from I through II, leaving III largelyundiscussed.

2.0 The Interpretation of Sign Meaning

We do not always have a one-to-onevelationship between sign form and sign meaningbecause of homonymy (see 2.1). So sign meaning hasto be interpreted, as a result of which we get asign message.

2.1 The Interpretation of Gramnatical Signs

The realization that the signifié of a word isdetermined by the linguistic system rather than bypreexistent ideas is a major contribution ofstructuralist linguistics. But when we give thesystematic nature of language too narrow anexplanation and demand that one and the samesignifiant must signify only one signifiti, we findourselves in great difficulty. What kind ofgeneral signifié can we give to light (not dark)and light ( not heavy) , flat (busted tire) andflat (apartment) , march and March, sun and air and

130

6

I

is not difficult to infer, because after thehearer has processed the term the prime ministerand concluded that the referent (WilliamGladstone) is a man, he refuses to take an oldwoman as a noun modified by an adjective, but goeson recursively to include old woman in one signand succeeds in obtaining a sign message: "a

person who complains too much and cares fortrivial things". Then the sentence message of"William Gladstone is a person who complains toomuch and cares for trivial things" goes throughthe mind and emerges unchanged as the discoursemessage.

On the basis of the analysis of thesesentences we can draw another diagram (FIGURE 2),which is a more detailed version of FIGURE 1,including the inferential processes at the levelsboth of single signs and of a whole sentence. Forthe sake of simplicity, we now take the hearer'spoint of view, presuming that in order tounderstand the speaker's encoding process, we haveonly to reverse the direction of the arrows.

FIGURE 2

FORM

r

i

1

CONTENT

129

son and heir?In other words, the theory leads us to think

that there should not be homonyms, and yethomonymy abounds. In Latin inflectional endings,we have -ae meaning genitive and dative singularand nominative plural ; -i meaning genitivesingular and nominative plural; and -o meaningdative and ablative singular in the firstdeclension; -is meaning dative and ablative pluralin the first and second declensions; etc. Amongmodern languages, very much the same is true ofthe Russian inflectional endings. In English, the/-s/ ending, attached to nouns and realizable as[-s], [-z), (-iz), has two different meanings:"plural" and "possessive".

Looking at the actual use of language, one hasto admit that homonymy is not incompatible withthe systematic nature of language. This is becauselanguage is used by human beings, who areconstantly exercising their intelligence vis-à-visthe outside world. Homonymy results when one andthe same form participates in different systems orin different specificity levels of the samesystem, or when a form is assigned a temporary,pragmatic, system. Human beings overcome thedifficulty created by homonymy by differentiatinglinguistic systems, specificity levels andpragmatic systems. This is where the mind comesin

Jakobson (1949) approached this problem on thephonological level when he discussed "overlappingphonemes":

In Danish this opposition strong/weak is implemented, for example, by tvs. d in strong position, and d vs. inweak position, so that the weak phonemein the strong position materiallycoincides with the strong phoneme in theweak position...if one should measurethe sound matter without reference tothe rule of dichotomy imposed upon it bylanguage, the conclusion would be thatthere are "overlapping" phonemes, in thesame way, as a physicist with hisacoustic instruments, according to H.Frei's felicitous comparison, fails toexplain why, in a given piece of music,F-flat and E represent two differentvalues.

1 31

8

In grammar, it is often not difficult todifferentiate the relevant systems. For example,by relying on the p/-s opposition in nouns werecognize the number system, by the same 0/-sopposition in verbs we recognize the person systemof present tense verbs, and by the p/-s (which isinterchangeable with of NP) we recognize-tUe casesystem in nouns. As a result of the distinctionof these three systems, we disambiguate the three-s's, and arrive at the sign message carried by -sin a given context.

Movement along the specificity scale in

grammar is also a common phenomenon. For example,the meaning "possessive" as expressed by /-s/ is ageneral concept which may be presented as ahierarchical structure, as shown in FIGURE 3.

FIGURE 3

"possessive"

ownership

by legal right

by right of creation

temporal possession without ownership

The general concept "possessive" is not onlyrequired by the theory to match the signifiant/-s/, but also for the interpretation of the signmeaning at the lowest level of specificity. In asentence like:

(4) The police have banned the suspect's things.

the general possessive meaning is applicablerather than any or all of the specific possessivemeanings listed in FIGURE 3, because some of the

132

9

It is difficult to imagine a signifié in which allthese meanings are combined together. However,presumably in a pair of sentences sucl. as:

(6) God be blessed. Here comes the man.

most of the oppositions (indicative vs.subjunctive, active vs. passive, third person verbform vs. non-differentiated person verb form)would be activated by the verb forms of be -ed and-s. But often only one of these oppositions isactivated in actual language use. In thesentence:

(7) His father worked in a bank.

only the tense opposition (past vs. present) isactivated. That is why, hearing such a sentence,one would tend to ask: "Where is he now?" or even"Is he dead?" In the sentence:

(8) I said he comes everyday, not he shouldcome everyday.

the opposition activated is indicative vs.subjunctive.

The actually activated sign meaning, then,represents the sign message that the hearerobtains through the processing of the sign usingintelligence.

2.2 Lexical Systems

The complexities of meaning relationships inthe lexicon have deterred linguists from talkingabout lexical systems. However, if we do notexpect lexical systems to present an exhaustivelist of one-to-on:2 form-meaning combinations, butare prepared to find systems which intersect witheach other, move along the specificity scale andmay be realized as ad hoc pragmatic systems, as wefind in grammar, then we can see that lexicalsystems share the same characteristics asgrammatical systems, showing only much greaterflexibility.

A good example of how lexical meaning isassigned by the system is provided by better.Instead of "gooder" or "more good", as thecomparative form of good is supposed to mean, theactual meaning is very often "less bad": Thepatient is better, but still seriously ill; The

1 34

10

forms of possession may not be implied (e.g. thesuspect may not have created anything), while someother forms of possession not enumerated in FIGURE3, e.g. possession of stolen things, may beimplied. However, consider a sentence like (5):

(5) Each of the painter's works was worth afortune already in his life time, but hedied in poverty.

In this case, the general concept of possession isnot applicable, so /-s/ has to be interpreted atthe higher specificity level of "ownership byright of creation".

The inflectional endings of finite verbs aretraditionally said to carry a whole series ofmeanings. The pair /-s/ and 0, for example,shares the meanings: "indicative, active,present", then repectively /-s/ implies thirdperson singular subject, and 0 first or secondperson singular and all three persons in theplural. These meanings are hierarchical as can beseen in the following diagrhm:

FIGURE 4

tnood

Indicative

active 1

present

past

singWar 1

plural

1 p. 02 p. 03 p. Vs/

1 p. 02 p. 03 p. 0

subjunctive

Imperative

passive

1 1

133

situation has become better, but still verydangerous. Although this may be usual for otherEuropean languagestoo, the two meanings aredifferentiated in Chinese: gen hao for "gooder",but hao xie for "less bad":

(9) bing ren hao xie le buguosick person good some PERF but

yijiu bing de hen zhong.still sick PRT very heavy

"The patient is better, but still seriously ill."

(9a) * bing ren gen hao le buguosick person better PERF but

yijiu bing de hen zhong.still sick PRT very heavy

It is therefore reasonable to suggest that themeaning of better is not "gooder", but "in animproved state", i.e. it forms an opposition inthe direction of improvement with some statealready referred to. Then worse would mean anopposition with a certain state in the directionof deterioration. Therefore, the weather can besaid to have become better no matter whether it iswarmer or colder, dryer or wetter. It can also besaie to have become worse in the same atmosphericconditions. The opposition is between a state andan altered state which is thought to be animprovement in the case of better, and a

deterioration in the case of worse.As meanings are determined by systems rather

than actual physical qualities or relations, a

sign form acquires multiple meanings when it takespart in forming oppositions in more than onesystem. From this angle, it is not difficult tounderstand the multiple meanings of light, forexample. When it forms an opposition with heavy,a system is formed in which we have a binarydivision of the semantic field of weight, andlight means "the opposite of heavy". Likewise,when the opposition is formed with dark, lightmeans "the opposite of dark".

The systematic natui:e of lexical meaning,however, will be best shown if ttoo identical wordsform or participate in forming differentoppositions based on the sime kind ofrelationship. In such a case, we can attribute

135

12

the difference in meaning to nothing other thanthe systems, i.e. the oppositions in relational,not material terms. We have a good example inblack and white. First, there is a system ofrelative meaning consisting of the opposition ofthese two terms: a binary division of a semanticfield of color. Thus, coffee is said to be eitherblack or white while actually it can-only bedifferent shades of brown. For the samereason--that in this system only a binary divisionis permitted-- day is always said to be white inChinese, night is black probably in all languages,but when there is need to talk about a night thatis not that black, as in Leningrad in summer, oneis left with the only alternative: a white night.On the other hand, black and white can certainlydenote actual, "physical" colors. In this case,they form a different, quite separate, system withother names for colors, such as red, orange,yellow, green, blue, etc.

With names for concrete things, thespecificity levels of a system are of paramountimportance. When we tan( about streets full ofcars, we are on zhe levr4l of means oftransportation in a city. The system concerned ismainly an opposition between cars and pedestrians.When we say a typical American family would own acar and a wagon, we are on a level of higherspecificity, using a system in which car isopposed to wagon, limousine, bus, etc. When myremark you've got a nice car evokes the responseYeah, it's a pretty good pick-up, I am beingreminded of the need to shift to a level of higherspecificity. The word book is used in its lowspecificity meaning in the sentence The room isfull of books, where iournals, pictorials, etc.are not differentiated from books. When we sayBooks are on the second floor and journals are onthe third floor, a higher specificity meaning ofbooks is evoked, in which -ournals are notincluded. Finally, when we say The accountant isbusy with his books, the book refers morespecifically to definite kind of book. It isalso due to system shifting that we have nodifficulty in determining that All men are bornequal does not exclude women from equalprivileges.

One striking feature of lexical systems isthat they are often 'ad hoc'. For example, peopletend to speak in terms of cardinal colors. Whenan orange colored jacket is referred to as a red

136

13

'acket (Do you see the man in a red jacket?), themeaning of the signal is that it is not white, notblack, not grey, not blue, not maize, in a wordnot the color one would usually think a jacket maybe, except what is referred to as red. The hearerwould match all the jackets in sight and decidethat the orange one fits best. So he wouldunderstand the speaker as referring to the orangejacket. Here we are said to be using red in arough sense.

2.3 The Interpretation of Lexical Signs

From the brief description of lexical systemsabove, it is clear that when we use ourintelligence to interpret a lexical sign we aretrying to determine the opposition it forms, thespecificity level at which it is used, and/or thead hoc meaning it acquires in the context. That is

to say, we go along much the same tracks as whenwe interpret grammatical signs.

3.0 The Role of Inference in Sign Interpretation

In the above we have tried to show that theinterpretation of both grammatical and lexic;a1signs can, in the final analysis, be described asthe determining of the system of oppositions thatis contextually activated. Now we go on to showthat logical inference underlies this process ofdetermination whenever ambiguity results fromhomonymy or lack of clarity in general.

3.1 Logical Inference at all levels ofLanguage Use

It has been mentioned in 1.1 that the Griceanpattern of the working out of conversationalimplicatures is at work at the various levels oflanguage use. Here we quote Grice's ownformulation: "He has said that g; there is noreason to suppose that he is rot observing themaxims, or at least the CP; he could not be doingthis unless he thought that g; he knows (and knowsthat I know that he knows) that I can see that thesupposition that he thinks that g is required; hehas done nothing to stop me thinking that q; heintends me to think, or is at least willing toallow me to think that g; and so he implicatedthat g." (Grice 1967)

It remains for us to illustrate the process

1 37

14

at the various levels of phonology, lexicon,syntax and discourse.

3.1.1 Inference at the Phonological Level

Like all other linguistic systems, thephonological system is not free from homonymy.The neutralization of the voiced and voicelessconsonants word-finally in such languages asRussian and German is a point at issue. Do wehave archiphonemes here, or a differentphonological system in _# position, or adifferent phonological systei wherever there is achange in the system? From the view thatlinguistic systems undergo changes in languageuse, the loss of voicing in word-final positionresults simply in a variant of the phonologicalform of the word to be interpreted, throughinference, as linguistically the same as the formwith the corresponding voiced consonant. Theinferential process briefly is like this: he hassaid [sat], but his utterance is meaninglessunless he means sad (garden), it is usual for theRussians to 1,ronounce the voiced consonants astheir voiceless counterparts, there is nothing toprevent us from taking (sat) as sad, thereforewhat he means is sad.

There are two remarks to make here:(1) The inferential process is, needless to say,unconscious, more so at the phonological levelthan elsewhere. Hearing (rot], a native speakerof Russian would assign it either to the morpheme/rod/ "species" together with other forms such as(rode) "species-GEN", (rodu) "species-DAT", etc.,or to the morpheme /rot/ "mouth" together withother forms such as (rta) "mouth-GEN", [rtu]"mouth-DAT", etc . He does this as ifinstinctively, but in the final analysis theassignment can only be the result of a judgmentbased on the fitness of either paradigm to thesemantic structure of the sentence.(2) The conditions of neutralization (word-finalposition and others) reflect the regularities ofthe pronunciation of native speakers, but they arenot essential for understanding. If only theminimal data (grammatical, lexical and discoursal)are there for inferences to go on, hearers canmanage to arrive at the message signalled withoutknowing the conditions for neutralization.

In many Chinese dialects, n and 1 do notcontrast, although they do in Mandarin.

138

15

Therefore, it is not uncommon for speakers ofthose dialects to say:

(10) ging chi ni, zhe ni hen hao chiplease eat ni, this ni very good eat

which means, for Mandarin speakers, "Please eatthe mud, this mud is very delicious." Here, theGricean inferential process would be triggered:"He has said ni 'mud', he could not be doing thisunless, because of the non-differentiation of niand li in his dialect, what he meant was li'pear', he has done nothing to stop me thinkingthat he meant 'pear' instead of 'mud', so he musthave meant 'Please eat the pear, this pear is verydelicious'."

Inferences may also be triggered by intona-tion. For example, in the question-answer pair:

(11) A: Where's the spaghetti sauce?B: On the shelf (rising).

the rising intonation after "On the shelf"primarily signals the speaker's uncertainty.However, when uncertainty is excluded (e.g. if Bis a housewife and has just been using the sauce),then the usual process entails: since B cannot beuncertain about the whereabouts of the spaghettisauce, the rising intonation can only mean anavoidance of the possible implication of self-assertion or rudeness of the falling intonation.Hence women's preference for it.

3.1.2 Inference at the Lexical Level

In the phrases in a bank and on the bank, theword bank is not homonymous from a systematicview. But the sentence:

(12) His aim was to consolidate the bank.

is ambiguous. What we do in order to disambiguateit is to make an inference on the basis ofcontextual knowledge, e.g. if, for example, thegeneral topic is to make preparations before therainy season, the sentence is not meaningfulunless it means the strenghthening of the riverbank. Again, because this inferential process isso familiar, it is usually thought that just bypointing to the context the word bank would bedisambiguated. Actually, the jump in conclusion

1 39

from preparations against flooding to theconsolidation of river banks can be made onlybecause the inferential process lies behind it.This can be shown by the fact that if it should sohappen that in the context the prevention of thecollapse of a financial institution is essentialto making preparations against the flood, thesentence may still mean the strengthening of thefinancial institution.

It is well-known that in irony or otherinstances of language use where strong emotion isinvolved words are often used to mean theiropposites: poor devil, a fine little beggar, etc.This explains why, when the speaker apparently hasno reason to like the person he refers to, theinferred message of a sentence like you are a niceguy is just the opposite of what it means.

Inference is indispensable in interpretingidioms and set phrases. One does not actuallyhave to, and an advanced learner often does not,consult a dictionary to learn a phrase like havesomething on one's conscience, although it may beuseful for dictionaries to contain it forconfirmation. What a learner usually does is toinfer: since conscience is not a substantialthing, it cannot have anything on it unless in afigurative sense and since conscience means aperson's natural goodness, to have somethingburdening it can only mean, figuratively, "to feelguilty for something".

The sentence He offered her his hand has to bedisambiguated, and, theoretically at least, onlywhen both the sense of offering her a hand to beshaken and the interpretation of offering her helphave been -..xcluded, can the idiomatic meaning of aproposal of virriage be decided on.

The :a-aiung of some idioms cannot be inferred,but a learner has to go through the process ofexcluding the lit,ral interpretation anyway. Thus,only after making sure that the literal meaning ofkicking a bucket makes no sense in the context,can a learner turn to a dictionary or to someoneelse for the meaning of kick the bucket. In thecase of It's raining cats and dogs, nointerpretation can be worked out to be excluded,but one has at least to go through the process ofreasoning that "literally the sentence doesn'tmake sense, so it must have some idiomaticmeaning".

In this way it is not surprising that acyclist should, as reported in Bolinger (1981),

140

17

Or

think of saying, with a twist of his wrist:

(13) I find that the easiest way to shift gearsis just to kick the trigger.

A hearer would interpret his kick as containingall the semantic feat) res of kick with theexception of (+foot), and consequently substitute(+wrist] fur it. This would be the result of areasoning process: "No one could convenientlyshift gears with his foot, therefore he could nothave said kick unless he meant a movement of hishand, he indicated this with a twist of his wrist,therefore what he meant was 'to make a movementwith one's hand that is like a kick'."

3.1.3 Inference at the Grammatical Level

Inference disambiguates homonymous grammaticalstructures and interprets grammaticalirregularities, just as it does lexical items.Disambiguation through inference begins with theidentification of the inflectional endings. Afterall, how do we know that in the Russian sentence:

(14) Ljublju otsa djevushki.love-lsg father-ACC girl-GEN

or its Latin equivalent:

(15) Puellae patrem amo.girl-GEN father-ACC love-lsg

dievushki and puellae, which according to theirform may either be genitive singular or nominativeplural, are in the genitive case, except byreasoning: since they could not be in thenominative plural, they must be in the genitivesingular (these two being the only possiblechoices)? Likewise, whether the Russian sentence:

(16) Otsa djevushki ljubjat.a. father-ACC girls-NOM love-3p1b. father-ACC girls-GEN (people) love-3p1

or the equivalent Latin sentence:

(17) Patrem puellae amant.a. father-ACC girls-NOM love-3p1b. father-ACC girls-GFN (people) love-3p1

141

18

means "the girls love (their) father" or "(People)love the girl's father" can only be worked out, oncontextual data, by applying the formula: either 2or g, since not 2, so g. When the choice is to bemade from more than two possible terms, then:since not 2, nor g, nor r ..., so s.

Inference, or rather inference based onlinguistic knowledge and contextual data, can alsodeal successfully with the interpretation ofdeviant forms of grammar. For example, in thefollowing two dialogues:

(18) Victoria: All the servantsnow.

Frederick: They haven't!--W.

have given notice

Somerset Maugham

(19) Sally: I've got a job.Olive: You haven't!...What's the job?

--John van Druten

haven't is ambiguous between a negation of thehave in the previous sentence and an emotionalresponse to this have by way of meaning theopposite of the negative form (e.g. They haven'tmeans They have! cf. nice guy -- bad guy in3.1.2). Contextual data--in Maugham's dialogue,Frederick has just come back, while Victoria hasbeen home all the time; in van Druten's Olive goeson to ask "What's the job?"--removP thepossibility of a literal interpretation of haven'tas the negation of have, therefore both Theyhaven't and You haven't in these two examplescarry the message of great surprise: "They have!"and "You have!"

This is not intuitive interpretation, becausenot only is the inference conducted in accordancewith the Gricean formula, but it is based on thelinguistically viable commutation of zhe semanticopposites.

3.1.4 Inference at the Lexico-Grammatical Level

In contemporary linguistic literature, therehas been a long tradition of regarding grammar asthe sole embodiment of the systematic nature oflanguage, leaving the lexicon a passive role ofnonsystematic, linguistically uninterestingcarriers of grammatical relations. Hence thediscrepancy between the learned view of the studyof language as mainly a grammatical discipline and

142

19

the popular idea of language as words and theiruse.

It will be remembered that Saussure in hisCourse in General Linguistics illustrates thenature of the linguistic sign and linguistic valuemainly with examples from the lexicon--words. Whythis practice was not followed in linguistictradition is not clear, but it may be a result ofthe attempt to have as little to do with meaningas possible and the abhorrence for concepts.However, if we take a look at Chinese grammarvithout preconceptions about autonomous syntax, wewill see that che argument for an active role ofthe lexicon is quite strong. In this section weshall see how lexical meaning may, throughspeakers' capacI*11 for inference, determine themessage of an utterance in default or even inspite of the grammatical structure.

As is well-known, there are many sentences inChinese which are either subjectless orobjectless, or both, and there are sentences whichare normally either active or passive. Suchsentences are usually said to be interpretedpragmatically. However, pragmatics as an outletfor difficulties in formal linguistics is growingunmanageable. One would like to see its scopenarrowed, and this can be done to a considerabledegree if it can be shown that signals to theexercise of intelligence and understanding maycome from the lexicon, as well as from thegrammar.

Take, for example, a sentence like:

(20) haizi bao qu le.child embrace go PERF

The grammar of Chinese produces an ambiguity ofinterpretation: either "The child has been carriedaway", or "The child has carried (sb./sth.) away".It has to be disambiguated according to whether itis said in anF,wer, for example, to "Where is thechild?", or "Where is the cat?" If the whereaboutsof the child is in question, then the inferencepresumably goes like this: "Since a qustion hasbeen asked about the child, haizi refers to thischild rather than any other child; and since thereis nothing in the context saying that the childhas carried away something or someone else, themessage must be: the child itself has been carriedaway." In case the whereabouts of the cat is thepoint of interest, then haizi bao au le is not

143

20

meaningful unless the action is performed by thechild toward the cat and not the other way around.The verb bao (carry by embracing) also plays itsrole, since it is typical of the way a child iscarried away or a child carries a cat. Therefore,the inferences are triggered by lexical items.

Such structures are regular, rather thanexceptional, in Chinese. Here are some furtherexamples:

(21) ji chi diao le.chicken eat away PRT(The chicken has been eaten up, or:The chicken has eaten up something)

(22) mama jiao guo le.

mummy call EXPER PRT(Mum has called, or: Mum has been called.)

(23) ni qu yiyuan kan guo le ma?you go hospital look EXPER PRT Q-PRT(Have you been to the hospital to have a look,or: Have you been to the hospital to be lookedat (to see the doctor)?)

When one of the interpretations ispragmatically not permissible--usually the activeone--then the sentence is unambiguous with itswrong grammatical form and no context is requiredfor the passive interpretation:

(2,4) xin yijin shoudao.letter already receive

"The letter has been received."

(25) hua jiao guo le.

flower sprinkle EXPER PRT"The flowers have been sprinkled."

(26) cheng gong po le.

city attack broken PRT"The city has been broken open under attack."

In examples (24-26) we have strong evidence thatsometimes lexical meaning may have precedence overgrammatical structure in determining sentencemeaning. These sentences not only have a passiveinterpretation, but the passive particles beigei, etc. are normally not used in them, exceptwhen the occasion is thought to be in some senseout of the ordinary. Thus (26) , being an out-

21144

standing event itself, can readily be turned into:

(26a) cheng bei gong po le.city PASSIVE attack break PRT

"The city has been broken open under attack."

(25) can have the passive particle only when someadditional meaning has been put in to make theoccasion noteworthy, for example:

(25a) hua qei yong kaishui jiao guo leflower PASS use boiling water sprinkle EMP PRT

"The flowers have been sprinkled with hot water."

As for (24), it is difficult even to imagine anoccasion when the passive particle would be used.

One may of course provide pragmaticexplanations for both the ambiguous andunambiguous sentences above. But, for on.! thing,it is advisable to avoid throwing everything intopragmatics. More importantly, a lexico-grammatical, rather than pragmatic, interpretationhelps to resolve two difficult problems aboutChinese grammar: 1. the extremely frequentoccurrence of topicalization, for which there isoften no particular motivation, and 2. theexistence of what may be considered the reverse oftopicalization: subject in object position. Forexample,

(27) tai shang zuo zhe zhuxituan.platform on sit STAT presidium"The presidium is sitting on the platform."

(28) lu pang ting zhe yi liang che.road side stop STAT one CL vehicle

"A car is parked at the roadside."

The difficulty that sentences like (27) and(28) create for Chinese grammarians lies in thefact that there seems to be absolutely nothing tosay about these subjects (zhuxituan and che)sitting in object position in order to save a wordorder-based grammar from going to shambles, exceptto call tai shanq (on the platform) and lu pang(at the roadside) sr"ects, which some grammariansactually do.

With the active role of lexical meaning andthe inferences it triggers taken into account,however, Chinese grammar seems to be saying, in a

145

22 z

reasonable way: since Lao Li da Lao Wang and LaoWang da Lao Li ("John beat Jim" and "Jim beatJohn") are both possible events in this world andone has therefore to differentiate the one fromthe other, of course word order is important here;but letters are always received by someone andnever receive anyone; flowers are sprinkled on andnever sprinkle; men sit and not are sat on; carscan only be made to stop, so why bother about wordorder? Regardless of word order, simplicity isachieved in (24-26), and a special form fcrdescriptive statements, with the subject sittingafter the verb, is made possible, as in (27) and(28).

Another argument for an active lexical rolecan be found in sentences like (29), where thereis a typical collocation between noun and verb:

(29) laoying diao qu le.hawk hold in mouth go PRT."The hawk has snatched (it) away."

Although (29) is structurally ambiguous like (20),actually it can only mean an active relationshipbetween laoyina and diao because diao typicallycollocates only with laoying and huangshulanq(weasel). So much so that in answer to a questionlike "Where is the chicken?", some Chinese woulduse (30) , which, grammatically speaking, is thepassive form of (29), to convey the same messageas (29).

(30) laoying gei diao qu le.hawk PASS hold in mouth go PRT

Literally: "The hawk has been snatched away";actually: "The hawk has snatched (it) away".

Here the typical active voice collocation betweenlaoyinq and diao, together with the inference thatthis must be a statement about the whereabouts ofthe chicken and not of the hawk, succeeds inneutralizing the passivization functirn of qei andassigning it a new meaning of emphasis.

The potential collocation of a verb can alsoinfluence the interpretation of a sentence wherethere is no subject. Therefore,

(31) xia le (or) zai xia ledown (verb) PRT PROG down PRT

has the unmarked meaning of "Rain is falling",

146 23

because xia typically collocates with yu (rain).It also collocates with xue (snow) or peoplecoming down from a mountain or some other high

places. But these are marked cases where eitherthe subject should be present, or a context isrequired. In a cooking situation, xia is oftentransitive with an implied object (food to beboiled in water, especially noodles). Therefore(31) in the mouth of a waiter would mean: "Thenoodles are being put into the water to beboiled."

If we look at these transitivity relationshipsin Chinese as eventually determined by inferencebased on lexical information, then we finduniformity between transitivity and such othermajor categories as number and tense. As is well-

known, the category of number is absent in Chinese

nouns. Whenever necessary, number is expressedlexically, either directly with numerals ("threetable", "two person"), or indirectly withadverbials, deictics, etc. ("table alldestroyed","these book", "a half guest alreadyarrived" (half of the guests have arrived) and so

on). There is available a plural particle men,but its use, except in pronouns, is limited tonouns meaning humans--and even then not obligatoryexcept in a few functions, such as in an addresswhere numerals or deictics are absent. Likewise,

the category of tense is absent in Chinese ("Hedies at 1789 year"). There are a number ofparticles to be used if necessary: quo (verbperfect particle), le (verb or sentence perfectparticle), etc.

The influence of lexical items on sentenceinterpretation is not non-existent in English,though much rarer than in Chinese and thereforeneglected. The following set of sentences, forexample, is given in Fillmore (1972):

(32)A. Harvey viciously took advantage of Melissa.

B. Melissa was viciously taken advantage of by H.

C. Harvey willingly took advantage of Melissa.

D. Melissa was willingly taken advantage of by H.

The adverb viciously in both A and B relatesto Harvey's participation in the act, whilewillingly in C and D relates to the participationin the act of the individual in the surfacesubject NP. Fillmore (1972) explains as follows:

147

sent. =>

certain adverbs may be introduced into asentence as ways of qualifying oneparticipant's role in the activity...Thus Manner adverbs of the type vicious-ly may appear only in sentences havingunderlying Agents, the scope of theadverb being unaffected by the ultimatechoice of surface subject.

This is to say that manner adverbs of the typeviciously always qualify the underlying Agent'sactivity. We are still not enlightened as to whythis is so, and what type viciously represents.

From the viewpoint of the influence of lexicalitems on the interpretation of grammaticalstructures, A, C, and D have interpretations, aspredicted by the grammatical structure. Thepredicted interpretation for B, however, isblocked by the incompatibility in meaning of theadverb viciously with the patient of the act oftaking advantage. The only alternative is torelate viciously with the.agent of the act. Theresulting interpretation works. So, it is nothingbut a process of: since not R, then g.

3.2 A Model for Inference in Sign and SentenceInterpretation

In order to r-ovide a unified model to explainwhat mechanism _Ls behind the performance of theinferential processes at the various levels oflanguage use discussed in 3.1, the following issuggested:

FIGURE 5

assign. ofsign message

grammaticalphonologicalknowledge

11A IIB

semantic collocation-check check

HC

pragmatic

check

worldknowledge - -

linguistic

knowledge

contextualknowledge

148

25

Here II, the working out of the sentence message,is seen as a check on the interpretation reachedat I. The interpretation is checked semantically,collocationally, and pragmatically. If it fails topass a chec:. (when no coherent message isobtained) , it is sent back through the dottedcurved line to I to be processed again. When acheck is successfully passed, the output, just asthe output of the sign assignment process at I,can only be either an unambiguous or an ambiguousinterpretation. The former is represented by >and the latter by 7.::)> (which indicates not only

2-way, but 3-way or multiple ambiguity). The

exclusive.nfarrows and are therefore mutually

erences take place at the variouschecks on the basis of the output of the previousanalysis or check with the help of the knowledgetypical of the current check: world knowledge atthe semantic check, linguistic knowlege at thecollocational check and contextual knowledge atthe pragmatic check. It is normal for the outputof the semantic or the collocational check to be

either ambiguous or unambiguous no mattter whetherthe input is ambiguous or unambiguous. The output

of the pragmatic check is expected to beunambiguous because we have reached the end of theprocedure of interpretaticn. If an ambiguousinterpretation nevertheless results, either alarger context is needed for its interpretation,or the sentence may be judged as incoherent.

A non-ambiguous sentence used in its literalmeaning, such as (1) (Shakespeare was a greatplaywright), goes through all the checks without

any changes. If however the sentence is used to

imply that Shakespeare was no great philosopher,then the hearer will have to go through a Griceanprocess of inference at the pragmatic check (IIC),triggered by the sentence's lack of relevance tothe topic in its surface meaning. He will inferthe implicated message on the basis of hiscontextual knowledge that Shakespeare's greatness

as a philosopher is the topic of the conversaion.The structural analysis of the Russian

sentence (eto sat) will give at I, in consequence

of the phonological rules concerned, either etosat or eto sad. The first variant is rejected by

the semantic check (IIA) as meaningless, thereforewe have eto sad 'This is a garden'.

The dialectal Chinese sentence ging chi nigives two variant interpretations both at I andIIA, because 'Please eat the mud' and 'Please eat

149

26

the pear' are both grammatically and semanticallynormal, but the collocational check (IIB) willbias the hearer strongly against the former.Therefore IIB sends to IIC a still ambiguous butbiased (towards the 'pear' interpretation), whichIIC eventually disambiguates in favour of 'pear'if there is no contextual evidence against it, butpossibly also in favour of 'mud when contextforces this interpretation.

(11B), On the shelf (rising), will producetwo interpretations at I: (a) "On the shelf?"(normal interpretation for rising tone), (b) "Onthe shelf + non-assertive tone (a sociolinguisticconnotation of rising tone)". Both will go throughIIA and IIB, until contextual knowledge excludesone of the variants through inference, e.g. if thespeaker is a housewife in her own home, besidesshe has just used the spaghetti sauce, then themessage cannot be (a), it must be (b).

Sentence (32) (His aim was to consolidate thebank), as already mentioned in 3.1.2 , has to bedisambiguated at IIC.

He has something on his conscience gives twointerpretations, one literal and one idiomatic, atboth I and IIA. The collocational check IIBdecides on the idiomatic exp lanation - -thecollocation of to have something on one'sconscience is only used in its idiomatic sense.He offered his hand, on the other hand, can onlyget some bias at IIB and has to be disambiguatedat IIC through inference with consideration ofactual contextual circumstances.

They haven't! in (18) is unambiguous at I, andremains so through IIA and IIB, until thisunambiguous interpretation of a negation of theprevious statement is rejected by IIC becauseFrederick is just back home, and therefore hecouldn't know better about the servants thanVictoria, who has been with them all the time. It

is sent back to I, where the lexical knowledge of

the commutation of antonyms in an emotional stateprovides the interpretation "They have!", whichsucessfully passes all the checks as theimplicated message of the sentence.

Xin yiiin shoudao (24) is an instance of howthe two-way ambiguous interpretation at I is

disambiguated at IIA with the help of worldknowledge: since a letter never receives anyone oranything, it must have been received.

1 50

27

4.0 Conclusion

It is certainly very important to study thelinguistic systems, because without a clearconception of these systems, it would beimpossible to really handle the mess that islanguage use. However, linguistic systems aloneare not enough to explain the verbal activities ofhuman beings. Humans use their intelligence incommunication, hence the importance of signinterpretation and logical inference in languageuse. The fact that the same kind of logicalinference is at work at all levels of language usemay be seen as a key to the problem that has beenpuzzling linguists for so long: why language is atthe same time so difficult for experts toformalize, yet so easy for children to learn.

Acknowledgment

This is a partial version of my paper"Linguistic System, Lexical Meaning and LanguageUse," written at the Chinese University of HongKong in Spring 1990. I am indebted to the CUHKEnglish Department for funding this research.

References

Allan, K. (1986). Linguistic Meaning, Vol. 1 and2, London: Routledge and Kegan Paul.

Austin, J.L. (1962). How To Do Things With Words.Oxford: Clarendon Press.

Bolinger, D. (1981). Aspects of Language, 3rdedition, New York: Harcourt Brace Jovanovich.

Fillmore, C.J. (1972). "Subject, Speakers, andRoles", in Davidson and Harman (eds.),Semantics of Natural Language, Dordrecht:D. Reidel Publishing Company.

Grice, H.P. (1967). "Logic and Conversation", in

P. Cole and J.L. Morgan (eds.). Syntax andSemantics, Vol.12. New York: Academic Press.

Jakobson, R. (1949). On the Identification ofPhonemic Entities. Selected Works I, 418-25.The Hague: Mouton.

Saussure, F. de (1916). Course in GeneralLinguistics, translated by Wade Baskin.London: Peter Owen, 1960.