Why Prefixes

tActa Linguisti.cdEungarica, VoI. 50 (1-2), pp. 155 135 (2003)

WHY PREFIXES? M AR, I ANNE M I TH I T N

I

II

I

Abstract A \,?dety of explanatiorrs have been offered for the obserr,'ed cross-lingrdstic preponderauce of sufrxes ovet prefixes. Many are couched in terrns of syncbronic advantages, such as the cognitive simplicity of cross-category harmony betrveen synta".( and morphologr, and pre{ereuces for processing the leical mea,Iring irr sterns belbre the $amrnatic&l material in afiixes. Brlt hypotheses about ftlnctional a.dvantages caunot constitrrte explanations in thenr,seh'es 11-itLout accounts of the mechanisrns by whicl the advantages are translated into glammati(.al stNcture. Here it is shown that the numerous exce.ptions to s[ch hypotheses can be explniued when the individual histories of the affixes are couside-r€d, inchxling both their soruces and the steps by which they develop.

It has long been recognized that suf;frxes outnumber prefixes cross-linguistically. As eaxly as 1921 Edward Sapir remarked, the use of prefixes, suftr<es, aud infixes-suffixing is "Of the three types of affxing much the commonest. Indeed, it is a fah guess that srmxes do more of the fornati\.e work of language than all other methods combinecl." (1921, 67)

I I I '

Several kinds of explanations have been offered for the suffixing preference. Some have focused on cross-category harmonl', proposing that speakers prefer consistent ordering of heads and dependents across sJmtax and morphology. Since more languages show head-final s]ryrtactic structure (OV), it is natural that they should also show head-final morphological structure (Stem-suffix). Other explanations have focused on processing, proposing that hearers prefel to process stems before a"ffixes,since they contain richer information. Still other explanations have focused on production, proposing that since spealers tend to elide the ends of words, morphemes occurring later in words are more Iikely to erode into affixes than those occurring earlier. But these proposa,is,even in combination, do not account fullv for the morpheme orders we find. There are numerols examples of prefixes in languages with robust, head-final (OV) sptactic order. The identiflcatiou of 1216 8076/03/S20.00€) 2003Akad,4miaiKi.arl6,Btul,altest

156

MARIANNE MTTHUN

speaker preferences is a reasonable first step towar:d possible explanation, but we canutot claim to explain particula.i. structures if we have not identified the actual rnechanisms by which such preferences might shape grammar. As noted by Gleerberg (1957), mechanisms might be identified in two areas: (i) origin, that is, circumstances that could lead to the development of certain structures, ald (ii) srrvival, that is, circumstances that could contribute to their stability. Here we shall examine some situations in which expla.nations based. on cross-category harmony, processing, and production fail to account for morpliological structure. These involve prefixes in languages with clea,r, headfinal (OV) syntactic structure. It will be shown that a key to explaining such structures lies flrst in distinguishing the kinds of affixes involved a"nd then urcor.ering the different paths by which they develop.

1. Correlations and explanations In grouud-breafting typological work, Greenberg (1963; 1g66) observed correlations on the one hand between word order in clauses and the placement of aclpositions, and on the other between the placement of adpositions and affixes. a. Larglrage€ rvith domiuaDt VSO order ar.e always prepositional. (universal B) b. With overwhelmiugly gr.eater. than cha.uce liequency, languages with normal SOV ordei are postpositioua.l. (universal 4) c. If a la.uguage is exclusively suflixing, it is postpositionat; if it is exclusively prefixing, it is plepositio[a.1. (universal 2?)

A rnore direct correlation between syltactic and morphological order was repolted by Wirrfred Lehmann (1978, 2t2). "A nru[ber of rnor.phological char.acteristics have bee[ identified lbr specific language tll)es. Promiueut aDroltg thcse is the placerue t of affixes, noiably those exprressing verb:r.l clualifiers. Irr VSO languages tLese pr.ecede the ceutral velb; in OV languageg or, the other LaDd, tlcr. lbllorv.

The ploposed correlations thus link predicate,initial clause structure (often abbreviated VO) with prefixes, and predicate.fina,l structure (abbreviated OV) I'ith suffixes.

VO prefixes-Stem

OV Siem-sufixes

These correlations were generalized,in a series of papers by Theo Venneman (1973; 7974,etc.) as the Natural SerializationPrinciple. According to this Acto Linguistica Hungarica 50, 2003

WHY PREFIXES?

r57

principle, languages tend to develop towa,rd consistent order between Operator and Operand, also termed d,1terminant and ditermin1, or Head and Dependent. Operator-Operand order is manifested in Verb-Object, Noun-Genitive, Preposition-NP, and Auxiliary-Verb orders, while Operand-Operator orcler is the reverse: Object-Verb, Genitive-Noun, NP-Postposition, and Verb-Auxilia.ry. The prevalence of Operand-Operator order across languages was traced to a cross-linguistic tendency for topical material, usually expressed by subjects, to be placed early in sentences, before predicates. If the notiol of Operator or Head is extended to affixes, then the predominance of SOV word order would result in a predominance of Stem-suffix order and explain the suffixing preference. John Hawkins and associates(Cutler et al. 1985; Hawkins 1988b; Hawkins -Cutler 1988; Hawkins Gilligan 1988) adopted the Venneman proposal, which they termed the HOP or Head-Ordering Principle, and added a second factor, that of language processing, to explain the prevalence of suffixes. HEAD-oRDERrNcenrNcrele (uoe) Hea.ds are identically PR O C E S S IN C

ordered relative to theh modifiers in both syntax and molphologl'.

P RE F E RE N CE

FOR SUFFIXING

Lexical recognition precedessyntactic processing, so langrrage lrsels will preler to process sterns before affixes. Stem-affix order provides the most emcient stmctrue for proccssilg.

These proposals taise several issues. The first pertains to the exact nature of heads in morphology. An extensive literature on the subject indicates that identification of heads in morphology is not straightforwa.rd (Williarns 1981: Selkirk 1982; Zwicky 1985; Hudson 1987; Scalise 1988; Bauer 1990; Corbett et al. 1993, and others). In early discussions of heads in syntax, the Noun was identified as the head of the Noun Phrase, and the Verb as the head of the Verb Phrase. The Noun and Verb were the essentia.lfoundations of their phrases, the more general, modified, subcategorized elements. They determined the syntactic category of the phrase. When the notion of head was extended to morphology the pa.rallels were not always clear. For some, headedness is primarily a semantic notion: the head is the more general, modified, subcategorized morpheme of a word. In a compound like newsltaper, the head is the root paper, and, in an adverb like unkimlly, the head is the root kirul,. For others, headednessis primarily a syntactic notion: the head is the morpheme that determines the lexica.l category of the word. The head of newspaperis strill paper, but the head of unkiruI,fuis identified as the suffix -/y, since it is the suffix that determines the lexica.l category of the rvorcl, the element that makes it into an adverb and identifies it as such. Even amonq AcLa Linquislica Hunguri.n 5t1.'/,01).1

158

MARIANNE MITHUN

those who view a.ffixesas heads, opinion is not uniform. Some consider only delivational amxes to be heads, since only they can change lexical category. Others, such as Williams (1981), have proposed that the rightmost affix in a word is always the head, even when it is inflectional, since it identifies the lexical category of the word. Such questions about the nature of headedness in morphology render principles of cross-category harmonv in this domain sornewhat more difficult to eva.luate. Another issue raised by all ofthese proposals is what constitutes adequate explanation. The identification of recurring patterns is an importa.nt step, but it only assemblesthe material to be expiained. Hypotheses about functional advantages to certair patterns or clusters of patterns might lead us closer to an explanation by suggesting motivations for their retention. A consistent head-ordering principle might offer cognitive advantages in ease of acquisition, ald stem-initial words might ofer processing advantages if spealrers do ildeed process lexical material before grammatical material. But perceived adl'a.ntages alone cannot account lbr the existence of particular structures in la,nguages until we have traced the precise mechanisms by which the structures enter the language and become established. The recognition that the diachronic dimension should be considered is not new. It has been known since the nineteenth century and before that the most common source of a.ffixes is independent words. On this basis, Giv6r (1971; 1979; 1984) proposed a general principle that the position of affixes within words mirrors the earlier order of words within clauses. Like Verreman, he hypothesized that all ianguagesoriginated with head-final (OV) syntax, due to the general tendency for speakers to place topical information, expressed by subjects, early in the clause. He maintained that affixes are always descended from syntactic heads, so languages characterized by headfinal clause structure (OV) will show head-final word structure (Stem-suffix). The ploposals that all languages were originally head-final (OV), and that all affixes a,re descended from heads, have not met with general acceptance, but the idea that explanation mnst incorporate the diachronic dimension is a crucial one. Though explicitly discussed in Greenberg (1952), it is still overlooked by many seeking explanations lbr cross-linguistic tendencies. Still, it is an implicit fouudation of the work of many others and explicit in some, such as Bybee (1988) and Hall (1988; 1992). In work discussing explanation in rnorphology, Hall considers at length the kinds of factors that might favor the suffixing preference. After detailed surveys of forma,l models of the lexicon ald results of psycholinguistic experiments involving a"ffixes, he a.rgues for an integration of psycholinguistic {actors (both processing and production) and Acta Lingui,stica Eungari,ca 50, 2003

159

WHY PREFIXES?

diachronic considerations (the mechanisms by which psycholinguistic factors are incorporated into the grammar). "The hypothesised uoiversal psycholinguistic dispreference 1br prefixing must, it seems, be instantiated in pa.r'ticular languages word by word by a mec.hanism rvhich, given the right conditions, 'blocks' the frsion of potential prefixes with Ire-e sterns. The essential triggedog conditions a,r'epreseDt at the point where the fiIst terrta" tive reanalysis of a pair of free forms a,s one bound + oue free form takes place in the mental lexicon. This phs.se, I sugget, is characterised by a 'flirting' process, in which a sema.ntically and phonologically decayed free form attempts to becolue bound to (i.e., 'flirts'with) a full free form on which it depends and with which it is habitually contiguous. When the dispreference is triggered with stfficient stre[gth. the pressure for a bound analysis is challenged and the lbrm does not reduce a6 la-r as aifix status." (Hall 1992, 166)

Bybee et al. (1990) undertook the task of untangling the respective roles of the original position of grammatica.l markers in syntax and differential resistence to the fusion of potential prefixes a.rrd suffixes. They surveyed 71 languages for the relative positions of a small set of verbal morphemes to determine whether the predominance of suffixes might be due to a general prelerence for postposed grammatica,l morphemes, or a greater tendency to a.ffix morphemes that are aheady postposed. Their results showed both. GR AM M ATICAL MA RKERS

AL L

F REE

BOU N D

Preposed Postposed

34% 66%

48% 20%

s2% 80%

In their sample, grammatical markers appeax after the verb twice as often as before the verb (66% aftet to 34Yobefore), suggesting a general preference for postposed grammatical morphemes. Among the gra.mmatical markers that followed the verb, there are four times as many affixes as independent words (80% bound to 20% free), while among those that preceded the verb, the proportions were about the same (52% bound 48Vo fuee). Together these two sets of figures suggest a greater tendency to fuse postposed molphemes. Bybee et a.l.did find some correlation between affix order and syntactic orcler. Among the languagesin their sample with basic predicate-final (-V) syntactic order, there were strong tendencies for both postposirlg and fusion. Among those with predicate-initial order (V-), tendencies for postposing and fusion were slight. Among those with predicate-medial order (-V-), there was a slight tendency towa.rd preposing but none for fusion. They conclude that the sufixing preference in their sample comes primarily from the prevalence of Acta Ling istica Eungarica 50, 2003

160

MARIANNE MTTHUN

predicate-final order languages. Of the 71 languages, 32 were predicate-final (-V), 31 predicate-medial (-V-), and just 8 predicate-.initial (VJ. At least some of the forces proposed so far, that is, cross-category harmony, processing efficiency, erosion due to production, a preference for postposed grammatical morphemes, and a greater tendency to fuse postposed morphemes, undoubtedly contribute to the structures we flnd, but they do not account for the full raJrge, even when considered together. We could simply accept the fact that langlages ale full of exceptions. Alternatively, we could tly to refine our predictions by a closer examination of the cases where current explanations appea.r to fail. These cases a,re the prefixes that occur in languages with clear head-final (OV) syntactic structure. Such cases a.re actually not rare. Prefixes regula,rly occur in languages with otherwise clea.r,head-final (OV) typologicat profiles. Navajo, for example, a la.nguage of the Athabaskan-Eyak-Tlingit family spoken in the American Southwest, shows clear, basic SOV constituent order, along with a robust inventory of postpositions but no prepositions. It is exclusively prefixing, however, with an inventory of over a hundred prefixes. Samples of these stucrrl es can be seen in (1). ( l)

Navajo: Dolly SouJ6.spealer p., . (a)

Nihijish biiMahohiil. dhi-jish b-iih-daiw-oh-l-uiil 2.pl-suitcase lintodistr-indeflt1ite.object-2.pl.subjectJ.p_gp4fgtrlgligld 1.our suitcases you all put things irto them 'Pack yoru' suitcases.'

(b)

Nihil da.llat'iilwod. dh-il dalu-d-rii-l-wod 1.pl-with ofi-away-distr-indefi nite.sub ject-completive-detr-qg1..l@L!I[ nith rs something rarr off 'It [the bus] just lett wiih all of us.'

(c)

Alhan6iit'aash. a-l-ha-ni,--iid-'aash reciprocal-willlFriative-al'ourdl.dual.sub ject-lqgral.rE '\\'e'll get together now aud theu.'

[..plgg

The interest of this morpheme order was noted by Keren Rice in her book on morpheme order in the Athabaskan languages. She treats it essentially as an aberration and posits an underlying structure in which all a.frxes axe suffixes. "A prima.ry idiosyncrasy of the Athapaskan velb is that the verb stem is located in tLe'wloDg'place iu the suface string. Iu the rernainder of this book, I assume a rnovernent-based accorurt along the lines proposed in Speas 1990, 1991, and Rice Actu Li,nguistica Hutuaarica 50, 2003

161

WHY PREFIXES? 1993, 1998. [...] The verb stem originates as a sister to the preverbs, both of rvhich axe within the scope of, a.nd therefore enter into senautic comp(xition as a ulit $dth, quaniificational elements [...]. What' is unusual about the verb is the srular:e position of the stem it moves from its position within the verb phrase to the right of the functional morphemes." (Rice 2000, 78) "The greatest idiosyncrasy in the verb is the position of the verb stem a d the markers of voice/ralence. Gilen the scope hypothesis, one expects these to occru rvithin the verb phrase. I have treatecl their actual placement as a quirk of the Athapaskan verb, achieved tluotgh the raising of verbs." (Rice 2000, 414)

We can refine our undeffitanding of the morpheme orders that occrrr, and the reasons behind them, by peeling apart the various diachronic forces that shape them. In most work so far, genera.lizations have been made over prefixes and sumxes considered as two homogenots groups. But affixes with different functions develop from different kinds of sonrces, by diferent hinds of diachronic routes. If we consider them individually, we may be able to develop a finer-grained understanding of the processesthat shape granllnar) and ultimately axrive at more accurate generalizations.

2. Pronominal affixes Among the most common kinds of verbal affixes a.re pronominal affxes. \,Iore often than not, these morphemes run counter to the prediction that languages with head-initial (VO) syntactic structure should have prefixes, and those with head-final (OV) order should have sumxes. Navajo shows basic OV constituent order, but it contains pronominal prefixes. (2)

Nara.jo pronominal prefixes: Dolly Soul6, speaker p.c.

(a) 6t strizte'e' nl6i shii nl6i 6i shi-zh6'f shij that l.sg.possessor-fatherprobably there 'Ir{y father had probably FortWingatedi glaajlniiyi Fort Wingate=di sh-aa.j!-niiyd. Fort Wingate:at lgg-to-!.19!i9q!-ore.go.perfective at FoIt Wingate he ca,meto me gone to Foft Wingate to visit me.' (b) A6di mg!!nhj. riri:di ni-shi-ni-01-ti there:at dom-!:q&&jg9!-term-3.subject-tr-handle.animate.object.pl.f ovcr i here he put me dowrt 'He took me there.' AeLo Linguistira Hungarira 5U. 2-00,3

I

I

L62

MARTANNE r{rrHUN

(The Nar"ajo 'fourth person' category, is used for generic mentions much like English'one', for respect, and often for the protagonist in narratives. In (2a) the speaker used it to refer to her father. When basic third person subjects and objects cooccur in transitive verbs, the object pi.onominal prefix shows a distinction between proxirnates, the more topical of the two, and obviatives. the less topical.) The reason lbr this apparent exception is clear. These pronominal prefixes sirnply continue the earlier syntactic order of their sources. With predicate_ filal word order, iudependent pronouns would have preceded the veib. When they lirsed with the verb, they became prefixes. There are numerous examples of the same appalent exception on the othe' side as well. All languages of the Salisha' familv of northwestern North America show clear predicate-initial (VO) order, so this order is easily recon_ structed for their common a.ncestorProto-Salish. yet pronominal suffixes a.re common, as in Halkomelem. (3) Htdkourelern (Sa.lishan lamily,BritishColumbia;Galloway1993,1761: lllr-cal

l6m-cax-

eo-!lg."QJ99! 'l eo.'

e-Z!c."!!J9!!

'You go.'

luay0-illrra-cal

ney0-6x)-cox'

lrelp-2.sg.object-1.sg.sub

'It'"lP-fl-

=--

iect

help-bs!bjss!-2rs@iee! 'You help rne.'

As in Navajo, the moder.n morpheme order continues the ea,rlier syntactic order'. But it would be premature to assume that modern morphological stlucture always echoes earlier syntax. Within the Salishan familv we also fi rrd plonorninal prefixes. (.I)

Cocru d'Alene (Salisha.u,Idaho; Reicha.l.d1938 cited in Kroeber 1999. 106): Lute hi-s-tap-sd6nt. Dot 1.sg-uoninalizer.-shoot-iutla[sitive 'I dicl not shoot.'

It is easy to explain the position of these prefixes once we uncover the route by wliich they developed. In all of the Salishan languages (and presumably Proto-Salish), dependent clausescan be formed by nominalization. The result is similar to English 4lljujng late 'is always a mistake or I loue walking along the beuchat ni,ght The subjects of the nominalized. clauses*" u*pr""sed *it-f, possessivepronominal prefixes, not unlike English My arriuing late uas a mistahe or I resent hi,s ualking along the beach at aight. Salish nominalized Acta Linguistica Hungarica 50, Z00a

163

WHY PREFIXES?

clauses are used regula.rly in certain constructions, such as negation, where the negative word serves as a predicate and the negated clause as its argr.rment. In many of the Salishan languages nominalized clause structures have been extended to use as independent sentences,often with special discourse functions. When this independent use increases sufficiently in frequencl', it ca.n compete with the earlier independent clause structure. In some of the la.nguages it has even replaced it.

3. Directional affixes Among the verbal affixes that would be termed 'verbal qualifiers' by Lehmann are directional affixes. Navajo contains an extensive inventory of such affixes, and all are prefixes, despite predictions made by hypothesized cross-category harmony principles, processing preferences,and resistance to initial grammatical markers and fusion. speaker,p.c. (5) Navajodireciionalpre6xes:Dolly Sou16, (a)

Hastiin adand.r{.tsaad hastiin ada-n6#Gtsaad

nd99' ni-de9'

ma.n dowa-again-3,subject-scoot.momentaneous.pfthat:hom 'When the man came lgql again . . . ' (b) habicycle bikaa' darh'iz'1q bikri,6' !q-ch'i-iz-'4:go ha-bicycle 4.poss-bicycle&on.top !q-bqIEg44V-3.object-4.subject-hafdle.solid.obje.tt.ltt 'he put it up on his bicycle' (c) T'66 hdd.hg66shjjandiidl66h. t'66 h6rihgooshjj'a.ni-d-iid-dl66h just really gy3X-repeatedly-orally-1.du.subject'laugh.imprf 'We were just laughing ourselvesto death ' (d)

anri.6.nd.sdzri.. la'jig66 a-n66.n6-s-d-y6 la':ji:g66 some:to:towa.rd gI9[-back-dur.result.l.sg.subject-detr-one'walk 'I went somevrhereelse.'

prf

An explanation of their position as prefixes is easy to find in their origins. The prefixes are descended from independent adverbs and nouns' some of t'hich persist in the modern language. The adverbs and nouns still oecrrr before the verb, just as they did in the parent language. The prefixes have simply continued the position of their ancestors. Acta Linguislica Hungarica 50, 2003

164 ({i)

MARIANNE MTTHUN Some adverb sources of Navajo directioual PNI'FIXES ada'downward' da'.,p' a'away out of siglrt' (a)

prefixes: Dolly Soul6, speaker p.c.

ADVERBS adah 'downwald' dah ,ther€, remote' 6"6

Ad6h nri.rinrid6. ad6li u6ri.n#0-d6 down rreverse-3.sg.srbje.ct-sit.perfective 'He came back dowrr again.'

(b) T'ah dah nt6ii still up there 'He was still up there ndishchi' bii'nji'n6'o. udishchi' b:ii'-ui-ji-d-na':go tree 3.proxir[ate.iu-a.rorurd-4.subject-detr-crawl.cout.prl:subordirute clirnbing around il the tree.' (c)

(7)

A6di

sid6.

t6i'. si-0-d6 l6f 6{:di there:at durative.sequel-3.subject-one,person.sit.prf ruir.ative 'He was sitting back there!'

Sorne rroun sources of Navajo directional prefixes: Young (2000) T'REFIXES N OU N S 'up iDto the aA' y6' )'ri'sky' a'i'into a hole or brurow' a'ri6n 'hole, burrow' d6'rik'e'into the field' d6' 6k' eh' comfi el d,fi el d' ,dirt, soil, le. 'into the ashes (to cook)' leezb ,water' taof the water, to shore' 'out t6

tsrlza-

'in the belly' 'into mouth'

-ts,i -z&'

'belly' 'mouth'

But not a.ll directional affixes have developed by this route. Kawaiisu, a Uto-Azteca.n language of California, shows the sa.rne basic heacl-final (OV) syntactic structure as Navajo, and this order can be reconstructed lbr its palent langauge as well. Kawaiisu also contains directional amxes, but these are suffixes. A basic translocative suffix -&aree-,thither'indicates motion away from the speaker or other deictic center, and a cislocative sufiix -,{,i- 'hither' indicates motion toward it.

Acta L irLgrislica Hungarica 50,2003

WHY PREFIXES? (8)

165

Kawaiisu directional su.ffixes:Zigmond et al. (1991) 'enter' "ig"'go in' "ig*tlg'come in' "iea-t!hutma'carry several' hu?ma-kwee-'takeseveral' h u "ma, - F ' br ings ev e ra l ' yaa'carry one' yae-bg 'ta,ke one' yaa-ti'bring oue'

The contrast in position is easy to explain if we consider the paths by rvhich these directional maxkers made their way into the grammars. Kawaiisu., like other Numic languages, shows extensive compounding of many kinds, including the combination of two verb roots to yield a new, compound verb. (9)

Kawaiisu Vert>Verb compounds: Zigmond ei al. { 1991. 163) ka?a-pagieat-walk 'walk along eating'

kaa-havi singlie

yaa-pid! carry-anlve 'bring'

pi"aa-"abigi be.pretty-talk 'say in a pretty voice'

The directional suffixes -ftuee 'away' and -,ti 'toward' originated as the seconcl members of Verb-Verb compounds. The first still persists as a verb root in the modern language, that still occurs on its own. (10) Kawaiisuverbrool -hwee'go':Zigmondet al. (1991,83) Ilana?okosamam!ketyee-d!-mi? reduplication-gg-realized-indicativepl when they 'When did they go?'

The source of the cislocative sufrx -ii 'toward' no longer persists as an independent verb root in Kawaiisu, but a probabie ancestra,l verb root is reconstructed for an earlier stage of the language, Proto-Uto-Azteca.n *kitn 'come' (Miller 1987). Descendents of this verb still appear as independent roots in a number of related languages. The best explanation for the fact that directional a.ffixes appeal as prefixes in Nalajo but sufi.xes in Kawaiisu thus cannot be the cross-category harmony principle, since both languages have developed from a.ncestorswith basic OV clause structure which continues in the modern la.nguages. It is Acta Lingr,isticaEungarica 50, 2003

i66

MARIANNE MITHUN

probably a.lsonot attributable to processing preferences. It is, however, easily explailed by the different kinds of sources from which the markers developed.

4. Aspect Among the most frequent verbal affixes are aspect markers. For la.nguages with head-initial (VO) clause structure, the cross-category ha,rmony principle would predict that they should be prefixes, while processing and production hypotheses would predict suffixes. The head-initial Sa.lisha"nlanguages show numerous aspectual prefixes, among them the Bella Coola stativeprogressive ?al-. (11) BeUa Coola stative.progressive prefix ?ol-: Nater (1984, 96) (a) )afayucmtim )al-zay-uc-m-tirn stative-progressive-exchange-speech-detraositivizer'3.pl.passive 'somebody was telling them to . . . ' (b)

'aguya*r

14-r5uy-aa+

stativeprogressil,e-lall.over-tree '[tree] lies fallen'

We could conclude that the force towa.rd cross-category haxmony is more powerful than processing and production factors, or we could look more closely at the paths through which such markers develop. Common diachronic sources of aspect markers are verb roots. These verbs may fifft erode in shape to becone verbal auxiliaties, then subsequently fuse with associated verbs to become affixes. Alternatively they may first fuse with other verbs to form compounds, while their full shapes axe still intact, then subsequently erode into affixes. In either case, in languages with basic predicate-initial clause structure (VO), matrix velbs precede their complements, so we would expect the aspect markers that descend from higher verbs to be prefixes, just as in Bella Coola. The origin of this stative-progressive aspect prefix can in fact be traced to the verb root ali- 'to be located, stay somewhere', which still survives as a root in the language. (12) Probable source: root oli- 'to be located, stay somewher.e':Nater. (1984, 49) ?alic 2alasultac -""ti-" ?al:sul:?ac stay-l.sg in:house:this 'I aln stayi[g in this house" Acta Lirrguisticd Eungarica 50, 2003

167

WHY PREFIXES?

(The same root also deveioped into a locative preposition ?al 'in, at', another common path.) For languages with head-final (OV) syntax' all proposed principles would predict suffixes: cross-category ha.rmony, production, and processing, as well as the recognized path of development from higher verbs to aspect markers. We might accordingly expect that in a language like Navajo, with the headfinal patterns throughout its syntax and that of its ancestor, aspect rnarkers should be suffixes. Nala.jo does exhibit an unusually rich inventory of aspectual distinctions, but there are no suffixes. Aspect is expressed by complex combinations of verbal prefixes and stem ablaut. An example is the Iterative in (13), formed with the prefix nri- and the Repetitive form of the verlr stem'ask'. (13) Navajo iterative a,spectprefix nri-: Do\

Sonl6' spea.kerp c

Nri'adishkido. nri-'a-dlsh-kid:go iterative.indefi oite.object-ora.lly-1.s9.subject-ask.lepetitive:sub 'I kept asking questions.' The Iterative prefix evolved from a prefix rui- 'again' . nrore concrete meaning:

of the same shape with

a sliglrtly

(14) Navajo prefix zti-'again': Dolly Soul6,speakerp.c Hastiin ada,n6,]itsaad. hastiin ada-n60-tsaad rnan down-ggg!4-3.s bject-scoot.momeutaneous.perfecti!€ 'The man came dowrl again.' This prefix can in turn once more'.

be traced to an independent

adverb

ndrind'again,

(15) Earlier source: aAvetbnddnd'auother, again': Youlg-X'Iorgan(1987,583) dilkos N6rinri y6ego di-l-kos n56nri y66':go glgg!1ll extremely=adv orally-detransitivizer-cough.imperfective shilni shi-l-Gni 1.sg-tG3.sub ject-transitivizer-say. imperfective 'I{e told me to coush hard once more.'

The modern prefix simply continues the position before the verb of its adverbial source. The development of a,spectual prefixes in Athabaskan ianAcla Linguistiro Hungartca 50. 200,t

168

M ARIANND

]\'I]TH U N

guages again illustlates the point that grammatical morphemes, even those with comparable functions, can develop from a va.riety of sources. Although there are no suffixes in modern Navajo, the stem ablaut that contributes to the aspectual marking probably originated from earlier combinations of roots and aspectual suffixes. The stem ablaut lbr aspect never affects the initial consolant of the stem, but it can involve changes in the color, length, and nasalization of the vowel, and add final consonants. The various forrns of the stenr nsed above for'a.sk', for example, ate -heed, -hi', -kil, and -h[id.

5. Manner arrd means A Lu'ge number of genetically unrelated languages in North America contain velbal tr,ffixesthat ildicate the means or manner by which an activity takes place. Both the cross-categoly harmony principle and proposed processing prelerences predict that such mar.kerswill be suffixes in languages with headfilal (OV) syntax. But this prediction is violated more often than not. Some ex;rrrlrlescatr be seeu irr Navajo. (16) Nlrajo adverbial prefixes: Dolly Soul6, speaker p.c. (tr) diueezjdd'o {g!-s-0-j66'=go I elaxirrg-tertrFdurative.seqrrel-3.subordirratejectnultiple.recline.prl:suborditrate 'l'heu everybody rvas ir bed' (b)

t'66 baay.irriizlj'o t'66 b-aa-y.i-rr-ii-zjj':go .just 3-about-&sharu to ridicule somebody' (c)

'altamstl'x* ?al-tam-stl'x' stative,progresive-rarrsative-behave. well 'to [rake somebody behave well :- to gi\€ sorDebody sound advice'

The source of the causative prefix still survives in the language as the root lanl-'make, construct'. (19) Source of Bella Coola causative pr.efi* loot ,dm- 'make, cormtruct, (a)

tamsultuurinu tnlD-sul-tu-mi-rlu i <e-house-benelactive1.sg.subject-2. sg.obj ect 'I will build a house lbr yod

(b)

tamyayaxiitun tan-yaya+ii-tu-m nake-toy-benel'active-detrausitivizer 'soruebody rxade hiD a toy'

For larrguages with head-final (OV) syntactic patterns, all principles would leacl us to expect causative suffixes. La.Iihota, a language o{ the Siouan family of the Great Plains of North America, shows a basic SOV constituent order, an order rvhich can be reconstlucted for its pa"rent,Proto-Siouan. As predicted, it contai s a causative sumx, presumably descendedfrom a higher verb, though this velb no longer edsts in the language as such. A(ta Linguxstica Hungarica50, 2003

1.7r

WHY PREFIXES? (20) La.khota expected causative suffix -ye: Sta,nley Redbbd, speaker p.c (a) khisl6 k"igle go.home 'he went home' (b) khigl6ye I{ rgre-lp go,home.causative 'he sent him home'

But Lakhota also contains some causative prefixes, among them yu-. (21) La,khotaunexpectedcausativeprcfix yo-: Stan Redbird,speakerp.c. (") p,iYa p?Ya drunl< 'he's drunk' (b) tgp4ya Jg-PaYa causative-drunk

'GEiEmdrunk' The existence of the La,khota causative prefixes goes against all predictions. Again an explanation comes from a more detailed look at the steps by which the causative prefixes entered the language. Lakhota contains a set of means/manner prefixes. (22) Lakhota meam/manner

prefixes

1.u-'involvingpulling' pa.-'involvingpushing' ka- 'involving sudden impact' ya- 'involving the mouth, biting, talking' lra- 'involving the foot or leg' wa- 'involving a sawing motion or knife' wo- 'involving action from a distance, shooting, blowing, pounding with the end of a stic.k' na- 'involving an inner force, heat, cold' pu- 'involving pressure' (no longer' productive)

Some examples of their use can be seen below.

AcLa Linguislica Hunganca 50- 2tJ01

L72

MARIANNE MTTHUN

(23) Lalrtrota veDbswith means/rnanner prefixes: StaJrRedbird, speakerp.c.; Buechel (1970) yg-bl6ya 'to spread out, unlbld, make level' pg-bliya 'to spread out, as dough: to mafte level; to iron (clothes)' ka-lrlf.ya 'to make level by beating' 'to open,e.g. a door' I!-f6 Da-bl6ir'a 'to cruuch by pressing,pushing,or sitting on, as glass' hi1-bl6itta 'to brea,ksomething brittle by striking, as a glass' These prefixes axe old a.nd can be reconstructed for Proto.Siouan. It is still possible to trace their origirs, however. In all of the Siouan languages, roots ca.n be combined to form new, compound stems. (2.1) Laldlota compounds:Boas Deloda (1941,70, 73) NOUNNouN ir'4-h1pa wood-moccasin : NouN vERB ih1-16 fuewood-gather : : vERB vERB rj?iura-milni sleep-walk

'shoe' 'G/h") gathersfirewood' '(s/he) is a somnarnbulist'

prefixes have developed from the first element of The modern means/mannel such courpounds. The source of at least one of the prefixes, po- 'by pushing', can be traced to a verb root, pa 'push', which still persists in the la"nguage. (25) LalJrota verb root pa 'push': Buechel(1970,422) Irlay6pa sni kihA, wait6 yelo. ru1-ya-Da 5ni kjha wa.5t6:yelo l.sg.patieut-2.sg.ageDt-pushnotif good:asseltive 'It's good if you dou't p!4 rDe.' (Pronominal prefixes in Lakhota, Iike those in Mohawk and Choctaw discussed below, categorize core participants as grammatical agents and patients, rather than subjects and objects. The basis of the systems is semantic rather than discourse-pragmatic, but it is fully categorical a.nd lexicalized, lea,r'ned vrith is each verb) and in sorne cases the rationale behind certain categorizations no longer transparent.) affixes in Lakhota and other languages Verbs formed with means/manner The cauoften ilclude an element of causation as paxt of their meaning. prefix gaprefix. sation is not an explicit feature of the The 'orally' does not necessarily add causation, as can be seen in (26a) 'deceive/tell a falsehood', but stems derived with it can have a causative sense, as in (26b) 'be angry/rnake angry'. (26) Lahhota occasionalcausative efiect: Buechel (1970) (a) gnriyq, Yl-g!6ya

'deceive, cheat' 'tell a lalsehood'

Acta L,ingui,stica Hungari,ca50, 2003

1J '1

WHY PREFIXES? (b) ihqz6ka yg.at'4zeka

'be angry' 'make angry by talking to'

The causative interpretation comes about through inference. Man;' situations described with means/manner amxes involve causation. If someonedies tbrough beating, for example, it can be inferred that he or she was caused to die by the beater. If someone is knocked down by kicking, it can be infered that she or she was caused to fall by the kicker. Because verbs are derived only as needed, the prefixes do not appear ila the lexicon and in use in equal numbers. Verbs involving hancl action are especially perlasive, since so many more actions are ca.rried out with the hands than by shooting or by sawing, for example. As is well known, the meanings of morphemes often become more general and abstract rvith ex-tensiveuse. Several kinds of processescan be involved. An important one is the metaphorical extension of common, concrete, lexical items to net', more abstract contexts. In English, for example, we easily say it's otr,t o.f mg han,ds, or the chance just slipped,through my f,ngers, even when no actual hand or flnger action is involved. The basic, concrete meaning of the prefix oz- 'by hand action, pulling' can still be seen in a number of derived verbs, inclucling many with causalive meaning. (27) La.khota causative gu- 'with discernible hand action' yggl6glo Xg-glo-glo by.pulling-redupli'ation-grunt 'make grunt' (a buffalo calf by catching it) But the felt this

in some derived verbs, the causative element has been reinterpreted as central meaning of the prefix. What was originally only inferred is now to be asserted. As a result, new causative verbs have been formed with prefix which show no element of hand action.

(28) Lafthotaextensionto abstrart catsation without physicalhand artion: Buechel(1970.656) Toksa, bluwiihakhikte,lo tokia wa-yg-wii" a,k"a:kte:lo rtttre:a"ssertive before.longl.sg.agent-causative-be.tme:fi 'In time, I will proveit.' Evidence that the original means/manner prefix yz- 'b3r pulling' has been ftlly reanalyzed as a causative in some uses ca.n be seen in the fact that it can norv be used with verbs already containing another means/manner prefix.

Acta Linglristico Hungatx.a 50. 200:)

774

MARIANNE MITHUN

(29) Evideuceof rea.ua.lysis: ccoccurencewith other mannerprefixes yru6ii JU-!!-.i causative-on.foot-be.stiff 'ttE.rp' Similar developments of causative prefixes from eaxlier means/manner prefixes iu head-final (OV) languages are discussed in Mithun (2002). The existence of both causative prefixes and causative sumxes within the same language shox's again that a lirll explaration of affix positions will ultimately require explolation of the complete range of possible histories of each kind of gra,mDatical maxker.

7. Applicatives Nlany languages contain derivational a.ffixes called 'applicatives' that add a core argument to the argument structure of verbs. The most common kinds of applicatives add a recipient or beneficia,ry ('cook' > 'cook-for'), an instrument ('write' > 'write-with'), a companion ('sing' > 'sing-with'), or a location ('jurnp' ' 'jump-over'). The added participant is usually a direct object, absolutive, or grammatical patient. Examples of applicative sufiixes can be seen in the Iroquoian languages of eastern North America. Among the applicatives in Nlohawk, for example, a.rethe benefactive applicative suffixes -az.riand -ni. (30) Itlohawk benefactive applicative suffixes (a)

(b)

rinata,hrcn6iwi ri-natahren.iii

wn"keuatalu'6:nen wake-nataluen-en l.sg.patient-visit-stati\€ 'I've visited'

LSg/m.sg-vr$[- Dell.sla ve 'I've visited him'

khthri.rha' k-hthar-ha' l.sg.agent-ta,lk-impd 'I'm talking'

rihthar6:ni ri-hthar-a--rri r.sg/m.sg-iau(-eF DeD.rmprr 'I'm talkiug to him'

The origins of both of these applicative suffixes can be seen in verb roots wlich still survive in the language: -awi,'give' and -ni 'lend'. (31) Lexi(al sotuces of X,Iohawk applicativesi verb root -drri 'give', -ni 'lend' (a) koni6.wihs korri-awi-hs 1.sg/2.sg-give-imperlective 'I give it to you' Acta Linguistica Hungarica 50, 2003

koni6:wi koni-awi1.sg/2.sg-give.stative 'I've given it to you'

WHY PREFIXES?

(b) t6ke4

775

wahiikeni' wa-hak&i-' Iactual-m-g/2.sg-!91g!-prf 'He lent it to rne'

take.ni

z.sg/Gg-&sl 'Lend it to me!'

Modern Mohawk does not have a syntactically-defined constituent order. The order of words in sentencesreflects their pragmatic status within the discourse rather than their grammatical role. A basic SOV order can be reconstructed for the ancestral language, however. The Mohawk applicatives, like those of related languages, are descended from the matrix verbs of complex seltences. Since matrix verbs usually follow their complements in langtages with head-final (OV) syntactic patterns, it is no surprise that the applicatives are suffixes. Navajo contains numerous applicative afrxes. Given the strong head-final (OV) syntactic patterns of both Navajo and its ancestor, and the recognizecl development of applicatives from matrix verbs, there is every reason to expect that the Navajo applicatives should be suffixes. But again, Navajo runs counter to expectation. All applicatives a.re prefixes. One example is the locative applicative prefix le'i-'on'. Added to the intransitive verb root 'gaze', it forms the transitive verb 'gaze on' > 'watch'. i32) Nar,ajo -'pli'gaze' with applicative i'f- 'on': Dolly 5o1116, speaker p.c :ii'gaze'

k'i-'lj 'on-gaze' > 'watch'

(a) Nl6igo desh'jj'. Dl6i:Co de_sh_'ii' yonder:to thematic-1.sg.subject-gaze,imperfective 'I'm just gazing over there.' (b) Awee' bik'id6sh'!i'. a.wee' bi-k'i-d6sh-'i/' baby 3-on-thematic-1.sg.subject-gaze.imperfective 'I'm watching the baby.'

Added to the intransitive verb'run' (actually based on a root meaning'flex', referring to the flexing of legs) the applicative derives the transitive verb 'mn o n'>'attack'. tlB) Navajo -tuod'run'with

applicative *'i-'on':

-wod'run'

Dolly Soul6, spea.kerp.c.

-k'i-wod 'on-run' > 'attack'

(a) Eelwod.

"o0+l-4 completive-3.sudect-detra.nsitivizer-run.perfective 'It ran.' A r La Lin g ui slira H ungarien i 0, 2-003

776

MARIANNE MITHUN (b)

Ndshd6itsoh shik'iilwod. shdoi-tsoh shi-k i-i-01-wod wildcat-big rne-rrpoD-completive-3,subject-detra.nsitivizer-r'un.p 'A rnourtaiD liol attacked rne.'

Again a consideration of the diachronic development of the markers provides an explanation. They evolved frorn separate words that occurred before the verb, postpositions with pronominal prefixes representing their arguments. The language still contains a laxge inventory of postpositions. Some of these show suggestive resemblances in form and meaning to applicative prefixes. (34) Porstposition origin of Nar',ajo applicatives:

Dolly Soul6, speakel p.c.

Sltik'i de' dahniyeeh shi-lfide' dah-0-ui-1'eeh ture-orrhither utrr3.object-2.sg.subject-ha.ndle.burden.imprf 'Put it on me' : 'Put it (a heaw sack of fotatoes) up on my back.'

The direction of the evolution from postposition to applicative prefix is clear. Where the postpositior sources of modern applicative prefixes still persist in the language, they are often more substantial in form. One can compa^re,for example, the postposition -'qq 'over' with the prefix -'q-, or the postposition -lcitih 'beyond' with the prefix -ld-. The development of the prefixes has been accompanied by phonological attdtion.

8. Headedness We have seen a number of failures of the hypothesized match between sufixes with head-final (OV) syntax, and there axe many more. It is not unlikely that obselved cross-category harmony is more often an a.rtifact of regular processes of lalguage change than the product of a synchronic force. We might find, as proposed by Giv6n, that affixes axe descended from syntactic heads, but that if syrtactic structure shifts, it may no longer coincide with morphological structure. But can we be certain that a,ffixes always develop from syntactic heads? In fact they do not. Both prefixes and suffixes can develop in a language from the same construction. The aucestor of Kawaiisu, Proto-Uto-Aztecan, is reconstructed with basic head-final (OV) sptax, which has been passeddown to most of its daughters. Both the parent and the daughters show extensive compounding of va.rious kinds. The compounds, like the syntax, would be considered right-headed (head-final) by all accounts. Acta Linguisti,ca Ilungarica50, 2003

777

WHY PREFIXES? (35) Kawaiisu right-headed compounds: Zigmond et aJ. (1991) sana-eoocozi ciga-roci hi?i-kamamo?ezigi?ataasiniya-kweenazipi-kwee-

pitch-bottle rough-head be.good-taste hand-wash slowly-go lainatego

'pitch bottle' 'taDgl.'haired' 'to taste good' 'to wash one's hands' 'to go slorvly' 'to go to the bathr'oom'

Some suffixes have indeed developed from the heads of such compounds, the right-hand members. As seen earlier, the directional su;ffix -huee developecl from a verb root which still persists as such in the language. (36) Root kwee'go': Zigmor'd et al. (1991, 83) Hana"oko samami ke!ye9-di-mi? reduplication-go-realized-iudicative-pl when they 'When did they go?'

It also appears as the right-hand member, or head, of verb-verb cornpouncls. From constructions of this type ('crawling-go') it has been reinterpleted as a directional suffix away. (37) -kwee'away': Zigmondet al. (1991,100) kahni-mkrva Togo$'a iirigrii-kwee-di house-rrnder rattlesna.ke qawl-gryg{-indicative 'The rattlesna.ke is crawling away urder the horme.'

It is in paradigmatic opposition to the cislocative suffix -fi 'hither, toward', which exists only as a sumx in modern Kawaiisu. (38) Tla,nslocative sufix -kwee'away''. Zigmond et al. (1991) "ig*-

'enter'

"iea-tlg-

'go in'

yaayaa-@g

'carry one object' 'take one object'

huemalrr?ma-$p-

'ca.rry several objects' 'take sevexal objects'

The tra.nslocative has in turn given birth to another sumx with slightl)' more abstract mea.ning, an inchoative with the meaning 'become'. Added to the terb 'be loose', for example, it derives the verb'come loose', or 'loosen'.

Acla Ltngunttra Hungartrn 5l). 200.t

178

MARIANNE MITHUN

(39) Irrchoativesulfix -&ueerZi$uoud et al. (1991) ha.ylrni'jpii hlia-1.e-e cabi

'to 'to 'to 'to

be loose' sleep' be aged, elderly' smash, deAt'

hay l m i - k w e e 'g e t l o o s e ' 'go to sleep' "ipii-btrq h!a-ye?e-kwee'get old, to age' caba-kwee'go flat (of a tire)'

It has also come to be used as a resultative. Added to the verb 'break,, it yields'broken'. (40) Resultative suffx -fruee:Zigmondet al. (1991,97) liolvqri-kwee-dj bt eali.rnom-resultative-indicative 'The tree limb is tloken off.'

pogewa:ika. 'uusu already branch-possessed:its

The verb root kwee 'go' has thus evolved, in its position as the right-hand member or head o{ verb-verb compounds, from a root to a translocative 'thither, away' then to an inchoative and resultative. The counterpart of the translocative suffix, tire cislocative sufix -&z 'hither', can be traced to a different verb root +tr;irn 'come' that appa,rently appeared in the same head position of compounds. This root no longer survives as such in Kawaiisu, but it is reconstructed for Proto-Uto-Aztecan. The cislocative sumx has in turn evolved still further into another kind of aspect ma.rker, a durative. Similarly, a third Kawaiisu directional suffix -rni 'along' ca.n be traced to a Proto-Uto-Aztecan verb root *mi 'walk' which has not survived as such in Kawaiistr. The Kawaiisu svffix -mi has also evoived further into an aspect marker', a habitual. There is thus robust evidence in Kawaiisu of the heads of verb-verb compounds, the word-fina,l roots, evolving into suffixes serving a valiety of functions. But that is not the v/hole story for Kawaiisu. The very same rootroot conpound structures that gave rise to suffixes also gave rise to prefixes. Among the verbs that can occur in the initial, non-head position of compounds is a root "aaga 'be stealthy'. (41) Iiau'aiisueooga'be stealthy':Zig:rrond et al. (1991,79) (a) siua"a-vi ?aagalnzi?a-u-pigadj:ina Coyote-absbe.stealthylook-urcmentaneous-perfective:him 'Coyotestealthily peekedat hirn.' (b) raqgt-'ablsibe.stealthy-talk 'whispe.r' This nor-head root has given rise to a prefix 2aa- 'quietly, stealthily'. Acta Linguistica Hungarica 50, 2003

WHY PR.EFIXES?

L79

ll2) Prefix ?aa-'quietly,stealthily': Zigmondet al. (1991,79) (a) Neeziaikardl. grrl sit 'The girl is sittirg.' (b) Neeziii "aa-ga4dj. quietly-sit girl 'The girl is sitting quietly.' prefixes Kawaiisn, like other Uto-Aztecan la.nguages, shows both additional and additional sumxes descended from the same compound constructions. Similar examples from Salisha.n languages axe discussed in Mithun (1997). It is thus clea.r that headedness does not determine which roots will become a.ffixes: both the heads and the dependents of compounds can evolve into a.ffixes. Other factors must enter into the formation of prefixes and suffixes. One of these factors can be surmised from the Kawaiisu examples just seen, a factor that has been recognized for some time. This is generality of rreaning. Verbs with general meanings like go and come. wili appear in large nnmbers of compounds, setting the scene for grammatical evolution.

9. Flequency The generality of meaning of verbs like 'go' and 'come' ensures that thel' 3vs likely to be used more often than verbs like'smash' and 'sleep'. As discussecl at length by Bybee Hopper (2001), a crucial factor which enters into the formation of affixes from roots is frequency of use. The role of frequency is illustrated in an interesting way in Choctaw, a language of the lUuskogeal family of Alabama, Mississippi, and Oklahoma, in the southern United States. Choctaw verbs contain pronominal afiRxesreferring to the core arguments of clauses. Grammatical agents, patients, and datives are representecl. Third persons aJe zero. The forms of the affixes can be seen below. (43) Choctaw pronominal a.ffixes (X{uskogean lhmily)r

1.sg 2.sg 1.pl 2.pl

Ul ch (1986)

ACENTS PATIENTS DATIVES sasanL -li l clc1mpimpiilha?-

haii-

haiim-

Normally pa.radigmatic sets of ma.rkers ali appear in the same area of the grammax. The Choctaw pronominal afrxes show a surprising arrangement. All of them are prefixes except for one, the first person singula.r agent -lz. Acta Lin,gl1i,tlica,Hunpnrica 5(.1,200,9

180

MARI,ANNEMITHUN

({4) Choctaw first and secorrdpersol pisalitok pisa-li-tok sec'1.sg.ageDt-past 'I sas' it' i,ssrrssotoL is-sa-ssctok 3!EIIESI!- 1.sg.p.tieDt-hi t-pa. t 'Yorr ilit rne'

plouorninal afExes: Ulrictr (19g6) saillaaii sa-illi-a:ti l.sg.patieDt-die-luture 'I arn going to die'

aida:ha 2sg.patient-be.tall 'You're tall'

Olce again, the surprising positions of the amxes can be explained by inves_ tigating the events that led up to their development, this time observing the lole of Ii'equency in use. The parent languag€ proto_Muskogean, like the mod_ ern dalrghter languages, contained auxiliaxies. As would be consistent with basic head-final (SOV) constituent order, the auxiliaries followed the content verb in sentences. The auxilia;ries were presumably descended from higher verbs. In modern Choctaw some of these auxiliaries have evolved into ve-rbal suffixes. The Choctaw past tense suffix -toA, visible above in ,I saw it, and 'you_hit me'for example, apparently developed from an auxiliary which itself developed from a Proto-Muskogea,n verb *ia[ ,finish, complete'. This verb also survives as a.n independent verb root in modern Choctaw. (45) Choctaw verb trx, ,firish,: Jacob et al. 1977 cited in Booker (19g0, 13g) DiPi ba?li-t ir_tah_li_tok o. rDcdt, [[-arlive-sal e.suLject 2.sg.agerrt-fiuish-artive_past q 'Did you finish cttting the rneat?,

Third person pronominal a,frxes are usually zero in the Muskogean languages. There rvas thus often no overt pronominal prefix on either verb in comolex senterces with the matr.ix verb ,finish'. As the construction became tighter, and the 'finish' verb developed into an auxiliary, it would be easy for leainers to reanalyze the position of the pronominal prefixes, when they did occur, from before the verb to before the verb phrase. In most cases the two anal_ yses would yield the same result. But first person singula,r agent prefixes were plobably considerably more frequent in daily conversation than plurals or secold persons. For the most par-t, inflected verbs, and especially inflected auxilia.lies, would not be assembled online as speakers spoke, but rather remernbered and selected as rrnits. The first person agent piefix was apparently so firmly attached to the auxiliaries that it remained there after the-auxiliary had evolved ilto a tense suffix. The result is a surprising paradigm, in whicl the first persol singular agent reflects the earlier order of-morphemes. 1111 The only explanation for its position is frequency of use. Acta Lituguittica Eutugarica 50, Z00J

181

WHY PREFIXES?

10. Conclusion

f a I J I

The search for explanations ofthe suffixing preference has resulted in a clir.'erse rt of hypotheses about the forces that shape gramma,r's. Nlany are couched b terms of synchronic advantages, such as the cognitive simplicity of crossqategory harmony, and the efficiency of processing lexical material belbre grammatical material. Certain issues remain unresolved in these accoults, however. A compelling definition of the notion 'head' in morphology has not 5et been established. The extent to which laboratory experiments on lexical recognition acurately model language processing in conte:'t remains to be demonstrated. They also leave numerous unexplained exceptions, even whel mnsidered tosether. But hypotheses about functional advantages cannot constitute explanations in themselves, without accounts of the mechanisms by which the aclvantages are translated into grammatical structure. If we hope to explain I'hy grammatical structures take the shapes they do, it males senseto unpack the steps by which they come into being. Very often the position of a.ffixeswithin words simply continues the syntactic position of the lexical items from which they a.re descended. Btt the development of affixes is often more than a simple process of formal fusion. hi order to explain the position of individual affixes, ra.'efirst need to identify their sources. In some cases this is straightforwatd, because their ftrnctions have remained little changed. Independent subject and object pronouns may develop into pronominal affixes, for example, as apparently happened in Navajo and Halkomelem. In many cases, however, the path of development is Iess straightforward. Sometimes the sources of pronominal subject prefxes were not independent subject pronouns at all. In a number of la.nguages,including some of the Salishan languages seen here, they originated as possessive prefixes on nominalized clauses. The formal evolution from independent word to affix is typicallJ' accompanied by functional cha.nge. Navajo and Kawaiisu directional adverbs u'ere seen to evolve first into directional a.ffixes and then into aspect ma.rkers. A single lexical source may spawn a variety of grammatical ma.rkers. The Bella Coola root 'be located, stay', for example, has yielded both a locative pr-eposition and an aspect prefix. Some kinds of affixes may develop frorr any one of several kinds of sources. The Bella Cooia catsative prefix developed from a verb root 'make', while the Lakhota causative prefix developed liom a means/manner prefix'by hand'. Causatives from both kinds of soul'cesmay even coexist within a single language, as in Lakhota The Mohawk applicaActa Li,nguisticaHungarica 50, 2003

782

MARIANNE MITHUN

tives have developed from matrix verb roots, while the Navajo applicatives have developed from postpositions. The lelative timing of particular func_ tional and formal c"ha.ngesmay vary. Lexical roots may develop first into irtdeperrdent grammatical words, such as auxiliaries, aduerbs, or adpositions, before they fuse with hosts to become affixes, like the Navajo directiolj prefixes. Alternatively, roots may first fuse with other roots in compounds, then undergo the functional abstraction and phonological reduction that make them into graanmatical affixes, as seen in Kawaiisu directional suffixes. So far most caleful studies of sufixing preferences have involved a very limited in_ ventoly of in-flectional categories. The case for cross-category ha.rmony as a motivating force behind the patterns we find might be strengthened if historical langrrage changes were ideltified whereby prefixes shifted position, hopping over stems to suffix po_ sition, in response to a syntactic shift to head-final clause structure. So iar, however', such situations are not well knovrn. In Navajo we saw the creation of verbal prefixes from postpositions, but there was no shift in order. The oost_ positions, along with their pronorninal object prefixes, simply fused with the verbs that immediately followed them: baby it-on gaze > baby it-on_gaze,She is watching the baby'. In Choctaw we saw a first person singular pronominal suffix that developed from a prefix, but the marker itself never actua.lly moved. Without an awa.renessof the individual histories of affixes, attempts at genera"lexplanations for their positions are bound to faii much of the time. But generalizations are not impossible. Some of the seemingly idiosyncratic developnents seen here reflect deeper, general principles. The development of adverbial prefixes and adverbial suffixes in Kawaiisu from difierent members of the same compound construction shows that headednessdoes not determine which roots will develop into a.ffixes. But the same development suggests that semantic generality and frequency in speech might. The f."qo"n"y iJ"to, has also shaped the Choctaw pronominal a"ffix paradigm in a complex way, t-esulting in pronominal prefixes for the entire pa,radigm except for the first person agents. Explanations of both the observed suffixing preference and the occLlrrence of prefixes are surely not beyond our grasp. Our generalizations will suffer, however, if we fail to tal<e into account the individual histories of the affixes we are seekins to describe.

ALta L,ingui,stic& Eungarica 50, p00S

183

WHY PREFIXES?

Abbreviations abs adv cmpl cont delr distr du dru ep imprf m

absolutive adverbializer completive contiNrative det,ransitir_izpr distribrtive dual dura.tive epenthetic vowel imperfective masculine

mom obj pl prf prog q sg sub srrbj ter.m tI

momentateous object plura.l perfe.ctive ploglessive questiol singular' srrboxlinate subject terlrinative tlaDsitivizer'

References Bauer, Laurie 1990. Be headilg the wold. Jorrnal of Linguistics 26 : 1-31. Boas, Ilanz - Ella Deloria 1941. Dakota grammar:. I!{emoirs of the National Acvlerul Sciences23.2. U.S. Government Printiug Office, Washingtou D.C. Booker, Karen 1980. Compalative 1980, Unirenitv of Kansas.

Muskogban: aspects ol Protol\Iuskogean

of

verb molphologr'.

Buechel, Eugene 1970. A dictiona.ry of the Teton Dakota Sioux Lurguage. Red Cloud lndian School and l{oly Rosary Mission, Pine Ridge SD. Bybee, Joan 1988. The diacluonic dimension in explanation. In: Ha$'kins (1988a, 35(1 71)). Bybee, Joan -Parul Hopper 2001. Introdrction. In: Joau Bybee Parrl Hopper (eds) Et'c1nclc1' and the emeJgenceof linguistic structure, 1 26. John Benjamins, Arnsterdarn. Bybee, Joan William Paglirca Revere Perkius 1990. O[ the asymmetries iu the afiixati(nr of gra"mmatica.l material. In: William Croft I{eith Denning Suzanne Kermuer (cds) Studies in typologr and diarluony, 1 39. John Benjamius, Arnsterdam & Philarlclphia. Corbett, Greville Norman Fra"ser- Scott Mcclashan 1993. Heads in gramrnatical tlxxrlr-. Cambridge University Press, Cambridge. Cutler, Anne John A. Harvkil,s Gary Gilligan 1985. The stffixiug prelbrence: a plot cssilg explanation. In: Linguistics 23:723 58. Galloway, Brent 1993. A $ammar of Upriver Halkomelerr Univelsity tions in LiDgldstics 96. Uliversity of California, Berkeley CA.

of Calilbruia

Prrblica-

Giv6n, Talmy 1971. Historical syntax atd synclrronic morpholo[ry: an archn€ologist's fickt tri1). ID: Proceedings of the Chicago Lingr stics Society 7:394 411-->. Giv6n, Talmy 1979. On urderstanding

gramma.r'. Academic Pless, New Yolk.

Giv6n, Ta.lmy 1984. Syntax a functional-typological introcluctiorr. Jolrn Benjaurins. Arrrsterdam. Greenberg, Joseph H. 1957. Order of affxirrg: a stt(ly irr ge[eral liuguistics. Irr: Joscph H. Greenberg (ed.) Essays in general lingnistics, 86 94. Univcrsit5, of Chicago, Chicagrr.

Aclo L,nq

^fxca

Hungonn J0, 200.1

184

MARIANNE MITHUN

Gr'eeub