678 REMARKSANDREPLIES

NavajoVerb Stem Positionand the BipartiteStructure of the NavajoConjunct Sector

Ken Hale

TheNavajo stem appears at the rightmost edge of the verb word. Innumerouscases it forms a lexicalconstituent with a preverb,occur- ringat theleftmost edge of thesurface verb word, much in themanner ofDutchand German verb-particle arrangements in verb-secondfinite clauses.In Navajothe initial and finalpositions are separated by eight morphemeorder ‘ ‘slots’’ recognizedin the Athabaskan literature (and describedin detail for Navajo in Young and Morgan 1987). A phono- logicalsolution to this and a numberof otherdeep-surface disparities isexploredhere, based on theinsights of earlierworks on theNavajo verb,including Speas 1984, 1990, McDonough 1996, 2000, and Rice 1989, 2000. Keywords: verbmorphology, morphosyntactic disparity, spell-out, Navajo,Athabaskan

1Verb Stem Position TheNavajo verb stem forms asinglelexical constituent with the prefixal, particle-like category called preverb inthe Athabaskan linguistic literature (see Rice2000). However, like particle- verbcombinations in Dutchand German verb-second clauses, the parts of thisconstituent in the Navajocounterpart are separated in the surface forms ofverbal projections in syntax. In the Navajoversion of thisarrangement, the preverb occupies the leftmost position in the verb word, andthe verb stem occupies the rightmost position. TheNavajo verb in (1) canbe used to illustrate the phenomenon. The verb is segmented intoits various parts in (2), and the components are identified in slightly greater detail in (3).

(1) Sila´o t€’o´o´’go´o´ ch’´õ shidinõ´€daøzh.(Young and Morgan 1987:283) ‘Thepoliceman jerked me outdoors (to get me out of afight).’ (2) ch’´õ sh d n-´ € daøzh PreverbObject Qualifier Mode/ SubjVoice V (stem)

Iwishto dedicate thisarticle totheNavajo Language Academy, asmall groupof linguists and teachers devotedto Navajolanguage research andeducation. And I thanka numberof peoplewho have helped me inmy efforts to learn aboutAthabaskan linguistics, some ofwhomhave graciously read orlistenedto these remarks, offeringvarying measures ofcriticism andadvice, including Jonathan Bobaljik, Leonard Faltz, Theodore Fernald, Victor Golla, Tom Green, Morris Halle, SharonHargus, Ken Hiraiwa, EloiseJelinek, James Kari, Jay Keyser, MichaelKrauss, Jeff Leer, Alyse Neundorf, EllavinaTsosie Perkins, Paul Platero, Bill Poser, Keren Rice, NorvinRichards, John Ritter, Haj Ross,Leslie Saxon, CarlotaSmith, Margaret Speas,Lisa Travis,Siri Tuttle, MaryAnn Willie, and Cheryl Zoll. I regret notbeing able to reflect fullythe contributions of these peopleat thistime, butI willdo so in future work I hopeto do on the Navajo verb.

Linguistic Inquiry, Volume 32, Number 4, Fall 2001 678–693 q 2001 bythe Massachusetts Institute ofTechnology REMARKSANDREPLIES 679

(3) ch’´õ Preverb ‘,out horizontally’ sh ‘firstperson singular object’ d Qualifier ‘thematicprefix relating to arms andlegs’ n-´ Mode/Subj‘ portmanteauof n-situationaspect, , and thirdperson ’ € ‘transitive’ dazh Stem ‘verbalroot, move in jerky manner; and momentaneous perfectivesuffixal ’ The preverb ch’´õ ‘outhorizontally’ belongs to thedisjunct (proclitic-like) sector of theNavajo verb,while the other elements belong to the conjunct sector, the system of nuclear elements (prevailinglyconsonantal in form) thatmake up theextended projection of the verb(see Grimshaw 1991on the concept ‘ ‘extendedprojection’ ’). 1 Allof these parts are assembled into a single phonologicalword (and so written in most cases). Theproblem addressed here consists in thecircumstance that the preverb and the verb stem form aunitin the lexicon, a factwell argued in Rice 2000, in stark contradiction to the above- notedsurface arrangement of the constituents of the verb word. Quitereasonably, in her excellent study of Athabaskan scope and morpheme order, Rice proposesa movementanalysis to account for thisdisparity. The underlying organization she assumesfor theNavajo (and general Athabaskan) verb word is shown in (4), in which Rice’ s principlesof lexical integrity and leftward semantic scope are expressed structurally.

(4) Mode/Subj

Qualifier Mode/Subj n-9

Object Qualifier d

Voice Object sh

VP Voice

Preverb V ch’í daÎ zh

1 Witha few exceptions,a (monomorphemic)conjunct consists of acoronalconsonant ([d, j, s, z, sh,zh, n, €,l]),an oral glide ([y, w (roundedvelar), gh 4 g (unroundedvelar)]), or alaryngeal([’ , h]).These functionas onset orcoda,depending in part on the nature of thesegment andin parton context. In the majority of cases, where theyare realized as onsets,conjunct are supportedby theNavajo default vowel [i], regarded as eitherepenthetic or basic, 680 REMARKSANDREPLIES

Toderive the surface configuration, two raising operations take place. The verb ( -dazh) raises andright-adjoins to itsstructural governor, Voice. Rice argues, correctly I believe,that the Voice elementis alightverb that selects the verbal category VP and,in its productive uses, defines the transitivityof the verbal construction (see Hale2000). Subsequently the Voice, now ladenwith its complement V, itselfraises and right-adjoins to the highest nuclear element, the portmanteauMode/ Subj,as shown in (5). 2 (5) Mode/Subj

Qualifier Mode/Subj

Object Qualifier Mode/Subj Voice d n-9

Voice Object Voice V sh daÎ zh

VP Voice (trace)

Preverb V ch’í (trace) Thealternative I willconsider here makes a differentclaim concerning the Qualifier and Mode/Subjnuclear elements. The first of these selects thesecond; consequently, the c-command andscope relations between these elements are opposite to those assigned to them by Rice. In addition,the functional projections of Navajoare taken to beleft-headed, rather than right-headed asinRice’ s analysis.The configuration defined by these assumptions is shownin (6).

dependingon the particular analysis— Iassume theformer forpresent purposes. Some conjunct prefixes are composite, assumingthe shape CVC, in which V isgenerally the default [i]; a few havethe surface formC- ¢ (consonantand ‘ ‘floating hightone’ ’ dockedon anassociated neutralvowel); a few appear tocontain the long vowel [ii] as theirvocalic nucleus (e.g.,the qualifierand the first person nonsingular subject); the optative appears tobe realized as therounded vowel,written [o] (long or short,high or low tone, depending on various factors; see Kari 1976);and the second person nonsingularsubject is [oh]in its basic form.The neutral vowel is veryoften assimilated tothe features ofan adjacent segment,giving the superficial impression that the conjunct sector permits all ofthe vowel qualities in the Navajo inventory.It is quitegenerally assumed, however,that underlyingly the conjunct sector permits onlythe default neutral vowel,whose quality is [i]when unassimilated —unlikethe disjunct sector, which allows all vowelqualities and all consonanttypes. Where [’]wouldbe expectedto occupy a codaposition at theend of aderivation,it is normallyabsorbed intothe adjacent (following)C, where itappears as ‘‘glottalization.’’ Thenotorious d- neversurfaces as acoda; instead,it surfaces (a) as theso-called d-effect, or(b) as anonset to one of therare V-initialverb stems. 2 Forthe purposes of thepresent discussion, I arbitrarilyset aside here analternative conception according to which Voice enters intothe Merge relation with Mode/ Subj. REMARKSANDREPLIES 681

(6) Qualifier

Qualifier Mode/Subj d

Mode/Subj Voice n-9

Voice VP

Preverb VP ch’í

Object V sh daÎ zh

Theobject is introduced as a complementto V, andthe preverb is introduced as an adjunct to thatcategory. Crucially, these are not heads (i.e., nuclear elements) in the extended projection of theverb. The heads, from lowestto highest, are as follows: V, Voice,Mode/ Subj,and Qualifier. Jointly,these define the conjunct sector of theNavajo verb word. Thequestion of interestnow ishowthesurface order of morphemes is achieved.In particular, canthe surface positions of theverb and its dependents be definedsimply and, preferably, without resortto movement (in the usually understood sense of theterm)? Ibelievethe answer to this question is positive, if we adoptthe theories of McDonough (2000)and Speas (1984) according to whichthe verb is expressed phonologically in theform of aminimaldisyllabic skeleton that must be filledout to definethe phonetic realization of aNavajo verb.3 Imaintainthat it is precisely this process of ‘ ‘fillingout the skeleton’ ’ (especiallythe receptor, orleft-hand,portion of theskeleton) that accounts for thesurface arrangement of the verb, itsdependents, and the nuclear elements in its extended projection. There is no headmovement in thetraditional sense. The processes involved are properly speaking phonological. Myproposal is this.The verb stem brings with it a structuralphonological skeleton whose right-handhalf is filled out (sometimes partially, sometimes fully) by thestem itself, as indicated in(7) byunderlining. (7) CVCCVC

3 Marginalexceptions to the minimal bipartite skeleton exist: for example, nõ´ ‘(he/she) says’and le’‘optative,let itbe/ become’(respectively from hypothetical bipartite * dinõ´ and *wo´le’).These consistof the stem portionof theskeleton alone.It is conceivablethat they are tobe regarded as particles orenclitics inthe Navajo of today.The future auxiliary doo (avariantof the full verb form doolee€ ‘future,it will become’ ) consistsof thereceptor alone, historically at least. 682 REMARKSANDREPLIES

McDonough’s andSpeas’ s bipartitetheory holds that the verb stem itself is onlyone part of the minimalstructure of theverb. There is aleft-handcomponent as well,the receptor. The phonetic expressionof thereceptor is notfixed for allforms ofaverb.Rather, it isvariously filled in by thephonological features of the nuclear elements appearing in the extended projection of the verb. Letus consider the derivation of the verb whose structure is shown in (6). The relevant phonologicalcomponents are set out in (8).

(8) d n-´ € CVCdaøzh ch’´õ sh Theleft-to-right ordering is as in (6), but the parts are arranged on two distinct planes, whose purposehere is primarilyexpository, to distinguishheads from nonheads.The preverb, being an adjunct,and the object, being a complement,are set apart as nonheads. In assuming that the objectinflection sh- isthe true object argument in Navajo and thus appears within the verbal projectionas acomplement,I followJelinek’ s PronominalArgument Theory (Jelinek 1984, and subsequentwritings). Therealization of the verbal skeleton is a matterthat concerns heads alone. In effect, it ‘‘builds’’ theconjunct sector of theverb word. The process is this:the phonetic features of the functionalheads are transferred successive-cyclically to the receptor within the verbal skeleton, satisfyingthe coda requirement first, if possible. Inthe case at hand, the voice marker € (likeother fricatives) seeks to assume the position ofthecoda, in the CVC receptorof theskeleton. Hence:

(9) d n-´ À CV€daøzh ch’´õ sh Nextthe features of theMode/ Subjcomponent are spelled out on the next available position in thereceptor. The Mode/ Subjcomponent is complex, consisting here of a nasalaspectual prefix (calledalternately a conjugationmarker or situation aspect) together with a floatinghigh tone correspondingto perfective viewpoint aspect (and, of course, the third person singular person feature,not expressed phonologically here). The nasal cannot assume the coda position (in any prefixalstructure) and must be anonset,and this must be supportedby epenthesis.The high tone isrealizedon the epenthetic vowel.

(10) d À À nõ´€daøzh ch’´õ sh Theremaining functional head, the Qualifier prefix d-, cannotfit into the minimal phonological skeleton.But this is a minimalskeleton, permitting further augmentation. The prefix d- is spelled outattheleft edge of whatis nowavailable of the conjunct sector of theverb. There it issupported byepenthesis.

(11) À À À dinõ´€daøzh ch’´õ sh REMARKSANDREPLIES 683

Theconjunct sector of theverb is nowfully spelled out. 4 Collapsingthe expository planar represen- tation,keeping the verb and its dependents in the original order, and applying epenthesis as needed,we obtain(12), the form thatcorresponds to the surface spelling of theconjunct sector oftheverb together with its dependents, the preverb and the object.

(12) ch’´õ shidinõ´€daøzh Ihavesilently eliminated the zeros corresponding to the functional elements prior to their overtrealization within the conjunct sector skeleton. I assumethat these are in fact invisible to phonology,though they correspond to real entities for computation.In any event, the process involvedin defining the conjunct sector is not head movement, implicating traces. It is apurely phonologicalprocess involving the spelling of morphological material into a skeletalreceptor andsatisfying a structuralrequirement of the Navajo verb.

2InformalRemarks on FillingOut the Skeleton For detailedtreatments of the phonology of the bipartite verb, see Speas 1984, 1990, and McDonough1996, 2000. What I myselfwill have to saywith regard to the phonology of actual spell-outprocesses, adapted to the particular viewpoint being explored here, is impressionistic andtentative. Likethe receptor, the stem (right-hand portion) of theskeleton is not fixed for allforms of agivenverb, being dependent not only on thebasic segmental shape of the stem as listed in the lexiconbut also on the various shapes assumed by stems inflected for aspect.Nor isit always thecase that the favorite CVC patternis realizedby the stem alone. Many verb stem forms lack acoda;a few lackan onset;and a few lackboth, consisting solely of thevocalic nucleus. Thus, theCVC patternis to be understood as amaximum,realized to the extent it can be, and in the mannerit can be, depending on the actual stem form involved.Thus, in the basic, underlying representationsof eachof the shown in (13), the stem portion of theskeleton is less than fullydetermined: in (13a) the onset and vocalic nucleus of the stem skeleton are satisfied, but thecoda is not;in (13b) the onset is missing;and in (13c) neither the onset nor the coda of the stemis satisfied.

(13) a. se´€hõø´ (surface form) se´ € CVChõø´C(nuclearplane) Mode/SubjVoice V ‘Ikilledit.’

4 Theobject (position IV inYoungand Morgan’ s (1987)template) are normallyregarded as belonging tothe conjunct sector (butsee Rice 1989:645–646).They are at theleft edgeof thatdomain, and the idea thatthey are inthe disjunct sector is notentirely out of the question. In addition to position IV objectinflections, there is another (phonologicallyidentical but separate) set thatappears clearly withinthe disjunct sector. Moreover, the object inflections inthe inchoative (see Youngand Morgan 1987 and Hale, Munro,and Platero 2000) appear sometimes as conjunct inflections,sometimes as disjunctinflections, depending on various factors. 684 REMARKSANDREPLIES

b. yisha´a´h (surface form) sh À CVCa´a´h(nuclearplane) Mode/SubjVoice V ‘Igo/walk.’ (as in taah yisha´a´h ‘Igetinto the water’ ) c. yiyaø´ (surface form) À CVCCaø´C (nuclearplane) Voice V y (complementplane) Object ‘He/Sheeats it.’

Nothingspecial needs to be saidabout (13a). The coda segment of thestem there is simplynot realized;the receptor is filled automatically, with [€ ] fillingthe coda and [se ´]filingthe vocalic nucleusand onset.The situation in (13b) is slightly more involved. The Voice element is nonovert andthus irrelevant for spell-out.The Mode/ Subjcomponent is realized overtly by [sh] (thefirst personsingular subject marker); this would normally fill the coda segment of the receptor (the left-handportion of theskeleton). Here, however,the onset requirement of thestem takes prece- denceand is satisfied by assigning [sh] ofthe Mode/ Subjportmanteau to the onset slot in the stem.This leaves the first part of the skeleton unsatisfied altogether, an impossible circumstance thatis regularly corrected by epenthesis— the default epenthetic vowel being [i], thenearest Navajoapproximation to schwa. Word-initial vowels are also disallowed, a conditioncorrected inNavajo by glide prothesis ([y] before[i], [w] before[o]). Together,these processes yield the surfaceform [yisha´a´h] of (13b). In(13c) no spell-outoperation takes place on thenuclear (head) plane, since the only overt nuclearelement is thestem itself. This occupies the V-slot (vocalic nucleus) in thestem portion oftheskeleton. The stem has no onset C, butan onset is obligatory here and must be supplied byglideepenthesis. The V-slot of thereceptor is filledby i-epenthesis.The third person object(3o), overt [y], is inthe correct left-hand position in relation to the bipartite skeleton, where itopportunistically fills the onset slot of the receptor. By epenthesis, a purelyphonological [y] isinserted in the onset (initial C-slot) of the stem. These operations result in the surface form [yiyaø´] of (13c). For themost part, once nuclear elements in the verbal extended projection are spelled out inthe bipartite skeleton defining the core of the conjunct sector, the attachment of additional prefixesis straightforward.By epenthesis,consonantal prefixes assume the shape CV whenattach- ingto a C-initialskeleton (as doesthe object [sh] in(12)), and vowel-final prefixes (like [ch’ ´õ ] in(12)) attachto the left edge of theconjunct sector without modification. This is the ordinary situation;for morecomplicated situations, see Speas 1984, 1990, and McDonough 1996, 2000, aswell as Kari 1976.I willillustrate here what transpires when a vowel-finaldisjunct prefix attachesto areceptor not suppliedwith an onset and vowel nucleus by virtueof elementsappearing REMARKSANDREPLIES 685 inthe nuclear plane. The verbs in (14) illustrate this situation.

(14)a. naashbe ´ (surface form) sh À CVCbe´C(nuclearplane) Mode/SubjVoice V na (adjunctplane) Preverb ‘Ibathe/swim.’ b. ya´shti’ (surface form) sh € CVCti’(nuclear plane) Mode/SubjVoice V ya´ (adjunctplane) Preverb ‘I speak.’

Ineach of these verbs, the preverb and the stem, by hypothesis, constitute a singlelexical unit; theyare nonetheless placed on separate planes for expositorypurposes, representing the fact that adjuncts(and complements) are invisible to the processes that spell out the nuclear elements in theextended projection of V, instantiatingthe bipartite skeleton. The derivation of the nuclear planein (14a) proceeds as intheearlier examples. The vocalic nucleus of thereceptor is supplied byepenthesis,as before, and we mustassume that it isunderspecifiedwhen inserted, permitting itto assimilate to thevowel of the preverb [na] ‘around,about’ — hencethe long vowel [aa]. This iswhat happens regularly when a low-tonedvowel (belonging to a disjunctprefix) abuts the vocalicnucleus in the receptor portion. The derivation of (14b) is essentially the same, except that a high-tonedprefixal vowel simply ‘ ‘takesup’ ’ theV-position in the receptor, leaving no evidenceof priorepenthesis in thatposition, with implications (which I willnot pursue here) for anaccount of rule ordering vis-a ‘-viswhat is tacitly assumed for thederivations of other verb forms, suchas (13b –c).

3Further Observationson NonnuclearPrefixes and the Orderof Elements

Ihavebeen exploring here the possibility that the relativeorder of nuclearand nonnuclear prefixes inthe Navajo verb word follows automatically from thefact that the nuclear prefixes (heads in theextended projection of theverb) are spelled out in an abstract bipartite skeleton that, in effect, isan extensionof theverb stem (V) itself.The verb stem— qua skeleton containing the spell-out ofthe nuclear functional categories belonging to the extended projection of the verbal nu- cleus—follows (appearsto the right of) allnonnuclear elements, given that the stem (by assump- tion)is the lowest and linearly final constituent in the verbal projection, as depicted in (6), for instance,exemplifying the prototypical pre-spell-out Navajo verb word of moderate internal complexity.The derivation of the actual verb word that (6) underliesis given in some detail in 686 REMARKSANDREPLIES section1, andthe derivations of (14a–b)are,to allintents and purposes, identical to that of (6), insofaras the resulting surface linear ordering of elementsis concerned. Theexamples of nonnuclear categories cited so far includethe preverb (an adjunct, by hypothesis)and the object (a complement).The following verbs illustrate two additional adjunct categories,the iterative [na ´] ([nõ´]beforeapicals) and the (distributive) plural [da]. These are categoriesknown not onlyfrom Navajobut also from mostif not allother Athabaskan languages. 5 Iwillemploy both arboreal and planar representations here.

(15) a. daacha (surface form) À À CVCchaC (nuclear plane) Mode/Subj Voice V da (adjunct plane) Plural They cry. b. Mode/Subj

À Voice Mode/Subj

À VP Voice

da VP Plural V CVCchaC Thisverb is in the zero imperfective with third person subject; hence, the Mode/ Subjnucleus isnonovert altogether. Voice in this verb is the so-called zero classifier, also nonovert. These elementshave no effect on the stem skeleton. The vowel nucleus of the receptor is supplied byepenthesis. The default vowel thus inserted assimilates to the vowel of the adjunct prefix [da],which procliticizes to the stem skeleton in the normal fashion, giving [daacha], with long [aa]. Amorecomplex example is givenin (16).

5 TheIterative [na´ , nõ´]andthe Reversionary (‘ returnto previous state’ ; see Youngand Morgan 1987:167) have thesame shapeand the same allomorphy.I followRice (2000:61–68, 102 –106)in taking these toberepresentatives of asinglemorpheme, glossed simply It(erative) forpresent purposes. REMARKSANDREPLIES 687

(16) a. yisd n dashidoo t (surface form) d w CVCt (nuclear plane) Qualifier Mode/Subj Voice V yisd n da sh (nonnuclear plane) Pv It Pl Obj They will get me back to safety. b. Qualifier

Qualifier Mode/Subj d

Mode/Subj Voice w

Voice VP

Preverb VP yisd to safety Iterative VP n

Plural VP da

Object V sh CVCt handle inanimate object Thisstructure requires some discussion. I willbegin with the derivation of thesurface form, formulatedin thestyle used in section1 (e.g.,(8)– (11)). The expository planar representation of (16)is repeatedhere.

(17) d w € CVCte´e´€(nuclear) yisda´ nõ´ da sh (nonnuclear)

Thelight verb [€ ] (i.e.,Voice), assumes the coda C-slot in thereceptor— like fricatives in general, itseeks first to occupy a codaposition if it can. The third person perfective and progressive Mode/Subjprefix ([w] inthis context), being a glide,must occupy the onset C-slot in the receptor, whereit is supported by epenthesis.The inceptive qualifier [d]— left over, so to speak, once the receptoris fully satisfied —isprefixed to the receptor and supported there by epenthesis. The defaultneutral epenthetic vowel [i], bothbefore and after [w], assumesthe rounding feature of thelatter, giving [o]. The glide itself drops, leaving the homorganic vowel sequence that is normallyresolved into the long vowel written [oo]— hence (18), in which the complete overt conjunctsector is correctlypositioned linearly to the right of the nonnuclear elements. 688 REMARKSANDREPLIES

(18) doo€te´e´€(surface form, nuclearplane) yisda´ nõ´ da sh (nonnuclearplane)

Thenonnuclear elements are successively procliticized to the left of the conjunct sector in the ordergiven (with epenthesis applying where needed, i.e., to supportthe object prefix [sh] inthis case).The result of theseprocesses is thesurface form (19),corresponding to theverb word fully spelledout.

(19) yisda´nõ´dashidoo€te´ e´€

Thesecond point of discussionrequires some terminological and conceptual clarification of certainaspects of theanalysis being promoted here. The tree diagram in (16b), on theface of it, beliesthe very point being made in this discussion, namely, the constituency hypothesis that placesthe preverb [yisda ´]andthe verb stem V (andits melodic skeleton, [CVCte ´e´€])togetheras a unit inthe lexicon. The preverb has the appearance of beingseparated from theverb. This is afiction,however, perpetuated by the two-dimensional depiction of thestructure, which, perforce, confusesand obscures the true relationships in the verb word. The true relationships involve three separate‘ ‘dimensions’’:

(20)a. the argument structure; b.the adjunction structure; c.the nuclear structure.

Relationssuch as structural proximity, sisterhood, domination, and command must be defined separatelyfor eachof thesedimensions. The relevant argument structure dimension is therelation betweenthe verb and its object; they are structural sisters. The nuclearstructure dimension amounts totherelation between the heads in the extended projection of the verb; each head is ‘ ‘adjacent’’ tothehead that selects it, if there is one.The adjunct dimension in the diagramin (16), on theother hand,consists of asetof elements(Preverb, Iterative, Plural) whose relationships are completely misrepresentedin the tree diagram as that is normally understood. In reality, the substructure containingthe adjuncts— that is, the tree segment extracted and shown in (21)—isa flat structure, properlyspeaking and despite appearances.

(21) VP

Preverb VP yisdá ‘to safety’ Iterative VP ní

Plural da REMARKSANDREPLIES 689

Whilethe adjuncts, as agroupand individually, asymmetrically c-command the verb, there are no asymmetricc-command relations among the adjuncts themselves. Structurally, they are equidistant from eachother and from theverb. Hence, the verb and the preverb are ‘ ‘sisters’’ inthe sense thatis relevanthere. Furthermore, the preverb is nofartherfrom theverb than the object is. One cannotspeak of relative distance from theverb here because adjuncts and arguments belong to distinctdimensions. Theonly ‘ ‘structural’’ relationthat holds among the adjuncts is precedence. It is therefore necessaryto ask, as Rice does in her detailed study of theAthabaskan verb (Rice 2000), why the orderis asit is.Within the system to whichthe elements in (21) belong, at least,Rice establishes thatsemantic scope is to the left, so that the iterative [na ´ , nõ´]andthe plural (her distributive plural)[da] havethe preverb within their scope. These elements also take the verb within their scope,by virtue of the c-command relation (see below). This is surely correct; and scope in general,as Rice expertly shows, is systematically related to the linear order arrangement of the elementsthat make up the Athabaskan verb word. I willdepart slightly from hertheory in propos- ingthat, among adjuncts, it is scope alone thatdetermines the order. And within the adjunction structuredimension, scope is rightto left, as Rice demonstrates convincingly and in greatdetail; thatis, an element with wide scope is to the right of elements appearing within its scope. On the otherhand, selection (by a functionalhead of itscomplement) is leftto right,in accordance with theparametric head-initial character of the extended projection of the verb. And scope follows suitwithin the systems to which selection and c-command are relevant: it is likewise left to right, sothat the selector has scope over its sister and all that appears within it. Interestingly,where scopal relations do not matter, it is possible to find variable ordering acrosslanguages within the Athabaskan family. This is thesituation that obtains in theordering ofIterativeand Plural: they appear in the order just given in Navajo,but they appear in the reverse orderin certain other Athabaskan languages. Rice argues that scope does not ‘ ‘fix’’ theorder in thiscase because the two elements belong to distinct systems of grammar: Iterative is relevant toevent structure, while (Distributive) Plural is relevant to argument structure. In other words, thesetwo categories do not ‘‘interact’’ inrelationto semanticscope. Variability here, like stability elsewherein the system, strongly supports Rice’ s generalthesis.

4TheDeictic Subject Prefixes

PositionV inthe prefix ordering given by Youngand Morgan (1987) is occupiedby the members ofa setof elements traditionally termed deicticsubjects inthe literature. In the relative linear orderingof prefixes,position V isto the right of Objectand to the left of Qualifier(with some phonologicallymotivated exceptions).

(22)a. shiji€ hozh sh j À € ghozh Object3a Mode/ SubjVoice V ‘He/She/They/Oneis tickling me.’ 690 REMARKSANDREPLIES

b.shi’ doo’ niid sh ’ d w d niid Object3i Qualifier Mode/ SubjVoice V ‘Iwas told(lit. someone told me).’ c. hodildo´o´h hw d À l do´o´h 3sQualifier Mode/ SubjVoice V ‘It(e.g., cloud) starts to drift.’

Thethree prefixes occupying position V areglossed, respectively, 3a, 3i, and 3s by Young and Morgan.The first of theseis also (and more often) termed the fourthperson andis thetopic of muchdiscussion in the literature (e.g., Willie 1991). For ourpurposes, it is simply a human subjectmarker construed with the third person in Mode/Subj.The prefix glossed 3i isanindefinite subjectmarker, also construed with the third person. It functions autonomously as a subjectbut alsoin the so-called agentive passive (not a truepassive (see Neudorf2000), but rather a fully transitiveimpersonal subject construction). 6 The3i prefix [’ ],inconjunction with the qualifier [d],is the grammatical subject in the agentive passive illustrated in (22b). The 3s prefix, again asubjectmarker, corresponds to the ‘ ‘spatial’’ or‘ ‘areal’’ subjectin a widerange of Navajo locutionsdescribing the movement, position, or qualityof an entityviewed as anarea,space, or timesegment (see, e.g., the entry for hodildo´o´h inYoungand Morgan 1987:450 and other relevant entriesalphabetized with that verb). Theseelements are problematic precisely because they are subject prefixes.But they appear inthe wrong position for subjects;that is, they are internal tothe verbal projection and lower thanObject. This is impossible within the theory being adopted here —basedultimately on the workof others, including Speas 1984, 1990, McDonough 2000, and Rice 2000. Theproblem is resolvedin thefollowing way. As Riceargues persuasively, the position V elementsare not invested with a personfeature. They are third person only in the sense that a fullDP orNPisthirdperson, like ashkii ‘boy’in (23).

(23)Ashkii shi€ hozh. ashkii sh À € ghozh boyObject Mode/ SubjVoice V ‘Theboy is tickling me.’

Thatis to say, the position V prefixes,despite their morphophonological lydependent surface status,are to be identified with adjunct nominal expressions linked to true arguments (subject, object,as the case may be) internal to the verbword, in conformity with the Pronominal Argument andDiscourse Configurational Theory of Navajogrammar (e.g.,Jelinek 1984, Willie and Jelinek

6 Likethe simple passive,the agentive passive requires the d-classifier inYoungand Morgan’ s positionIX (Voice). See Neundorf2000 and Hale 2000for some discussionof theidentification and role of the d-classifier inthe two Navajo ‘‘passive’’ constructions. REMARKSANDREPLIES 691

2000)that I amassuming here. The actual third person subject argument is to be found in the Mode/Subjportmanteau (a compositeof Youngand Morgan’ s positionsVII andVIII). Thisproposal would lose its force and credibility completely if third person within this compositewere consistentlyzero, the prevailing exponent of thirdperson subject. But this is not thecase. There is one environment in which third person is phonologically distinguished from thelocal (first andsecond) persons. The relevant environment is in the presence of the gamma [gh]situation type aspect marker in combination with the perfective and progressive viewpoint aspectswhen, in addition,this [gh] prefix, a velarglide, is flankedby low-toned neutral epenthetic vowels (i.e.,appears in the environment VghV, Vtheneutral vowel). In this situation the glide isrealized as a certain‘ ‘coloring’’ oftheflanking neutral vowels. While the glide never surfaces assuch in this environment, being realized only as its effect on the vowels flanking it, it is reasonableto assume that, abstractly, the gamma perfective marker has three allomorphs: rounded [w], plain[gh 4 g],andhigh fronted [y]. The rounded allomorph appears only inthethird person, asin the future (24a) and progressive (24b). (24)a. shidoo€ hosh sh d w € ghosh ObjectQualifier 3rd:Perf Voice V ‘He/Shewill tickle me.’ b. shoo€te´e´€ sh w € te´e´€ Object3rd:Perf Voice V ‘He/Sheis carrying me along.’ TheNavajo progressive is rendered by means of the gamma perfective situation aspect prefix, togetherwith zero-viewpoint aspect and the progressive stem. The future is renderedby this same complexof elements preceded by the inceptive qualifier [d]. The environment of theallomorph [w] ishighly specific (as isthatfor thefirst person singular). And the total picture makes it clear that[y], not [w], isthe default allomorph. I takethis to mean that the category ‘ ‘thirdperson’ ’ actuallyhas an overtexponent here and therefore that the actual position of thethird person, like thatof localperson subjects, is withinthe Mode/ Subjportmanteau. If thisis correct,then the so-called deictic subject prefixes of positionV belonggrammatically tothe same category of elements as full DP (or NP) arguments,like ashkii ‘boy’in (23). They are,in some fundamental sense, ‘ ‘adjuncts’’ tothe clause, that is, to theconstruct containing the extendedprojection of Vandits preverbal modifiers (Preverb, Iterative, Plural), where these are present.To account for the actual surfaceposition of the deictic subjects, I willassume that somethinglike the analysis in Hale 2000 is correct —thisis a modificationof Speas 1990, in whichinflectional morphology, originating in positionsexternal to the verb, is ‘‘infixed’’ tothe verb.I restrictthis ‘ ‘infixation’’ process(tentatively) to the deictic prefixes alone, however. In short,the deictic subjects are adjuncts in the abstract representations of clauses, but they are spelledout as prefixesto the conjunct sector of theverb (or, perhaps more narrowly, as prefixes tothe Qualifier element, as suggestedin Hale2000), rather than to the verbal clause as a whole. 692 REMARKSANDREPLIES

Theydiffer from ‘‘fullDPs/ NPs’’ inthis respect; the latter are spelled out as adjuncts to the verbalclause as a whole,and they assume there a linearordering determined by discoursefactors (inthe manner described in Willie and Jelinek 2000). On this account, the so-calleddeictic subjects arein no sense exceptions to general and established principles of Navajogrammar.

5ConcludingRemarks TheNavajo verb presents a ‘‘mixedsystem,’ ’ inasense,given that the principles that determine thelinear ordering of elements appear to work from leftto right in one part of the verb word, andfrom rightto left in another. Withinthe conjunct sector the ordering of elementsfollows (a) from theselectional relations holdingamong the nuclearelements, and (b) from thehead-initial parameter defining the structural relationbetween functional heads and their complements. The ordering of theobject within the conjunctsector follows from thefact (assuming it to be a fact)that a lexicalhead takes its complementto the left (giving the head-final order in the innermost VP projection). 7 In this respect,incidentally, the light verb Voice conforms to thebehavior expected of afunctional(i.e., inflectional)element, taking its complement (VP) tothe right, not to the left in the manner of a ‘‘full’’ verb(V). Withinthe adjunct domain, by contrast, the order of elementsappears to conform purely to thedirection determined by semantic scope, which, as Rice demonstrates clearly, works from rightto left in those situations where semantic scope is in fact relevant (i.e., where it shows interactioneffects of thetype detailed by Rice)and autonomous (i.e., where it isindependent of selection).

7 Anaspect ofmorphemeordering that is completelymasked inmy analysis is the internal structure of theMode/ Subjnuclear element, which,I haveasserted, is acompositeconsisting of fromtwo to three distinctmorphemes (situation typeaspect, viewpointaspect, andsubject person). I haveclaimed thatthese elements forma portmanteau,suggesting thatthere is nodeterminableordering of thecomponent morphemes. This is somewhat misleading,because itis possible toargue, at least, thatperson follows situationaspect withinthe Mode/ Subjsubstructure (as isdone traditionally, on the basis ofperspicuous cases). Myanalysisdoes not predict this ordering and must simply stipulate it. By contrast, Hargus andTuttle (1997) propose a system ofrankedalignment constraints that determine theordering of morphemes here and elsewhere inthe Navajo verb. Anotherproperty of thissystem ofallegedportmanteaus is theirallomorphy. In the perfective viewpointaspect, the Mode/Subjparadigms display two sets ofallomorphs depending upon the identity of theVoice element (i.e.,the traditional Classifier ofAthabaskanlanguage scholarship). One set ofallomorphsoccurs with the €-andzero classifiers; theother occurswith the d- and l-classifiers (thelatter beingin fact acompositeof d- and €-).This raises thequestion of when VocabularyInsertion applies. In the model I am assuming,the vocabulary items belongingto the perfective Mode/Subj paradigmsare insertedin their base position(i.e., into the terminal nodesof the diagram in (16)) and the appropriate allomorphyis determined at thatpoint, implying that Mode/ Subjis properlysituated in relation to Voice soas topermit thecorrect allomorphyto be effected. Thiscould be verywrong, however, and it might be necessary toapply a ruleof Lowering,prior to Vocabulary Insertion, in order to account for this allomorphy. If so,my claim thatthis is, so to speak, a‘‘nonmovement’’ theoryof the internal structure of the Navajo verb is considerably weakened (see Embickand Noyer 2001for much relevant discussion). My command of the principles of Distributive Morphology (as represented,e.g., in Halle andMarantz 1993),which I am assuminghere, is notsufficient at presentto permit me todo anything more than pointout the potential problem. Rice (2000:169,178) questions whether this is allomorphyin the usual sense. Rather, theperfective viewpointaspect marker is simplymissing in the context of the d- and l-classifiers, causingMode/ Subjto berealized as ifit were imperfective.The problem for my analysis remains, however,because thisimpoverishment of theperfective takes place inan environment that must be properlyidentified. REMARKSANDREPLIES 693

References Embick,David, and Rolf Noyer. 2001. Movement operations after syntax. LinguisticInquiry 32:555–595. Grimshaw,Jane. 1991. Extended projection. Ms., Brandeis University, Waltham, Mass. Hale,Ken. 2000. Remarks on the syntax of the Navajo verb. In Dine´bizaadnaalkaah: investigations, ed.Theodore Fernald and Kenneth Hale, 55 –97.Working Papers on Endangeredand LessFamiliar Languages 3. MITWPL, Departmentof Linguisticsand Philosophy, MIT, Cambridge, Mass. Hale,Ken, Pamela Munro, and Paul Platero. 2000. The Navajo form. In Dine´bizaad naalkaah:Navajo language investigations, ed.Theodore Fernald and Kenneth Hale, 97 –116.Work- ingPapers on Endangered and Less Familiar Languages 3. MITWPL, Departmentof Linguistics andPhilosophy, MIT, Cambridge, Mass. Halle,Morris, and Alec Marantz. 1993. Distributed Morphology and the pieces of inflection.In The view fromBuilding 20: Essaysin linguisticsin honor of SylvainBromberger, ed.Kenneth Hale and Samuel JayKeyser, 111 –176.Cambridge, Mass.: MIT Press. Hargus,Sharon, and Siri Tuttle. 1997. Augmentation as affixation in Athabaskan languages. Phonology 14: 177–220. Jelinek,Eloise. 1984. Empty categories, Case, and configurationality. NaturalLanguage & LinguisticTheory 2:39–76. Kari,James. 1976. Navajoprefix phonology. New York:Garland. McDonough,Joyce. 1996. Epenthesis in Navajo. In Athabaskanlanguage studies: Essays in honorof Robert W. Young, ed.Eloise Jelinek, Sally Midgette, Keren Rice, and Leslie Saxon, 235 –257.Albuquerque: Universityof New MexicoPress. McDonough,Joyce. 2000. The bipartite structure of the Navajo verb. In TheAthabaskan languages: Perspec- tiveson a NativeAmerican language family, ed.Theodore Fernald and Paul Platero, 139 –166. Oxford:Oxford University Press. Neundorf,Alyse. 2000. The Navajo simple passive. In Dine´ bizaadnaalkaah: Navajo language investiga- tions, ed.Theodore Fernald and Kenneth Hale, 117 –128.Working Papers on Endangeredand Less FamiliarLanguages 3. MITWPL, Departmentof Linguisticsand Philosophy, MIT, Cambridge, Mass. Rice,Keren. 1989. Agrammarof Slave. Berlin:Mouton de Gruyter. Rice,Keren. 2000. Morphemeorder and semantic scope: Word formation in the Athapaskan verb. Cam- bridge:Cambridge University Press. Speas,Margaret. 1984. Navajo prefixes and word structure typology. In Papersfrom the January 1984 MassachusettsInstitute of Technology Workshop in Morphology, ed.Margaret Speas and Richard Sproat,86 –109.MIT WorkingPapers in Linguistics 7. MITWPL, Departmentof Linguistics and Philosophy,MIT, Cambridge, Mass. Speas,Margaret. 1990. Phrasestructure in natural language. Dordrecht:Kluwer. Willie,MaryAnn. 1991. Navajo and obviation. Doctoral dissertation, University of Arizona, Tucson. Willie,MaryAnn, and Eloise Jelinek. 2000. Navajo as a discourseconfigurational language. In TheAthabas- kanlanguages: Perspectives on aNativeAmerican language family, ed.Theodore Fernald and Paul Platero,252 –287.Oxford: Oxford University Press. Young,Robert, and William Morgan. 1987. TheNavajo language. Albuquerque:University of New Mexico Press.

E39-320 MIT Cambridge,Massachusetts 02139 [email protected]