I. In an important article, 'Two models of grammaticaldescription',l C. F. Hockett sets out and contrasts gram-matical statement made in terms of Item and Arrangement(IA) and that made in terms of Item and Process (IP), withexamples drawn from grammars already published as well aexamples specifically constructed for the purpose. These twomodels are selected as 'ideal' or ' cardinal' types of state-ment, to one or the other of which actual descriptions oflanguages recently published in the United States conform invarying degrees. In both types the minimal grammatical itemis the morpheme, and this is the fundamental unit for allsubsequent grammatical analysis.

At the beginning of his article Hockett remarks, however,that if one surveys the whole field of linguistic description thetwo models he discusses are not the only models possible oractually made use of. A third type or model of description is

, admitted to have had a long period of employment and a veryrespectable antiquity, that of Word and Paradigm (WP), andis familiar, in varying approximations, in the school bookpresentations of Latin, Greek, and Sanskrit, as well as ofseveral modern European languages.

Hockett mentions WP only to refer it, almost apologetically,to a possible separate consideration at a later date, remarkingthat the bulk of contemporary American grammatical descrip-tions can be brought under the IA or IP models. He admitsfurther that neither IA nor IP is as yet fully satisfactory, andthat WP deserves the same consideration as the other two.

It is the feeling of the present writer that this reconsidera-tion of WP as a valid model of grammatical description islong overdue, since neither Hockett nor anyone else seems as

1 Word 10 (1954), 21~4, also in Readings in Linguistics, ed. M. Joos,W&8hington, 1957, 386-99, and Linguistics Today, ed. Martinet and Wein-reich, New York, 1954,90-114.

yet to have taken up the suggestion made in the article justmentioned. It may well be that it offers the means of linguisticdescriptions, at any rate of some languages, that are by variouscriteria preferable to those available by either IA or IPmethods, a~d that linguistics would be impoverished if analystswere exclusIvely or even principally to subject their materialto treatment by one of these two models.l. II. WP as a :r.n0delfor contemporary grammatical descrip_

tlon may be saId to have acquired certain advantages anddisadvantages from its very antiquity and traditional statusin the presentation of familiar languages.

On the one hand, linguistics is more conscious, perhaps,than some other sciences, of the extensive and ahnost violentchanges in methods and criteria that have characterized andmade possible its progress in this century. This is nowheremore true than in America, where explicit rejection of non-formal criteria in grammatical analysis is preached (and on the,:hol~ ~ractised) by ahnost every linguist and in ahnost everyhngulstlcs textbook. The main objection to the older grammarsand grammatical treatments was their reliance on non-formalcriteria, either derived from the misapplication of the cate-gories of other languages (usually English or Latin) or fromalleged' meanings' or conceptual categories, and often on asloppy mixture of both of these. And since these oldergrammars were ahnost all, in so far as they consistentlyfollowed any model, of the WP type, WP itself became,though unjustifiably, associated with the inefficiencies andunscientific method that linguists see in the work of theirpredecessors and now, ahnost iconoclastically, strive to avoid.Indeed, as Hockett points out,2 IA is often more favourablyrated than IP just because Process is an older term and still

1 It must be remembered throughout that the P of IP stands for Process,and .the P of WP stands for Paradigm. If this is kept in mind the slightamb~valence of symbols will not cause trouble, and has the advantage ofbreVIty and also of conformity with Hockett's usage (now adopted by someother writers).

• op. cit. 213-4.

, smells of tradition as well as of the earlier confusion ofsynchronic and diachronic analysis.

On the other hand, however, the long maintained traditionalemployment of WP type grammatical statement in Europe(and WP is the model implicit in the grammars of DionysiusThrax and of Priscian) must argue for a certain conformity ofWP with some of the commonsense intuitive ideas of grammarand grammatical structure of which speakers are themselvesaware. In suggesting that this conformity, in so far as it isfound, is an argument in favour of WP, we must be quiteolear on what it is that is being put forward. No one wouldseriously support the view to-day that linguistic analysis andstatement should be based on categories taken from sentimentZinguistique or the native speaker's intuitions. But it mayfairly be claimed that a model in terms of which grammaticalanalysis can be firmly established on formal and overt criteriais in one respect preferable to other models equally admittingformal analysis, if the presentations taken from it conform tosome extent to the unformulated intuitions of native speakers.In somewhat different words, intuitions of grammatical struc-ture may be the result of a maximal convergence of variousfeatures in the language on particular units and associationsof units, which are therefore likely to yie~da strong foundationfor the systematic organization of the grammatical statement.

III. In one sense WP is not a model in opposition to IAand IP in the same way that these two are in opposition toeach other. It is claimed for IA and IP that one can accountfor all grammatical forms and constructions, from singlemorpheme to complete sentence, as either the result of thearrangement of items or of processes applied to items;nobody would suggest that word and paradigm can likewiseCover the whole field of grammar. The salient differencebetween WP and both the other two models is the centralityit accords to the word as a fundamental unit in the grammaras a whole and as the basic unit of syntactic structure. IA andIP both start from the morpheme as the minimal grammaticalelement and also the basic syntactic unit, passing through the

word as relatively unimportant, and consequently regardingthe traditional division between morphology and syntax asunnecessary or even misleading.' This division, with morpho_logy covering the formation of words and syntax the structureof sentences stated principally in terms of the relationsbetween words and word groups, remains central to WP, incontradistinction to both the other models. On the other handIA stands at a greater distance from WP than does IP in on~respect, in that WP is prepared to make use of ' process' asa term in grammatical description (cp. VI, p. 134).

It was certainly a weakness in the classical grammariansand in many later writers who followed their example, tha~they barely recognized any grammatical unit below the levelof the word, and certainly never set out with any rigour theestablishment of the morphemes of a language. A formalizedversion of WP must recognize the morpheme as the minimalgrammatical (not semantic !) unit of a language, but it makesthe word the unit that carries in its paradigmatic and syntag-matic associations the main weight of grammatical description.

!o be acceptable any model must be able to identify itsUllltSformally and state its operations in terms of which thegrammar of a language may be analysed. WP can beformalized just as fully as IA or IP. Its basic units are Words,many of which can be grouped into Paradigms of morphologic-ally associated words, and words enter into constructionsthe maximal construction (at present and by tacit agreement)at the grammatical level being the sentence.2

Below the word in terms of size, the morpheme is recognizedand established by the now well known techniques (includingthe grouping together of morphemic elements in com-plementary distribution, often called 'allomorphs'); but

1 cpoz. S. Harris, "Methods in Structural Linguistics Chicago 1951 262281. ' ",

2 'Discourse analysis' very rightly seeks to pursue structural statementbeyon~ the sentence .level, but would scarcely be regarded as strictly andexclUSIvelygrammatICal (cp. Harris, ' Discourse analysis' Lang. 28 (1952)1-30). "

me notoriously' difficult' cases of morpheme analysis mayso . d thObe simplified by the less fundamental status asslgne to ISunit in WP (cp. VI, pp. 132-3).

Formal criteria of the word as a unit are by now well known.Bloomfield's use of the' minimum free form' as a centr~lcriterion of the word is familiar,' but perhaps exaggerated. mview of the contextual unlikelihood of many words ap,Pearmgas isolated utterances. More significant is the st~ble mternalcomposition of the word, either absolute or relatIve, as com-pared with its relatively free mobility within larger sequences.This general criterion subsumes such ~th~r features .asuninterruptability by pause or parenthesIs m normal dIS-course, general immobility of the order of its internal mor~ho-logical constituents or segmental morphemes (any exceptIOnsbeing far fewer than the general mobility of words t~er:'~elvesin word groups), and limited and regulated extens~b~l~tyatany point as against the almost indefinite extensIbIlIty ofword groups by the addition of other words.2 To the wordunits of this sort there correspond to a large extent the, minimum free forms' or smallest potential utterances in alanguage, and in many, though by no means in all, languagesphonologically marked stretches delimited ?y accentualfeatures of stress, pitch or the like, and sometImes by otherprosodic marks.3 Since the word is by definition a gram-matical abstraction, any such phonological criteria mustremain logically secondary, even though in certain languagesmarginal cases may be decided one way or the other byreference to phonological criteria where the primary, gram-matical, criteria are conflicting or ambiguous.

1 L. Bloomfield, Language, London, 1935, 178. . ,• cpo C. E. Bazell, ' On the historical sources of some structural umts ,

Misceldnea Homenaje a Andre Martinet, La Laguna, 1957, 19-29; Robms,Proc. Eighth Int. Congo Ling., Oslo, 1957, 379-80. . .

General immobility of internal morphemic order prOVIdes the mamcriterion of word unity in languages like Eskimo, where, as envisaged byJ. H. Greenberg, Essays in Linguistics, Chicago, 1957,32, there is in theoryno limit to the extensibility of bases by derivational morphemes.

• cpo J. R. Firth, 'Sounds and Prosodies " TPS 1948, 127-52; W. S.Allen, , Zero and Panini " Indian Linguistics 16 (1956), 108.

Words may be divided in most languages into variable wordsand invariable words, with the concomitant identification ofroot or base morphemes, forming generally large and open-ended classes, and by which variable words are broughttogether, and of affixes, forming smaller and closed classesand by which the different forms of variable words ar~individually distinguished. The total set of word forms sharinga common root or base are grouped into one or more para-digms, and further paradigms may be similarly set up ofcorresponding sets of forms built on non-minimal or extendedbases or stems.

A cross-classification of less importance to the grammar isthat of monomorphemic words (i.e. free morphemes) andpolymorphemic words. Monomorphemic words may be mem-bers of a paradigm of variables (unless extensive and undesir-able use is made of zero morphemes to obviate this: e.g. dog,dog-s), and many polymorphemic words are invariable (notmembers of a paradigm: e.g. without, midway 1). Poly-morphemic words and bases include some constituent mor-phemes that are irrelev'lint to the grammatical class of thewhole and do not of themselves carry any grammaticalcategory or limitation (e.g. English: receive, recapture, prewar,preelect, precursor). The constituent morphemes may be inpart or wholly bound as in the examples just given, or freeand linked together by phonological features (e.g. unitarystress) and/or a positional relation not found with the corre-sponding separate words in the language (e.g. blackberry,a comeback, etc.). To this general type belong the so-calledcompound words (those written with more than one character)in the Chinese-derived part of Japanese vocabulary. Thetreatment of such words and bases has little effect on anymodel, except that in a morpheme based syntax their unionwill be stated as the lowest layer of immediate constituents,

1 That is to say, paradigms in the usual sense of limited sets of affixesass?ciated with sets (often unlimited) of bases or stems. In a wider, mainlyle:X:ICal,se~se: any.polymorphemic word could be said to fall into a paradigm(WIMut, lmthm, lmthal " midway, midstream, midday, etc.).

while a word based syntax will take their composition forgranted. In either case their analysis and classification belongsprimarily to the lexicon of the language. . . .

IV. As has been said above, WP preserves the dlVlSlOnofmorphology and syntax more clearly than the two morphemebased models, and attaches more importance to it. Construc-tions, the province of syntax, of which the maximal con-struction is the sentence, are principally made up, in WP, ofwords in the first instance, and the places in a syntacticstructure are filled primarily by words and word groups.lSentence types and construction types are analysed formallyin the now familiar terms: exocentric, endocentric, subordin-ate, co-ordinate, superordinate, by the application of the usualtests of cohesion and substitutability (distributional parallel-ism) of part and whole, with the distinctive provision thatthe cohesion of a construction or group rests on its sub-stitutability for or by a single word (of the same class as oneof its components in an endocentric construction, of a differentclass in an exocentric construction).

It must be emphasized that by the inter-word relationswithin constructions one understands the interrelations ofwords not as individual lexical items (which is the provinceof collocation), but as members of classes that occupy placesand fulfil functions in a construction by virtue of their classmembership and any inflectional categories they may bear.In complementation to this the inflectional formations andthe categories of which they are the exponents are relevantto the grammar of a language by virtue of the syntacticstructures into which they enter and the constructions andstructural relations within constructions that they mark.

In this respect syntax may be considered the core ofgrammar; a language may dispense with some morphologicalcomplexity, as in the case of modern English as comparedwith Anglo-Saxon, or some modern Romance languages ascompared with Latin, and it may reduce its inflectionalapparatus almost or quite to vanishing point, as have some

1 For reservations to this general statement see VII, pp. 140-2.

far-eastern languages. 1 But in every language words asmembers of word classes cohere in groups with specificdistributional and functional relations with each other andwith other groups to form sentences of different syntactictypes. And the formal description of these constructions andsentence types (whether the starting point is made the wordor the morpheme) does not differ markedly in its inventoryof basic syntactic terms as between the so-called inflectionallanguages with complex morphology and languages of the typerepresented by Chinese.

Hockett suggests in passing that WP is 'incapable oforganizing efficiently the facts of a language like Chinese'. 2

To say that paradigms are not part of a model applied to alanguage whose words do not exhibit grammatical paradigmsis labouring the obvious,s but it may still be urged that theword as formally established is the most profitable unit to betaken as basic in the statement of the sentence structures ofsuch languages; and this indeed is the case with two recenttreatments of different varieties of Chinese by writers bothof whom reject the facile and time-worn equation' word =

morpheme = segment written with a single character', andrecognize the word as the unit of sentence composition andas comprising simple (monomorphemic) and compoundwords.4 It may be said that it is in its grammatical structuremarkers rather than in its basic grammatical structures thatlanguages of the Chinese type differ most sharply from thoseof other types, and that therefore a word based syntacticanalysis of such languages is still, though as an extreme case,on WP lines.

1 For the view that Chinese was once an inflectional language seeB. Karigren, 'Le proto-chinois, langue f1exionelle', Journal Asiatique 15(1920), 205-32.

• op. cit., 210.3 There are, in fact, a few rudimentary paradigms in modern Chinese,

such as wo3, wo3menO, I, we; ni3, ni3menO, you (s. and pl.); t'al, t'almenO,

he, she, it, they. Vietnamese may be the 'purest' language of this type.« M. A. K. Halliday, 'Grammatical categories in Chinese', TPS 1956,

177-224 (esp. 211-2); W. A. C. H. Dobson, Late Archaic Chinese Toronto1959, esp. pp. 4-7, 18. ' ,

V. In terms of WP it has been argued that syntax is thecore of grammar, that words are to be treated as the primaryunits of syntax, and that morphological formations (or morestrictly inflectional formations 2) in languages exhibiting themserve as the markers of syntactic structures and groupingsand the relations within and between them. This last proposi-tion brings the paradigm into prominence in grammaticaldescription. The paradigms of traditional presentations,whatever semantic glosses are attempted either for a wholeparadigm or for individual members of it, bring together ~he

, inflectionally different forms into closed sets each of whIChmarks one or more of the syntactic functions of the wordclass in the constructions into which it enters; paradigmsthus represent interlocking systems of grammatical oppositionswithin particular syntactic fields. Paradigms are primarilyand mainly of single words, but where short groups of wordsor phrases (e.g. Latin, and some Greek, perfect passives) aresyntactically comparable to single words in the correspondingplaces of a different paradigm they are obviously to beincluded in paradigms themselves.

From what has been said about the centrality of syntax, itfollows that word classes are primarily distinguished by theirdifferent syntactic functions; in inflecting languages thesemay in some (but not all) classes be associated with differe~tsets of morphological paradigms. In these latter cases Itmatters little whether the class is predicated of the root orstem less inflection or of each of the words brought togetherin the paradigms.

Examples of the syntactic criteria by which paradigmcategories are formally established are obvious and need nomore than mention: Number in English, relation betweensubject (usually preverbal) noun or pronoun and present tenseverb; case in Latin, relation of noun or pronoun to preposi-tion (exocentric construction) or verb (endocentric construc-tion); gender in French, relation between noun or pronoun

1 Derivational formations are syntactically relevant only by determiningthe class of the resultant word or stem.

and adjective in endocentric (' modifying ') construction andexocentric (' predicative') construction, in German only inendocentric adjective + noun constructions.

The marking of a particular set of syntactic relations is thecriterion by which a large number of inflectional forms sharinga single base or stem is divided multidimensionally amongseveral crossing paradigms.! A familiar example is theorganization of the Latin adjectives of each grade (positive,comparative and superlative), which may exhibit upto 15 different forms, into a paradigm of number withtwo members, a paradigm of gender with three members, anda paradigm of case with six members, each paradigm coveringa different set of syntactic relations of the adjective formswith other words in syntactic groups. As a result each formof the word class is uniquely specifiable as the meeting pointof three interlocking paradigms.

Certain comments are needed here:1. Diagrams can only clearly exhibit two dimensions (across

and down) at a time. T.hus an inflected word involving morethan two paradigm categories (or three, if one resorts toperspective representation for the third dimension) mustrepeat particular paradigms under different cross headings.In the example just given this is done by duplicating the caseand gender diagram under both numbers. This representa-tional necessity, which is more noticeable with word classesinvolving more paradigm categories, for example verbs inmany languages, does not affect the logic of paradigmaticorganization.

2. Only inflectional paradigms ha ve so far been considered.Derivational formations with a common base form can also beset out in paradigms,2 though these are mostly less regular

1 The converse, of course, is nOt true. Some syntactic relations (more insome languages, e.g. English, than others, e.g. Latin or Greek) are markedby w:ord order or by the class of words involved alone, as is wholly thecase m languages of the Chinese type, in which particular classes of words(often designated' particles ') operate as construction markers.

• cpoB. Bloch and G. L. Trager, Outline of Linguistic Analysis, Baltimore,1942,56.

and offer more variation as between individual bases in theclasses involved. Some paradigms can be said to be on ornear the borderline as between inflection and derivation. Thecategory of grade in the Latin adjective is borderline in thisrespect; almost (though not quite) universal in their applica-bility to the class of adjectives, the comparative and superla-tive forms share the majority of the syntactic functions andrelations of the positive forms, though a certain few construc-tions are unique to each. In the example of paradigm treat-ment given above grade was regarded as derivational and sonot included in the (inflectional) paradigms of the adjective;the picture could easily be adjusted in terms of four inflectionalparadigms instead of three.

3. To what extent should symmetry determine the assign-ment of single forms to more than one place in a multi-dimensional frame of paradigms? This must be decided bythe degree to which the multiplication of grammatical homo-phones is held to be compensated by the greater overallsimplicity of a generally valid frame. The general symmetryof singular and plural case paradigms in Latin nouns may bea justification for continuing to recognize separate dative andablative cases in the plural even though no formal distinctionis found in any declension. A like argument can be used tojustify distinguishing nominative and accusative cases inneuter nouns and adjectives by reference to the formallyseparate corresponding masculine and feminine forms.! Onthe other hand it would hardly be reasonable to posit threegenders in the plural paradigms of German adjectives andarticles, where no formal distinction is found in any ofthe cases, merely to provide symmetry with the singularparadigms.

4. Irregular, defective and suppletive paradigms, which bydefinition are always a minority, are established subsequently

1 A somewhat similar situation makes for the recognition of six cases inthe singular of every Latin declension though two (or more) case forms arehomophonous in everyone.

(logically and temporarily) to the regular paradigms, princip_ally on the basis of syntactic parallelism with the corre-sponding members of regular paradigms (e.g. French aime,aimes, aime, aimons, aimez, aiment, and therefore vais, vas, va,allons, allez, vont).

VI. The main distinctive characteristics of a formalized WPmodel of grammatical description would then be the following:the word is taken as the basic unit of both syntax and morpho-logy, and variable words are grouped into paradigms for thestatement of their morphological forms and the listing of theirvarious syntactic functions. It will now be reasonable toconsider some of the advantages that word based WP offersin certain cases over morpheme based IA or IP.

In the first place in a grammar considered as part of acomplete structural description, the central position accordedby WP to the word may lead to a greater' congruence oflevels' 1 in languages wherein the word is itself the domainof phonological features, prosodic or other, and therefore afocal point for the statement of phonological structuring. Thisis not stressed here, however, for two reasons: 1. There areseveral languages in which the word as a grammatical unitdoes not appear to correspond clearly with any phonologicallymarked or significant unit, and 2. It is as a model of strictlygrammatical description that WP is being examined.

In these terms WP seems to offer advantages in formaldescription of two principal types: 1. The word as a unityis more easily susceptible to grl1mmatical statement than isthe individual bound morpheme, and 2. Certain problems ofmorphemic composition (morphophonology or (American)morphophonemics) become less tiresome in WP.

These may be considered in turn :

1. As has been said, the morpheme must be recognized asthe minimal element of grammatical structure; but this doesnot imply that it is the most suitable element to bear the

1 cpo Firth, 'A synopsis of linguistic theory' in Studies in LinguisticAnalysis, Oxford, 1957, esp. pp. 22-32.

assignment of all the grammatical functions fulfilled by thewords into whose composition it enters. Some entities thatare clearly to be assigned morphemic status may be seen inseveral languages to bear conflicting and even contradictorygrammatical functions when considered in isolation. This isoften accompanied by conflicting' meanings' also, but as onemust insist that grammatical methods are not determined bysemantic considerations, and that the morpheme is not to bedefined as a semantic element in the first instance, thisconsideration is to be ruled out as irrelevant at this point.On the other hand words anchored, as it were, in the paradigmsof which they form a part usually bear a consistent, relativelysimple and statable grammatical function. The word is a morestable and solid focus of grammatical relations than the com-ponent morpheme by itself. Put another way, grammaticalstatements are abstractions, but they are more profitablyabstracted from words as wholes than from individualmorphemes.

Examples of this may be considered from a number oflanguages:

In Sundanese one finds an extreme example of a languagewhose individual affix morphemes are very few in number,but in which most of them combine with stems and with eachother in forming a considerable array of grammatically differentword forms, giving a very rich set of possibilities of wordstructure. Most affixed words are built up from combinationsof root (most of which occur also as free forms withoutaffixation) and one or more of the following: -an, -kyn, -yn,ka-, pa-, parJ- (marJ-), pi- (mi-), sa-. Reduplication of thewhole root or of its initial syllable is also freely employed;other affixes are rarer both in variety of uses and in frequencyof occurrence with roots. 1

These affixes are found in both derivational and inflectionalformations, the former being the more numerous and diverse,

1 Sundanese is a language of West Java. A full treatment of this topicmay be seen in Lingua 8 (1959) pp. 337~9.

and paradigms of derivational possibilities for roots of thenoun and the verb classes can be set up, though with con-siderable lexically determined limitation on the use of aparticular formation on a particular root. As an example,-an may form denominal nouns (laut, sea; lautan, ocean),denominal verbs (tanduk, horn, tandukan, to be horned),deverbal verbs (asup, to enter; asupan, to put in), anddeverbal nouns (JrJ[fYs,to be complete; JrJg¥san, end); andsimilarly diversified derivational functions of other affixes arefound. Combinations of more than one affix exhibit a separatederivational power from that shown by either used singly.Thus parJ- + verb root yields a deverbal noun (derJe,to hear;parJderJe,what is heard); -kYn suffixed to an intransitive verbroot forms the corresponding transitive verb (sare, to sleep;sarekYn, to send to sleep); but parJ- and -kYn together affixedto a verb root produce a ' benefactive' form (bawa, to carry ;parJbawakYn, to carry (something) for someone.

One sees that by themselves several of the most commonlyused affixes have no set grammatical function or non-contradictory group o( functions, but serve to form mor-phemically complex words possessed of very definite gram-matical functions and belonging to definite word classes. It isto be noted that one is not dealing with whole series ofgrammatical neutrals; though grammatical neutrals (wordforms serving both as nouns and verbs) are found fairly freelyamong roots and occasionally among derivative formations,the great majority of the formations referred to above havea single and perfectly precise grammatical function: and thisis a property of the word as a whole unity, and not predictablefrom its component morphemes in isolation (unless anintolerable degree of homophony among the affixes werepermitted).

In written Japanese a distinction of categories central tothe verbal system is that of Conclusive and Attributive,conclusive forms corresponding somewhat to the indicativetenses of verbs in other languages and mainly occurring insentence final position, attributive occurring before nouns to

form endocentric nominal groups, and also before certainparticles and in some other positions in the sentence; thus:conclusive, kono koto 0 sirubesi, (one) should know this(thing); attributive, sirubeki koto, a thing that (one) shouldknow. 1 This distinction is marked by the final morpheme ineach verb form, but among the morphemes most frequentlyso used -ki and -si are each found as both attributive andconclusive. Thus we find: attributive yukisi, yukitarisi,(which) went, and yukazarisi, (which) did not go; conclusiveyukiki, yukitariki, went, and yukazariki, did not go; butattributive yukubeki, (which) should go, and yukitaki, (which)wants to go; conclusive yukubesi, should go, and yukitasi,wants to go. In the inflected adjective, which in Japanesebelongs to the class of verbals both in morphology andsyntactic function we find: attributive hayaki, early,hayakarisi, (which) was early, and conclusive hayasi, is early,hayakariki, was early.

None of the words quoted is grammatically ambiguous(though there are some homophonous attributive and con-clusive forms, e.g. yuku, goes, which goes); but the gram-matical significance of the morphemes-ki and -si is onlydetermined within the complete word in its paradigms ofcategory oppositions.

Similar situations may be seen in English and otherEuropean languages where a single morpheme shape, or classof such shapes in complementary distribution, may as anisolate serve quite different grammatical functions, but withoutgrammatical ambiguity in the whole words of which they formpart. Thus we have in English -er serving as comparativeadjectival suffix and as ' agentive ' deverbal noun formative,and -(e)s ([-s], [-z], [-iz]) as third person singular present tenseBuffix in verbs and as plural suffix in nouns; and in German

1 In the Nipponsiki romanization here used si ([lli]) = ski, ti ([~i]) =chi, and 0 = wo of other transcriptions in the examples cited here andelsewhere in this paper. For further details of the grammar see B. H.Chamberlain, A Simplified Grammar of tke Japanese Language, revisedMcilroy, Chicago, 1924.

the prefix ge- acts both as the formative of collective and?ener~l nouns (e.g. Geste~n, Gerede) and (often with otherInflectIOns) as past partIcle marker in verbs (gekommgernaeht). en,

Latin and anci~nt Greek are well known for the difficultiesthey reveal, unlIke languages of the type represented bJapanese and.Sundanese men~i?ne~ above, when one attemp~to find a umque segmentabilIty Into serial morpheme seg-ments; most morphemically complex words admit a numbf h' ero ~orp ~mlC analyses almost equally justifiable by COm-

parIson wIth other forms elsewhere in the language.l But inthese languages we find morphemic elements which in isolationare grammatically a~biguous or even contradictory but helpto form words revealIng a clear-cut grammatical function int~e set of oppositions maintained by the members of theirdifferent paradigms.

In Greek verbs with a sigmatic aorist and future tense -a-may be abstracted as the tense morpheme. If the boundmorpheme.s are to be~r grammatical descriptions individually,-c::- X:lUst eIther be assigned a hazy' non-present' grammaticalSIgnificance only clarified in the word,2 or at the cost of anuneconomic homophony of different morphemes, a distinctfuture morpheme and aorist morpheme, both with the shape-~-, must be posited. A similar problem arises with thesI~gular members of the perfect and aorist indicative para-dlgm~ of some -JL~verbs (O€OWKa, etc., €DWKa, etc.; T€BHKa,etc., EB'Y)Ka, etc.).

~f, however, the word is taken as the basic grammatical~It of de~cription. (as in the traditional grammars), thenA,vO'w and EA vO'a, WIth the rest of their paradigms, are respe~-tively the bearers of the categories of future and aorist tensetogether with the other relevant categories, and likewis~

1 cpo C. E. Bazell, Linguistic Typology (inaugural lecture) London1958,8. ' ,

• M. S. Ruiperez, 'Neutralization of morphological oppositions' Word 9(1953), 246, suggests that the' momentary' feature of the aorist 'stem as~ontrast~d ';ith the 'durative' feature of the present stem is neutralizedm cOlhbmatlOnwith the future category.

8€OWKa and EowKa, as whole words, bear the categories ofperfect an~ aorist respectively.l Their mor~hemic co~positionas includIng -0'- and -K- morphemes IS recogmzed, butcategories need not be assigned at that level to these mor-phemes, as would be necessary if the morpheme is to beregarded as the fundamental unit of grammatical description.

A number of related problems have been left untreated here,including the possible morphemic identification of the -0'-

element with parts of some non-sigmatic (strong) aorists andfutures, the morphemic segmentation of the rest of the wordscited, and the morphemic status of the augment and reduplica-tion; but these omissions do not affect the significance ofthe examples in comparing a word based and a morphemebased grammatical analysis.

A similar case of morphemic ambiguity (among others thatcould be mentioned) arises in the future and imperfect tensesof Latin verbs of the first and second conjugations (amabo,amabam, etc.), unless use is made of the morphemic distinc-tion, adopted by A. A. Hill,2 produced by taking -b- as thefuture tense morpheme, and -ba- as the imperfect tensemorpheme, which runs counter to the entirely parallel syllabicstructure of all the word forms concerned.

2. We now turn to the second reason for preferring, in somecases, a WP treatment to an IA or IP one: WP avoids someof the difficulties in morphophonology (morphophonemics),in the relating of grammatical structuring to phonologicalstructuring, which beset IA (and to a lesser extent IP).

In languages of an almost pure agglutinative type, of whichTurkish is often taken as an example, it is usually possibleboth to segment polymorphemic words into serial segmentalmorphemes without difficulty and to assign each to a par-

1 The names given to these grammatical categories are here used aslabels only; the grammatical significance of the categories lies in their, sequential' and other syntactic relations with the rest of their sentences,rather than in the temporal and aspectival meanings that may be assignedto them, important as these are in the total analysis of the language.

2 Introduction to Linguistic Structures, New York, 1958,466-7.

ticular grammatical category as its exponent. Other languages,such as Latin, Greek, and Sanskrit, and to some extentEnglish, freely admit cumulation of categories in a mor-phemically indivisible segment.

It has already been made clear that a formalized versionof WP must admit the morpheme as the minimal unit ofgrammatical formation, below the word level; but if the wordis to be the focus of grammatical statement, the segmentationinto morphemes need not take into account any need for aparallel representation of the grammatical categories applic-able in every word in any class (though, of course, it may doso if a clear statement on these lines is possible). Whateverthe morphemicization of amabamus and bonarum, thecategories of tense, mood, voice, number, and person, andof case, number and gender are exhibited by the verb andnoun forms as whole words taken together. Phonologicalstructure does not cut across morphemic analysis in words likeEnglish took and geese, or French au. The grammaticalcategories are borne by the words in their paradigms, and noobjection need be take~ to paradigms varying in morphemecomposition from word to word within a word class. Thusrang (the bell) and r~nged (the bird's leg) are both past tenseforms in the paradigms ring, rings, ringing, rang, rung, andring, rings, ringing, ringed; and rang and rung, like tookand geese, are ' perfectly respectable' monomorphemic words,as their whole phonological composition argues.! The' tacticalparallelism' between took and baked that prevents Hockett 2

admitting their morphemic non-parallelism is not binding onWP as in this model it is the word that is the (syn)tacticalunit; and we are spared the subterfuges of zero suffixes withconditioned allomorphic variation, infixes (a formation other-wise unknown in English), morphemes 'consisting of vowelchange " and suffixes that are not suffixed.3

If morphemes are the Items that have to be Arranged, then

1 cpo Hazell, 'The correspondence fallacy in structural linguistics',Studies by Members of the English Department, University of Istanbul 3(1952),4-5. • 'Two models', 224. • Hill, Introduction, 140-1.

olearly something must be found comparable to the oed ofbaked in took; but if took is just part of the paradigm of theverb take, no such need arises. Among paradigms of whollypolymorphemic words one may plausibly regard Latin first

, and second conjugation futures (amabo, monebo, etc.) ascontaining one more morpheme than the futures, grammatic-ally equivalent in every way, of the third and fourthconjugations (regam, audiam, etc.).

The number of morphemes in words belonging to compar-able paradigms may vary, and categories may be serially orcumulatively expressed. In consequence the ' empty morph 'as an oddity disappears.! If the segmentation requires or isfacilitated by a recurrent partial without any grammaticalcategory applicable to it, this is of no consequence; it is partof the word to which it belongs and which, as a whole,exhibits the categories relevant to it as a whole.

Process and Arrangement both have their place in agrammatical description. Process is most naturally applicableto the morphological formation of words, and arrangement tothe syntactical relations (not linear order, though this is often,in varying degrees, a mark or exponellt of syntacticalrelations) of words as members of classes bearing specificgrammatical categories in constructions. . .

Process adequately covers the ,variety of formal aSSOCIatIverelations found between root and stem forms (III, p. 121) andmembers of the paradigms, and provides a single means ofdescribing the formation of these words by reference to aconstant or common element (except in suppletive paradigms)underlying them, whether or not the root or stem form itselfserves as a paradigm member or not. The processes employedin languages include affixation (fix-ed, €-7T{aTw-aa), alterna-tion of vowel and/or consonant structure (sing sang, bring,brought), reduplication in whole or part of base form (AI-Au-Ka,Malay akar-akar, roots), compounding bUrek-berry, Japanesesan-sui, landscape (mountain-water)), stress difference

1 cpo Hockett, 'Problems of morphemic analysis', Lang. 23 (1947),321-43.

(export, export) and tone difference (Ibo isi--, head, ok" _rat, isi OkE-- __ , rat's head 1). - ,

Su.ttraction is also a possibility,2 though the advantages ofkeepmg to the constant (in regular paradigms), and thereforeshort~r, form ~s the descriptive base probably outweigh theoccaslOnally bnefer statement of formation rules made avail-able by recognizing subtraction.

It may be said that in affixation at least, Arrangement isas appropriate a descriptive term as Process. To this onereplies that the intra-word bond is always much closer thanthe b?nd between words, in the restriction or completeexcluslOn of further affixation of bound morphemes, and inthe almost complete fixity of internal order, whereas syntacticwor~ groups are often variable in relative position, usuallyadmIt the further insertion of other words, and are almostalways further extensible in one or both directions by otherwor~s. This difference between affixation and word groupinghas. Its counterpart on the phonological level in that junctionor Juncture features are more frequently admissible (if notalways found) between/rather than within words. Affixation,even where the fusion of the elements is at a minimum (as inth~ purest agglutinative type languages) is a much tighterUllIon than syntactic juxtaposition, and this justifies itsinclusion in the class of processes describing the morphologyof members of paradigms. Conversely, in 'allegro forms'where syntactically separate words are phonologically fused(e.g. English [wount] (won't), ['aiuana 'get it] (I'm going toget it», there is almost always the grammatically equivalent(and semantically near equivalent) non-fused sequence avail-able ([wil not], [ai am 'gouiU ta 'get it]).3

1 See further I. C. 'Vard, An Introduction to the Ibo Language, Cambridge,1936, chapters 4 and 8; An Introduction to the Yoruba LanguaJe, Cambridge,1952, p. 30 and Appendix I, cpo Eugenie J. A. Henderson, •ColloquialChin as a pronominalized language,' BSOAS 20 (1957) 324-5; otherexamples in K. L. Pike, Tone Languages, Ann Arbor, 1948, 22-3.

• cpo Bloomfield, Language 217.3 In this way words like English won't are descriptively contractions in a

sense in which French au is not a contraction of ale.

In keeping intra-word Process and inter-word Arrangementas separate terms, it would also seem better to restrict, exocentric' and 'endocentric' to syntactic (inter-word)structures, , derivation' and' inflection' being more suitabletermS with reference to word structure and covering approxi-mately the comparable situations; and the collocation of, semantic' with 'endocentric' and 'exocentric' in theclassification of compound words 1 only brings confusion byreintroducing semantic criteria into what should be formalanalysis and classification.

VII. We must now consider the possible objections thatmay be made, and the disadvantages that may accrue toWP, or to a model of grammatical analysis wherein the wordis the unit of morphological description and the basic unit ofsyntactic analysis.

Objections may be either theoretical, and in fact almostaesthetic, or practical.

Firstly it may be argued that with its use of Processterminology, WP, like IP, implies a historic perspective andconfuses synchronic linguistics with diachronic linguistics.This should by now have lost its force. 'Process' as usedto-day in any model of descriptive grammar is simply a meansof relating formally one word or form to another word orform by the most economical statement, and implies no sortof time dimension.2

Secondly, and more seriously, WP would appear at firstsight less tidy and economical in requiring both Process andArrangement as separate terms (in morphological and syn-tactic description respectively) than either IA or IP withtheir exclusive use of one or the other. The tidiness andeconomy of these models, however, is more apparent thanreal when it is seen that Arrangement in IA includes (by

1 As in, for example, Hockett, Course in Modern Linguistics, New York,1958, 185; W. Bright, The Karok Language, UCPL 13, Berkeley and LosAngeles, 1957, 70.

3 cpo A. R. Diebold, reviewing Bright, The Karok Language, Lang. 35(1959), 392.

whatever devices, with or without zeroes) internal vowel andconsonant alternation and all accentual differences, and thatProcess in IP is made to cover such patently diverse thingsas 'past tense formation', 'predication', 'addition' and, modification', as well as the markers thereof such as, vowel change' and the like.!

Thirdly it may be felt that a descriptive model is intrinsic_ally less desirable if it makes a non-minimal element, theword, basic in the hierarchy of structures at the same generallevel of analysis, as against both the other models that makethe minimal grammatical element, the morpheme, also thebasic element of structure. The failure of the ancient gram-marians of Latin and Greek to recognize the morpheme, andof their traditional followers of more modern times to takeproper account of it, shielded them from the need to considerthis possible objection to their method and so weakened itsappeal.

The answer to this has in effect already been given. Inmany ways, and quite apart from any phonological markers,the word is a unique entity in grammar, and not just a stagein the progression 'from morpheme to utterance'. As agrammatical element the word is unique in its relative fixityof internal morphemic structure, its focal status in relationto syntactically relevant categories, and, in inflected words,the stability of its paradigms. All of these factors make it astrong basis for grammatical description, both morphologicaland syntactic. The assumption of a simple ascent in order ofsize from single morpheme to complete sentence, ignoring orblurring the distinction of morphological structuring andsyntactic structuring, achieves its apparent simplicity at thecost of neglecting or distorting patent structural features oflanguages.

The most serious objection to WP is the practical one,namely that at least in some languages the word is not infact the best unit for statements of syntactic structure, and

that word boundaries repeatedly cut across any reasonablenalysis into immediate constituents.

a It is agreed that immediate constituent analysis underliesost schemes of syntactic statement, and that the basis of

m uping units into constituents is principally replace-gro . Iability or syntactic equivalence between a gr~up and .a smg eunit.! If the morpheme is taken as the smgle umt, t~enmany examples appear of immediate constituents cuttmgacross word boundaries; for example, in large books, as .atypical English adjective noun construction, large book- ISreplaceable by book- in almost every case, whereas the. wordbooks (or any other pluralized noun) is only replaceable m theimportant preverbal (subject) position by a single morphemefrom the very small class of monomorphemic pluralia tantumlike cattle. But WP takes the word as the basic syntacticunit, and the question to be asked is to what extent are wordboundaries cut across by immediate constituents when syn-tactic equivalence to a single word is taken as ~he m~incriterion of constituent status. Wells indeed makes It a pomtof principle that the word should be treated as a unitaryconstituent in all cases except when this clearly engenders aconflict or an avoidable complication.2

One may compare the possible immediate constituentanalyses of the same sentence on the basis of the morphemeas the minimal constituent and on the basis of the word asthe minimal constituent. Harris's example' My most recentplays closed down' offers the following possibilities 3 :

I r:L2 r-tI r4-:,my most recent play-s close-a downI y, I Y

1 cpoBazell, , Correspondence fallacy', 23-4 .• R. S. Wells, 'Immediate constituents " Lang. 23 (1947), 81-117.3 M ethads in Structural Linguistics, 278-80.The line diagramming used here seems the most suitable for. present

purposes; it can, of course, be easily and uniquely converted mto theQuine system of dots as given in Harris's text.

The union of play and -s and of close and -d in the secondanalysis is accounted for elsewhere in the grammar, in themorphology of the noun and verb paradigms; for the restthe two analyses are equally efficient and the word basedone is justified on the same general ground, substitutabilityof the group with the single unit (in this case the word).My may be treated as a more' outer' constituent than mostrecent by virtue of its virtually closing the endocentric nominalconstruction in one direction; only all or both and someadverbs like only and exclusively can precede the possessivepronouns in a larger endocentric construction with the finalnoun as head. Harris appears to treat my in this way mainlyto preserve a parallelism between order of constituency andorder of morphemes in the text; this breaks down in theverb group where he has to admit -d as an infix (a categoryavoided in English grammar by a WP type analysis). How-ever, in some languages, e.g. Latin, on either methodimmediate constituent structure and positional order of unitswould bear scant correspondence.

In many cases the immediate constituent analysis on botha morpheme and on a word basis will be very similar apartfrom the treatment of intra-word composition elsewhere, witha consequently less complex constituent structure. In somecases however a co~struction may be exocentric by oneanalysis and endocent'ric by the other. A sentence in writtenJapanese yields the following possible analyses:

, II

III A. I -C:teki gun-ta~ 0 kakoma-ba ti-ten

enemy troop object if (you) positionunit particle surround II

I

~I~lo sir-as-e

obj. let (us)part. know!

_~~~I II I

(if you surround the enemy troop unit report its position)

The curved lines link lexically united morphemes in com-pound words (III, p. 121) and need not be further considered

here; siras- is a causative stem formed on the verb root sir-, toknow. In the upper (morpheme based) analysis teki --- kakoma-is syntactically equivalent to a single verb base (mi-, see,yuka-, go, etc.), and the whole clause teki .-- kakomaba is re-placeable by a monomorphemic adverb (e.g. ima, now) and isconsequently exocentric, since neither constituent, verb baseand expansion, or hypothetical suffix is substitutable for thewhole clause. In the lower (word based) analysis the wholeclause is replaceable by any hypothetical (subordinating) verbform (miba, if (you) see, yukaba, if (you) go, etc.), and theclause is therefore endocentric. Within the nominal group ofthe English example above, most recent play-s is exocentric ona morphemic analysis and endocentric on a word analysis.Probably exocentric constructions will always be morenumerous in a morpheme based constituent system than ina word based one.

In general one may claim that WP permits an equallyefficient and sometimes simpler immediate constituent analysisin grammatical statement. Constructions that are ambiguousin isolation in another model are also ambiguous in WP;old men and women is resoluble one way or the other eitherby context, linguistic or non-linguistic, or is made explicit bypause features not ordinarily present: old, men and women;old men, and women'!

There remain, however, in several languages some hard casein which immediate constituent analysis must cut across wordboundaries. In English there are sentences on the pattern ofthe by now notorious example the king of England's hat, theman I saw's daughter, commoner in colloquial spoken Englishthan in more formal discourse or in writing. The's is mostsuitably treated as a bound form and as part of the word to

1 cpo Hockett, 'Two models', 219-20. It may be pertinent here toprotest at the tendency among certain writers to-day to assert that juncturaland other phonological features are always present at certain grammaticalboundaries and in particular constructions, and can therefore serve as theircriteria. Such features are better treated as merely available and inter-mittently employed as additional marks of grammatical differences.

which it adheres, as it is neither itself a potential free formnor the syntactic equivalent of one, nor is there availablea corresponding form consisting of more than one consonantas there is with the smallest ' weak forms' of other Englishwords.' But this morpheme, excluded from the category ofwords, may belong grammatically, as in the examples abovein immediate constituency to a noun disjoined from it in th~sentence. It is interesting to see that a corresponding situationis found among the Scandinavian languages in Danish andNorwegian but not in Swedish.2

In still more colloquial English one finds almost nonce formslike extract-of-beefed his bread,3 where beef could not appearalone as a transitive verb.

Expressions such as bandy-legged, heavy-fisted, and the likeare rather border-line cases; they are normally stressed assingle words and are hardly extensible like word groups.Dark and shaggy-bearded man would ordinarily be construeddark and slfaggy-bearded man,4 and sequences of this type withI I i I r.-...:.......J Ithe first adjective not caUocable with the final noun, thoughjust possible, are unlikely (sandy- and shqggy-bearded man).

I " \ I IIEnclitics in other languages (e.g. Latin and ~reek) present

similar difficulties for word based immediate constituentanalysis, unless their relative freedom of position (other thanin sentence or clause initial position), which is greater thanthat of English's, and their word-like segmental structuremay be held to justify word status despite their accentualdependence on the immediately preceding word. The inflected~apanese form -rasi-, looks like (spoken form -rasii), is similarill this respect. It is a bound form, cannot occur initially,

1 The form [-iz] (horse's mane) is, of course, only a contextually boundvariant (allomorph).

2 cpo K. Togeby, ' Qu'est-ce qu'un mot " TOLO 5 (1949), 106.• This example is from Punch magazine.• For this' tripartite' constituent see Wells, ' Immediate constituents'

103. '

and is accentually treated as part of the word it immediatelyfollows. In a sentence such as (spoken Japanese) kodomorasii,it looks like a child, -rasii coheres grammatically withkodomo; in kodomo norasii, it looks like a child's, it coheresgrammatically with the group kodomo no (and norasii wouldnot be a possible group by itself), but is phonologically boundto no. Bloch and Wells, despite their general principle of wordbased constituents, do not regard -rasii as a word, but asa suffix of unique structural patterning.'

In Yurok, a native language of North California,2 a crosscutting of word boundaries by immediate constituents doesoccur, but in one type of construction and then as a free, butless frequent, variant of sentences without the cross cutting.kic toktomoyel ki ?umuc megetolkwepel, they are now oldenough to look after themselves, is in free variation withkic toktomoyel ki muc ?umegetolkwepel, where ?u- in bothsentences coheres grammatically with the rest of the verbform megetolkwepel, to give a pronominal prefix verb form, asis used in subordinate clauses. Thus we have:

ki ?umuc megetolkwepel and ki muc ?umegetolkwepel.I' L;= , I I '. ' '

If these types of cross cutting were frequent in languagesit would be a serious and indeed fatal objection to a wordbased immediate constituent analysis, and a word basedsyntax, as a whole. But in the languages cited, those elementsgiving rise to this cross cutting are in a small minority, andin default of further evidence it would seem that they are besttreated separately, and that marginal elements should not beallowed to disturb the choice of a model of description thatdisplays considerable advantages over most of the system.

A possible example of a language ill adapted to a WP typeof grammatical analysis is Mixteco as described by Pike.3 In

1 B. Bloch, 'Studies in coloquial Japanese: II Syntax', Lang. 22 (1946),205; Wells, 'Immediate Constituents', 101-2.

2 See Robins, The Yurok Language, UCPL 15, BerkeleY and Los Angeles,1958,57-8.

• K. L. Pike, , A problem in morphology-syntax division " Act. Ling. 5(1945-9), 124-38, and further refs.

that language sequences of morphemes belonging to differentimmediate constituents may all cohere phonologically into uni-tary stretches, and grammatical and phonological boundariesconstantly fail to coincide. However, a different picture ofthis language might emerge if phonological features were nottreated as definitive, and the word were defined in grammaticalterms as a morphemic sequence with positional mobilitywithin the sentence and absence or restriction of morphemesinsertable within its boundaries, and with potential occurrenceas a free form (or complete sentence) or having a regularlexical correspondence with such a free form (many lexicalitems are said to appear in two forms, one as used alone andin certain constructions, and a shorter form as used in certainother constructions).

VIII. Of the three models, lA, IP, and WP, no one has asyet been worked out to be equally suitable for every part ofa grammatical system in every language, 1 a fact hardlysurprising in view of the immense complexity of language.But, if no model is wholly' right " no model that can bestated and applied in f~rmal terms is 'wrong', in the sensethat a model of grammatical description relying on intuitionsand on classifications based on unverifiable categories of, meanings' is, in this day and age, 'wrong '. WP is aworkable model of formal grammatical description, offeringmany advantages in analysis, not to mention greater clarityin the allied field of pedagogical presentation.2

This is not to say that traditional paradigm presentationsexhibit the morphology of their languages in the best possibleway; there is clearly scope for the examination of severaldifferent ways of organizing the material of a language into

1 cpoHockett, 'Two models " 233.2 Nothing is said in this papcr of the effect of applying transformational

analysis on thc choicc of grammatical models (scc N.Chomsky, SyntacticStructures, 's-Gravcnhage, 1957). Chomsky, whose work, of course, post-dates Hockctt's 'Two models " seems himsclf to operate more or less inmorpheme bascd IP terms with transformations as the major class ofProcesses; but it would appear equally possible and fruitful to applytransformational proccdures to a word based syntactic frame.

paradigms. It may also be that while each of the modelsdiscussed in this paper is feasible with every language, one ofthem is more appropriate with certain languages; possiblyMixteco, at least on Pike's analysis, is not a' WP language',and certainly on the evidence we have considered somelanguages are less suitably' IA ' or ' IP languages'.

WP when reworked in terms of current formal criteriadeserves proper consideration as a means of stating andanalysing grammatical systems. It is due for a moresympathetic hearing than it has received in recent years, andincreased representation among the descriptive grammars oflanguages newly made available to scholarship.