Lingua 71 (1987) 203-222. North-HoUand
203
ABSOLUTIVE ZERO: PARADIGM ADAPTIVITY IN SACAPULTEC MAYA John W. DU BOIS* UCLA, California, USA
1. Introduction
The fact that the phenomenon is called 'ergativity', and the languages exhibiting it 'ergative', highlights the distinctive status of the ergative category {A}, to the possible neglect of the absolutive category {S, 0}. 1 The ergative is indeed the marked member of the opposition; and in many theoretical approaches there is more profit in formulating rules in terms of the marked than the unmarked. Thus many have focused on accounting for the marked ergative (e.g. by a rule stating special conditions under which ergative casemarking is assigned), while leaving the unmarked absolutive as the 'elsewhere' case. But from time to time it is worth extending special scrutiny to the unmarked category, precisely because it is unmarked. This is especially so from a discourse point of view, where the unmarked category typically has a special relation to the most prevalent, or preferred, patterns of language use. If 'grammars code best what speakers do most' (Du Bois (1985a: 363)), then, given that it is the unmarked category which typically provides the most * The field research for this study was carried out in Guatemala during the seasons of 1974-1980, with support from the Survey of California and Other Indian Languages. In developing the ideas presented in this paper I have benefited greatly from the comments of many individuals; I wish especially to thank Judith Aissen, Jon Dayley, R. M.W. Dixon, Mark Durie, John Haiman, Terry Kaufman, Tom Larsen, Tony Naro, Will Norman, Michael Silverstein, and Sandy Thompson for their contributions to this paper. Final responsibility for any remaining infelicities rests with me. i Abbreviations used include the following: Abs--absolutive (e.g. Abs2sg absolutive second person singular); Abs3 = third person nonplural absolutive; Crop--completive aspect; D i m = diminutive; Era = ergative; Foc = focus particle; IF = intransitive phrase-final suffix; Inc-~ incompletive aspect; Mvt = movement inflection; N = noun, noun phrase; T f = transitive phrase-final suffix; V = verb; . . . . pause (not used for ellipsis); ..--very brief pause; : = phonemic length; = = prosodic lengthening, and " = devoicing rule does not apply to previous segment (usually reflects a deleted vowel). 0024-3841/87/$3.50 © 1987, Elsevier Science Publishers B.V. (North-Holland)
204
J. Du Bois / Absolutive zero
economical coding, it will be crucial to find out what discourse functions speakers most often carry out via, or in conjunction with, this category. The need to understand the role of unmarked grammar in discourse has defined the orientation of the research presented below. In section 2, I lay out the background of ergative--absolutive cross-referencing in the language of this study, Sacapultec Maya. In section 3, I describe a discourse pattern which is isomorphic to the ergative-absolutive grammatical pattern, and then try to show how the absolutive discourse category interacts with the absolutive grammatical category - and ultimately, influences the grammaticization of the system of grammatical relations. In section 4, I conclude with a look at some of the wider implications of my analysis.
2. The grammar of cross-referencing in Sacapultec Maya I will begin by presenting the structure of the cross-referencing system of Sacapultec Maya, giving special attention to the role of the absolutive zero. Sacapultec is a Mayan language of the Quichean branch, and is spoken in highland Guatemala (Du Bois (1981b, 1985b)). (It is genetically and geographically close to the Mayan language Quich6, treated by Larsen (this volume: 3359).) Sacapultec is squarely ergative in morphology (see below), and shows evidence of what has been called syntactic ergativity, in a typical Mayan pattern (Smith-Stark (1976,1978), Norman and Campbell (1978), Dayley (1981)., England (1983), Larsen (this volume: 33--59)). It is the morphological aspect of Sacapultec ergativity which will concern us here. Sacapultec has two prefixal paradigms which cross-reference both of the direct arguments of a transitive verb (A role and O role), and the single direct argument of an intransitive verb (S role). The A role argument of a transitive clause is cross-referenced with one from a paradigm of six ergative personnumber-role prefixes, while the O role argument of a transitive, and the S role argument of an intransitive, are cross-referenced by one from a paradigm of six absolutive person-number-role prefixes. In a transitive verb, the ergative prefix immediately precedes the verb stem, and the absolutive prefix precedes the ergative prefix. (In addition, an inflectional prefix indicating motion optionally appears between the absolutive prefix and the ergative prefix; an aspect prefix appears at the beginning of the whole verbal complex, in most aspects.) Each of the following transitive verbs constitutes (potentially) a complete sentence.
J. Du Bois / Absolutive zero
205
(1) g-at-ri-6'iy-arj Cmp-Abs2sg-Erg3sg-hit-TF 'He hit you.' (2) ~-0-a:-6'iy-arj Cmp-Abs3-Erg2sg-hit-TF 'You hit him.' In an intransitive verb, the only person-number-role prefix to appear is the absolutive.
(3) g-at-ak-ek Cmp-Abs2sg-enter-IF 'You entered.' (4) g-0-ak-ek Cmp-Abs3-enter-IF 'He entered.' Note the defining pattern of ergativity: the prefix a t - is used for second person singular in the O role (in (1)) or the S role (in (3)) (i.e. for the absolutive), while a : - is used for A role (in (2)) for second person singular (the ergative). Similarly, the absolutive prefix position is filled by 0- to cross-reference the O role (in (2) or S role (in (4)) for third person singular, while the ergative position is filled by ri- for the A role third person singular (in (1)). The full paradigms for the ergative and absolutive person-number-role prefixes are shown in table 1. Note that the ergative prefixes show some morphological alternation governed by position, i.e. preconsonantal vs. prevocalic. Table I Sacapultec cross-referencing paradigms. Ergative
Absolutive
Precons.
Frevoc.
ni- ~ ina:ri-
inwa:wr-
inat0-
qai:ki-
qi:wk-
axi[e:- ~ e?-
Singular 1 2 3
Plural 1 2 3
206
J. Du Boia / Absolutive zero
Before proceeding further, explanations are needed for certain terms and categories. I will use the term argument to refer to the 'core' or 'direct' arguments A, S, and O (as defined in this volume and in Dixon (1979)), but not obliques. In the grammar of Sacapultec the boundary between arguments and obliques is maintained consistently, and the difference is straightforwardly recognizable: an oblique noun phrase is consistently dependent on (and headed by) a preposition, while an argument is dependent on a verb. The clausal argument group is a subset o f the clause consisting of the verb plus its arguments (i.e. excluding prepositional phrases and other adverbials). Distinctions of form are crucial in this paper, at both grammatical and substantive (phonetic) levels. The category of free form includes full noun phrases (lexical forms) and independent pronoun phrases or pronoun words, while the label bound is applied to the cross-referencing affixes. A mention is an entity reference consisting of either a bound form (the 'cross-referencing' affix, as in sentences (1)-(4) above), or an overt free form plus its cross-referencing bound form within the same clause (as in (5) and (6) below). That is, a co-referential pair consisting of an overt free noun phrase and its cross-referencing affix in the same clause are counted as one mention rather than two. A mention (entity reference) consisting of a bound form only will be termed (obviously) a bound mention, while one consisting of a lexical form plus a bound form will be termed a lexical mention. A lexical argument is a full noun phrase in argument position. Distinction will also be made between morpheme types according to their substance. A positive (or nonzero) morpheme is one whose realization has phonetic substance, while a zero morpheme is one whose realization does not. (See below for the distinction between zero morpheme and zero allomorph.) Positive marks in Sacapultec encompass free noun phrases, free pronouns, and cross-referencing affixes other than zero. The term overt implies the presence of a noun phrase or free pronoun phrase in a position where it is optional, in contrast with a nonovert argument (what some would call 'deleted'). (Other terms will be explained as they come up.) As is typical of the Mayan languages generally, the third person nonplural 2 absolutive is indicated by the noticeable absence of any positive mark in the syntagmatic position of its paradigm, that is, by what is called a zero morpheme (Jakobson 0939), cf. also Bybee 0985), Jelinek (1984)). Given the strict syntagmatic ordering of the absolutive paradigm prefixes relative to the 2 The absolutive zero in Sacapultec is properly analyzed as nonplural (that is, unmarked for the positive feature + plural) rather than singular, because under certain conditions it may have a 'neutralized' value which can encompass plural third person cross-referencing (cf. section 3.2, example (11)).
J. Du Bois / Absolutive zero
207
other morpheme classes that make up the verbal complex (including the ergative), it is straightforward to locate the position for the absolutive in a given inflected vei'b. This, in conjunction with the fact that all the other prefixes of the absolutive paradigm are positive (nonzero), means that the third person nonplural absolutive can be unambiguously indicated by simple absence - what I refer to as absolutive zero. Since the third person nonplural absolutive is realized uniquely by zero, this is not simply a case of a zero allomorph, such as would alternate with one or more positive allomorphs, but of a true zero morpheme. The zero morpheme must be distinguished from simple phonetic attenuation, assimilation, or deletion, as happens elsewhere in the grammar of Sacapultec. For example, the completive aspect marker .~- coalesces under certain circumstances with a following third singular ergative prefix r-, resulting in a single segment (either ~- or r-, depending on the speaker's dialect). But in such phonologically- or morpho-phonologically-conditioned aUomorphic silence, the underlying form of the morpheme is not zero, the morpheme is not realized uniquely by zero, and any absence of audible marking is the result of secondary processes. This is distinct in principle from the true paradigmatic zero morpheme. Nor must the zero morpheme, as established by opposition within a strictly inflectional paradigm like that of Sacapultec cross-referencing, which is intra-clausal in scope, be confused with 'zero anaphora', as found in many or all languages, which is of inter-clausal scope. Being a relatively loose relation, zero anaphora can reach across boundaries of non-adjacent clauses which otherwise have little interconnection grammatically, in the sense of one clause placing grammatical constraints upon the other. Because paradigmatic substitution in exactly equivalent structural contexts would otherwise make confusion of reference a real possibility and in fact a frequent occurrence, it is necessary, if each of the prefixes in this kind of crossreferencing inflectional paradigm is to be distinctively symbolized, that only one prefix be marked by zero. So it is that, from the perspective of the ecology of paradigms, the value which is to be 'assigned' to this unique zero takes on a special significance (section 3). While the so-called cross-referencing affixes often appear as the only marking of a referent within the clause (as in (1)-(4) above), it is also possible for a free referring form to appear overtly along with its cross-referencing marker within the same clause. (It is such cases which give some legitimacy to the term 'cross-referencing', which I use here for lack of anything better.) Overt noun phrases and free pronouns are most abundant in elicited (as opposed to naturally occurring) sentences, such as those in (5)-(6) below. To
208
I. Du Bois / Absolutive zero
make clear the cross-referencing relations in these sentences, I have drawn a line connecting the overt noun phrase or pronoun with the prefix that crossreferences it. Further, because of the significance for thisstudy of the contrast between positive and zero elements, I have placed above each positive referring form (whether phrase, word, or affix) a single solid line, while above each zero referring form (that is, in this ease always the absolutive zero morpheme) I have placed a double solid line.
(5)
t
!1
±
e: 1 c'e? ki-0-r-tix ak' Foc the dog Inc-Abs3-Erg3sg-eat chicken 'The dog eats chickens.' Thus in (5), the (positive) noun phrase e: I c'e? 'the dog (focus)', s in A role, is cross-referenced with the (positive) third person singular ergative marker r-," while the (positive) noun phrase ak' 'chicken(s)', in O role, is cross-referenced with the (nonpositive) third person absolutive marker 0-.s Sentence (6) illustrates an optional free pronoun phrase. (6)
e: ra ax ka-O-qa-tix ti?ex Foc the we Inc-Abs3-Erglpl-eat meat ' W e eat meat.' In (6), the A role free pronoun phrase e: ra a x 'we (focus)' is cross-referenced with the first person plural ergative prefix qa-, while the O role noun phrase ti?ex 'meat' is cross-referenced with the third person absolutive zero 0-. 3 I use the gloss 'focus' here as nothing more than a place-holder in lieu of a proper analysis of the actual discourse function of the nominal particle e:, which in the end will probably be found to involve focus-of-contrast (Chafe 0976)) and other factors. 4 The vowel in underlying ri- here has been deleted to give r-. 3 In this contrastive sentence, the AVO constituent order is marked. Sacapultec allows most orderings of major constituents, with different pragmatic values (e.g. contrastiveness) associated with each. Unmarked order in (elicited) sentences is verb-initial: V NPabs NPerg. That is, for (elicited) transitives it is VOA; for intransitives it is VS.
J. Du Boi~ / Absolutive zero
209
While elicited sentences like (5) and (6) are useful for presenting the grammatical potentialities of the cross-referencing system in accessible terms, the configuration of overt surface elements which they exhibit should not necessarily be taken as representative of the structures that speakers tend to produce. What speakers actually do most often is taken up in the next section.
3. The absolutive in discourse and grammar
I will first need to briefly summarize my findings on the ergative-absolutive patterning of discourse (section 3.1), before I can show what the absolutive zero contributes to ergativity (section 3.2).
3.1. The ergative-absolutive patterning of syntax in discourse As part of my long-term research on the discourse basis of ergativity (Du Bois (1981a, 1981b: 285f., 1985a, forthcoming, a), cf. Cumming and Wouk (this volume: 271-296)), in 1977 I showed a brief film (now widely known as the Pear Film, Chafe (1980)) to a group of Sacapultec speakers, after which a Sacapultec interviewer asked each one to tell what happened in the film (Du Bois (1980)). The resulting narrations constitute the corpus for my studies on discourse and ergativity, including the present one. ~ (All examples cited henceforth are from this corpus.) Here I will briefly summarize the relevant discourse patterns, of which I have given a fuller account elsewhere (Du Bois (1987a)). Most important for the present inquiry is the finding that there is a systematic preference in discourse for one particular configuration of lexical arguments in surface syntax. This is part of what I have called Preferred Argument Structure (Du Bois (1981a, 1985a, forthcoming, a)), which is a complex of constraints on quantity and on role, paralleled in the domains of syntax and pragmatics, as realized in discourse. In terms of the appearance of overt noun phrases in surface syntax, in spontaneous narrative discourse in Sacapultec there is a strong tendency for clauses to contain no more than one lexical argument (the quantity constraint). Further, this single lexical argument does not appear with equal frequency in all argument positions, but shows a marked preference for the S role of an intransitive clause or the O role of a transitive clause, while it avoids the A role of a transitive clause (the role See Du Bois (1987a) for details on this corpus.
210
.L Du Bois / Absolutive zero
constraint). As a result of these two tendencies, the predominant surface syntactic structure for the clausal argument group in Sacapultee discourse is V N{s,o) - that is, a verb followed by a single (or no) overt noun phrase in S or O role (but not A role). (Since the constraints as stated apply only to the clausal argument group, any prepositional phrases within the clause are of course not governed by them. Obliques in fact constitute another position where noun phrases appear freely.)We can illustrate these patterns with examples taken from the corpus of Pear Film narrations. 7 First, an intransitive clause. (7)
L
I
... 0-0-pc: xun ax-la:b' .. ~ix bisikle:ta Cmp-Abs3-come a Dim-boy on.back bicycle '... a little boy came .. on a bicycle,' Here, a single lexical argument (xun ax-la:b' 'a little boy') appears in the S role of intransitive -pe: 'come'. (A lexical mention also appears here outside the argument group, i.e. in a non-argument role: oblique bisikle:ta 'bicycle' is headed by the preposition ~ix 'on (back)', and bears no direct grammatical relation to the verb.) The significance of the cross-referencing patterns in these sentences will be taken u p i n section 3.2. Another commonly found intransitive pattern is that of a verb with no lexical argument.
(8) ... tik'ara? 0-0-qa:x-u:l , then Cmp-Abs3-descend-hither '... Then he came down,' (This typically arises when a one-place predication is made of a referent which has already been mentioned, as in this predication about a man who had been introduced two clauses earlier.)
To allow the interested reader access to the full discoursecontext if desired, I have drawn all the discoursecitations in this paper (7)-(11) from a Pear Film narration which I have published in its entirety (Du Bois (1987b)).
J. Du Bois/Absolutivezero
211
In a transitive clause there are two argument positions, each of which could in theory hold a lexical mention. But as I have noted above, what actually predominates in discourse is sentences with just one overt argument, on the following pattern.
(9) ... ~aq ~-0-a-r"-k'am-a? xun ~ke~ pe:ra just Cmp-Abs3-Mvt-Erg3sg-take-Mvt a basket pear '... he just came and took a basket of pears.' Here, there is just one lexical argument, xun c~ke~ pe:ra 'a basket of pears', which falls in the O role position; and no lexical argument appears in the A role position. The syntactic aspect of Preferred Argument Structure is paralleled by the pragmatic aspect, embodied in constraints on the quantity of new information which can appear within the clause, and the role in which it can appear. (This is to be expected, given the association of full noun phrases with new information, see Chafe (1976).) New information is limited to one referent within the clausal argument group (the quantity constraint). This single new information reference tends to appear in the S or O position, but strongly avoids the A position (the role constraint, Du Bois (1981a, 1985a), of. Larsen (1981)). In sentence (7) above xun ax-la:b' 'a little boy' is new information, being the first mention of this referent, and constitutes the single new referent in this clausal argument group, appearing in the S role. In (9) (which in this speaker's narration is the next clause spoken after (7)), the single new information referent xun ?ke~ pe:ra 'a basket of pears' falls in the O role. The A position contains no new information; the boy, having been introduced in the S role of the previous clause, appears here as given information encoded in a cross-referencing affix, the ergative third person singular r(i)-. In summary, the dominant pattern in discourse for the clausal argument group is V N{s,o}. The {S, O} position accommodates full noun phrases and new information, while the {A} position tends not to accommodate them. (Although prepositional phrases also freely accommodate noun phrases and new information, insofar as these oblique noun phrases do not bear a direct syntactic relation to the verb, they are not eligible to participate in the grammaticization of the structure of grammatical relations, which is what is at issue.) Clearly, the Preferred Argument Structure pattern in discourse corre-
212
J. Du Bois / Ab$olutive zero
lates with the grammatical alignment represented in the category of absolutive. The absolutive accommodates new information, and receives most argument position noun phrases. While I have argued elsewhere that Preferred Argument Structure in itself sets the pattern in discourse for ergative-absolutive grammar (Du Bois (1985a, forthcoming, a)), there are other important factors to be reckoned with, factors which can reinforce this grammaticization motivation. Most important are motivating forces deriving from the paradigmatic structuring of crossreferencing systems, which may serve t o fix the absolutive alignment in grammar. Given the right alignment of syntagmatic relations with the paradigmatic zero, the incidence of what I will call 'double positives' can be sharply reduced. This reinforcing motivation for ergativity is what I will try to elucidate in the next section, by means of a series of text counts on the corpus. 3.2. Absolutive zero versus the double positive
Once we have observed that there is a pattern to where lexical arguments appear, we can consider whether there is a pattern to what kind of crossreferencing typically correlates with them. To this end, I divided the twelve cross-referencing affixes into two classes according to their substance: the class realized positively, containing all six ergative affixes and all but one of the absolutive affixes; and the class realized by zero, containing just one absolutive affix (the third person nonplural). Then, for each lexical argument in the corpus, I recorded whether it was cross-referenced with a positive class affix or a zero class affix. For example, in (7) above, the S-role lexical argument xun ax-la:b' 'a little boy' was cross-referenced on the intransitive verb with absolutive zero. Similarly, in a transitive clause the one lexical argument may be cross-referenced with zero, as in (9), where the O-role lexical argument xun ~ke~ pe:ra 'a basket of pears' is cross-referenced by the absolutive zero. It is also possible for a lexical argument to be cross-referenced by a positive morpheme:
(lo)
1
f
~-e:-pe: e: i~eb' al"?-o:m Cmp-Abs3pl-come P1 three boy-Pi 'three boys came,'
J. Du Bois / Absolutive zero
213
Here the S-role lexical argument e: i~eb' al"?-o:m 'three boys' is cross-referenced with the third person plural absolutive prefix, the positive morpheme e:-. We can now ask: when one examines the entire corpus, is there a systematic preference for one type of cross-referencing over the other? The results for the corpus are shown in figure 1, where the first column represents the percentage of all lexical arguments which are cross-referenced with a zero, and the second column represents the percentage which are cross-referenced with a positive morpheme. While on an a priori basis one might expect there to be more positive cross-referencing, since there certainly are more positive morphemes in the cross-referencing paradigms, s it turns out that in discourse the reverse is true. As seen in figure 1, a noun phrase in argument position is much more likely to be cross-referenced by zero (84.4%) than by a positive morpheme (15.6%). 9 100 84.4 80
per cent of lexical arguments (N=218)
60.
40 -
20-
o
15.6
I I
Zero Positive (N = 184) (N = 34) Cross-Referencing Type
Fig. 1. The zero correlation: lexical arguments receiving zero cross-referencing vs. receiving positive cross-referencing. s Even considering just the third person, the paradigms contain an equal number of morphemes in each class, i.e. one each. Figure 1 represents only lexical free arguments, and excludes personal pronoun free arguments (e.g. contrastive first and second person free pronouns). If we add in the small number (17) of personal pronoun free arguments in this corpus, we see that the zero correlation for all free arguments is slightly lower than for lexical arguments alone: 80.0% of free arguments have zero cross-referencing (N = 188), while 20.0% have positive cross-referencing (N -- 47).
J. Du Bois / Absolutive zero
214
While this shows that the pattern of lexical cross-referencing as exemplified in (7) and (9) predominates, there are some cases like that in (10), where a lexical argument is cross-referenced positively (even if in the S or O position). But what happens if we consider, not the substance Of cross-referencing, but its grammatical category? The same basic pattern of skewing appears - only more extreme. For this tabulation, the classes of cross-referencing morphemes are distinguished along grammatical lines, as absolutive vs. ergative. For example, in sentence (10) the lexical argument e: i~eb' al"?-o:m 'three boys' is cross-referenced absolutively, even if with a positive morpheme. The results for the corpus are presented in figure 2. The first column, which represents all lexical arguments which are cross-referenced with an absolutive morpheme (whether zero or positive), predominates still more strongly (95.4%) over the second column, which represents overt lexical arguments cross-referenced with an ergative morpheme (4.6%). ~° This skewing underscores the observation made in section 3.1, that an argument which is realized with a lexical mention 100-
95.4
80-
per cent of lexical arguments (N=218)
60-
40-
20-
0
I Absolutive (N--- 208)
[
Ergative (N = 10)
Cross-Referencing Type Fig. 2. The absolutive correlation: lexical arguments receiving absolutive vs. receiving ergative cross-referencing. 1o Again, figure 2 represents lexical free arguments, and excludes personal pronoun free arguments (N = 17). If we add these in, the absolutive correlation is slightly lower: 90.2% (hl = 212) of free arguments have absolutive cross-referencing, while 9.8% (N -- 23) have ergative cross-referencing.
J. Du Bois / Absolutive zero
215
will strongly tend to be in the absolutive category - and hence show absolutive cross-referencing. The significance of the zero cross-referencing of lexical arguments is made clear if we consider its effect on the economy (in discourse terms) with which referents are marked. Notice that, for a given argument reference, the number of positive marks within the clause can vary in principle from two to none. For example, if there is an overt argument (i.e. a full noun phrase or free pronoun argument) in the free position, and a positive morpheme in the crossreferencing position, this will result in what I will call a double positive, as in sentence (10) above. A single positive will occur either when an overt argument in the free position is accompanied by a zero in the cross-referencing position - as in the S-role argument of (7) or the O-role argument of (9) - or when there is no overt argument in the free position (so-called 'deletion'), but a positive morpheme in cross-referencing position - as in the A-role argument of (9). Finally, if there is no overt argument in the free position and a zero in the cross-referencing position (i.e. neither a positive nominal nor a positive cross-referencing morpheme - nothing but a zero cross-referencing morpheme), this is what I call a naught lr positive, as exemplified in (8). 60.1
60-
per cent of argument mentions (N-- 646)
4032.4
207.6
I i Naught Positive (N = 209)
Single Positive (N--- 388)
Double Positive (N=49)
Argument Marking Fig. 3. Double positive avoidance: frequency of naught positive, single positive, and double positive among argument mentions. 1 The word 'zero' is avoided here due to its obvious potential for confusion.
216
J.
Duaois/Absolutivezero
"
Of the three possibilities - naught positive, single positive, or double positive which is the most prevalent among mentions of arguments in discourse? Figure 3 presents the tabulations for the corpus, revealing what the observant reader will by now expect. Single positives (60.1%) and naught positives (32.4%) predominate, with the former more common in this particular corpus. But double positives occur only very rarely, in fewer than 8% of the argument mentions. We may entertain the hypothesis that a marking strategy is at work, to the effect that a mention of a referent should receive no more than one positive mark. This criterion is met, in this corpus, above 90 per cent of the time. What would be the significance of such a double positive avoidance for ergativity, and specifically for the absolutive category? To approach from another direction: why is the unique zero in the paradigm used to index precisely the S and O roles of the third person? If the double positive is to be avoided, it should be clear that this puts a premium of sorts on the assignment of grammatical value to the unique zero cross-referencing morpheme. The zero can contribute significantly to the avoidance of double positives, if it is judiciously 'assigned' to cross-reference the syntactic roles which will in any case already have one positive mark in the clause, in the form of a nominal that is, if assigned to cross-reference the category of full noun phrases. As we have already seen, this category is the conjunction of S and O: the absolutive. In evaluating the contribution of optimal zero valuation to the avoidance of double positives, it should be made clear that the attested frequency of double positives is not simply the result of a low frequency of bound position positive marks taken in conjunction with a low frequency of free position positive marks. If we examine all argument mentions in the corpus (N = 646), we find that the bound position of 37.0% ( N = 239) contains a positive mark, while the free position of 38.2% (N = 247) contains a positive mark. If it were the case that the occurrence of a positive mark in bound position and the occurrence of a positive mark in free position were independent of each other, we might expect that double positives would occur with a frequency equal to the product of the individual frequencies for positive marks in the two positions. By this account the expected frequency of double positives would be 0.370 × 0.382 = 0.141. But the actual frequency of 0.076 (figure 3) is just over half of this, suggesting that there is indeed some factor that effectively reduces the incidence of double positives (see appendix). According to my hypothesis, this factor is not an active 'rule' - the first solution that comes to mind for the modem rule-oriented grammarian - but a static (synchronically speaking) fact of grammatical value: the absolutive valuation of unique zero.
J. Du Bois / Absolutive zero
217
One fact of Sacapultee grammar that is clarified in light of the above discourse pattern concerns the unmarked value for plurality of the absolutive zero, which I have commented on above. When a third person lexical argument which is semantically plural appears overtly in an absolutive role, it need not be cross-referenced on the verb with the plural affix (which could be the positive morpheme e:-). Instead, it can be cross-referenced with the absolutive zero - which is sometimes mis-glossed as third person 'singular' absolutive, but is herewith shown to be unmarked for the positive feature + plural: (ll) L g-0-a-r =-...-6'up-o? nik'yax pe:ra-s. Cmp-Abs3-Mvt-Erg3sg-...-pick-Mvt some pear-Pl '.*. he went and ... picked some pears.' Here the plural O-role argument nik'yax pe:ra-s 'some pears' is cross-referenced with the absolutive zero rather than the marked (and positive) plural e:-. The obvious consequence of this is one less instance of the double positive. The use of absolutive zero to cross-reference plural referents is more frequent with non-humans than humans. Of course non-humans, being typically less thematically (topically) continuous, more frequently require overt lexical mentions (Du Bois (1985a)), and so would more frequently run the risk of double positives, were the absolutive zero not unmarked. (It is of interest that the incidence of double positives in this corpus would be cut still further - nearly in h a l f - if the e:- morpheme were always replaced by zero. Significantly, several of the Mayan languages have in effect done this, by broadening the value of zero to cover third person absolutive in general, without regard to number. Of course, this comes at the expense of lessening the amount of information carried in the cross-referencing affixes.) 12 One can envision a rough, preliminary reference parsing heuristic for Sacapultec speakers whereby each appearance of a positive mark in a clause is provisionally hypothesized to correspond to a distinct referent. Of course this could only be a preliminary strategy, since ultimately the strict grammatical 12 For a variety of reasons, doublepositivesin this corpus werealmost evenlydivided betweenS and A role arguments, with only two double positives in 0 role arguments (both by the same speaker).
218
J. Du Bois / Absolutive zero
values of the cross-referencing morphology would have to be attended to in order to verify the correct parsing; and the initial parsing hypothesis would on occasion be reversed in the final verification. While such a rough-and-ready strategy could be accounted an adaptive motivation for the double positive avoidance constraint, the psycholinguistic methodology for testing whether it is in fact operative for Sacapultec speakers lies outside the scope of this study. I would not try to claim at this stage that the double positive avoidance constraint represents a necessary, or a universal, tendency in discourse. In languages without full-fledged verbal cross-referencing (e.g. English) it is no great feat to avoid double positives, since there is little in the inflectional system to give rise to them. Even in languages which do have cross-referencing, but for only one rather than both of the arguments in a transitive clause, the ecology of the cross-referencing paradigm in relation to discourse will be significantly different. In a language which marks grammatical relations on nouns (case-marking), the presence or absence of noun phrase arguments will obviously have consequences for referent marking quite different from those detailed here. Thus, it will be necessary to investigate carefully both the discourse patterns and their specific grammatical correlates in languages of each grammatical type. Nevertheless, we may present as a hypothesis for further investigation that double positive avoidance constitutes a type-specific motivation operating within languages having the relevant typological features (e.g. at least the Mayan languages), with the power to shape the grammars of these languages. It should be pointed out that the corpus reported on here consists of third person narratives, and does not contain much in the way of first and second person reference. From some limited conversational data outside this corpus, however, it appears that the free pronouns (exemplified in the elicited sentence (6)) are usually absent except in contrastive contexts. Thus in noncontrastive contexts most reference to first and second persons - - w h e t h e r a b s o l u t i v e o r e r g a t i v e ~ is carried out with just the cross-referencing affixes. This is of course possible because the speaker and the hearer are always given information (Chafe (1976:31-32)). What this means is that first and second person do n o t have the lopsided distribution (absolutive!) that overt free mentions do in third person - - since they don't have many free mentions at all, and there is no particular reason for those few that do occur to show a preference for S and 0 over A. It is the skewed distribution within the argument positions which in the third person motivates the absolutive grammatical alignment, in conjunction with the absolutive zero. While up to this point I have emphasized the importance of assigning to the unique zero the grammatical relational
J. Du Bois / Absolutive zero
219
value of {S, O}, it should now be clear that this importance attaches specifically to the third person, but not necessarily to the first or second, which are neutral in this respect. In fact, the discourse pressures motivating ergativity would appear to fade considerably outside the domain of third person lexical reference. Given the weakening of the absolutive zero and Preferred Argument Structure motivations outside the third person, the competing force of topic continuity accommodation stands unopposed. Topic continuity in Sacapultec is no different from topic continuity in an accusative language like English, in respect of running in the grooves of the {S, A} category (despite what has been said about the {S, O} category (or the patient) being the 'topic' for speakers of ergative languages (cf. Mallinson and Blake (1982: 109ft.), among others), a claim which appropriate discourse analysis shows to be completely untrue for Sacapultec (Du Bois (forthcoming, a)). This elucidates the vulnerability of the first and second person to intrusion of accusative patterning, and hence to split ergativity (cf. Silverstein (1976)). In languages which do not exhibit this kind of ergativity split (including Sacapultec), the presence of the ergative patter n in first and second person must presumably be ascribed to generalizing the pattern of the dominant third person throughout the rest of the paradigm (Bybee (1985: 51ft.)). (A full-scale treatment of first and second persons in discourse in relation to split ergativity is called for by the above hypothesis, but space limitations ensure that this will have to wait for another occasion.)
4. Conclusions
In this paper I have tried to show how a pattern in discourse - in the aggregate of tokens - can have profound effects on the grammaticization of types, shaping the very structure of the system of grammatical relations. Given the {S, O} pattern of lexical arguments in discourse, the unique zero in the cross-referencing paradigm most effectively contributes to the avoidance of double positives if it is assigned the grammatical relational value of {S, 0}. In speaking to the issue of the adaptivity of grammatical paradigms, I seek to raise the more general issue of the relationship between discourse and grammar. Does discourse, or speaking, just instantiate grammar? My discovery of this connection between discourse and ergativity suggests rather that the patterning of the aggregate of acts of speaking - that is, parole - can have profound effects on grammatical structure as an aspect of langue. While marking has figured in this analysis, it bears pointing out that my use of the notion differs significantly from a more traditional Praguean approach
220
J. Du Bois / Absolutive zero
to markedness. The most effective application of markedness theory to explaining ergative phenomena (especially splits) has been that of Silverstein (1976, 1981). In traditional analysis, markedness is a structural relation between elements in a paradigm, and as such is a phenomenon which exists essentially in the domain of types (though the type theory is flexible enough to recognize that specific instantiated grammatical-semantic configurations can accrue specialized meanings in particular oppositional contexts). While Silverstein (1976) contains several statements about relative likelihood of a referent appearing in various semantic roles, there is less pursuit of systematic generalizations about surface form and information status across the token aggregate, which forms a central feature of the present methodology. The present approach, given its focus on the aggregate of tokens, makes a more literal interpretation of marking, and in attending to actual phonetic substance sticks close to the physical surface to a degree that would no doubt appear potentially problematic in traditional type-oriented theories (which encompasses most grammatical and semantic theories around nowadays). In making this comparison I do not intend to promote one approach at the expense of the other, but rather to suggest some of the differences in theories and in the kinds of issues they are designed to address. As I have indicated elsewhere (Du Bois (1985a: 359ff.)), in the end both approaches are needed, since to understand language we must understand the actual processes of inter-accomodation between the token aggregate and the type system - between discourse and grammar. We should recall that the discourse patterning of lexical arguments and new information described in section 3.1 (and more fully in Du Bois (1987a)) already in themselves constituted a motivation for ergativity. This motivation is apparently prior to, and independent of, such facts about the substantive realization of ergativ¢ and absolutive paradigm members as those described in section 3.2. The fact that the 'right' alignment of zero can mesh optimally with the ergative-absolutive patterning of surface syntax must be seen as an additional factor overlaid on the deeper and more basic motivation for ergativity provided by Preferred Argument Structure. This convergence of motivations is typical in multiple causation (Malkiel (1967, 1983), Du Bois (1985a)). The motivation provided by absolutive zero alignment is nevertheless especially significant, because it involves a grammaticization which becomes embedded in inflectional morphology. If we give up the notion popular in recent decades that morphology is among the most superficial levels of grammar, we may relearn the earlier wisdom which places it at the most fundamental level of language structure (Sapir (1921)). We can then appreciate
221
J. Du Bois / Absolutive zero
that, once the ergative pattern is embedded in the concrete substance of an inflectional paradigm, which moreover intermeshes tightly with a natural discourse pattern, the mutually reinforcing forms of discourse and morphology confer on their shared ergative structure a stability which can resist disruption over time. In this light it is of some interest that, during four millenia of diachronic development within the Mayan family, amidst numerous phonetic and morphological changes that have affected the cross-referencing paradigms, there are two things which have remained constant: ergativity, and absolutive zero.
Appendix Table A I Co-occurrence of positive and zero morpheme types in bound and free positions (N = 646; ~z = 51, dr= 1, p < 0.001). Bound + Free +
49
198
@
190
209
References Bybee, J. U, 1985. Morphology: A study of the relation between meaning and form. Amsterdam: Benjamins. Chafe, W.L., 1976. Givenness, contrastiveness, definiteness, subjects, topics, and point of view. In: C.N. Li (ed.), Subject and topic, 25-55. New York: Academic Press. Chafe, W.L. (ed.), 1980. The pear stories: Cognitive, cultural, and linguistic aspects of narrative production. Norwood, N J: Ablex. Dayley, J.P., 1981. Voice and ergativity in Mayan languages. Journal of Mayan Linguistics 2, 382. Dixon, R.M.W., 1979. Ergativity. Language 55, 59-138. Du Bois, J., 1980. The search for a cultural niche: Showing the pear film in a Mayan community. In: W.M. Chafe (ed.), I-7. Du Bois, J., 1981a. Ergativity and preferred argument structure in Sacapultec discourse. (MS, University of California, Santa Barbara.) Du Bois, J., 1981b. The Sacapultec language. Berkeley, CA: University of California. (Unpublished Ph.D. dissertation,)
222
J. Du Bois / Absolutive zero
Du Bois, J., 1985a. Competing motivations. In: J. Halman (ed.), lconicity in syntax, 343-365. (Typological studies in language, Vol. 6.) Amsterdam: Benjamins. Du Bois, J., 1985b. Mayan morpheme structure and the final vowel question in Quichean. Journal. of Mayan Linguistics 5, 1-71. Du Bois, J., forthcoming, a. The discourse basis of ergativity. To appear in Language. Du Bois, J., forthcoming, b. The pear story in Sacapultec: A case study of information flow in narrative. In: L. Furbee (ed.), Mayan voices: Mayan text and discourse. Mayan Texts 4, Native American Texts Series. Berlin: Mouton. England, N.C., 1983. Ergativity in Mamean (Mayan) languages. International Journal of American Linguistics 49, 1-19. Jakobson, R., 1939. Sigue z~ro. In: M~langes de linguistique, offerts a Charles Bally, 143-152. Geneva: Georg. (Reprinted in: E.P. Hamp, F.W. Householder and R. Austerlitz (eds.), Readings in linguistics II, 109-115 (1966). Chicago, IL: University of Chicago Press.) Jelinek, E., 1984. Empty categories, case, and configurationality. Natural Language and Linguistic Theory 2, 39-76. Larsen, T.W., 1981. Functional correlates of ergetivity in Aguacatec. Berkeley Linguistics Society 7, 136-153. Malkiel, Y., 1967. Multiple versus simple causation in linguistic change. In: To honor Roman Jakobson: Essays on the occasion of his seventieth birthday, Vol. 2, 1228-1246. The Hague: Mouton. Malkiel, Y., 1983. On hierarchizing the components of multiple causation. In: From particular to general linguistics: Selected essays 1965-1978, 297-319. Amsterdam: Benjamins. Mallinson, G. and B.J. Blake, 1982. Language typology: Cross-linguistic studies in syntax. Amsterdam: North-Holland. Norman, W.M. and L. Campbell, 1978. Toward a Proto-Mayan syntax: A comparative perspective on grammar. In: N.C. England (ed.), Papers in Mayan Linguistics, 136-156. Miscellaneous Publications in Anthropology no. 6. Columbia, Missouri: Museum of Anthropology, University of Missouri. Sapir, E., 1921. Language. New York: Harcourt, Brace and World. Silverstein, M., 1976. Hierarchy of features and ergativity. In: R.M.W. Dixon (ed.), Grammatical categories in Australian languages, 112-171. New York: Humanities Press. Silverstein, M., 1981. Case-marking and the nature of language. Australian Journal of Linguistics 1,227-244. Smith-Stark, T.C., 1976. Some hypotheses on syntactic and morphological aspects of ProtoMayan (*PM). In: M. McClaran (ed.), Mayan linguistics, Vol. I, Los Angeles, CA: University of California, American Indian Studies Center. Smith-Stark, T.C., 1978. The Mayan antipassive: Some facts and fictions. In: N.C. England ted.), Papers in Mayan linguistics, 169-187. Miscellaneous Publications in Anthropology no. 6. Columbia, Missouri: Museum of Anthropology, University' of Missouri.