Object asymmetries in comparative Bantu syntax

ing away from the effects of the relative animacy of passivized subject and object markers, we see that object markers can generally cooccur with passives. This is not true in Chiche wia: the object marker never appears on a passive verb (Sam Mchombo (personal communication)). Therefore, we summarize the essential typological difference between Kichaga and Chichew'a as in table (23): (23) Kichaga Chicheiwa OM-vpas / * 1.5.2. Unspecified Object Deletion with Passives. In Kichaga, unspecified object deletion of one object can cooccur with passivization of another object: (24) a. M-ka n-a-4'-lyi-i-o. NPben Vpas (NPpt) 1-wife FOC1 S-PR-eat-AP-PAS 'The wife is being eaten for/on.' 0 b. M'a-w'oko -ya-4-1y1-f-o'. NPins Vpas (NPpt) 6-hand 6 s-PR-eat-AP-PAs X 'Hands are being eaten with.' 0 In contrast, in Chiche 'wa, passivization of one object makes unspecified object deletion of another object impossible, even though it is possible with the active form of the verb (Alsina and Mchombo (1989, ex. 40b)). The contrast is schematized in table (25): (25) Kichaga Chichewx a NP Vpas (NPpt) / 1.5.3. Unspecified Object Deletion with Object Markers. In Kichaga unspecified object deletion of one argument can cooccur with object marking of another: (26) a. N-a-i-mh-lyi-f-a. oMb,,-Vst,,, (NPpt) FOC1 S-PR1 o-eat-AP-FV X 'He/She is eating for/on him/her.' 0 b. Na-'iya'-Iy if-i-a'. omi,sVst,em (NPpt) FOC1 S-PR-6 o-eat-AP-FV 'He/She is eating with them.' 0 In Chiche wxa, in contrast, it cannot. A verb that otherwise allows unspecified object deletion disallows it when another argument is object-marked (Alsina and Mchombo (1989, ex. 40a)). Table (27) schematizes this contrast: (27) Kichaga Chichew'a OM-Vstem (NPpt) / * 4> 0 156 JOAN BRESNAN AND LIOBA MOSHI 1.5.4. Cooccurrence of Reciprocals with Passives. Reciprocal verbs can be passivized in Kichaga: (28) Sh'f-mi'i' sh--i'-k6-r-an-o (nai) wa-ch'aka. Vrcp-pas 8-firebrand 8 s-PR-burn-AP-RcP-PAs (by) 2-Chaga 'Firebrands are being used by the Chagas to burn each other.' This is impossible in Chichewx a: 13 (29) *Mi-kondo i-na-meny-er-an-idw-a' ndi ai-lenje. *Vrcp-pas 4-spears 4 SB-REC PST-hit-APPL-RECIP-PASS-IND by 2-hunter 'Spears were used by the hunters to hit each other.' (30) Kichaga Chichewxa Vrcp-pas I * 1.5.5. Cooccurrence of Reciprocals with Object Markers. In Kichaga reciprocalization and object marking of two different arguments of the same verb are possible (compare (15b)): (31) W?a-chaka wa-i-shi-kor-i-an-a'. OM-Vrcp 2-Chaga 2 S-PR-8 o-burn-AP-RCP-FV 'The Chagas are burning each other with them (i.e., firebrands).' In Chiche wxa, in contrast, an object cannot be eliminated by reciprocalization in the presence of an object marker (Sam Mchombo (personal communication); compare (14b)): (32) *A-lenje a-na-i-meny-er-an-a. *OM-Vr(P 2-hunter 2 SB-REC PST-4 OB-hit-APPL-RECIP-IND 'The hunters hit each other with them (i.e., spears).' (33) Kichaga Chichewx a OM-Vrcp I * 1.5.6. Cooccurrence of Reciprocals with Unspecified Object Deletion. Finally, in Kichaga unspecified object deletion can cooccur with a reciprocal verb: (34) W-and wa--Y-kor-i-a'n-a'. NP Vrcp (NPpt) 2-child 2 s-PR-cook-AP-RcP-Fv X 'The children are cooking for each other.' In Chichewx a it cannot (Sam Mchombo (personal communication)): (35) *Ana a-ku-phfk-ir-an-a. *NP Vrcp (NPpt) 2-child 2 s-PREs-cook-APPL-REcIP-IND X 'The children are cooking for each other.' 0 "1 Our Chichewa example (29) is based on Alsina (1988). OBJECT ASYMMETRIES IN COMPARATIVE BANTU SYNTAX 157 (36) Kichaga Chichew'a NP Vrcp (NPpt) I * We have just examined all six pairwise combinations of object marking, passive, reciprocalization, and unspecified object deletion in Kichaga and Chiche w'a. In every case Kichaga allows the combination and Chiche wia prohibits it. This evidence indicates that the covariation of object properties in symmetrical and asymmetrical Bantu languages is pervasive and systematic. 2. Previous Theories The problem of explaining the variation between symmetrical and asymmetrical object properties has been tackled repeatedly by different theorists over the past twelve years. However, no proposed solution has yet succeeded: the proposals that do succeed in reducing the differences between the symmetrical and asymmetrical object types to a single parameter of variation and thus explaining why they should covary have been shown to have descriptive inadequacies, whereas the proposals that solve the descriptive problems postulate multiple independent differences in the grammars of the two types of languages, thus failing to explain the covariation. We support this claim with a review of five different approaches to the problem. 2.1. Gary and Keenan (1977) In their classic study of Kinyarwanda, Gary and Keenan (1977) propose a solution to this problem of object variation, based on the idea that the universal relational hierarchy-Subject > Object > Indirect Object > Oblique > . . . -is collapsed in some languages by omitting a separate category of Indirect Object and allowing multiple instances of Object. Given universal syntactic properties of Object, this proposal successfully explains the covariation of differences we have seen between Kichaga and Chiche wia. However, Perlmutter and Postal (1983), Dryer (1983), and De Guzman (1987) have all pointed out problems with this proposal, the most telling being that even in symmetrical object languages some syntactic processes distinguish objects from indirect objects. We can see this in Kichaga and Chichew'a as well. First, the basic word order constraint on transitive sentences in Chichewx a is that the object is adjacent to the verb (Bresnan and Mchombo (1987), Bresnan and Kanerva (1989)). Since Chiche w'a is a head-initial language, the object then immediately follows the verb in the verb phrase. However, the word order constraints appear more complicated in ditransitive applicative constructions: the applied NP must be adjacent to the verb if it is a beneficiary or recipient, but the patient NP may be adjacent to the verb if the applied NP has any of the other thematic roles (Baker (1988b, 370), Alsina and Mchombo (1989, exs. 3, 4, 44)). In Gary and Keenan's (1977) terms, the indirect object, 158 JOAN BRESNAN AND LIOBA MOSHI if there is one, is adjacent to the verb; otherwise, the object is adjacent to the verb. Such a distinction between direct and indirect object should disappear in a symmetrical object language like Kichaga, if Gary and Keenan's proposal is correct, but in fact Kichaga shows the same word order generalization as Chiche w'a. The patient can be adjacent to the verb if the applied argument has any thematic role other than the "indirect object" roles of beneficiary (maleficiary) or recipient (compare (2), (3)): (37) a. *N-a-i-lyi-i-a k-elya m-ka. *V NPp, NPben FOC-i S-PR-eat-AP-Fv 7-food 1-wife 'He is eating food for/on his wife.' b. N-i-'-lyi-i-a k-ely'a mi-ww!oko. V NPpt NPins FOC1 s-PR-eat-AP-Fv 7-food 6-hand 'He/She is eating food with his/her hands.' c. N-i-i-lyi-i-a k-elya m-ri-nyi. V NPpt NPI,0 FOC1 S-PR-eat-AP-Fv 7-food 3-homestead-LOC 'He/She is eating food at the homestead.' d. N-d-i-lyi-i-at k-elya' nja''a. V NPpt NPmot FOC1 S-PR-eat-AP-Fv 7-food 9 hunger 'He/She is eating the food because of hunger.' Animacy is not significant in determining these word order patterns, although it is important in other Bantu languages, such as Sesotho (Morolong and Hyman (1977), Hyman and Duranti (1982), Machobane (1987)) and elsewhere in Kichaga as we have shown above. For example, the Kichaga sentence Mshaw1 na-i-lyi-[-'4a mika mana 'The witch is eating the child for the woman' requires the beneficiary to be adjacent to the verb; reversing the order of 'woman' and 'child' simply changes the interpretation so that the child becomes the beneficiary of the witch's eating the woman. Similarly, in N-a-l!e-reng-i-a ki-tima ku'-zrende 'She/He carved a leg for the chair', the inanimate beneficiary 'chair' must be adjacent to the verb; again reversing the order of postverbal objects simply changes the interpretation to 'She/He carved a chair for the leg'."4 Word order patterns with applied instrumentals, locatives, and motives remain unchanged when the patient is animate: reversing the order of the animate patient NP and the applied NP is both grammatical and meaning-preserving. A second piece of evidence that indirect objects must be distinguished from objects in symmetrical object languages comes from extractions. In Chichewx a there is a restriction against long-distance extractions of beneficiary and recipient objects, which does not apply to other objects, such as patients or applied instrumentals and locatives (Baker (1988a, 289-302; 1988b, 355-356, 374-376), Alsina and Mchombo (1989, exs. 23, 24, 49)). In other words, there is a constraint against long-distance extraction of indirect objects. Once again, such a distinction between direct and indirect objects should disappear in Kichaga, if these grammatical relations are collapsed; yet Kichaga has exactly '4 We are grateful to Carolyn Harford for suggesting this crucial point. OBJECT ASYMMETRIES IN COMPARATIVE BANTU SYNTAX 159 the same restriction: (38) a. *M-ka a-i-lyi-i-a k-elya nyi-ichu. 1 -wife 1 s REL-PR-eat-AP-Fv 7-food copI this 'The wife for whom he is eating the food is this one.' b. K-e'lya' ai-i-lyi-i-a 'm-k'a k'i'-pas'u. 7-food 1 s REL-PR-eat-AP-Fv 1-wife 7-rotten 'The food which he is eating for the wife is rotten.' c. Ki-shui a-i-freng-if-a ki-tima ki-6h'i. 7-knife 1 s REL-PR-carve-AP-Fv 7-chair 7-sharp 'The knife with which he is carving the chair is sharp.' d. M-ri-nyil a-i-lyi-i-a k-'elya ch!i ko'-ky'e pfo. 3-homestead-Loc 1 s REL-PR-eat-AP-Fv 7-food NEG FOC 17-his NEG 'The homestead at which he is eating the food is not his.' In sum, Gary and Keenan's (1977) proposal is attractive for its simplicity in explaining object symmetries, but it fails to account for the asymmetries between direct and indirect ob