Pragmatic Skewing in 1 # 2 Pronominal Combinations in Native American Languages

by Jeffrey Heath
Pragmatic Skewing in 1 # 2 Pronominal Combinations in Native American Languages
Jeffrey Heath
International Journal of American Linguistics
Start Page: 
End Page: 
Select license: 
Select License


1. Introduction. Many Native American languages have rich verbal mor- phology including "agreement" for both subject and object of transitive clauses. In some languages the subject and object (or ergative and absolu- tive) pronominal markers occur in separate morphological slots and do not interact; an example is Nahuatl, where subject and object series appear to reflect recent agglutination of formerly clitic pronominals. However, in a remarkable percentage of languages, several subjectlobject combinations are fused, opaque (difficult to segment), or unusually complex.

Some of the irregularity can be explained by classical morphological markedness theory. Most obviously, in many languages the pronominal subject markers of intransitive verbs express a full set of number distinc- tions (perhaps sgldulpl), but some of these distinctions are neutralized in transitive combinations (e.g., semantic du -+ du, du -,pl, pl -+ du, and pl

+ pl all expressed as morphological "pl +pl").

Other agreement systems revolve around a person hierarchy of the type 11, 2) versus 3, where speech-act participants 11, 2) outrank the third or "nonperson."' Other features (animacy, number) may be used to rank-order pairs of 3d-person markers. In "direct-inverse" (= "hierarchical") systems, the order of subject and object markers is determined not by grammatical relation but by these hierarchies, so "direct" (1, 2) + 3 combinations like 'I saw her' differ in structure from "inverse" 3 + { 1, 2) combinations like 'she saw me'.2 Markedness considerations are again at work here, since 'I saw her' is less marked, as a combination, than 'she saw me'. If (as in most

'The following notations should be distinguished in the body of this paper: "(1, 2)" with set-theoretical curly brackets denotes the set of 1st- and 2d-person markers; "12" denotes 1st inclusive; and "112" denotes a single morpheme used for either 1st or 2d (but not 3d) persons. In languages with a 12 category, the residual lpl category is to be interpreted as exclusive.

My thanks to the many students of Native American languages who have provided infor- mation, offprints, and literature tips on this project, and to three IJAL referees.

?In true "directlinverse" systems there is a nonzero Inverse morpheme in 3 -{I, 2) com- binations which helps differentiate them from direct 11, 2) + 3 counterparts. In some recent papers one notices extensions in the use of these terms, as "inverse" is applied to forms lack- ing an Inverse morpheme. Such mission creep has destroyed the usefulness of many once valuable linguistic terms, including "diglossia," and should be resisted in this case.

[IJAL. vol 64. no 2. April 1998, pp 83-1041 % 1998 by The Unlverslty of Chlcago All rlghts reserved 0020-707119816402-0001%02.00

such languages) the higher-ranking element comes first, we interpret the ordering as the grammaticalization of a topic-first ordering principle. Most direct-inverse systems (e.g., Algonquian) result in surface combinations that are moderately difficult to segment, but their basic organization can be explained in functional-typological terms (along the lines of Silverstein 1976).

Armed with markedness theory and a few simple functional notions (such as topic-first), we can motivate many of the apparent irregularities we find in 3 -+ 3 and {l, 2) ++ 3 combinations in various languages.3 However, the remaining quadrant, the 1 ++ 2 combinations where both sub- ject and object are speech-act participants, has attracted little attention from typologists. This is hardly because these combinations are transparent and regular; on the contrary, in language after language they are opaque and irregular. But there seems to be no method in their madness. The "ir- regularities" of asymmetrical { 1, 2) -3 combinations turn out to show recurrent cross-linguistic characteristics which have been successfully ana- lyzed by cognitive-functional typologists, but in 1 -2 combinations the formal irregularities differ so much from language to language that no (pos- itive) cross-linguistic pattern emerges.

A decade after putting the final coat of paint on the last of several Aus- tralian grammars in which I had wrestled with the formal morphological analysis of transitive pronominal paradigms, it dawned on me (on the road to Damascus) that the correct cross-linguistic generalization is a negative one, namely, that transparent 1 ++ 2 combinations are avoided (Heath 1991). In other words, maximally transparent 'I saw you', 'you saw me', etc., tend to form negative or taboo targets and are often replaced by more opaque sur- face forms.4

This preference for opacity resembles pragmatic restrictions in many languages on the use of transparent 2sg pronominals (thou, tu, etc.), which may become bluntly "familiar," hence inappropriate in polite discourse, where they are replaced by impersonal, third-person, or morphological "2pl" forms. In Japanese, Indonesian, and other Asian languages deeply influenced by court traditions, pragmatic restrictions apply to I st- as well as 2d-person

jThe double-headed arrow notation P ++ Q denotes the set {P-Q, Q + PI.Therefore "1 ++ 2," used passim in this paper and its title, represents the set {I -2, 2 + 1).

4~venthe title of this paper has been modified for taboo reasons. Referees complained about "Amerindian" in the original version, though for different reasons; one was "offended by this term (though not by "American Indian"), another worried about associations with the controversial Greenbergian "Amerind" phylum. The point of using a substitute expression like "Native American" is as much the avoidance of the negative target as the positive semantic content of the replacement.

pronouns; Japanese is famous for largely avoiding 1st- and 2d-person pro- nouns in conversational discourse.

In languages with rich pronominal agreement, pragmatic restrictions do not focus on independent pronouns, which are omitted unless required for discourse effect (e.g., contrastive topic). Instead, restrictions focus on the verbal agreement system, and particularly (I suggest) on the transitive 1 * 2 combinations. Here we get both of the pragmatically sensitive pronomi- nal categories, and in an asymmetrical subjectlobject relationship to boot. In this light, I recall that during my first fieldwork on (Mississippi) Choc- taw in 1972, I had difficulties eliciting 1 *2 forms of transitive verbs (even though they turn out to be formally regular in that language). My first informant cheerfully translated 'he hit her', 'he hit them', 'I hit him', etc., but when it came to 'you hit me' he balked, saying "We Choctaws don't talk like that; it sounds like I'm accusing you." I treated this, probably with little grace, as another bit of informant obstinacy to overcome; it would have saved me some grief later on had I recognized the wisdom in his words.

A maximally transparent system would be one where all 1 *2 combi- nations are of the schematic type 'lm-2m' consisting of a 1st- and a 2d- person marker ("m"). The relative ordering and even adjacency of the two markers is subject to language-specific adjustment. Each marker may be internally multimorphemic, maximally, e.g., '1-Fem-P1-Agent' with person, gender, number, and case components. To be fully transparent, both the '1m' and the '2m' markers should be identical to the intransitive '1m' and '2m' markers and/or to the corresponding independent pronouns, and should maintain the same numberlgender distinctions. This is based on the assump- tion that intransitive affixes and independent pronominals have unequivocal categorial values, and that native speakers used them as touchstones in ana- lyzing transitive pronominal combinations.

The departures from this idealized maximal transparency that were noted, with Australian examples, in Heath (1991) can be organized into the following "strategies":

    marker disguised by partial phonological distortion
    one of the two markers expressed by isolated suppletive allomorph
    one of the two markers (elsewhere nonzero) expressed by zero
    number neutralization, sometimes including use of "pl" for seman- tic "sg"
    1st or 2d marker merged with (or replaced by) 3d-person marker
    entire combination expressed by unanalyzable portmanteau
    entire combination expressed by zero (special case of portmanteau)
    inclusive (12) marker replaces 1st or 2d marker, or entire combination

Schematically, using "?" to indicate fuzziness in morpheme identification, and "n" as a numberlgender variable, we could represent (1) and (2) as 'ln?-2n' or 'ln-2n?, (3) as '0-211' or 'ln-0', (4) as '1-211' or 'ln-2', (5) as '3n- 2n' or 'ln-3n', (6) as 'X', (7) as '0',and (8) as '1211'. The Native American data surveyed below confirm these types and add some new ones, of which the most typologically significant appear to be the following (with one or two salient examples, to be described later, shown in parentheses):

    merged 112 marker is part of both 1 *2 and 2 -1 combinations (Salishan: Lummi)
    subject and object markers compete for a single slot (Guarani, Carib, Delaware)
    co-occurring 1st and 2d markers are widely separated (Salishan: Kalispel)
    combinations with identical segments differ in tones (Oto-Manguean: Popoloca)

Schematically, we could represent (9) as '[1/2]n-2n' or 'ln-[1/2]n', (1 0) as a choice between 'In' and '2n', and (11) as 'In- . . .-2n'. Type (12) involves a tonal overlay on one of the other systems, such as (6).

Some of the avoidance strategies have the effect of reducing the surface "transitivity" of the morphology (disguising or deleting a subject or object marker, or fusing the two into a portmanteau). However, the motivation here is not the discourse-pragmatic backgrounding emphasized by Hopper and Thompson (1980), rather social-pragmatic suppression.

Since the driving force is avoidance of a negative target, and since many avoidance strategies are available, we may think of the systems in question as subject to more or less Darwinian principles. Assuming an idealized, maximally transparent proto-system, chance "mutations" due to rapid-speech mechanisms, speech errors, or sporadic situational phenomena are "selected for" in the more socially delicate pronominal combinations, and become grammaticalized.

Pragmatic skewing ranges from slight uncertainty about morpheme identifiability to actual loss of information, and at this point the pragmatic forces favoring opacity run into communicative forces favoring at least a minimal level of referential clarity. A possible ascending scale of seri- ousness of information loss is this: neutralization of number marking < neutralization of person marker < homophony among semantically disjunct combinations. The Native American data reveal several cases of homoph- ony between 'l(sg) -+ 2(sg)' and its inversion, '2(sg) +l(sg)', which pushes the system to its referential limits; see Penutian (Siuslawan), Tarascan, Mixe-Zoque (Zoque of Copainalh), and Carib (Surinam and Apalai), below.

In morphologically complex languages, information lost or obscured in the core pronominal agreement markers is sometimes recoverable (inferra- ble) from clues in other co-occurring morphemes. If we could track the morphologies over a few millennia we might discern a tendency toward equilibrium between (pragmatically favored) opacity and (communicatively favored) transparency. Of course, one can always use optional free pro- nouns to resolve ambiguities.

We now embark on a whirlwind Levi-Straussian tour of the Western Hemisphere, from north to south. Readers are free to skip around the fol- lowing sections in small doses, rather than trying to read the paper as a narrative, which may induce grogginess. I reproduce the transcription of original sources unless otherwise noted, except that morpheme breaks are often added. I use Sapirean groupings like "Hokan" and "Penutian" for bureaucratic convenience rather than out of megalocomparativistic fervor. I apologize in advance for the inevitably cursory treatment of individual lan- guages, many of which deserve amplified synchronic/historical discussion by competent specialists. There is always a tension between "general" the- orizing and devoted language-specific analysis, but precisely this confron- tation is the font of most wisdom in Linguistics. I would encourage readers to remain focused on the general point, consider whether the array of data surveyed here suffices to support it in a broad way, and assess its relevance (if any) to the languages they work on. General readers whose interest is piqued by one of the languages surveyed here should consult primary sources or contact specialists, rather than citing data from this paper as au- thoritative. In addition to this and the original programmatic paper (Heath 1991), I plan further regional surveys on this topic for other parts of the world.

2. Eskimo-Aleut. As our exemplar we take West Greenlandic Eskimo, whose verbs have inflectional suffix combinations including subject-object agreement (Schultz-Lorentzen 1945:54-55, with morpheme breaks-some- times fairly opaque-added). In lsg -, 2sg indicative -va-v-kit or -va-git, we have indicative transitive -va-, then an optional -v- that shows up else- where in 2pl (and some lpl) combinations, and a final portion -kit (or len- ited -git) that resembles the most common lpl ending -gut but whose final segment could also be identified with 2sg -t, -it. Certainly there is no clear 'lsg' morpheme here. The 2sg -+ lsg indicative -va-r-ma is likewise opaque; the -r-appears elsewhere in combinations with lsg, lpl, or 2pl (but not 2sg!) subject on 3sg object. The ending -ma occurs nowhere else in the ver- bal agreement system, though it is obscurely related to -ma, the suppletive ergative-case counterpart for lsg possessive -ga or -ra (note that in 2sg + lsg the lsg marker is absolutive, not ergative). So we have an obscure 'lsg'


allomorph (strategy 2) and no discernible '2sg' morpheme (verging on strategy 3). There is a 1 + 2pl indicative -va-v-se with subject number neutralized ('Ilwe + you[pl]', strategy 4), ending with 2pl marker -se. The remaining forms (lpl + 2sg, 2sg + lpl, and those with 2pl subject) are fairly transparent.

For 2sg -+ lsg the interrogative form is -vi-nga (with 2d-person inter- rogative -vi-) and the imperative form is just -nga added directly to the stem. Both have the regular lsg marker -nga and are therefore more trans- parent than the indicative forms (and the "relative preterit" and "relative fu- ture" forms, not shown here but also rather messy). This makes me wonder whether the pragmatic-skewing model proposed here might be sensitive to the specific pragmatic effects of particular moods. Since the interrogative is not ordinarily assertive ('Did you see me?' as opposed to indicative 'You saw me'), transparent 1 tr 2 combinations are not dangerous here and per- haps are not selected against. On the other hand, blunt directness is appro- priate in commands and transparent pronominal forms there are positively selected for.

3. Algonquian. Data are from Bloomfield (1946), citing morphemes in their Proto-Algonquian forms. The most memorable (and overrated) fea- ture of these languages is the simple 2 > 1 > 3 rule for access to the single prefixal slot in the indicative mode. The prefix is absent from the conjunct order, and even in the independent order it is largely redundant, since the elaborate suffixal system makes almost all of the necessary distinctions. Transitive animate verbs take "theme" suffixes that indicate the broad sets "1 object," "2 object," "direct," or "inverse." In the independent order, the "1 object" theme *-i-is used only in 2 -+ 1 combinations, and the "2 ob- ject" theme *-e0-is used only in 1 + 2 combinations, since 3 -+ (1, 2) combinations use the inverse theme. As a result, *-i-and *-e0-function as portmanteaus in this mood. (In the conjunct order, *-i-and *-e0-are used in all (2, 3) + 1 and (1, 3) + 2 combinations, respectively, instead of the inverse theme.)

The theme suffixes are followed by one or more additional suffixes that complement the prefix and the theme suffix, marking subject person and (if applicable) plurality of subject or object. Therefore, the personlnumber cat- egory of each transitive participant is recovered by addressees by piecing together information distributed over the prefix, the (inner) theme suffix, and the outer suffixes. Although such a cumbersome system does not neatly fit into one of the simple "strategies" listed above, it inevitably lacks the transparency that we would get with separate subject and object slots and seems relevant to the current study.


In Delaware (Goddard 1979:136; 1 add morpheme breaks), the conjunct order shows a further development, in that the 2sg conjunct subject marker (PA *-(y)an) has spread to 1sg -+ 2sg -91-an, which is therefore distin- guished from 2sg -+ 1sg -a-yan only by the theme suffix. (Delaware -21- and -a- reflect PA theme suffixes *-ee- and *-i-.) The post-theme slot in 1 * 2 combinations is filled by a 1pl morpheme if relevant (lpl -+ 2, 2 + lpl), and by the relevant 2d-person morpheme (2sg or 2pl) if the first-person pro- nominal is singular. This specific subsystem of Delaware therefore related to strategy

4. Iroquoian and Caddoan. In Northern Iroquoian (Tuscarora and the Five Nations languages, data checked with W. Chafe, personal communica- tion, 1994), the usual lsg -+ 2sg form k-Q-consists of the regular '1' mor- pheme k- and an unglossable nasalized-vowel element. The { ldu, lpl) + 2 combinations are exemplified by k-ni- '1-Dual-', used for any 1 + 2 combi- nation involving one or two duals (but no plural). In none of the 1 + 2 forms is there an identifiable '2' marker, so these forms fit nicely into our framework (verging on strategy 3). The 2 -+ 1 combinations, on the other hand, are transparent sequences of the type '2-1- . . . ' (nonsingular number markers follow the person markers).

To be sure, Mithun's (1976) dissertation on Tuscarora ventured a for- mal analysis of the pronominal paradigms in which k-~-(represented as k6:-or a variant thereof, with v as the nasalized vowel) was analyzed as underlying //k-e-a-// ('1-2-OBJ-), but there is no real support from the rest of the morphology for either '2' //-e-// or-in the relevant morphemic position-an objective case allomorph //-a-// (Mithun 1976: 156, 178, 185). This then-fashionable mode of analysis reflected the generativist quest for "deep" symmetry lurking under a fragmented and epiphenomena1 "sur- face." (I wasted my youth doing exactly similar formal analyses of Austra- lian pronominal paradigms, so I cast no stones or aspersions.) Within the framework offered here, irregular and opaque morphological combinations have a positive pragmatic value and need no apologies.

In Caddo (Wally Chafe, personal communication, 1994), the 2 -+ 1 pre- fix combination yah-ku- is transparent ('2agent' + 'lpatient'), but the 1 + 2 form is a portmanteau th-, unrelated to 'lagent' ci- and '2patient' si(strategy 6).

5. Siouan. For the large Mississippi Valley subgroup, we can cite sev- eral opaque 1 * 2 combinations from Boas and Swanton (191 1:909, 913,

I thank an IJAL referee for clarifying the Delaware material.

which covers "Dakota" (= Lakhota) and has some non-Lahkota compara- tive data. The lsg + 2sg combination is either an unanalyzable portman- teau (Lahkota Ehi-, Ponca wi-) or morphologically reduced (Winnebago nin-, i.e., [ndj-1, consists of the bare '2patient' morpheme, without the usual lsg agent marker ha-). These illustrate, respectively, strategies 6 and 3. The other 1 -2 combinations are generally transparent, but in Ponca the 2sg + lpl combination wa-ea-, while regular, is homophonous with 2sg + 3p1, hence must be labeled 2sg + { lpl, 3pl). "6" is unexplained but appar- ently represents [a] (referee comment). Though no actual morphological transformation of the 2sg -+ lpl combination occurs, the surface effect is similar to that of strategy 5. For Lakhota, Rood and Taylor (1996:468) additionally indicate that for the small class of "stative transitive" verbs, 2sg + lsg and lsg + 2sg are both expressed by ni-ma- (2sg-lsg); the morphemes are transparent but the rigid ordering blurs the subjectlobject relationship.

John Koontz (personal communication, 1994) believes that the synchro- nically portmanteau lsg -+ 2sg forms in many Siouan languages (including the Ehi-, wi-, and ndj- mentioned above) can be derived from proto-forms *w-yi- and nasalized *w-yt-, readily analyzable as 'lagent-2patient', though the historical phonology necessary to derive the attested forms is quite messy. If he is right, one could conclude that all 1 *2 combinations were once transparent and regular, and disrupting phonetic changes are respon- sible for the synchronic mess. Even if this is historically correct, my spin on this is to emphasize that the languages have not undertaken restructur- ing~to restore morphological transparency undermined by regular sound changes, and indeed seem to have actively selected for rapid-speech muta- tions that accentuated the opacity.

In Biloxi, which belongs to the southeastern subgroup, 2 + 1 and 3 -+ 1 prefixes are homophonous; thus ya-xtedi 'you[sg]/he hit me' (Einaudi 1976: 73), cf. strategy 5.

6. Salishan. A few of the languages in this family show interesting be- havior. For Kalispel, Vogt (1940:35-36; morpheme breaks added) notes that the lpl prefix qe'- occurs in lpl -+ 3 qe'-[verb]-an-tam but is inexpli- cably omitted from lpl + 2 0-[verb]-an-c-t, both of which contain com- pletive -an-. The endings -tam and -t can perhaps be glossed 'lpl', but they occur only in a few transitive combinations and one could argue about segmentation and glossing. By contrast, qe'- is a very clear 'lpl' morpheme also used with intransitive verbs and its omission seems pragmatically rel- evant (strategy 3).

The morphology of predicate possessed nouns of the type 'X (is) father of Y' also appears to indicate avoidance of close juxtapositions of lpl qe'- and 2d-person markers. In ku-p.oxu't 'thou art our father' and p-p.oxu't


'you[pl] are our parents', the only surface prefixes are the 2sg and 2pl in- transitive subject markers, respectively. Contrast qe'-p'ofut 'he is our fa- ther', which shows the lpl prefix (here in possessive function). Reversing the roles of lpl and 2d-person markers, we get 'we are thylyour parents' which could likewise, in theory, result in close juxtaposition of lpl and 2d-person prefixes. As it happens, this surfaces as qe'-p'ofut-fal-twith the regular lpl prefix. Instead of the expected 2d-person prefix, we get a rather opaque suffix complex that can be shakily segmented into 'lpl' -fa[-and '2' -t (with number neutralized) by comparison with qe'-p'oRut-fal-s'we are hisltheir parents' (Vogt 1940:25-26). Overall, the lpl *2 forms are rather opaque in this language, making use of suppletive allomorphs (strategy 2), deletion of lpl marker (strategy 3), number neutralization (strategy 4), and perhaps physical separation whereby an expected prefix surfaces as a suffix (strategy 11).

In Lummi (Jelinek and Demers 1983:168), the interesting detail is a '112object' suffix -ogas-shared by all 1 * 2 combinations, following the transitive suffix -t-and preceding the subject marker: xc'i-t-ogas-san'I know you', xc'i-t-ogas-sxw'you know me' (strategy 9).

7. Penutian. I take no position on the genetic validity of Sapir's "Penu- tian." Quite a few of the languages show opaque 1 -2 combinations.

Chinook (Boas 1911:580-84, cf. Silverstein 1976:132) has a basically well-behaved transitive agreement system with ergative and absolutive pre- fix slots. The irregularities occur in 1 -+ 2 combinations. In lsg-+ 2 forms, instead of the regular lsg prefix n-we get a suppletive ya-, hence lsg -+ 2sg ya-m-, lsg -+ 2du ya-m-t-, and lsg + 2pl ya-m-c-(arguably y-am-, etc., with a different segmentation). The 2d-person components -m-, -m-t-, -m-c-are regular; one might alternatively segment the combined forms as y-am-, etc. In combinations of 1st nonsingular subject on 2d object, the usual 1st nonsingular ergative markers (e.g., lpl exclusive ergative ntc-k-, including ergative case marker -k-) are replaced by qa-, which seems to be an indefinite subject (ergative) marker q-plus a linking vowel. Boas further indicates (191 1584) that a 1st-person agent is omitted in ditransitive verbs when a 2d-person dative object is present: 0-t-am-1-a't-a 'I shall give them to you' (0-3pl-2sg-to-give-Future). So we have three distinct ways to avoid clear expression of 1st-person agents with 2d-person complements: supple- tive allomorph (strategy 2), replacement by impersonal morpheme (variant of strategy 5), and omission of normally nonzero morpheme (strategy 3).

In Takelma (Sapir 1922:170-72), the transitive agreement markers are usually transparent, with pronominal suffixes showing the order subject- object-. However, (2, 3) + lsg forms express lsg object by a stem-inter- nal vowel change (o + u, a + e, vacuous for other stem vowels) instead of by a suffix (Sapir 1922:59). The 2 + lsg forms are expressed by this (overtly or vacuously) ablauted stem plus an intransitive 2d-person marker of the intransitive (stative/resultative) class 11; the intransitivity suggests that the lsg element (stem ablaut) is not treated as though fully present morphemically (cf. strategy 3). The reduced salience of a (sometimes vac- uous) stem-vowel change, as opposed to an easily segmented affix, is simi- lar to the low visibility of pronominals marked only by tonal distinctions overlaid on affixes; see discussion of Popoloca (Oto-Manguean), below).

In Coos (Frachtenberg 1922a:350-51), verbs have a single pronominal prefix and many transitive verbs also have a suffix. In transitives, the prefix slot is filled by a { 1, 2) person marker in preference to a 3d-person marker (regardless of case), and by a 3d-person subject in 3 + 3 combinations. In 1 tr 2 combinations, subject number is neutralized (strategy 4) and the subject is expressed only opaquely in the form of portmanteau suffixes -dis '2 -+ 1' and -am? '1 + 2', which are not used elsewhere in the morphology. If the object in a 1 * 2 combination is plural, it is expressed in the pre- fix. In both lsg + 2sg e'-[verb]-ami and 2sg -+ lsg e'-[verb]-ais, the prefix is 2sg el-. This system shows elements of strategies 6 (portmanteau) and 9 (component morpheme shared).

In Siuslawan (Frachtenberg 1922b:468, 472-75), transitive verbs are suffixed with (a) very general "objective element" markers, either '3obj' or 'lI2obj' (cf. strategy 9), next (b) specific object markers (like lpl), and finally

(c) specific subject markers. In several 1 * 2 combinations, the expected three-part sequence of -112obj-0-S, with "objective element" followed by specific object and subject markers, actually surfaces only for combinations where the object is plural (2sg + lpl, 2pl-+ lpl, lsg +2p1, and lpl +2pl). With nonsingular subject on singular object (ldu + 2sg, lpl + 2sg, 2pl + lsg), the specific singular object marker is omitted (strategy 3), giving -112obj-S. In the remaining (sg + sg) cases, we get -iits-(a)-nx (-112obj-2sg, with intervening linking vowel) in both senses ('lsg -+ 2sg' and '2sg + lsg'), which can only be disambiguated by adding (optional) independent pro- nouns. Although this paper adduces other cases of merger of 'l(sg) +2(sg)' and '2(sg) + l(sg)' ('I see you' = 'you see me'), this creates communicative strains, and in Siuslawan it may be motivated idiosyncratically by avoidance of homophony with forms elsewhere in the ~ystem.~

Klamath (Barker 1964:239, 242) has, in addition to ordinary independent pronouns, exactly two special portmanteaulike forms (strategy 6) for sub- ject-object combinations: 2sg + lsg 'ins (or 'is) and lsg + 2sg min. Note

Since 3sg is a zero pronominal in Siuslawan, the sense '3sg + lsg' is expressed regularly as -112obj-lsg(-0). This may explain why the sense 'Isg + 2sg' is expressed not by the expected *-112obj-lsg (which would be homophonous to the '3sg + lsg' just described) but rather by -112obj-2sg (also used for '2sg + lsg'). Given the omission of Isg and 2sg object markers in 1 ++ 2 combinations, some homonymy is unavoidable, but the language can "choose" where the homonymy will occur.

that these are precisely the two most pragmatically sensitive combinations. The first of them is vaguely segmentable as a fused contraction of 2sg 'i, lsg ni, and objective marker s. The second is more thoroughly unanalyz- able, though again the alveolar nasal could perhaps be connected with the 1 sg morpheme.

In the 1 ++ 2 forms of Southern Sierra Miwok, we can usually segment the morphemes but there are some difficulties in glossing them (Broadbent 1964:43-44, 93). In 1sg -+ 2sg -mu-s.u and lsg + 2pl -mu-tok-su the -s.u is probably a '2' morpheme (cf. '2 intransitive' si.) and the -tok- is clearly a '2pl' marker. The remaining morpheme -mu- is perhaps a '1' allo- morph (cf. lsg intransitive ma- in alternation with zero). The 2 + 1 forms are even more opaque, especially 2sg + lsg -mu., which appears to con- sist just of the '1' allomorph with no overt subject marker (strategy 3; for 2sg + lpl we get -muh-me., ending with a 'lpl object' morpheme). The preceding forms are from the indicative "series 3" used in present imper- fect and present perfect. Among the "series 4" forms used in the impera- tive, we may mention 1sg + 2sg -ni (the morpheme can be tentatively identified as a minor '2' marker) and 2sg + lsg -. (i.e., lengthening of preceding vowel) or -tY.. The latter variant most closely resembles the in- transitive 'linclusive pl' marker (see strategy 8).

8. Hokan. This is a notoriously controversial genetic grouping. I have spot-checked a few of the languages, but not many have fused transitive pronominal agreement systems.

An exception is Karuk (= Karok), where the archaic-looking transitive pronominal prefixes (occasionally circumfixes) show extensive number neu- tralization and syncretism, so the forms often require supplementation by independent pronouns. Data are from Bright (1957:64); cf. also the mor- phological reanalysis in Macaulay (1992). In the indicative positive, we get the following syncretisms involving 1 ++ 2 combinations: (a) 1 +2sg = 1pl -+ 3 nu:; (b) { 1, 3) -+ 2pl ki.(k):..-up; (c) {2sg, 3sg} + 1sg na-; (d) (2p1, 3pl) + lsg kana:; and (e)(2, 3) -+ lpl = 3pl + 3pl kin:. In all cases, one of the two component markers in the 1 ++ 2 combination is merged with a 3d-person marker (cf. strategy 5), and except for (a) all the mergers are in the subject category. In the 1 + 2 (but not 2 + 1) forms, subject number ('Ilwe') is always neutralized (strategy 4). Since the transitive-prefix sys- tem seems to be residual, a formal morphological analysis is intrinsically slippery, but we may hazard a few comments. There is no sign of a 2sg morpheme (cf. pronoun 'i.m) in any of the 1 ++ 2 combinations. Form (a) consists of a lpl morpheme, so we have number neutralization by which the plural form generalizes to include singular usage. This violates Jakob- sonianIGreenbergian markedness principles but is nicely in line with prag- matic skewing, since the reader will agree with us (!) that 'lpl' is less intense pragmatically than 'lsg' (strategy 4). Form (c) is a lsg marker. Neither (d) nor (e)occurs as such in the intransitive paradigm, so neither is a pure per- son marker. We might analyze (d) kana': as containing (c) na'-'lsg', whereby the residual ka-..-is taken as indicating plural subject (with 1sg object). The morphemic glossing of (e) is also tricky; it is mainly a 'lpl object' marker but extends also to '3pl + 3pl'. -In Washo (Jacobsen 1996:33 and personal communication, 1997), the 2 1 proclitic combinations are 1 + 2 mi-1'- (the superscript indicates e-coloring of a following vowel) and 2 + 1 le-m-. These consist of the reg- ular independent pronouns 16: and mL: (minus the latter's word-final length- ening), plus the regular subject markers also used with, e.g., intransitive verbs. The system of categories is already minimal, since the Dual and Plural morphemes which can be added to independent pronouns are not used in these fused combinations. Also relevant is the fact that the precon- sonantal variant of mi-1'- is not the expected #mi-di- with the usual precon- sonantal lsg-person subject allomorph di-, rather a truncated mi- with zero expression of the 1st person (strategy 3), as in mi-da'mal-6:s-i 'I do not hear you'. Another instance of the same strategy occurs in the imperative. The usual Imperative prefix is ge- (or allomorph), but just in the 1st-person ob- ject (i.e., 2 -+ 1) form this prefix is omitted. Instead, we get the prefix used (in the indicative mood) for 3 + 1, hence la-da'mal 'listen to me!', distin- guished from la-da'mal-i 'he hears me' only by absence of Imperfective -i.

9. Kiowa. This language is related to the Tanoan languages of the pueb- los; Sapir's grouping of Kiowa-Tanoan with Uto-Aztecan remains to be definitively proved. Inflected verbs begin with a slot for an internally com- plex pronominal prefix. In Watkins's analysis (1984: 109-45), the prefix be- gins with one full pronominal marker, from either the "agent" or "patient" series, and ends (if applicable) with an indication of the number category of a 3d-person "object" (a category distinct from "patient"). Watkins's ab- stract underlying representations (intended, more or less, as internally re- constructed proto-forms) bear only a distant resemblance to the surface forms; e.g., underlying 2sg intransitive llb-id-// metamorphoses into dm-.

By definition, a system like this with only one full pronominal does not allow the combination of overt '1' and '2' person markers, so it fits nicely into the pragmatic-skewing model. The choice of '1' or '2' person marker at the beginning of the various 1 *2 combinations depends on various fac- tors, with a nonsingular patient getting first priority, and 2d-person nonsin- gular agent next highest priority. For 1sg + 2sg, one option is dm-, which is identical to the 2sg intransitive and so is analyzable as just '2' morpho- logically (strategy 3). Another option is the referentially overlapping 1sg + {2sg, 3sgJ gya-, which is arguably based on a special '1' allomorph also seen (but with different tone) in 'lsg agent with 3sg object [t.patient]' gya-.


In each of the seven remaining 1 * 2 combinations, for one of the two components we get a merger {1, 3) or 12, 3) involving third person (cf. strategy 3), though the morphemic composition is often not transparent, as in {2sg, 3sg) -+ 1sg e-.

    Keresan. In Acoma (Miller 1965:lOO-101), the whole system of agreement prefixes is messy, since even the intransitive prefixes occur in three main series (nonmodal, dubitative, hortative) with many irregularities. For 1 -+ 2 the forms for the three series are ?a-/ca-/?a-, which seem loosely connected to the parallel intransitive '2' forms ?-/c-/{F-, Fuu-, Fuu-); this is in the region of strategies 1 through 3. The three 2 + 1 forms are dyu-/dyu-/gu-, for which no clear morphological identifications emerge, so they are arguably portmanteaus (strategy 6).

    Tarascan. This Mexican isolate distinguishes '112indicative' -kafrom '3indicative' -ti-based on whether the subject is or is not a speech-act participant. This opposition helps resolve ambiguities due to syncretisms in the pure pronominal affixes and enclitics (lpl = 3pl subject ksi, 2pl = 3pl object ksi-ni; Foster 1969:44, 54, 56). Of more direct interest here is another combined 11, 2) morpheme, namely, the enclitic c'e for '112object' (often but not always indirect object; Foster 1969:125). When '112indicative' is combined with '1/2object', the result has 1 -+ 2 and 2 -+ 1 readings: //a- ri=c'e-a-ka// 'you will tell me; we will tell you' (the last two morphemes are '-Fut-112indicative'; Foster 1969:35). As noted earlier, the merger of 1 + 2 and 2 + 1 forms can cause communicative problems, and presumably re- quires the occasional addition of free pronouns.

    Mayan. In most languages of this large family, ergative pronomi- nal~are prefixed but absolutive pronominals are suffixed, so there is no mor- phological fusion or other interaction between the two series and the pickings are thin for our pragmatic-skewing model. However, in some of the languages we get adjacent ergative and absolutive prefixes, and a few "chemical reactions" among them occur. In Tzotzil, absolutive pronominal markers are prefixed (to the left of the ergative prefixes) if an aspectual prefix (e.g., completive) is also present, otherwise the absolutive markers are suffixed. The intriguing irregularity is that in the 2 + 1 combination, the 1st-person absolutive is suffixed, even when an aspectual prefix is present (Aissen 1987:44-45). One could interpret this as an instance of strategy 11 (physical separation of '1' and '2' markers). However, a con- tributing factor motivating avoidance of a prefix sequence * labs-2erg-. . . could be avoidance of phonological problems; the sequence would be -iplus -a(v)- and the first prefix would risk being phonetically absorbed by the initial vowel of the second.

    Mixe-Zoque. For this family, inspection of a few grammatical sketches suggests that opaque 1 o2 combinations are common; I presume that free pronouns are regularly used for disambiguation. In the languages examined, verbal prefixes indicating pronominal person (or transitive per- son combinations) are supplemented by suffixes, mainly '112pl' and '3pl'.

For the Mixe branch, in ~o~oluca~ the

of Oluta (Clark 1981:126-27), prefix t~5-is used both for 1 + 2 (without suffix) and 2 -+ 1 (with a further suffix "-'c," i.e., [-'k]).This -'k is the only pronominal suffix not marking plurality. The prefix ttts'-contains the regular '1' prefix t~-.The s' shows up elsewhere only in 1 -t 3 and 2 -+ 3 forms (and there only in the dependent mode) and is difficult to gloss; it has no resemblance to the '2' morpheme, which is mi-.

In Zoque of Francisco Le6n (= Magdalena), if I interpret Engel and Al- hiser de Engel (1987:378-82) correctly, the only transitive pronominal prefixes are (a)3 -+ 1 0-(zero); (b) 3 -+ 3 Iy-I; (c) (1, 2) -+ 3 and 1 -t 2 IN-I; (d) 12 -+ 3 and 2 + 1 nd@-; and (e) 3 -+ 2 lNy-1. Phonetically, Iyl represents palatalization of the stem-initial consonant, and IN1 surfaces as a homorganic nasal. Form (d) is presumably the combination of the nasal archiphoneme in (c) plus '12' (inclusive) t@-(this is the 12 intransitive subject form). We can therefore generalize that /N-/ is always present for { 1, 2) transitive subject (it is not used in the intransitive paradigm). Form

(e) looks superficially like a composite of forms (6)and (c), but /Ny-/ is also used as the '2' subject marker in intransitives and so may be a single morpheme, and 3 -+ 2 does not fit the categorial specification just given for use of /N-/. Plurality of subject is indicated by '112pl' and '3pl' suffixes. Of special typological interest here is the use of '12' (i.e., 1st inclusive) morphology for the 2 -+ 1 combination. Instead of appearing as separate pronominal markers ('2subject' plus 'lobject'), the 1st- and 2d-person ele- ments are, for morphological purposes, additively combined into inclusive person (strategy 8). This is a logician's (and syntactician's) nightmare but a pragmaticist's idyll. Note also that form (c) subsumes 1 -+ 3 and 1 -+ 2, giving the familiar 1 -+ (2, 3) pattern (strategy 5).

In Zoque of CopainalA (Harrison, Harrison, and Garcia 198 1:425-26 and Wonderley 1951:140-41), the facts are similar except that the 2 -+ 1 prefix is /N-/ with no additional 'linclusive' morpheme, so 2 1 is syncretic


with 1 -+ 2. Although this syncretism can cause communicative problems, the residual prefix systems in Zoque languages are so reduced overall that they are no longer heavily relied on to express pronominal categories.

14. Oto-Manguean. In Popoloca of Metzontla (Veerman-Leichsenring 1991:126, 152-54), there is a distinction between (morphological) one-

'Unrelated to Popoloca (Oto-Manguean, see below)

place ("S") and two-place ("SO") verbs. The "0"(object) in SO is usually a human object (indirect object, or object of certain verbs like 'want', 'owe', and 'bite'), so most syntactically transitive verbs ('hit', 'kill', 'eat') are sim- ple S verbs morphologically. Inflected verbs take a pronominal suffix ("desinence"); there are also clitics for various plural pronominal catego- ries. Suffixes, like syllables of stems, can have lexical tones, and there are some tonal interactions between stems and suffixes. For S verbs, the basic suffixes are: zero for '3', -a (no inherent tone) for 'l', and -a (macron in- dicates inherent mid tone) for '2' (1991:129). For some S verbs, the first syllable of the stem has one tone for (1, 2) subject, a different tone for 3d-person subject.

A distinct set of suffixes is used with the SO verbs. If we disregard the tones and focus exclusively on the segments, we have -7e for (2, 3) -+ 3, -nu for 12, 3) + 1, and -?a for both ( 1, 3) -+ 2 and 1 + 3. From this infor- mation we infer that the basic senses of these morphemes are -3e = '3object', -na = 'lobject', and -?a = '2object', though the latter is also extended anom- alously to 1 + 3. Identifying these morphemes as object-person markers is confirmed by comparison with possessive suffixes for the corresponding persons on nouns, which have the same segments.

Overlain suffixal tonal distinctions complement these suffixal-segment oppositions, with the result that all seven nonreflexive person combina- tions are differentiated one way or the other. For example, what is shown above as a triply ambiguous suffix -?a actually materializes as three distinct forms: 3 -+ 2 -?a, 1 -+ 3 -313, and 1 -+ 2 -?d. Perhaps these tonal distinctions historically reflect the interaction between the original lexical tone of the (object-marking) suffix and anticipation of the tone of a following (subject- marking) clitic, at least for 1st- and 2d-person subjects (Popoloca has no nonzero 3d-person clitic). However, I can detect no clean synchronic corre- lation of subject person categories with suffixal tones: for 1st-person sub- ject we get low (1 -+ 3) and high (1 -+ 2) tone; for 2d-person subject we get mid (2 -+ 1) and high (2 -+ 3); and for 3d-person subject we get low (3 -+ 1) and mid (3 -+ 2 and 3 + 3).

To return to this paper's focus on 1 -2 combinations, the point to make is that the Popoloca 1 ++ 2 set distinguishes 2 -+ 1 from 3 -+ 1, and 1 -+ 2 from 3 + 2 (and 1 -+ 3), only by the tone of suffixes. Without entering into a psycholinguistic disquisition, it seems reasonable to argue that pitch dis- tinctions are perceptually less conspicuous under normal listening condi- tions than distinctions made by -CV (i.e., fully syllabic) segment strings, so that in the two-place verbs the object is strongly marked while the sub- ject is faintly marked. In Popoloca, this is all the more so since there is no one-to-one correlation of tones with specific subject persons. Although close in spirit to strategy 1 (person marker disguised by partial phonologi- cal distortion), this is really a different mechanism (strategy 12) taking advantage of the differential perceptual accessibility of segmental and supra- segmental material.

    Aymaran. In Jaqaru (Hardman 1983:92-106), we get fairly reduced transitive suffixes that seem to consist of a single clear person morpheme, along with an onset vowel that is unglossable and dubiously segmentable. For the realis tenses, 1 -+ 2 -ima is basically a '2' morpheme seen in pos- sessive -ma 'your'. 2 + 1 -uta resembles a different '2' morpheme seen in 2d intransitive and 2 + 3 -ta but is also similar to the basic '1' suffix (in- transitive, also 1 -+ 3) -tha. Taking the Future as representative of the irre- alis tenses, the 2 -+ 1 form -(u)tumata can be segmented as -(u)tu-mata, which ends in the 2 intransitive and 2 -+ 3 Future marker -mata and per- haps begins with a minor 'Irrealis lobject' marker (cf. 3 -+ 1 Future -(u)tuni). However, the Future 1 -+ 2 form -mama looks like a doubled '2' morpheme. Overall, the 1 -2 combinations consist of various '2' allo- morphs and contain no discernible '1' morphemes (approaching strategy 3).

    Npi-Guarani. Intransitive verbs are of two types, corresponding to what many typologists call "active" and "stative" intransitives (for a functional-typological analysis, see Payne 1994). For these two classes in Guarani, Gregores and SuArez (1967:131-32) give paradigms of active ("subject") and stative ("personal reference") prefixes.8 Transitive verbs have so-called object prefixes; I would prefer to analyze them as subject1 object combinations. To be sure, some "object" prefixes do basically mark object category and are identical in form to corresponding stative intransi- tive pronominals: X -+ lsg Jle-, X -+ lpl (exclusive) ore-, and X -+ 12 (in- clusive) yane-, each consisting of a simple 1st-person marker. However, we could alternatively claim that 2 + 1 merges with 3 -+ 1 (to form X -+ 1) due to zeroing of 2d-person agent markers (the '3' marker is zero anyway); cf. strategy 3. Moreover, the merger of 3 + 2 and 1 + 2 into X -+ 2, expected if these prefixes were simple "object" markers, does not occur. While the 3 -+ 2 forms do indeed match the corresponding stative 2d-person markers (2sg and 3 2sg ne-, 2pl and 3 -+ 2pl pene-), for 1 -+ 2 we get special


prefixes 1 -+ 2sg ro- and 1 -+ 2pl po-, distinct from the 2d-person stative markers. These are arguably related to active 2sg re- and 2pl pe-, respec- tively, on the basis of shared consonants, but the vowels are different and we would expect true "object" markers to match the stative, not active, series. Moreover, 1 -+ 2sg ro- is more exactly identical to active 'lpl' (exclusive) ro-, which makes sense if we take the 1 -+ 2sg morpheme as designating the

*The stative prefixes are also substantially identical to possessive prefixes on nouns. and to free pronouns.


agent (rather than object), and furthermore merging subject number in favor of the plural form-strategies 3 and 4. The ambiguous morphemic identity of 1 +2sg ro- ('lpl' or '2sg') is positively evaluated in a pragmatic-skewing model, though of course this example depends on a language-specific mor- pheme-similarity quirk and is not here beatified as a typological "strategy."

Aside from the activelstative contrast in intransitives, the Guarani sys- tem has structural affinities to the Carib systems discussed below, since both use strategy 10 (single-slot transitive agreement) supplemented by one or two transitive portmanteaus. Tupi-Guarani and Cariban are thought to be distantly related genetically.

17. Quechua. In the Huallaga variety of Quechua (Weber 1989:77-79, cf. 96), a "transition" is a suffix complex including both pronominals and tense markers, fused together to varying degrees. Weber argues that the ba- sic structure of a "transition" is object-tense-subject, but he proceeds to rec- ognize a "subject marking anomaly" rule (1989:97) by which the 3d-person subject marker in 3 -+ 12 (inclusive) and 3 -+ 2 combinations is replaced by a subject marker matching the category of the object. Hence, 3 -+ 12, which should show up as '12-tense-3', actually shows up as morphemic '1-tense-12'; likewise 3 -+ 2, which should show up as '2-tense-3', actually shows up as '2-tense-2'. Inspection of the paradigm of transitive transitions suggests a partial reanalysis. To begin with, since 3d-person object is ex- pressed by zero, X + 3 forms for any X can be disregarded since they are identical to intransitive-subject X prefixes. Remaining transitive transitions, except 1 + 2, begin with either 'lobject' -ma- (used in 12, 3) + 1 and 3 + 12) or -shu- for 3 -+ 2. Neither -ma- nor -shu- occurs elsewhere in the pro- nominal morphology, and since -shu- occurs only in the 3 + 2 form I prefer to just label it as '3 -+ 2' rather than (with Weber) more ambitiously as '2object'. The -ma- or -shu- is followed by a tense morpheme (Present -n-, Past -ra-). The third (rightmost) morpheme position may be either a subject or object marker, according to a person hierarchy (12, 2) > 3 > 1. We get '12' -nchi: only in the 3 -+ 12 forms (there is no *2 + 12), e.g., Past -ma-ra-n~hi:.~ 2 (Past -shu-ra-yki) and

We get a '2' morpheme in both 3 + 2 + 1 (Past -ma-ra-yki), and a '3' morpheme only in 3 + (lsg, lpl) (Past -ma-ra-n). So far there is no especially strong indication of pragmatic skewing of 1 *2 combinations.

However, the tripartite transition structure does not apply to 1 -+ 2 forms, which lack the initial morpheme. The Present tense 1 -+ 2 form is just -:(lengthening of preceding vowel), which lacks even the Present

9I am inclined to detect '1' (especially 'lsg') morpheme -:(see below) in -nchi:. though of course this violates Weber's tripartite analysis.

tense marker -n-and consists of a '1' (or perhaps 'lsg') morpheme seen in lsg possessive -:. Clearly there is no '2' morpheme here. The absence of the Present marker distinguishes 1 +2 -:from 1 +3 -n.l0However, in the Past tense a single form -ra-: is used for 1 + 2 and 1 + 3, so we have an- other syncretic 1 -+ (2, 3) combination. Strategy 3 therefore permeates the 1 +2 forms, but whether it produces actual loss of information depends on whether the tense marker is also deleted.

18. Carib. (I use I for the back unrounded vowel.) Strategy 10, where the agreement slot permits a single morpheme (representing subject, object, or a portmanteau), is typical of Carib languages (see Gildea 1992 for a historical and typological survey). Faced with such a system, a formal syn- tactician instinctively looks for a geometrical principle based on arboreal structure, such as a constraint on paths of feature percolation or of raising (Alexander 1990). On the other hand, a functional typologist's instinct is to look for a hierarchical principle explaining why the agreement slot is occupied by the subject in some combinations and by the object in others. Both approaches have been essayed. Neither has been fully satisfactory and neither has any account (going beyond pure parsimony) for the existence of single-slot systems.

In the case of 3 -+ 3 and (1, 2) *3 combinations, surface forms con- sisting of no more than one nonzero pronominal marker are typologically common even in multiple-slot agreement morphologies, since 3d person is often expressed by zero. So the most interesting aspect of single-slot systems is that 1 *2 combinations are expressed by either '1' or '2' instead of as a nonzero sequence -1-2- or -2-1-. I suggest that strategy 10 is simply one device among many to avoid such transparently bipartite 1 *2 com- binations. Single-slot systems prohibit (by definition) juxtaposed person markers and also entail a considerable amount of neutralization of subjectlobject categories, though excessive neutralization can be avoided by adding a few special transitive portmanteaus. Specific details (which do not matter much in the current theory) vary from one Carib language to an- other, but the basic single-slot structure (which does matter) has been rela- tively consistent from the proto-language on. This suggests that the real "function" of the system is at least as much to avoid transparent 1 * 2 combinations as to "express" a specific person hierarchy.

In Hixkaryana (Derbyshire 1985: 188), there are eight transitive agree- ment prefixes (two of which have phonologically based allomorphs). Six of them also occur in intransitive uses, from which we infer their core pro- nominal senses; the remaining two are candidates for the status of transitive

'"his could be analyzed as Present -n-plus a phonetically vacuous '1' marker. on the as- sumption that vowel lengthening applies vacuously to a preceding consonant.


portmanteaus. However, one of the "portmanteaus" 0.'-10-) occurs in place of the usual 3 + 3 prefix ni- when the verb is immediately preceded by an object NP; we can think of 0- (before consonants) as absence of a prefix, and of y-(before vowels) as a ligature." This leaves ro- '3 + 1' as the only clear transitive portmanteau.

Since there are no portmanteaus in the 1 ++ 2 part of the transitive sys- tem, we can identify all relevant prefixes with a specific personlnumber value (by comparing them with the intransitive paradigm). The sole 2 + { 1, 3) prefix mi- is a '2' allomorph used with certain intransitive verbs (in- cluding motion verbs); note the by-now familiar merger of object person (and number); cf. strategy 5. The lsg 2 prefix kl- is identifiable as the


(intransitive) 'lsg' morpheme; curiously, it has a very different transitive sense 3 -+ 12 with inclusive object. The remaining 1 ++ 2 category is lpl -+ 2, which is expressed by o-, the primary '2' allomorph in intransitives (it is also used for 3 -+ 2). There are hints, but only hints, of a hierarchical system here-there is a partial preference for '2' person markers (whether subject or object), but kl- ('1') for lsg -+ 2 is a counterexample, and the preference for '2' markers may simply reflect the convenient availability of two distinct allomorphs (o-, mi-) for this category. The impression one gets is that the single-slot system is motivated by pragmatic masking, but that within this structural framework the language strategically deploys its person markers and their allomorphs to maintain a decent level of referen- tial clarity (i.e., somewhat more than one would get with a rigid hierarchy such as 2 > 1 > 3 without compensating allomorphic variation).

The details differ in other Carib languages. In the "Carib" from Surinam described by Hoff (1968:26; cf. the functional-typological reanalysis in Gildea 1994), ki- (cognate to the '1' marker of Hixkaryana) turns up as a '12' (inclusive) marker in intransitives and is used for both 1 -+ 2 and 2 -+ 1 in transitive contexts. Its core sense in this language is therefore 'l2', and its use in 1 + 2 and 2 -+ 1 transitive combinations illustrates strategy 8 as well as 10, with the 1st and 2d markers being morphologically added to produce '12' in defiance of grammatical logic. A similar system occurs in Apalai (Koehn and Koehn 1986:108), where k(i)- is used for 1 -+ 2, 2 -+ 1, 3 -+ 12, and (with nouns) for 12 possessor, and hence seems to have '12' as core meaning; however, in intransitive verbs the 12 subject marker is s(l)-, not found elsewhere in the system.

19. Concluding remarks. Despite the brevity of the descriptive sec- tions above, I hope readers are persuaded that Native American languages with rich agreement amply confirm the same tendency toward opacity of

"The combination of object NP plus verb. like that of possessor plus possessed noun, approaches a compoundlike character in this language.

1 ++ 2 combinations which had been observed earlier in Australian lan- guages. This opacity takes a myriad of forms, which the twelve strategies listed above only crudely summarize.

I anticipate difficulties, on the part of both cognitive-functional typ- ologists (at least those who are more cognitive than functionalist) and formal syntacticians, in absorbing and applying the present framework. The whole point of those approaches is to positively motivate attested lin- guistic structures. I have argued, on the contrary, that 1 * 2 agreement combinations delight in messiness. Structures that make the most sense cognitively or formally are actually avoided when they denote pragmati- cally sensitive pronominal combinations. Opaque structures, I suggest, are selected for and may be stable over long periods of time.

As I comb through grammars for this project, I am continuously struck by the tendency of linguists to be bothered by the morphological disarray that confronts them. One way to defeat the messiness is to set up ideal- ized, transparent, and symmetrical underlying forms (which are then sub- jected to surface disfigurement and rearrangement by morphological and phonological rules). A similar intellectual comfort can be achieved by reconstructing a "golden age" proto-language, where the transparency and symmetry were visible on the surface (deep and surface structure being identical) before undergoing sound changes or restructurings. One also encounters efforts to impose order on the 1 ++ 2 subsystem by elaborating (1, 2) > 3 . . . hierarchies as either 1 > 2 > 3. . . or 2 > 1 > 3 . . . -whichever works better for a given language-but often at the cost of artificial segmentation and labeling of surface morphemes in opaque 1 *2 combinations, and at considerable risk of missing the general point.

My immediate concern is to establish the plausibility of the general idea of negative-target avoidance by demonstrating the frequency of nontrans- parent subjectlobject combinations in Australia (Heath 1991), the Ameri- cas (the present paper), and eventually other regions (work in progress). In this rhetorical context, I have tended to treat all of the avoidance strategies as functionally equivalent. If the basic model achieves a degree of initial acceptance, as the regional surveys progress it will be possible to shift the focus to lower-level issues. The typology of avoidance "strategies" is of course evolving; the present paper has added strategies 9-12 to those from the original Australian sample, as well as describing further variants on some of the eight originals. It is too early to attempt quantitative study of the distribution of these strategies, either globally or in correlation with typological characteristics, but in the long run such a study would be worthwhile, in spite of coding difficulties and all the technical problems associated with "random" sampling of languages. In strategies 3 and 10, where just one of the two markers is omitted, it may turn out that there are


asymmetrical preferences for omission from the set (subject, object}, from the set ( 1 st, 2d), or from both.

Another line of attack would be a quantitative cross-linguistic study of the relative frequencies with which particular 1 ++ 2 combinations display categorial merger and/or formal skewing. On theoretical grounds, it might be predicted that skewing is most typical of the pragmatically delicate sg -+ sg types 'I saw you[sg]', 'you[sg] saw me', but this is at present merely a hypothesis.

Another topic deserving further study is interactions between 1 ++ 2 skewing and tense/aspect/mood (TAM) categories. I suggested above, in con- nection with Eskimo, that the social delicacy of these pronoun combinations might vary from one mood category to another and that this might be occa- sionally reflected in the morphology. It is possible that with a larger sample of languages, a sufficient number of systems with pronominal/^^^ interactions can be found to permit a more general investigation of this issue.


AISSEN,JUDITH.1987. Tzotzil Clause Structure. Dordrecht: Reidel ALEXANDER.

MATT. 1990. Agreement configurations: argument structure in modular grammar. Ph.D. dissertation, Massachusetts Institute of Technology. BARKER.MUHAMMAD.

1964. Klamath Grammar. UCPL 32. Berkeley and Los Angeles: Uni- versity of California Press. BLOOMFIELD.LEONARD.1946. Algonquian. Linguistic Structures of Native America. ed.

C. Osgood, pp. 85-129. New York: Viking Fund.

BOAS, FRANZ. 191 1. Chinook. Handbook of American Indian Languages. pt. 1. ed. F. Boas. pp. 559-677. Bureau of American Ethnology Bulletin no. 40. Washington. D.C.: Smithso- nian Institution.

BOAS, FRANZ. AND JOHN SWANTON. 191 1. Siouan (Dakota). Handbook of American Indian Languages, pt. 1. ed. F. Boas, pp. 875-965. Bureau of American Ethnology Bulletin no. 40. Washington. D.C.: Smithsonian Institution.

BRIGHT.WILLIAM.1957. The Karok Language. UCPL 13. Berkeley and Los Angeles: Univer- sity of California Press. BROADBENT. 1964. The Southern Sierra Miwok Language. UCPL 38. Berkeley and

SYLVIA. Los Angeles: University of California Press. CLARK,LAWRENCE.

1981. Diccionario Popoluca de Oluta. Mexico City: Instituto Lingiiistico de Verano. DERBYSHIRE. 1985. Hixkaryana and Linguistic Typology. Dallas: Summer Institute


of Technology. EINAUDI,PAULA. 1976. A Grammar of Biloxi. New York: Garland. ENGEL, RALPH, AND MARY ALLHISER DE ENGEL. 1987. Diccionario Zoque de Francisco Le6n.

Mexico City: Instituto Lingiiistico de Verano. FOSTER,MARY. 1969. The Tarascan Language. UCPL 56. Berkeley and Los Angeles: Univer- sity of California Press. FRACHTENBERG,

LEO. 1922a. Coos. Handbook of American Indian Languages. pt. 2, pp. 297-

430. Bureau of American Ethnology Bulletin no. 40. Washington. D.C.: Smithsonian Institution.

. 19226. Siuslawan (Lower Umpqua). Handbook of American Indian Languages, pt. 2, pp. 431-629. Bureau of American Ethnology Bulletin no. 40. Washington. D.C.: Smithso- nian Institution.

GILDEA,SPIKE. 1992. Comparative Cariban morphosyntax: on the genesis of ergativity in independent clauses. Ph.D. dissertation. University of Oregon. , 1994. Semantic and pragmatic inverse: "inverse" alignment and "inverse voice" in Carib of Surinam. Voice: Form and Function, ed. Barbara Fox and Paul Hopper, pp. 187-

230. Amsterdam: Benjamins. GODDARD,IVES. 1979. Delaware Verbal Morphology. New York: Garland. GREGORES,EMMA,AND JORGE SUAREZ. 1967. A Description of Colloquial Guarani. Janua

Linguarum, Series Practica. no. 27. The Hague: Mouton.

HARDMAN.MARTHA. 1983. Jaqaru: Compendio de estructura fonol6gica y morfol6gica. [Re- vision and translation of 1966 Mouton edition in English.] Lima: Instituto de Estudios Peruanos.


H. 1981. Diccionario Zoque de Copainala. Mexico City: Instituto Lingiiistico de Verano. HEATH, JEFFREY. 1991. Pragmatic disguise in pronominal-affix paradigms. Paradigms: The Economy of Inflection. ed. Frans Plank. pp. 75-89. Berlin: Mouton de Gruyter. HOFF, BEREND. 1968. The Carib Language. Verhandelingen van het Koninklijk Instituut voor Taal-. Land- en Volkenkunde, vol. 55. The Hague: Martinus Mijhoff. HOPPER. PAUL, AND SANDRATHOMPSON.1980. Transitivity in grammar and discourse. Lan-

guage 56:25 1-99. JACOBSEN, 1996. Beginning Washo. Carson City: Nevada State Museum.

WILLIAM. JELINEK,ELOISE,AND RICHARDDEMERS.1983. An agent hierarchy and voice in some Coast Salish languages. IJAL 49: 167-85. KOEHN. EDWARD, AND SALLYKOEHN. 1986. Apalai. Handbook of Amazonian Languages, ed. Desmond Derbyshire and Geoffrey Pullum. vol. 1. pp. 33-127. Berlin: Mouton de Gruyter. MACAULAY.

MONICA. 1992. Inverse marking in Karuk: the function of the suffix -up. IJAL 58:182-201. MILLER. WICK. 1965. Acoma Grammar and Texts. UCPL 40. Berkeley and Los Angeles: Uni- versity of California Press. MITHUN.MARIANNE.

1976. A Grammar of Tuscarora. New York: Garland. PAYNE,DORIS. 1994. The Tupi-Guarani inverse. Voice: Form and Function, ed. Barbara Fox and Paul Hopper. pp. 313-40. Amsterdam: Benjamins.

ROOD. DAVID. AND ALLANTAYLOR.1996. Sketch of Lakhota. a Siouan language. Handbook of North American Indians, vol. 17. Languages. ed. Ives Goddard, pp. 440-82. Washington, D.C.: Smithsonian Institution.

SAPIR, EDWARD. 1922. The Takelma language of Southwestern Oregon. Handbook of Ameri- can Indian Languages. pt. 2. pp. 1-296. Bureau of American Ethnology Bulletin no. 40. Washington. D.C.: Smithsonian Institution.

SCHULTZ-LORENTZEN,CHRISTIAN.1945. A Grammar of the West Greenlandic Language.
Meddelelser om Gr@nland, vol. 129. no. 3. Copenhagen: Reitzels. [Reissued in 1967.1
SILVERSTEIN. 1976. Hierarchy of features and ergativity. Grammatical Categories in

MICHAEL. Australian Languages, ed. R. M. W. Dixon. pp. 112-71. Canberra: Australian Institute of Aboriginal Studies.

VEERMAN-LEICHSENRING, 1991. GramAtica del Popoloca de Metzontla. Amsterdam: ANNETTE. Rodopi.

VOGT. HANS. 1940. The Kalispel Language. Oslo: Det Norske Videnskaps-Academi.

WATKINS, 1984. A Grammar of Kiowa. Lincoln: University of Nebraska Press.

LAUREL. WEBER, DAVID. 1989. A Grammar of Huallaga (Huanuco) Quechua. UCPL 112. Berkeley and Los Angeles: University of California Press. WONDERLEY, 1951. Zoque 111: morphological classes and verbs. IJAL 17:137-62.


  • Recommend Us