4nn. Rev AnihmpoL 1981 ia27-62Copynght'S- 1981 by Annual Reviews Inc Ail rights reserved

MEANING-TEXT M O D E L S : •9670A Recen t TTend in Soviet Linguistics

Igor A. MeVcuk

Departemetit de lmguistique, Umversite de Montreal, Montreal H3C 3J7,Quebec, Canada

SOME HISTORY AND GENERAL REMARKS

The Meaning-Text Theory (henceforth MTT) was put forward by Alexan-der K Zholkovsky' and the present wnter in 1965 (86), 2 years later, amajor presentation of the theory appeared (87) and was soon translated intoEnghsh and then into French Shortly afterwards, Junj D Apresjan joinedus, and thus was formed the nucleus of what was called the MoscowSemantic Circle Over a 10-year penod, some 20 people contributed to thework on a Meaning-Text Model of Russian Basic g^ieral readings, inaddition to the two titles just mentioned, include (45,47, 50-52, 56,67, 88)A number of papers and books dealing with more specific topics will beindicated later

MTT did not flounsh officially m the USSR Its adherents were onlyrarely and reluctantly admitted to conferences and colloquia, it was nevertaught at any of the major umversities; and the leading professional journalsdid not accept papers on MTT for publication The main reasons were thatit was so un-Marxist, with its stress on establishing formal correspondences

'There are problems m transcnbmg Russian names into English For example, Zholkovsky'sname will be wntten ^olfcovsky whtn it is transhterat«d du'ectly from Cynlhc type, moreover,vanous erraac spdlings are found in American publications, e g Zholkovsku, or ^olkovsky,which had to be adopted in the literature cited The same applies to Apresjui (sometimesspelled Apresyan), atid my own name {MeVduk-Melchuk-MeVchuk, etc)

27

28 MEL'CUK

between meanmg and sound (thus separatmg both), its extensive use ofmathematical and quasi-mathematical apparatus, its heavy rehance onWestern hngiustics, and so on Later, the situation detenorated sharplywhen the founders of the trend participated in a letter campaign protestmgthe Sinyavsky-Daniel tnal and then failed to take the "nght" stand towardthe Soviet invasion of Czekoslovakia Eventually I was forced to leave theUSSR forever (the mterested reader can find more details in the Bntishreview Survey, 1977-78,23 2,126-40,1979,24 2,213-14) Two years later,Zholkovsky followed me At present, there remain a few supporters of theMeanmg-Text theory grouped around Apresjan m Moscow My personalhope IS that the theory will recover from its transplantation to NonhAmenca and perhaps be ennched through dn-ect contact with Amencanlinguistics

Conceived and developed as a general theory of human language, Mean-mg-Text Theory is based on the following two postulates

Postulate 1 Every speech event presupposes three mam components{a) content or pieces of information to be commumcated, which are calledmeaning(s), (b) certam forms or physical phenomena to be perceived,which are called text(s), (c) a many-to-many correspondence between aninfinite set of meanings and an lnfimte set of texts, which constitutes lan-guage proper (or "language in the narrow sense of the term") This postulatecan be diagrammed as follows

1- l a n g u a g e p r o p e r , ^

MEANING j ^ j - < =^-{_TEXT V

LANGUAGE

A natural language is thus viewed as a logical device which estabhshes thecorrespondence between the infinite set of all possible meanmgs and theinfinite set of all possible texts and vice versa This device ensures theconstruction of lmguistic utterances which express a given meanmg, l espeakmg, and the comprehension of possible meanmgs express^ by a givenutterance, l e the understandmg of speech

Postulate 2 Hypotheses about devices of the type illustrated in Postulate1 can be formulated as functional^ or cybernetic models, with the actual

^The lenn functional, which is now a buzz word in Imguistics, as used here has nothmg todo with functional sentence perspective (topic-comment problems) nor with grammaticalfunctions (such as "subject of," "object of," etc) We say that X is a functional or cyberneticmodel of an object Y if and only ^ X is a system of rules simulatmg the functioning, the"work," or the behavior of Y, with no claims as to its observable structure

MEANING-TEXT MODELS 29

language considered a "black box" where only the inputs and outputs canbe observed but not the intemal structure Such modek (called MTM) aresystems of rules approximating the Meaning 4=4 Text correspondence Thischapter will discuss only one specific class of MTMs

Proceedmg from Postulate 1 and 2, the MTM can be characterized bythe foUowmg important properties

First, the MTM is not a generative but rather a translative or purelytransformational system It does not seek to generate (enumerate, specify)all and only grammatically correct or meanin^ul texts, but merely to matchany given meamng with all texts having this meaning (synonymy) and,conversely, any given text with all meanings the text can have (homonymy)

Second, the MTM is no more than a fragment of the full-fledged modelof human lmguistic behavior

Rea l i ty r ^ ^ l Meanings>^=^^ Texts V <=^ < L i n g u i s t i c Sounds

I II III

Only Fragment II, i c Language, is deemed to be the subject of linguisticsproper and should be represented as an MTM Fragment I is the subjectof vanous fields such as philosophy and psychology, including what is calledartificial lnteUigence Fragment III is the subject of acoustic and articula-tory phonetics However, even Fragment II is not represented by MTM infull To simplify our task, we made abstraction from a number of relevantaspects, properties, and phenomena of natural language—m the hope thatwe will obtain a clearer and more insightful picture of what remams underour lenses This is the only justification for many dehberate omissions

In their actual form, MTMs observe the foUowmg six limitations

(l) Functions of natural language other than the communicative oneare not considered at all Language is treated exclusively as acommumcation system

(ll) The extremely important problem of language acquisition and de-velopment IS deliberately ignored

(ill) No attonpts have been made so far to relate the MTM expenmen-tally with psychological or neurological reality An MTM is no morethan a model, or a handy logical means for descnbmg observablecorrespondences

(lv) The correspondaice between meanings and texts is presented stati-cally, 1 e M a correspondence between some elementary fragments

30 MEL'CUK

of meaning and equally elementary fragments of text The proce-dures for actually movmg from complex meamngs to texts and viceversa, or the operations that have to put together those elementaryfragments to produce actual hfe-size representations, are left out ofconsideration because they are believed to he outside the domamof lmgmstics ^ For this reason, within the MTM the problem of ruleordenng does not anse

(v) The possible feedback between meamngs and texts in the actualprocess of speakmg or hstemng (changes m the ongmal messageunder the lnflu^ice of the text already constructed and put out, etc)IS not taken mto account

(vi) The anal^is of the meaning itself go« beyond the scope of anMTM, a different type of device is needed for this purpose (By thisI mean the detection of semantic anomalies such as contradictions,tnvialities, absurdities, etc ) Thus, if confronted with such deviantmeamngs as "Colorless green ideas sleep furiously" or "John ate upthe sincenty of his car," the MTM should provide for them perfectEnghsh (or French, or Russian ) sentences So it should be ableto detect the formal, or grammatical, anomahes of a text but it doesnot deal with the semantic ones (This is done on purpose in orderto reflect the essential asymmetry of meanmgs and texts)

Only a very sketchy descnption of the MTM can be given I do notmention any other feasible solutions and draw no parallels with the closelyrelated models proposed by transformational grammar, generative seman-tics, stratificational hnguistics, case grammar, and other more recent devel-opments, lndudmg relational grammar Suffice it to state that all these havesignificantly influenced the author, a qualified reader will easily recognizesome common pomts and perceive the differences

UTTERANCE REPRESENTATION LEVELSIN THE MEANING-TEXT MODEL

I assume that both meamngs and texts can be represented m an exphcit,discrete manner, using formal languages devised specially for this purposeAs for texts, this is more or less obviom phonetic tratiscnptions—or, forthat matter, ctmventional spellmg systems—have existed for a long timeThe approximation they ensure for the continuous acoustic flow of speechIS generally d^med satisfactory Though no semantic transcnption existsfor meanmgs, nothmg should prevent us from believmg that such a tran-

%ut a MT model does of course enumerate the possible kmds of changes between adjacentlevels

MEANING-TEXT MODELS 31

scnption IS possible and from trying to construct one, if only as a workmghypothesis A version of semantic transcnption, or, as we call it, semanticlanguage, is propo^d within the framework of MTT and is used hereinThus MTM deals with formal representations of meanings and texts ratherthan with "genume," "real" meanings and texts Therefore, Formula I canbe rewntten as 3

\ MTV # V\OT. (• e t l c '• ' I3err,.:antic; ^'epresertatio") ,}•<=> ^ or "^CeBresertatior) V

From now on, I speak only of representationsOne of the most basic facts about natural language is the following

4 A given meaning of sufficient complexity can normally be expressedby an astronomically large number of texts

Consider sentence 5, quoted (with slight modifications) from Ne-wsweekmagazine (Sept 15, 1980)

5 The Food and Drug Administration has seriously cautioned expectantmothers to avoid one of life's stmple pleasures a cup of coffee

The meaning of that sentence can be expressed by 6 as well

6 Pregnant women have been earnestly warned by the FDA against drink-tng coffee, one of the small pleasures of life

Except for the function words (articles, auxilianes ) and such technicalterms as FDA, coffee, and pleasures of life, all the other words have beenchanged, as has the overall structure Yet the meaning is preserved intactWith additional vanations, the number of English sentences synonymouswith 5 and 6 runs into thousands *

^ 0 see how thousands of [»iraplirases for 5 and 6 can be produced, consider the followmgSentence 5 can be broken into roughly eight semantic fragments, each of which will berepresented by one column in the chart, page 32, each column contains some (nearly) synony-mous variants that express the fragment in question (the figure in boldface under a columnrepresents the number of variants in it)

Any variant from one column combines with nearly any other variant from another column(i; The FDA has addressed a stem caution to ladies to the effect that while expecting a baby

they should abstain from coffee, tbat belongs to life's small joys(ii) The FDA made public a strong warning addressed to mothers-to-be they should not

indulge m consuming coffee, one of the simple pleasures of life, etcrhis allows us simply to multiply the number of vanants m each column I X 4 X 9 X 7 X6 X 6 >< 3 X 8 = 217,728

True, some of the paraphrases will be sifted out by selectional restnctions and other con-straints, however, a native speaker can easily think of new vanants. so it becomes evident thatsynonymy is really nch

32 MEVCUK

Natural language is a system capfU l̂e of producing a great many synony-mous texts for a given meamng; the mastery of a language boils down tothe control of its synonymic means

The MTM has to match a given meamng with many different texts, anda great many different texts have to be reduced to the same meamngrepresentation This makes it almost impossible to establish a direct corre-spondence between semantic and phonological representation, and so twointermediary levels of utterance representation have to be introduced—syntactic and morphological— t̂he former aimed at the sentence as a struc-tural object and the latter deahng with the word All levels—with theexception of the semantic one—are spht into two sublevels. a deep one,geared to meamng, and a surface one, determined by physical form Thisgives a total of seven representation levels

1 Semantic Representation (SemR), or the meamng,2. Deep-Syntactic Repres^tation (DSyntR),3 Surface-Syntactic Representation (SSyntR),4 Deep-Morphological Representation (DMorphR),5 Surface-Morphological Representation (SMorphR),

^Continued from

1

FDA

2

seriouslyearnestlysternlystrongly

page 31

3

cautionwarnforewarncounselput on guard

issue a warning

address a caution

make public/•a cautionVa wammg

4

pregnant womenexpectant mothers

mothers-to-beduringpregnancywhileexpectinga baby

women/ladies

5

to avoid

to abstam fiom

sh uld f^^°^^V abstain

not toshould not

6

coffee

, _, , - rdrink(u^) 1 „indulge in) { T ^ f coffeeVconsum<ing)J

njp of coffee

7

one of

that belongs to

that constitutes

8

smail 1 rpleasuresl of life/simpl^ \joys / hfe*s

MEANING- TEXT MODELS 33

6 Deep-Phonetic Representation (DPhonR, or what is commonly called"phonemic representation"),

7 Surface-Phtmetic Representation (SPhonR, which is called in the litera-ture simply "phonetic represoitation"), or the text

A repres^itation is a set of formal objects called STRUCTURES, one ofwhich IS distinguished as the mam one, with all the others spectfymg someof Its charactenstics Each structure depicts a certam aspect of the itemconsidered at a given level Utterance representations are wntten m formal-ized languages d^ned by the researcher

We can now present Formula 3 in full

The top line in 7 is a sequence of the utterance representations of all sevenlevels, with the correspondences between any two adjacent levels shown bytwo-headed arrows The bottom line shows the components of the MTMand their functions Thus semantics provides for the correspondence be-tween the semantic representation of an utterance and all the sequences ofdeep-syntactic representations carrying the same meaning, etc (See sectionon Design of the MTM)

FoUowmg are examples of successive representations for sentence 5 andsome synonymous sentences, starting from the meaning and working up-ward toward the text

Semantic RepresentationA sample Sem(antic) R(epresentation) simphfied for illustrative purposesIS given m Figure 1 Notice that the letters A, B, C and the numbers 1, 2,3 that appear at the left side and top of the diagram are not part of therepresentation they are used merely for refemng to locations m the dia-gram The shaggy hnes mark the boundanes of theme and rheme (cf below)This SemR can be worded approximately as follows

8 The FDA has mtensively attempted to commumcate [A 1-2] to allpregnant women [B2] that they should not [A2] dnnk coffee, one ofthe simple pleasures of life for them [A2-3, C2-3], the FDA behevesthat drmkmg coffee is dangerous for pregnant women [B2] Thisattempt took place bdbre the moment of speech yet its results persistuntil this moment [A2-3, m this way the Enghsh present perfect isrepresented]

Let It be emphasized that Figure 1 does not lay claim to absolute semanticpr^ision Many flaws exist in the semantic language used, but hopefully itserves the purposes of illustration

34 MEL'CUK

Figure J 8 IS the SemR of 5 and 6 as wdt as of all the other sentences {or sequences ofsentences) synonymous with them

The SemR of an utterance consists of two structures the semantic struc-ture and the semantico-communicative structure

The Sem(antic) Structure} speafies the meaning of the utterance inde-pendent of Its linguistic form The distnbution of meaning among thewords, clauses, or sentences is ignored; so are such linguistic features as theselection of specific syntactic constructions and so on At the same time, theSemS tnes to depict the meaning objectively—leaving out the speaker andhis intentions, which are taken into account in the second structure of theSemR Formally, a SemS is a connected graph or a network

The vertices or nodes of a SemS are labeled with semantic units, orSEMANTEMES meanmgs which can be either elementary (SEMES) orcomplex, complex meamngs consist of semes or less complex semantemesA complex semanteme can be represented by a semantic network, whichspecifies its semantic decomposition For example, the semanteme "earher"found in 8, A3, can be decomposed as follows

MEANING-TEXT MODELS 35

'Tionent 1 '

Semantic decomposition can go on, as deep as we need or want it, untilsemes are reached A d^nitive list of semes, or semantic primitives, is notavailable, but some likely candidates are "(some)thing," "more," "say,""this speech event," "not," "set" (in the mathematical sense), "space,""time," '*and," "or " [For a different view of semantic pnmitives see Wierz-bicka (79, 79a) where 13 semantic primitives are proposed and argued for ]Since a complete analysis into semes would make the semantic networkunreadable, complex semantemes are used for the most part The names ofthe semantemes are word senses picked from a good dictionary For theillustration here I have taken English word senses from The AmericanHeritage Illustrated Dictionary of the English Language, thus "persist 3"m A3 IS "continue in existence until " rather than "to be obstinatelyinsistent in " or "hold steadfastly to despite obstacles "

Two major classes of semantemes are distinguished

1 FUNCTORS, further subdivided into PREDICATES (relations,properties, actions, states, events, etc). LOGICAL CONNECTIVES("if. "and", "or", "not"), and QUANTIFIERS ("all", "there exist",plus all numbers).

2 NAMES (OF CLASSES) OF OBJECTS, lncludmg proper namesBoth types of semantemes can receive arcs or arrows, but only afunctor can head an arrow The arrows on the arcs point from functorsto their arguments

The arcs or arrows of a SemS are labeled with numbers which have nomeaning of their own but only serve to differentiate the vanous argumentsof the same functor For instance,

' corniin ca te '

means that A is the first argument of "communicate" (who communicates).

36 MEL'CUK

B Its second argument (what is commumcated), and C the third argument(to whom the information is passed) The exact role of each argument isspecified by further decomposition of the functor- "A commumcates B toC" = "A, who IS aware of B. exphcitly causes C to become aware of B "A deeper decomposition (bearmg, e g, on "aware" and "cause") wouldreveal more subtle links between the functor "communicate" and its argu-ments

The Sem(antico)-Comm(unicative)S(tructure) specifies the intentions ofthe speaker with respect to the orgamzation of the message The samemeanmg reflecting a given situation can be encoded m different messagesaccording to what the speaker wants So the Sem CommS must show at leastthe following contrasts.

(a) Theme (topic, designated 1) vs rheme (comment, designated d ), l ethe starting pomt of the utterance, its source, as opposed to what iscommumcated about the topic In 8,1 is "the caution given by theFDA and aimed at pregnant women," and U is that "pregnantwomen should not dnnk coffee smce that is dangerous for them " Ifwe mterchange the symbols 1 and H , the message becomesdifferent

9 Mothers-to-be had better keep clear of coffee, which can be harmful tothem—that ts the stern warning issued by the FDA

1 and d indicate the ltmerary through the situation, which is up to thespeaker to choose Note that there can be different layers of topic vs com-ment division For example, within 1 m 8, 1 and U of second order canbe indicated

12 = "the FDA," d 2 ~ "senously cautioned pregnant women "

(b) Old, or given (known to both interlocutors), vs new, l e communi-cated by the speaker

(c) Foregrounded (expressed as a main predication) vs backgroimded(relegated to an attnbute)

(d) Emphatically stressed vs neutral

Semantico-communicative information stands in approximately the samerelationship to the semantic network (SemS) as do suprasegmental prosodicphenomena to the segmental phonemic stnng that makes up a sentence In

MEANING-TEXT MODELS 37

the simplified examples given here only the topic-comment contrast isshown, and that on the first stratum only For more details on the SemRsee (52, pp 53-77, 56)

Deep-Syntactic RepresentationTo make the contrast between a SemR and a D(eep)~Synt(actic) R(epresen-tation) more vivid, I will give two different DSyntR^ for two different butfully synonymous sentences, 5 and 10

10 The FDA has issued a stern warning to pregnant women, they shouldnot drink coffee, which is one of life''s small pleasures.

Both sentences have of course the same SemR, namely, the one shown inFigure 1 Their DSyntRs appear as Figures 2 and 3, respectively

A D(eep)-Synt(actic)R(epr»entation) [of a sentence] consists of fourstructures the deep-syntactic structure, the deep-syntactico-communica-

II

Magn

Figure 2 11 is the DSyntR of sentence 5

38 MEL'CUK

12

Magn ^ - R

NAP

Figure S 12 is the DSyntR of sentence 10

tive structure, the deep-syntactico-anaphonc structure, and the deep-syn-tactico-prosodic structure

The D(eep)-Synt(actic)S(tructure) is a dependency tree (see 58, pp 3-21)It represents the syntactic organization of the sentence m terms of itsconstituent words and relationships between them

A node of a DSyntS is labeled with a generalized lexeme of the langu^eA generalized lexeme is one of the following four items

1) A full lexeme of the language Semantically empty words, bke(strongly) governed prepositKms and conjunctions or auxiliary verbs,are not represented thus in 11 has [cautioned], to [avoid] and [cup]of [cojffee] are absent

2) A fictive lexeme, l e a lexeme presupposed by the symmetry of thedenvational system, yet nonexistent (There are no fictive lexemes inmy examples, but this kmd of umt has been widely discussed mAmencan Imguistics)

MEANING-TEXT MODELS 39

3) A multilexemic idiom, e g hit ifoff "h&ve good rappon" or pull a fastone on someone "gam an advantage over sa unsuspectmg person bysubterfuge " I n 11 and 12 we find two examples of idioms one is TheFood and Drug Administration, ^oriened to FDA and represented bya single node; the other example is expectant mother

4) A lexical function- see below

The symbol of a generahzed lexeme must be subscnbed for all the mean-ing-beanng morpholo^cal valu^, such as number m nouns or mood, tense,and aspect in verbs (Syntactically conditioned morphological values, suchas person and number m verbs, are not shown in a DSyntS )

A (standard elementary) LEXICAL FUNCTION (LF) f is, roughlyspeaking, a relation which connects a word or phrase W—the argument off—with a set f( (K) of other words or phrases—the value off—in such a waythat

fa) for any W^ and ff̂ , if f(ff") and t{W^) exist, both f{W^) andf( W^) bear an identical relationship with respect to the meaning andsyntactic role to W^ and W^. respectively

ib) in most cases, f( W^) ^ f( W^), which means that f( W) is phraseologi-cally bound by W

In the MTM, about 50 standard elementary LFs are used Some examplesof LFs

Syn(ro shoot) — to fire [synonym],to shoot) = to shell, to machine-gun [narrower synonym],

= defeat [antonym].include) = to belong to [conversive, as in This paradigm in-

cludes the locative case = The locative case belongs to this paradigm],So(to despise) = contempt,Ao(xtt«) = solar [denved substantival/adjectival],M»gfk(need) = great, urgent, bad, M&gp{settled [area]) = thickly, Magn

(to illustrate) = vividly, Mngj^behef) = staunch ["very", an lntensifier],Oiperiianalysis) = to perform, Op&[i(attention = to pay. O^Tx{step) =

to take, Opetiifavor) = to do ["be the subject o f ] ,Opvxjianalysis) = to undergo, O^r^attention) = to receive, Oper2(con-

trol) ~ to be under ["be the object of] ,l^eaiiipromise) — to keep, to make good, R&d^attack) = to fall to

["fulfill the requirements of X, which is the argument, bemg respectively thesubject (for Reali) or the object (Real2)" of X,

40 MEL'CUK

Son{cow) ~ to low. to moo, Son{windowpanes) — to jingle, to rattle[typical sound]

Furthermore, there are COMPLEX LFs. e g AntiRa^^romise) — torenege on, AnHReak^attack) — to beat back, JncefOferiiflre) = U) open,YinO^eri^control) ~ to get out of, etc To avoid overburdenmg the exposi-tion, I will not touch on so-called NON-STANDARD LFs

LFs play a crucial role m covenng restncted lexical cooccurrence. In 11we see a Magn(ro caution), l e seriously; m 12 there are three LFs S^towarn) = warning, Operi{waming) — to issue, and Mafftiwaming) — stemFor more details about LFs see (8, 42, 67, cf 38 also).

A branch of a DSyntS is labeled with the name of a DEEP-SYNTACTICRELATION (DSyntRel) The DSyntRels are deemed to be umversal

A DSyntRel is one of nine binary relations> 1, 2, 6 are predicative relations connectmg a semantically predica-

tive lexeme with its 1st, 2nd, 6th arguments, respectively, for instance-

13 SELL

I who'']s e l l e r merchandise

[for how much'']

• ATTR IS the attributive relation, which covers all kmds of modifiersand attributes (m the broadest sense of the term).

• COORD IS a relation that accounts for all coordinate or conpmedconstructions hnkmg the nght conjunct to th^ left one, which is the gover-nor, e g

JOHN PETER AND PAUL

COORD COORD 2

• APPEND IS an "appendancy" relation that subsumes all parentheti-cals, lnteriections, adresses, etc, linking any one <^ these elements to the topnode (mam verb) of the correspondmg clause

Let It be emphasized that there is no linear order of nodes withm theDSyntS Word order is taken to be a means for encoding syntactic structureinto speech stnngs and therefore it is banned from the syntactic structure

The D(eep)-Synt(actico-)Comm(unicative)S(tructure) specifies the divi-

MEANING-TEXT MODELS 41

sion of the sentence represented mto topic and comment, old and new, etcIn our sunpbfied examples the DSynt-CommS shows the first stratum oftopic-comment division only m much the same manner as withm the SemRI will not discuss the differences between the DSynt-CommS and Sem-CommS any further

The D(eep)-Synt(actico-)Anaph(onc)S(tructure) cames the informationabout coreferentiahty cf the broken-line arrows in 11 and 12 that is, theunderstood subject of avoid is the expectant mothers who have been cau-tioned, etc

TheD(e€p)-Synt(actico-)Pros((xiic)S(tructure) represents all of the mean-ingful, prosodies that appear at this level—intonation contours, pauses,emphatic stresses, and the like—^where these are not syntactically condi-tioned. In 11 and 12 it is the acronym NAP, standing for N(eutral) A(ser-tive) P(rosody), that exemplifies the DSynt-ProsS

Surface-Syntactic Representation

Only sentence 5 will be represented at the surface-syntactic level (see Figure4) The S(urface)-Synt(actic)R(epresentation) of a sentence consists of fourstructures correspondmg to those of the DSyntR but replacmg D(eep) byS(urface) The SSyntS is also a dependency tree, but its composition andlabelmg differ sharply from those of the DSyntS

A node of a ^ y n t S is labeled with an actual lexeme of the language First,all the lexemes of the sentence are represented, including the semanticallyempty ones Second, all the idioms are expanded mto actual surface treesThird, the values of all the lexical functions are computed (on the basis ofa lexicon called an Explanatory Combmatonal Dictionary, see below) andspelled out m place of the LFs Fourth, all pronominal r^lacements anddeletions under lexical or referential identity are earned out

A branch of a SSyntS is labeled with the name of a SURFACE-SYN-TACTIC RELATION (SSyntRel)

A SSyntRel belongs to a set of language-specific bmary relations thatobtain between the words of a sentence, each descnbing a particular syntac-tic construction The inventory of SSyntRels for a language is estabhshedempincally(cf58,pp 91-150) In 14 the reader can see examples of EngbshSSyntRels Tentative lists of SSyntRels are found for Russian m (5) and (52,pp 221-35), for EngUsh in (63), and for some other languages m referencesgiven in (58, p 97)

As IS the case with the DSyntS, the nodes of the SSyntS are not orderedlinearly This enables us to keq) stnctly apart two basically different "or-ders" syntactic hierarchy and hnear ordenng, which serves to express thishierarchy

3c" "Ci; fes -"once-

COFtEE^^ SIMPLE

NAP

Figure 4 14 is the SSyntR of the Enghsh sentence that corresponds to the DSyntR 11, l eof 5

The SSynt-CommS, the SSynt-AnaphS, and the SSynt-ProsS are analo-gous to their deep counterparts A detailed discussion of important diver-gences cannot be undertaken here because of lack of space

Deep-Morphological RepresentationOnly one DMorphR will be offered for sentence 5, see 15

15THE FOODg AND g ADMINISTRATION^

HAVE , , SERIOUSLY CAUTION EXPECTANTindic,pres,3sg ppas

TO

PLEASURE. liONE

^̂A C U P . _ OF COFFEE

15 IS the DMorphR (rf the sentence which corresponds to the SSyntR 14,1 e to the SSyntR of 5

MEANING TEXT MODELS 43

The D(eep)-Morph(ological)R(epresentation) of a sentence consists oftwo structures the deep-morphological structure, and the deep-mor-phologico-prosodic structure

The D(eep)'Morph(ological) S(tructure) is a stnng of D(eep)-Morph(o-logical) R(epresentations) of all the wordforms that compose the sentenceThe DMorphR of a wordform W is the name of the lexeme to which Wbelongs sutecnbed for all its morphological values Thus the DMorphR ofW unambiguously specifies this particular W (up to wordform hom-onymy), while the DMorphS of the sentence unambiguously specifies itsword order (up to free vanatirai)

The D(eep)-Morph(ologico-)Pros(odic) Structure J indicates pauses, into-nation contours, and the like In 15 the vertical bars stand for pauses ofvarying length, and the arrows represent the pitch contouR

Representations at Other Levels

Levels 5 through 7 the Surface-Morphological [Morphemic] Representa-tion, the Deep-Phonetic [Phonemic] Representation, and the Surface-Pho-netic [Phonetic proper] Representation, will be omitted here because theyare of minor lmponance, but see (41, 43, 47) The aim of all these repre-sentations, with different structures within each, is to descnbe all the dif-ferent aspects of an actual utterance as separately and autonomously aspossible That is, a MTM does not aim for a homogeneous representa-tion, on the contrary, it explicitly keeps apart phenomena that appear tobe different

THE DESIGN OF THE MEANING-TEXT MODEL

As stated above, a MTM has the task of estabUshing correspondencesbetween the semantic and the (surface-)phonetic representations of anygiven utterance through the five intermediate levels listed in the precedingsection Accordingly, the MTM consists of the following six basic com-ponents

1 The Semantic Component, or semantics for short2 The Deep-Syntactic Component, or deep syntax3 The Surface-Syntactic Component, or surface syntax4 The Deep-Morphological Component, or deep morphology5 The Surface-Morphologica! Component, or surface morphology6 The Deep-Pbonetic Component, or phonemics

44 MEL'CUK

The Surface-Phonetic Component, which provides for the correspondencebetween a surface-phonetic representation and actual acoustic phenomena,falls outside the scope of the MTM model m the stnct sense.

Each component of the MTM is a set of rules havmg the tnvial formX <=^ Y\C, where X stands for a fragment of utterance rq>resentatton atlevel n, Y for a fragment of utterance representation at level n -h 1, andCis a set of conditions (expressed by Boolean formulas) unds* which thecorrespondence X *=* Y holds The two-headed arrow must be mterpretedas "corresponds," not "is transformed into." Thus, when the transitionfrom a meanmg 'X' to a DSyntR Y is performed, *X' itself is not changed:nothmg happens to *X' while Y is bemg constructed by semantic rules underthe control of 'X' The relation between a representation n and an "adja-cent" representation n-hl is the same as that between the blueprint of ahouse and the house itself. The bluepnnt is by no means transformed intothe house, but dunng construction, it is the bluepnnt that guides theworkers

Another important peculianty of the rules in the MTM is that they arelogically unordered All relevant information about the language is ex-pressed explicitly, 1 e. by symbols within the rules rather than by the orderof the latter The philosophy behmd this decision is that findmg the bestorder of rule apphcation in a specific situation goes far beyond the task oflmguistics proper

The rules themselves are conceived of not as prescnptions, or instructionsof an algonthm, but rather as permissions and prohibitions, or statementsm a calculus Basically each rule is a filter siftmg out wrong correspon-dences The idea of definmg the structures m the MTM m terms of sets ofrules representing well-formedness conditions was mtroduced into themeamng-text theory by L. N. Iordanskaja (24,25) and is now firmly embed-ded in the research on MTMs

The Semantic Component of the MTMThe semantic component establishes the correspondence between the SemRof an utterance and all the synonymous sequences of DSyntR's of thesentences that make up that utterance (cf 7). To do that, it performs thefollowmg eight operations

1) It cuts the SemR into subnetworks such that each corresponds m itssemantic "size" to a sentence (In fact, the ongjnal SemR is not cot; simply,the semantic component constructs alongside it another SeniiR, equivalentto the ongmal one, but consistmg of a sequence of smaller SemRs.)

2) It selects the correspondmg lexemes by means of s^nuitico-lexicalrules of the type illustrated below

MEANING-TEXT MODELS 45

[who] U-homj [that ]

CAUTION

'dangerous 2 ' •reason 1'who] [whoir]

16 IS a semantico-lexical rule in the stnct sense (Ci and C2 stand for differentsets of conditions that might determine the choice of the lexeme) Essen-tially, 16 represents two dictionary entnes for two (roughly) synonymousverbs to warn J and to caution. The variables X, Y, etc in the left-hand partof the rule represent semantic subnetworks that "hang" from the nodeslabeled X, Y, , respectively, the vanables X', Y', etc in the nght-handpart of the rule stand for Engbsh lexical items that express X, Y, etc,respectively In rules 18-20 the vanables are used in the same way

17!woman' EXPECTANT-MOTHER

MOTHER-TO-BE

•pregnant'

This IS a semantico-idiomatic rule specifymg two idioms for the meaning"pregnant woman," namely, expectant mother and mother-to-be Thismeamng can of coui^ be expressed also by the free phrase pregnant woman,which IS ensured by two separate straightforward semantico-lexical rules

18 1intensive *

This IS a semantico-functional rule responsible for the lexical function M a ^m the DSyntS (cf m 11 and 12 seriously cautioned, stern warning)

3) It supphes meamng-beanng morphological values of lexemes bymeans of semantico-morphologtcal rules like the following one

MEL'CUK

' e a r l i e r '

• this speech — ^ ^sres.perf JT is a vert

' r esu l t s

Here the meanmg of the English present perfect is roughly rendered as "theevent X took place earlier than the moment of this speech event but itsresults persist up to this moment"

4) It forms a DSyntS (a tree) out of the lexemes it has chosen5) It introduces the DSynt-AnaphS; that is, it indicates coreference etc

for the lexical nodes that have atq>eared as a result of the duphcation ofsome semantic nodes

6) It computes the prosody of the sentence on the basis of semantico-prosodic rules, e g

'truth value'

P IS a statement'—•V'commu- • •<-^ • J

'speaker'

('The speaker wants the hstener to communicate to him the truth value ofthe statement P' is rendered—along with other means of course—by thespecific mterrogative pr(»odics) As we see, all the meanhig-beanng pro-sodic phenomena are represented in the SemR in exactly the same way asany other meanings, but in the DSyntR their representation is differentsince prosodies and words are so different

7) It provides the DSynt-CommS (topic-comment, etc) from the datacontamed in the Sem-CommS

8) For each DSyntR produced the semantic component constructs allthe synonymous DSyntRs that can be exhaustively descnbed m terms oflexical functions (Other types of hnguistic synonymy are accounted for bysemantico-lexical rules that can group bundles of semantemes mto lexemesm different ways ) This is thieved by means of a paraphrasing system thatdefines an algebra of transformations on such DSyntR's where the DSyntScontains LF symbols The paraphrasing system, first descnbed m (87, seealso 67), includes rules of two classes

MEANING-TEXT MODELS 47

Lexical Paraphrasing rules (about 60 of them, vahd for any language)represent either semantic equivalences or semantic lmphcations Examples

21

22

Equivalences

(a)

(b) K'

Implication

PerfCaus (X) => Pernncep {X)

The set contains [W\ the point M** The point M belongs [Convjiro the set

He warned [W] them^He issued [Oper, (Noa warning [NQ(KO] to them

He followed [Real2(H0] heradvice [W] to enroll'=> He enrolled on [Adv|gher advice [W]

John started [PerfCaus (run II)]the motor => The motor started[Perflncep (ran 11)]

Syntactic Paraphrasing Rules indicate what restructunng of a DSyntS isneeded when a particular lexical rule is applied Smce there are only fourbasic syntactic opei^tions at the deep level (merger of two nodes, splittingof a node, transfer of a node to another governor, and renaming of a branch)and only nine deep-syntactic relations, the number of elementary deep-treetransformations is finite and not very large Any particular syntactic ruledefined on DSyntSs can be represented in terms of those specific transforma-tions

To operate lexical paraphrasing rules (21), the following syntacticrules are needed

23.a

lexical rule 21a)

i; (for lexical rule 21c\ i ;

48 MEL'CUK

Generally speaking, a syntactic rule, due to its abstract character, may serveseveral different lexical rules A formalism has been devised for descnbingunordered dependency tree transformations—so-called A-grammar (see17-19, 52, pp 163-66)

The Deep-Syntactic Component of the MTMThe deep-syntactic component establishes the correspondence between theDSyntR of a sentence and all the alternative SSyntR's which correspond toIt To do that, it performs the following five operations

1) It computes the values of all lexical functions by means of rules hke24

^^ ^CAUTION .yj ^CAUTIONf^j

IATTR *? > c lrcutns t a n t i a t iveI •*Mltgn •SERIOUSLY

^ ^ I2 •< ^ Jlst completive

•lA'ARNING

Such rules belong, of course, to the dictionary2) It expands the nodes of idioms mto corresponding surfiace trees

25 FDA ADMINISTRATION^

determinatT ve-/ ^s^omposi t i ve

TTHEcoordinatlve

AND

coordinate-,,conjunctional

Rules such as 25 belong to the dictionary, too3) It eliminates some DSyntS nodes that occur m anaphonc relations

and should not appear in the actual text, thus the occurrence of MOTH-i under AVOID m 11 is deleted by the following rule (a kind of Equi)

MEANING-TEXT MODELS 49

26 X

X* 1st comp-/ \ 2 n d comp-

.,̂ => letive>^ N^letiveA T

sV

A

4) It constructs the SSyntS by means of three types of transformations

27

Replacement of a D3yntRel by a SSyntRel

X Y -L If Y

_ • t V JI i n • •Ipredicative |ATTR^=> Imodifleative

• t •

X Xc. • •

ATTn i; > Iquantitatxve

• t

^(Nun) (Num)

Replacement of a DSynt-node by a SSyntRel

• (N)

\ATTR

I2 lappositive

^(N) (N)

Replacement of a DSynt^el by a SSynt-node

j ^ [2nct completive

• •TOJprepositional

50 MEL'CUK

The notation X(V,3|TO]) indicates a lexeme whose 3rd DSynt-valence slotmust be filled m the SSyntS by the mfinitive marker TO, the correspondinginformation is stored m the dictionary entry of X (cf CAUTION in 11 and14)

5) It processes the other three structures of the SSyntR

A descnption of a small fragment of the DSynt-component for RussianIS found m (52, pp 237-59)

The Surface-Syntactic Component of the MTMThe surface-syntactic component establishes the correspondence betweenthe SSyntR of a sentence and all the alternative DMorphR's that arerealizations of it It performs the foUowmg four main operations

1) Morphologization of the SSyntS, l e it determines all the syntacticallyconditioned morphological values of all the words, such as the number andperson of the verb

2) Lmeanzation of the SSyntS, l e it determines the actual word orderof the sentence

3) SSynt-ellipsis, l e it carnes out all kinds of conjunction reductions anddeletions that are prescribed by the language in question, e g

30 George will take the course, and Dick might take the course, too^ * George will take the course, and Dick might too

4) Punctuation, l e it determmes, on the basis of the DSynt-ProsS as wellas on the basis of the resulting SSyntS, the correct prosody, which, in thecase of pnnted texts, is rendered by punctuation

The basic tool of morphologization and lmeanzation of the SSyntS is thesyntagm, or SSynt-rule, which can be illustrated with the following threerules for Enghsh

not31

• ^ ' ' '^"TOl O t j + +X

(predicative

X+ -fY.not obn

r(Y,X)

v(Y,X)

The rule says that to build a predicative construction with a NP as thegrammatical subject, one can put the subject either before the verb if thestandard function^ "obligatory inversions of the subject and the verb" does

'A standard function (in programming, also a standard subroutine) is a closed set of rulesused in solving a specific class of problems or in carrying out a specific class of transformationsIn a lmguistic model, standard iiinctions provide for different types of agreement, inversion,coordination, eUipsis, and the like

MEANING TEXT MODELS 51

not apply, or after the verb if a similar function ("possible or obligatoryinversions of the subject and the verb") applies, the subject must not be inobjective form (relevant only for pronouns me, us, ), and the verb mustagree with it in accordance with yet another standard function "agreementof the verb with the subject noun " The NP and the standard functions aredefined elsewhere since they appear m many different syntagms

c i r c u T P S t a n t i a t i v e < ^ 7 - ^ X Y

v 1r ( N , t e p p ) TI

This rule means that a noun Y with the syntactic feature *temp(oral)' whichIS an adverbial modifier of a verb can be placed before or after the verb ifY Itself has a dependent Z other than an article Last week [Y] we finishedour job. He wtll go there next year [Y]

Y(N , not prop , hurran ;

apposTtive • < > X + + Y I X + Z

t deterpr ( N } •

detern' i natT ve

^ i , oers)

The rule states that a noun Y govermng THE or a personal adjective {my,your, ) and being an appositive to a human nonpronominal noun X canbe placed after X in such a manner that its article or the personal adjectivefollows X immediately Jack [X] the [Z] editor [Y]

A nearly complete list of &ighsh syntagms is found in (64), an enlargedand revised English edition is now being prepared for publication by Benja-mins Publishers Cenam types of Ru^ian syntagms have been descnbed mdetail in [(22, 23, 76), cf also a list of basic Alutor syntagms (65)]

52 MEL'CUK

Besides syntagms, surface syntax uses at least four additional types ofrules

• Word order patterns for elementary phrases; e.g all those beautifulFrench magazines vs * French those beautiful all magazines.

• Global word order rules, which compute the best word order possiblefor the given SSyntS on the basis of vanous data* the syntactic propertiesof some words marked m the lexicon; the relative length of different partsof the sentence, topicalization, emphasis, and the hke; possible ambiguitiesproduced by specific arrangements, etc These rules try to minimize thevalue of a utility function that represents the "penalties" assigned (by thehnguist) to certain unfehcitous arrangements llie rules do this by reshufl9-lng the constituent phrases within the limits of what the syntagms allowA fairly full description of such global word order rules for Russian, alanguage famous for its involved word order, was published in (40) and (52,pp 268-300)

• Ellipsis rules• Prosodic or Punctuation rules

The SSynt-component of the MTM for Russian is characterized at somelength in (5)

Other Components of the MTMThese will not be dealt with m this review The interested reader is referredto (47, 50, 56)

SOME LINGUISTIC RESEARCH DONE WITHINTHE FRAMEWORK OF THE MEANING-TEXTAPPROACH

It IS impossible to hst here all the studies conducted and pubhshed by theadherents of the Meaning-Text approach. I therefore mention only a few,picked more or Iras at random Nichols (68) draws parallels between theMTM approach and Modem American linguistics; the paper offers anexcellent perspective for a beginner

SemanticsA new trend m modem semantics, at least m the USSR, was <^ned m theearly 1960s by the work of the Machine Translation Laboratory of the MThorez Institute of Forragn Languages m Moscow Headed by V Ju Ro-zencveig, a group of young Soviet s^nanticists, with A. K. Zholkovsky inthe forefront, started research m this hitherto neglected field, and in 1964published an issue of die Laboratory's tnilktm MaSinny) perevod i priklad-naja hngvistika devoted entirely to semantics* the now famous MPiPL 8This modest-lookmg volume was to play a crucial role for the Soviet seman-

MEANING TEXT MODELS 53

tic school Many papers in it (77, 80-82) are still of enormous interest forsemantics, but they can be summarized here only in outhne It was therethat the idea of a "pure," language-mdependent semantic representationfirst emerged and brilliant examples of semantic decomposition of lexicalmeanmgs were produced The pnnciple of using highly involved syntax inSemR's, as opposed to hsts of semantic features, was exphcitly stated andapphed to hundreds of semantically very difficult Russian words (withmeanmgs such as "goal," "proceed from," "attempt," "manage," "help,"etc) In a word, the semantic research of 1961-1965, spearheaded by AZholkovsky, laid the foundation for the MTM approach [cf (73), wheremost of the MPiPL 8 anicles are repnnted in Enghsh]

Apresjan's book (1) sums up much of his work done in the late 60s andearly 70s Particularly important is Apresjan's treatment of synonymy andantonymy, of conversives (such as X is behind ¥=Y is injront ofX), ofregular polysemy, of semantic metalanguage, and of modal frames as panof lexical meanmgs

A recent book by the same author (3) discusses nontnvial semanticfeatures, standard lexicographic definitions of lexemes, grammemes, andsyntactic constructions, syntagmatic lnteractioti of meanings, etc

Among the minor studies are these

(a) A careful analysis of two important semantic famihes (feelmgs andperceptions) is found in (28, 29)

{b) The problem of the scope of some "operator" meanings, such as nega-tion or "only," is explored m depth m (12, cf also 13)

(c) The highly important distinction between lingmstic anomaly and log-ical contradiction is convincmgly dealt with m (2)

{d) Connotation m hnguistic semantics is dealt with m (30)

Lexicography

Lexicography is a major concern of the meamng-text approach since all theinformation about the relevant properties of mdtvidual words vital for thesuccessful functKmmg of an MTM must be stored in a dictionary As soonas the work on the MTM of Russian be^in m 1965, much of our effort wasconcentrate on the so-called Ei^danatory Comlnnatonal Dictionary(ECD) c^ Modem Rmsian Dunng a penod (^ about 12 years between 10and ^ petite took part m our dict»mary {H'CQect, wu^ sc»iie hundreddictionary entries w«e published (39) (Presoitly, a foUer vo^on of theRussian ECD is beng prepared for pubhcatK>ii m book form by Lingu»ticResearch, Inc, Edmcaiton, Canada)

The lexicographic th«ffy belund the ECD proposed in (87) was furtherelaborated and expounded m (6-8, 52, pp 113-33, 62)

54 MEL'CUK

Beginning m 1978, a Modem French ECD project was launched at theDepartment of Linguistics, University of Montreal (62) In addition, somepioneer work has been done on £CDs for other languages, e.g Pohsh (35),Somah (83, 84), Enghsh (89), French (90), and German (71, 72) Recently,a hmited dictionary of Enghsh synonyms has appeared (74), edited byApresjan and orgamzed following some basic prmciples of ECDs More-over, a dictionary project for Huitchol (a Uto-Aztecan language in Mexico)IS bemg earned out by J Gnmes (Cornell University, Ithaca, NY) withextensive use of the ECD model

PhraseologyThe notion of the lexical function and the hohstic concept of a lexical unitembodied in the BCD have contnbuted to the theory and practice ofphraseology The place of LFs in linguistics is descnbed in (42) and (52, pp78-109); (27) is devoted to a specific class of idioms denotations of emotion-ally conditioned physical reactions (eg. His eyes started from his head—astonishment, fear, His legs gave way—fear) See also (7)

SyntaxA senes of concrete syntactic descnptions was completed and pubhshed inthe early 1970s bearmg mostly on Russian and Enghsh, many of these havebeen mdicated above The following syntactic topics deserve special men-tion

(a) A general characterization of the dependency system used m MTMtheory can be found in (52, pp 207-302, 58)

{b) A full-fledged descnption of the syntax of a particular language, namelySomah, is given m (85) This book is, in fact, an excellent illustrationof how our theoretical pnnciples can be apphed to an exotic hnguisticstructure

(c) For a systematic survey of the types of rulw and information used mthe surface-syntactic component of an MTM for Russian, see (5)

(</) Criteria for the types of SSynt-relations are proposed m (58, pp 9 1 -150, 60)

{e) See (75) for a discussion of comphcations ansmg when conjunction isdescribed m the MTM framework, an mgemous solution is suggested

if) Problems of grammatical subjecthood and the ergative construction areanalyzed m (57, 58, pp 23-90, 61) The ergative construction is inter-preted 9& a predicative construction where the surface-syntactic (gram-matical) subject IS marked by a case other than the nominative(frequently by the ergative), while the agent phrase m the ergative casein such languages as Dyirbal or Lezghian is taken to be an agentive

MEANING-TEXT MODELS 55

compl^nent rather than the subject, so that no ergative construction isrecognized in these languages

(g) For general questiims of grammatical agreement, with special referenceto Russian, see

To this hst one could add many other topics of mterest such as syntacticzero (53), syntactic amt»guity (26), the comparative construction m Rus-sian (76), grammaUcal voice and diathesis (21, 36, 66), and a meaning-beanng sjrntactic [quotative] construction m Ri^sian (31), among others

MorphologyIn the dommn of theoretical morphology, work has been m progress for anumber of years on some basic morphological notions, such as conversion[like a footnote-to footnote (48)], suppletion (46), lesser or greater semanticand formal ccni^)lexity of linguistic itrans (44), grammatical case (55), etcMost of this work is summ^ up (m Bngbsh or German) m (54) Thisauthor's book (59) offers a coherent syst»n of formal morphological con-cepts pertammg to the expression plane of morphology (such as linguisticsign, representabihty, linear segmentability, morph, morpheme, root, affix,apophony, reduplication, etc).

As for descnptive morphology, a few formal morphological models havebeen produced for different languages Spanish conjugation (41), Russiannominal inflection (15), Hungarian noun inflection (43), Tartar declensionand ccmjugation (32-34), Alutor (Chukchee-Kamchadal stock) conjugation(49) An interesting mathematical formahzation of MT morphologicalmodels is suggested m (10, 11) A detailed descnption of the grammaticalcategones of the Enghsh verb is proposed m (69, 70).

CONCLUDING REMARKS

Possible Applications of the Meamng Text TheorySome ideas of the MTM approach may find unexpected apphcations m quitedifferent thewetical frameworks withm linguistics itself. Thus m (9) theconcept of LF h^>s to solve some specific problems of an otherwise trans-formati(»ial descnption of Russian impersonal sentences (cf also 9a)

The whole of MT theory is strongly practice-onented, and we considerIt an asset that this theory is readily apphcable m vanous practical fielcULet me mention only four such po^ible applications.

LANGUAGE TEACHING AND LEARNING LmgUlStlC descnptionsbased on MTT seem to be adaptable to language teaching or learmng Thelexical functions and ECD-type dictionaries might be especially useful (cf38, 78)

56 MEL'CUK

AUTOMATIC TEXT PROCESSING Since the MTM formahsms lendthemselves easily to computer programmmg, it is natural that several text-processing computenzed systems at various hnguistic levels, including syn-tax, dictionary, and morphology, have been developed in the USSR withinthe MTT framework As an illustration, two automatic translation systems(English- and French-to-Russian) could be mentioned (4, 14, 37) Amongthe more recent developments in the USA, we find the apphcation of someMTT ideas in information retneval [in panicular, use of the lexical func-tions (16) and E Fox, "Lexical relations enhancing effectiveness of infor-mation retneval systems" (unpublished progress report. Department ofComputer Science, Cornell Umversity, Ithaca, NY)]

TRANSLATION PRACTICE AND THEORY As MTT IS primarily con-cemed with the translation of meanings into texts and vice versa, its rela-tionship to translation in general is direct from the viewpoint of MTT, thewhole of linguistics is nothing more than the science of translation Specifi-cally, an explanatory combmatonal dictionary could be used as an effectivetranslator's tool (see 56, 78)

ANTHROPOLOGICAL RESEARCH MTT was testcd on the Alutor lan-guage (49, 65) As I discovered in the summer of 1980, some ideas andpostulates of MTT are taught in the courses offered by the Summer Instituteof Linguistics (University of Oklahoma, Norman) This clearly shows thatspecialists in field linguistics and exotic languages do find handy and helpfultools in the realm of MT theory

Perspective and SummaryIt IS next to impossible to set an outline of MTT in a background readilyunderstandable to readers in related fields To do so, one should show howMTT compares with "traditional" linguistics, the structuralist school, andtransformational grammar However, such an analysis is precluded by con-siderations of space Instead, I will mdicate two sources of the MT approachand five basic methodological pnnciples, arnved at in the course of a15-year research effort, which constitute, in my view, the most general andcogent advances

SOURCES OF THE MT APPROACH On the one hand, the most immediatefoundations of the MTT are to be found m Chomsky's theory I believe thatthe MT approach can be viewed as an outgrowth and natural continuationof the generative-transformational approach

On the other hand, MTT has developed under the strong influence ofautomatic text processing The problem of automatic translation puts theemphasis on the transfer of meaning and thus bnngs forth the task ofrepresenting meanmg, extracting it from the input text, and expressing it

MEANING-TEXT MODELS 57

in the output text That is, the meanmg-text approach is virtually requiredby automatic translation

FIVE BASIC PRINCIPLES OF THE MEANING-TEXT APPROACH Thecredo of a righteous practitioner of the MT approach (l e my own credo)can be reduced to the following five operational principles

1 Description of a specific language rather than the quest for its bestpossible grammar MTT concentrates on languages, not on grammars Theonly legitimate question to be asked withm this theory is "How do youexpress the meamng X m the language L''" or "What does the expressionX of the language L mean''" It seems to us to be more rewarding to wntea good grammar (mcluding the lexicon) at least for a language than wasteone's time discussing the properties of the best theory of Language mgeneral

2 Establishing specific facts of a specific language rather than hunting forgeneralizations or discovering linguistic universals To state a generalization,one needs something to generalize from, therefore, well-organized collec-tions of facts must perforce precede generalizations lest the latter be scien-tifically void Hence the particular interest of MTT in lexical entnes, i e mthe dictionary, a grammar is considered to be nothmg more than a set ofgeneralizations over a good dictionary MTT is thus a dictionary- ratherthan a grammar-onented theor>' Not only every language, but every lexemeof a language, is an entire world m itself All this means that the pnmaryobject of MTT is particulars, not universals

3 Cybernetic modeling of native linguistic intuition rather than a mathe-matical study of formal systems representing language The only goal of aMTT hnguist is to exteriorize his or her own intuition about language(probably checked against other speakers' intuition) Language is viewed asa "black box," not as a formal mathematical system (semi-Thue, or what-ever) Hence no rigorous proofs are possible withm the MTT Good coun-terexamples can show that a description is not sufficient, but one can neverprove—m the stnct sense of the term prove—that a descnption or a solutionIS sufficient, let alone necessary What a MTT linguist really does is suggestpractical names, useful labels, etc to pass his own hnguistic intuition on toothers m a most unambiguous and graphic way He builds a model, and ifIt functions correctly, there are no arguments to prove or disprove itsacceptability

4 Formalisms serve the linguist rather than the other way around Theformality of all representations is strongly emphasized in MTT, but noformalism is allowed to become a fetish Nobody should be afraid of replac-mg 17 symbols by 439, if need be Why should we be slaves to oversimplifiedand overly narrow mathematical formalisms'' Formalisms are nothing more

58 MEL'CUK

than nec^sary tools, and it is up to the linguist to create tools that he feelsare necessary (Let the mathematicians accommodate our formalisms, ifthey so wish, to their theones ) As far as the technical nature of MTMformalisms is concerned, dependency systems, with their emphasis on rela-tions rather than constituency, are preferred, for they detach linear orderfrom structural representation Moreover, it is believed that the relationrather than the groupmg is the pivot mechanism of natural languages at alllevels

5 Different formal languages for different levels of representation ratherthan the quest for a homogeneous representation Language is a multidimen-sional, multifanous, highly heterogeneous formation, and there is no pointin trying to squeeze it into the strait jacket of a single formalism Therefore,in MTM a special formal language is devised to represent differently eachone of the different levels of natural languages, and at each level differentformal languages are used for its different aspects and strata

ACKNOWLEDGMENTS

The first draft of the present review was carefully read and commented uponby L L Elmtsky, L N lordanskaja, R I Kittredge, J Nichols, and PRoberge J D McCawley read the final version Their cnticisms, remarks,and suggestions have helped me considerably to improve the presentationI wish to express my appreciation and gratitude for their efforts

ABBREVIATIONS USED IN BIBLIOGRAPHY

AJCL Amencan Journal of Computational LinguisticsAN SSSR Akademija Nauk SSSRBSLP Bulletin de la Societe de linguistique de Paris. Pans

KhncksieckIRJa Institut Russkogo JazykaIRSL International Review of Slavic Linguistics. Edmonton, Alberta-

Lingmstic Research, IncIVSLiAP Informacionnye voprosy semiotiki, lingvistiki i avtomatides-

kogo perevoda. Moscow Vses Inst NauCn Texn Info.Izv IzvestijaMPiPL MaSinnyj Perevod i Pnkladnaja Lingvistika. Moscow Mosk

Gos Ped Inst Inostr Jazykov lm M TorezaNTI Naudno-Texnideskaja Informacija Moscow Vses Inst

Nau6n Texn InfoPGfePL Problemnaja Gnippa po eksperimental'noj l pnkladnoj hngvis-

tikeVJa Voprosy Jazykoznanija Moscow AN SSSRWSA Wiener Slawistischer Almanack Vienna

Literature Cited

MEANING-TEXT MODELS 59

1 Apresjan. Ju D 1974 LeksideskajaSemantika. Sinonimideskie SredstvaJazyka Moscow Nauka 367 pp

2 Apresjan, J u D 1978 Jazykovajaanomalija i logi(:^koe pronvore€ie InTekst'Jezyk-Poetyka, pp 129-51 Wro-claw Ossolmeum

3 Apresjan, J u D 1980 Tipy Informa-en Dlja Pijverxnostno-semantideskogoKomponenta Modeh "Smysl i=^Tekst" Wien WSA (Sonderband 1)119 pp

4 Apresjan, Ju D , Boguslavskij, I M .Iomdin, L L , Krysin, L P , Lazurskn,A V , Percov, N V, Sannikov, V Z1978 Lingvistideskoe Obespedenie v Sis-teme Avtomatideskogo Perevoda Tret-'ego Pokolenija. Moscow AN SSSR47 pp

5 Apresjan, Ju D , Iomdm, L L , Percov,N V 1978 Ob"ekty l sredstva modelipoverxnostnogo sintaksisa nisskogojazyka. IRSL 3(3)249-312

6 Apresjan, Ju D , 2o&avskij, A K ,MerCuk, I A 1968 O sisteme semanti-deskogostntezn Ul Obra2cy siovamyxstatej AT/Ser 2, No 118-21

7 Apresjan, Ju D , 2olkovsktj, A K ,Mereuk, I A 1969 C^ odnom sposobeizuCemia soCetaemosti slov Russkijjazyk vSkole, No 6 61-72

8 Apresjan, Ju D , Mel'Cuk, I A , 2olk-ovskij, A K 1969 Semantics and lex-icography towards a new type of unihn-guat aicttooary In Studio in Syntaxand Semantics, ed F Kiefer, pp 1-33Dordrecht Reidel

9 Babby, L H 1978 Lexical functionsand syntactic constructions Russianexistential sentences In CLS Parases-sion on the Lexicon, pp 26-33Chic^o Umv Chicago Press, 364j^

9a Babby, L H 1980 Existential Sen-tences and Ne^tmn in Russian. AnnArbor, Mich KAROMA 180 pp

10 Bider, I G , Boi'Sakov, I A 1976 For-malizaaja moribl^aeskogo kom-ponenta mo<teli "Smysl «=^ Tekst" IPostanovka probiemy l osnovnye pon-jatija. Texn. kibem. No 6 42-57

U Bider, I G , Bol'$akov, I A 1977 For-malizacija moTfotogi6eskO0O kom-

Bmenta taoddi "Smvil ^=^ Tekst" IImamika morfofogi(^cskix preo-

brazovamj Texn. Kibem No I 32-5012 Boguslavskif, I M 1977 Osonastiaes-

kom c^usami russkix ckepn6astij neo-predelennost' lh mnogoznadnost'^ IzvAN SSSR, Ser jaz. i lit 36(3) 270-81

13 Boguslavskij, I M 1978 Otncanie vpredloiemjax s obstojatel'stvami v nis-

skom jazyke In Studia grammaty-czne 2 122-36 Wroclaw Ossolmeum138 pp

14 Erastov, K O , Kulanna, O S, Mel-'^uk, I A (971 Ob Odnoj VozmoinojSisteme Majfmnogo Ferevoda. MoscowIRja AN SSSR (PGfePL. vyp 21)72 pp

15 Es'kova, N A , Mel'duk, I A , Sao-nikov, V Z 1971 Forma/'na/a Model"Russkoj Morfologiu I Formoobrazova-nie Su^estvitel'nyx i Fnlagatel'nyx.Moscow IRjaANSSSR(PGEPL,vyp15) 71 pp

16 Evens, M W, Smith, R N 1979 Alexicon for a computer cjuestton-answenng system AJCL (microfiche83)

17 Gladkij, A V, Mel'Cuk, I A 1969Tree Grammars. Int Conf ComputLing Stockholm (preprint No 1) 7pp

18 Gladkij, A V , Mel'duk, I A 1971Grammatiki derev'ev I Opyt formal-lzacu pref^razovaiuj smtaksideskixstruktur estestvennogo jazyka IV-SLiAP 1 16-41

19 Gladkij, A V , Mel'Cuk, I A 1974Granunatiki derev'ev II K postroenijuA-grammatikidljarusskogo jazyka IV-SLiAP 4 4-29

20 lomdin, L L 1979 E3de Raz o Stntak-sideskom Soglasovanii v RusskomJazyke. Moscow IRJa AN SSSR(PGfePL, vyp 122) 45 pp

21 Iomdm, L L 1980 SimmetridnyePredikaty v Russkom Jaz^ i F^lemaVzaimnogo Zalt^a. Moscow IRJa ANSSSR ( P G 6 P L , vyp 131) 38 pp

22 Iomdm, L L , MeI'duk, I A , Percov,N V 1975 Fragment modeh russkogopoverxnostnogo sintaksaa I Predika-tivnye smtagmy NTI Sa 2, No 7 3 1 -43 [French transl Analyse et validationdans I'etude des donnies textuelles, edM Bonllo, J Virbel, 1977, pp 83-122Pans CNRS 289 pp ]

23 Iomdm, L L, Percov, N V 1975Fragment modeh russkogo poverxnost-nt^o sintaksisa. II K(»npletivnye tpnsvjazoCnye konstnikcu NTI Sw 2,No 1122-32

24 Iordanskaja, L N 1963 O nekotoiyxsvojstvax pravil'noj sintaksid«kojstruktury (na matenale russkogojazyka) VJa No 4 102-12

25 Iordanskaja, L N 1964 Svojstva pra-vil'noj smtaksi^eskoj struktury l al-gontm ee obnaru^enija (na matenalerusskogo jazyka) FrobL kibern11215^14

60 MEL'CUK

26 Iordanskaja, L N 1967 Smtaksites-kaja omonimija v russkom jazyke (sto£ki zremja avtomatii^eskogo analiza lsmteza teksta) NTI No 5 9-17

27 Iordanskaja, L N 1972 Leksiko-grafideskoe opisanie russkix vyraiemj,obozna^ajus^ fizi^eskie simptomyiuvstv MPiPL 16 3-30

28 Iordanskaja, L N 1973 Tentative lex-icographic definitions for a group ofRussian words denoting emotions InTrends in Soviet Theoretical Linguistics,ed F Kiefer, pp 389-410 DordrechtReidel 438 pp

29 Iordanskaja, L N 1979 The semanticsof three Russian verbs of perceptionvosprintmat' '(to) perceive', oSduSdat''(to) sense' and duvstvovat' '(to) feel'Linguistics 17 825-42

30 Iordanskaja, L N,Mel'i;uk,I A 1980Konnotacija v lmgvistifeskoj seman-Uke WSA 6 191-210

31 Iordanskaja,L N,Mel 'duk,I A 1981On a class of Russian verbs which canintroduce direct speech (constructionsof the tj-pe "Ostav'te menja'" - is-pugalsja bufetdik lexical polysemy orsemantic syntax'') In FestschriftS0rcnscn, Copenhagen

32 Isxakova, X F 1968 Avtomati^eskijsintez form suSJ^estvitel'nogo v tatar-skom jazyke NTI Ser 2, No 3 23-26

33 Isxakova, X F 1971 K voprosu o for-mal'nom opisami morfologu tjurkskixjazykov Sovetskaja tjurmlogija No6 32-44

34 Isxakova, X F 1971 Model' tatar-skogo spijaienija (obrazovame sinteti-6eskix fcmn glagola v sovranennom ta-tarskom jazyke) In Sinxronno-sopost-avitel'nyj analiz jazykov raznyx ststem,ed 6 A Makaev, pp 108-26 MoscowNauka 279 pp

35 Janus, E 1971 Pjaf pol'skix slov^piyxstate] KLESKA, PASAZER,WALKA. ZWYCIESTWO, ZWYCIE-ZAC NTI Ser 2, No 11 21-24

36 Korolev, t I 1968 O klassifikacuvozvratnyx glagolov l zapisi lx v slo-vare NTI Ser 2, No 3 19-22

37 Kulagma, O S, Mel'tuk, I A 1967Automatic translation some theoreticalaspects and the desim of a translationsystem In Machine Translation, ed ABooth, pp 139-71 Amsterdam NorthHolland 529 pp

38 Lecd, R L , Nakhimovsky, A D 1979Lexical functions and language learn-ing SL and East Eur J 23 104-13

39 Materialy k Tolkovo-KombinatornomuSlovarju Russkogo Jazyka. 1970-^1976Vypuski NN° 2,7,14, 16,17, 23, 28, 29,

34, 35, 37, 38, 42, 62, 63, 80, 85, 86Moscow IRJa AN SSSR (PredvantPublikacu PGfiPL)

40 Mel'iuk, I A 1965 Poijadok slov pnavtomatideskom sinteze russkogo teksta(predvantel'noe soobSfienie) .V77 No12 36-44 French Transl T A Infor-mations, 1967, No 2 65-84

41 Mei'Cuk, I A 1967 Model' spqaiemjaV lspanskom jazyke MPiFL vyp10 21-53 English transl see Ref 73.I 43-94, Ref 54, pp 210-57

42 Mel'euk, I A 1967 O "vneimx" razh-eitel'nyx elementax (o semantiCeskixjjarametrax) In To Honor RomanJakobson. Essays on the Occasion of his70th Birthday, pp 352-63 s'Graven-hage Momon

43 Mel'euk, I A 1968 Model' sklonenijavengerskix suSeestvitel'nyx In Pro-blemy Strukturnoj Lingvistiki-1967, edS K Saumjan, pp 344-73 MoscowNauka 411 pp

44 Mel'euk, I A 1969 Ob opredelemibol'sg men'Sej/smyslovoj sloinosti pnslovoobrazovaternyx otnoSemjax IzvAN SSSR. Ser lit i jaz. 28(2)126-35

45 Mel'tuk, I A 1970 Towards a func-tionmg model of language In Progressin Linguistics, ed M Bierwisch, K EH e ^ l p h , pp 198-207 The Hague-Pans Mouton 344 pp

46 Mel'Cuk, I A 1972 O su|^letivizmeIn Problemy strukturnoj lingvistiki-1971 ed S K gaumjan, pp 396^38Moscow Nauka 554 pp Enlarge, re-vised Enghsh version Linguistics 1976,No 170, pp 45-90

47 Mel'Cuk, I A 1973 Towards a linguis-tic "Meaning ^ = ^ Text" model InTrends in Soviet Theoretical Linguistics,ed F Kiefer, pp 33-57 DordrechtReidel 438 pp

48 Mel'Cuk, I A 1973 Konversija kakmorfologieeskoe sredstvo Izv ANSSSR, Ser lit ijaz. 32(1) 15-28 ^=>

49 MerCuk, I A 1973 Model'SprjaienijaV Aljutorskom Jazyke Moscow IRJaAN SSSR (PGfePL, vyp 45-46) 132PP

50 Mel'Cuk, I A 1974 Esquisse d'unmodele linguistique du type "Sens <=^Texte" In Problemes actuels en psy-chohnguistique Collogues internation-aux du CNRS 206 291-317 PansCNRS

51 Mereuk,I A 1974 Ob odnoj hngvisu-£eskoj modeh tipa "Smysl ^=> Tekst"Urovni predstavlemja jazykovyx vys-kazyvamj Izv AN SSSR, Ser lit tjaz.33(5) 436-47

MEANING-TEXT MODELS 61

52 Mel'Cuk, I A 1974 Opyt Teoni Ling-- 68ytstideskixModelej "Smysl Tekst"Semantika, Sintaksis Moscow Nauka314 pp

53 Mel'fuk, I A 1974 O sintaksiCeskomnule In Tipologija passivnyx konstruk- 69fj/ Diatezy i zalogi, ed A AXolodovie, pp 343-61 LetungradNauka 383 pp English version Proc5th Ann Meet Berkeley Ling Soc. ^01978, pp 224-60 Berkeley Umv Calif

54 Mel'^uk, I A 1976 Das Won Mun-chen Fmk 461 pp

55 Mel'6uk,I A 19^7 Lecas Rev Etudesslaves 501, 5-36 " '

56 MeRuk, I A 1978 Theone de Ian-gage, th^nedetraducnon Meta23(4) ''^271-302

57 Mel'Cuk, I A 1978 Toward a defini-tion of the concept Ergative Construc-tion' Proc 22nd Int Congr LingVienna, pp 384-8"^

58 Mel'Cuk, I A 1979 Studies in Depen-dency Syntax Ann Arbor, Mich 'Karoma 163 pp

59 Mel'Cuk, I A 1981 Toward a Lan- .^guage of Linguistics. Munchen Fink

60 Mel'Cuk, I A 1981 Types de depend-ance syntagmatique entre les mots-formes d'une phrase BSLP In press -,,

61 Mel'Cidt, I A 1981 Grammatical sub-fcctand the problem of the ergative con-struction in Lezgian !n \on-SlayicLanguages of the Soviet Union, ed BComne Edmonton Ling Res In press -,7

62 Mel'Cuk, I A , lordanskaja, L N , Ar-batchewsky-Jumane, X 1981 Un nou- 7^veau type de dictionnaire le Diction-naire explicatif et combinatoire du fran-gais contemporain (six entrees dedictionnaire) Cah lexicol In press

63 Mel'duk, I A , Percov, N V W^i Po -79verxnostno-syntaksideskie Otnosemja vAnghjskomjazyke Moscow IRJa AN -•()„SSSR (PGEPL. vyp 43) 80 pp

64 Mel'cuk, I A , Percov. N V 19^5Model' Anglipkogo Poverxnostnogo Sin- gQtaksisa Moscow IRJa \ N SSSR(PGfePL, vyp 64-66) 232 pp si

65 Mel'Cuk, I A , Savvma, E N 19^8 To-ward a formal model of Alutor surfacesyntax predicative and completive con- ^2structions Linguistics, Spec Issue, pp5-39

66 Mel'cuk, I A , XolodoviC. A A l*>70 K32alog (Opredelenie IsCislenie) MarodvAzii I Afriki No 4 11 J-24

67 Mel'cuk, I A , Zoikovskij, A K 1970Towards a funononmg Meaning-Text S4model of language Linguistics '̂ "̂10-47

Nichols, J 1979 The meeting of Eastdnd West confrontatjon and conver-gence in contemporary linguistics Proc5th Ann. Meet. BLS, pp 261-76 Ber-keley Umv Cahf 660 ppPercov, N V 1976 O GrammatideskixKategonjax Anghjskogo Clagola Mos-cow IRJa AN SSSR (PGfePL, vyp 90)62 ppPercov, N V 1977 Granunatij^eskoepredstavleme anglijskqj giagol'nojformy dlja celej avtomatieeskogoperevoda Linguistica Silesiana No 2139-63Reuther, T 1978 Pladoyer for dasWorterbuch Linguist Ber 57 25-48Reuther, T 1980 Nemeckie frazeologi-ceskie slovosodetanija tipa in voUigerVerzweiflung sein 1 ix russkie ekvival-enty WSA 5 207-19Rozencvejg, V Ju , ed 1974 Essays onLexical Semantics, Vols 1. 2 Stock-holm Sknptor 404̂ pp , 282 ppRozenman, A I , Apresjan, Ju D 1979Anglo-Russkij SinonimiCeskij Slovar'Moscow Russkij Jazyk 543 ppSannikov, V Z 1979-80 Sofimtel'nye1 sravniternye konsrrukcii ix blizost', ixsintaksiceskoe predstavleme I-II WS44 413-32. 5 221^2

Savvina, E N 1976 Fragment modelirusskoro poverxnostnogo sintaksisaIII Sravnitel'nye konstrukcii (srav-nitel'iiye otsojuzuye sintagmy) V77Ser 2, No 1 38^3SCeglov, Ju K 1964 Dve gruppy slovrubskogo jazyka MPiPL 8 50-66'Slodzian, M 1979 Paraphrase", sy-nonymie" et traduction In lie Colloguede hnguistique russe (Paris. 22-24 avril19^7), pp 147-51 Pans Institut d'e-tudes slavesWierzbicka, A 1972 Semantic Primi-tives Frankfurt Athenaum 235 ppWierzbicka, A 1980 Lingua Mentalis

The Semantics of Natural LanguageSydney Academic 367 ppZoikovskij, A K 1964 PredislovieMPiPL 8 3-162olkovskij, A K 1964 O pravilaxNcmanti^eskogo analiza MPiPL 817-322oikovskij, A K 1964 Leksikacelesoobraznoj dejatel'nosti ^fPiPL 867-103Zoikovskij, A K 1967 K leksiko-grafit^eskomu opisaniju somaliiskix su-sdestvitel'ny\ Narody Azii i Afriki No1 93-1022olkovbkij, A K 1970 Matenaly krussko-somahjskomu slovarju MPiPL13 35-63

62 MEL'CUK

85 2olkovsktj, A K 1971 SintaksisSomali (Ghibinnye i PO}KrxnostnyeStruktury) Moscow Hauka. 263 pp

86 Zolkovskij, A K , Md'^uk, I A 1965O vozmoiann metode l uutnimentaxsemantiCeskogo smteza NTI No523-28

87 Zolkovskij, A K., Mel'euk, I A 1967Osemanbdeskomsmteze ProbL kibem.19177-238 English Trans SystemsTheory Research, VoL 19. New York

88 Zolkovskij, A K, Md'Cuk. I A 1969

K postroeniju deistvujujee) modelijazyka "Smysl ^=^ Tekst" MPiPL 11

89 Zolkovskij, A K, Pivovarova, T V1972 IitU) 1 C^fo/Ocsematitieeskomutolkovamiu dvux anriijskix pre^ogov)IVSLiAP 3 86-96

90 2olkovskij, A K, Razlogova, E E1974 O sisteme semanticeskogo sm-teza VI Obrazcy slovarnyx state] IV-SLUP 4 36-62