1. Two Agreement Parameters In Baker (In press), I propose that the syntax of agreement can vary across languages in the following two respects:
The Direction of Agreement Parameter (DAP):
F agrees with DP/NP only if DP/NP asymmetrically c-commands F (Yes or No).
The Case Dependence of Agreement Parameter (CDAP)
F agrees with DP/NP only if F values the Case feature of DP/NP or vice versa (Yes or
In general, a functional head can search upward or downward through the syntactic tree in order to find a DP/NP that it can agree with (I claim, see Baker In press: ch. 2, 3 for much discussion). The DAP says that this freedom is restricted in some languages, such that functional heads only search upward. Most Niger-Congo (NC) languages have the positive setting of the DAP, whereas most Indo-European (IE) languages have the negative setting. The CDAP concerns not the configurational aspect of agreement, but its relationship to case features. It states that in some languages agreement between a functional head and an NP/DP depends on the two entering into a Case-valuation relationship, whereas in other languages it does not. Most IE languages have the positive setting of this parameter, whereas most NC languages have the negative setting. I go on to argue that these are parameters in the classical sense of Chomsky (1981): they are variations in the syntactic principles that define a language, not reducible to the featural specifications of individual lexical items or classes of lexical items. Thus, in general all of the functional heads in a given language will, if they agree at all, show the same syntactic behavior with respect to (1) and (2).
The DAP and the CDAP were originally motivated by a close comparison of the NC languages and the IE languages. As such, they may seem far removed from the normal concerns of Austronesianists, whose languages are generally not as rich in agreement as the Bantu languages are. But if it is right to think of the differences between NC languages and IE languages in terms of parameters embedded within a theory of Universal Grammar, then those parameters should apply to Austronesian languages just as well as they do to the languages that first led to their discovery. With this in mind, Baker (In press) went on to the validity of the parameters in (1) and (2) on a sample of 108 languages from around the world, consisting of the core sample defined in the World Atlas of Language Structures (WALS) (Haspelmath et al. 2005), plus a few others. The purpose of this paper is to discuss how these parameters apply to the Austronesian languages in that sample, going into a bit more detail on this than was possible in Baker (In press). I also extend the discussion by discussing another interesting pattern of agreement found in some languages of the Philippines, brought to my attention by Paul Kroeger and Mark Donohue.
The WALS core languages sample includes eight Austronesian languages: Chamorro, Tukang Besi, Fijian, Paiwan, Indonesian, Malagasy, Rapanui, and Tagalog. Of these, the last five do not have much for agreement phenomena (at least according to WALS itself), and Chamorro agreement has already been thoroughly studied from a generative perspective by Chung (1998). But Fijian and Tukang Besi prove very interesting. In section 2, I show how the less familiar settings of the DAP and the CDAP that are used in Kinande and other Bantu languages are active in Fijian as well. This counts as an important replication of one of the more novel parts of my overall approach to agreement. Then I go on to consider Tukang Besi (section 3), which shows some of the same behavior as Fijian, but unlike Fijian it has overt case marking on its noun phrases. The interaction of case and agreement is thus much richer in Tukang Besi. Some of this new data supports the CDAP, but some of it also looks problematic—until one integrates in a view of how case is assigned in Tukang Besi. Finally, I consider Kapampangan, which is like Tukang Besi in some respects, but is maddeningly different in others (section 4). I claim that the difference between the two is not only in the agreement parameters, but also involves differences in the movement processes that create the structures over which agreement is defined. Together, these three case studies serve as an illustration of how a small number of agreement parameters can interact with other aspects of grammar to characterize (much of) the rather rich diversity of agreement systems that we observe in languages of the world. I hope that they also illustrate the value of there being meaningful interaction between general theorists/typologists and those who are familiar with the intricate details of a particular language family.
2. Replicating the Bantu Parameter Settings in Austronesian: Fijian The first Austronesian language that I discuss will be Fijian. In that it has no overt case marking and no complex Philippines-style voice system, Fijian is rather like Kinande and the other Bantu languages that originally motivated the parameters in (1) and (2). Hence, it provides an opportunity to see how these parameters are supposed to work.
2.1 The Direction of Agreement Parameter Consider first the DAP, stated in (1). Perhaps the most obvious effect of this parameter being set positively in the Bantu language Kinande is that finite T always agrees with the phrase that has moved to Spec, TP, a position from which it c-commands T. In the most common sentences in this SVO language, the moved NP is the agentive subject, which results in normal subject agreement (see, for example (11a) and (18a) below). But Kinande also allows locative expressions to move to Spec, TP, in locative inversion sentences like (3a). When this happens, T agrees with the locative expression. Furthermore, Kinande allows the object to move to Spec, TP when there is contrastive focus on the subject (subject to other, ill-understood conditions). When this happens, T agrees with the fronted object ((3b)). Or it is possible for no NP to move to Spec, TP, as in (3c). Then T does not agree with any NP in the clause, but the agreement slot is taken by a special expletive element ha.
a. Oko-mesakw-a-hir-aw-a ehilanga. (Locative Inversion)
b. Olukwi si-lu-li-seny-a bakali (omo-mbasa). (Object Fronting)
wood.11 NEG-11S-PRES-chop-FV women.2 LOC.18-axe.9
‘WOMEN do not chop wood (with an axe).’
c. Mo-ha-teta-sat-a mukali (omo-soko). (Expletive subject)
AFF-there-NEG/PAST-dance-FV woman.1 LOC.18-market
‘No woman danced (in the market).
In none of these sentence types can T agree with anything that has not moved to Spec, TP. Kinande thus differs markedly from English and other IE languages, in which T can agree with the nominative subject even when Spec, TP is occupied by something else. (4a) shows this for locative inversion in English, (4b) shows it for object fronting in Yiddish, and (4c) shows it for expletive constructions in English.
…az vayn kenmen makhn fun troybn oykh. (Object fronting, Yiddish)
that wine can one make from grapes also (Diesing 1990:44)
‘(You should know)…that one can make wine from grapes also.’
c. There are/*is some peanuts on the table. (Expletive subject)
These differences follow if one says that DP must c-command T for T to agree with DP in Kinande but not in the IE languages, a special case of the DAP.
Consider now Fijian. Finite verbs generally agree with their subjects in Fijian—except in existential constructions. Thus, there is third person plural agreement on the intransitive verb in (5b), but not on the existential verb in (5a).
a. e sō na vūlagi (Schütz 1985:329)
3sS be.some DEF villagers
‘There were some villagers.’
b. era yaco māī e sō na vūlagi
3pS arrive DIR 3s some DEF villagers
‘Some villagers arrived.’
This contrast follows from the DAP being set positively in Fijian, plus the assumption that the Spec, TP position can be filled by a null expletive (comparable to English there) in existential sentences only. Hence, the phrase ‘some villagers’ must raise to Spec, TP and trigger agreement on the c-commanded head T in (5b), but not in (5a). The same contrast is seen in Chamorro, and is analyzed by Chung (1998: 68-69, 182-183). Inasmuch as there is no agreement in (5a), Fijian is more like Kinande ((3c)) than like English ((4c)) in this respect.
There is also evidence that v can agree with the object in Bantu languages only if the object c-commands v. In some of these languages, including Kinande, object agreement appears on the verb only if the object is dislocated to the edge of the clause:
a. N-a-(*ri)-gul-a eritunda. (Baker 2003)
‘I bought a fruit.’
b. Eritunda, n-a-ri-gul-a.
‘The fruit, I bought it.’
In other Bantu languages, like Zulu and Swahili, the evidence is a bit more subtle. There is no difference in word order between (7a), which does not have object agreement on the verb, and (7b), which does have object agreement.
a. Ngi-leth-el-a umfundisi incwadi (Doke 1963:299)
1sS-bring-APPL-FV teacher.1 book
‘I am bringing a teacher a book.’
b. Ngi-ya-m-lethela umfundisi incwadi
1sS-DISJ-OM1-bring-APPL-FV teacher.1 book
I am bringing the teacher—the one who told me to do so—a book.
There is a difference in interpretation however: the agreed-with object in (7b) must be understood as being definite (or some related notion), whereas the unagreed with object in (7a) is not. Following the leading idea of Diesing 1992, I take the strong/definite reading in (7b) to be a sign that the object has moved out of VP, the domain of existential closure, and into Spec, vP in (7b) but not in (7a). This “object shift” has no effect on surface word order in Zulu, because the verb independently moves past v into T:
a. [TP Tense+bring [vP (*AGRi+)v [ [VP teacheri book ]]] (=(7a))
b. [TP Tense+bring [vP teacheri AGRi+v [ [VP ti book ]]] (=(7b))
The object shift does affect agreement, however. In (8b), the shifted object c-commands v and v can agree with it, in accordance with the positive setting of the DAP. In contrast, v cannot agree downward with the unshifted object in (7a)/(8a), just as T cannot agree downward with the unmoved subject in (3). Again, there is no similar requirement that agreement be upward in IE languages. Ormazabal and Romero (2006:18) argue that les in a sentence like (9) from a leista dialect of Spanish counts as an object agreement. If so, then it is agreement with an undislocated and indefinite NP that remains inside VP—just the kind of agreement that is not allowed in (6a) and (7a) in the Bantu languages.
(Yo) les-llevé a unos jóvenes al pueblo.
I 3pO-carry ACC a.PL youngsters to.the town
‘I gave some young people a ride to the town.’
The fact that v can agree downward into VP in IE but not in NC is parallel to the fact that T can agree downward into vP in IE but not Bantu—a testimony to the generality of the DAP.
Consider now Fijian. Dixon (1988) shows that transitive verbs in Fijian normally bear a special “transitivity suffix” that varies depending on whether the object is singular or plural; the singular version is seen in (10b). This is a kind of (relatively impoverished) object agreement, presumably associated with v. However, the agreement-bearing transitivity suffix is absent in (10a), which has an indefinite subject, in what is traditionally analyzed as a noun incorporation construction.
a. [E’au.i vola mai] a cauravou. (Dixon 1988:49)
deliver letter to.here the youth
‘The youth is delivering letters.’
b. [E’au-ta -- mai] a-i-vola yai a cauravou.
deliver-TR.3sO to.here the-letter this the youth
‘The youth is delivering the letter.’
I assume that Massam’s (2001) analysis of pseudo-noun incorporation in Niuean also applies to this alternation in Fijian. According to this view, the bracketed constituent in (10a) is not a word-like unit, but a full VP, its right edge marked by the adverbial particle mai. An indefinite object stays inside VP, just as in Zulu, whereas a definite object-shifts out of VP, moving to a higher position. This accounts for the morpheme order difference between (10a) and (10b), as well as the difference in the interpretation of the object, given Diesing’s Mapping Hypothesis.1 Taken together with the idea that the DAP is set positively in Fijian, it also explains the fact that the agreement-bearing transitivity marker is present in (10b) but not in (10a), since the object moves to a position that c-commands v only in (10b). The contrast between (10a) and (10b) in Fijian is thus very similar to the contrast between (7a) and (7b) in Zulu, whereas Fijian is different from Spanish in this respect.
Next consider agreement on prepositions. Kinyalolo (1991:111) shows that in Kilega P does not agree with its object when the object remains in situ. The object of the P can however undergo focus movement, or can move to the subject position in a passive. When it does so, then the P does agree with the moved object. (11) replicates this observation in Kinande.
a. Kambale a-ka-kanay-a na-(*bo) abasyakulu
Kambale 3sS-PRES-speak-FV with-2 2.old.people
‘Kambale is speaking with the old people.’
b. Abasyakulu si-ba-li-kan-ibaw-a na-bo.
2.old.people NEG-2S-PRES-speak-PASS-FV with-2
‘Old people are not spoken with.’
This contrast too can be attributed to the DAP being set positively in Kinande (and Kilega). The unmoved object does not asymmetrically c-command P (rather there is mutual c-command), so P cannot agree with the unmoved object. From its final position in a sentence like (11b) (or more likely from an intermediate position along the way), however, NP does asymmetrically c-command P and agreement is possible in this structure. In contrast, Ps can agree with their complements even when the complement has not been moved in Welsh, showing that the DAP is not set positively in this IE language.
about-1s me about-2s you about-3sF her about-3p them
Again, Fijian is like a Bantu language and not like an IE language in this respect. Dixon (1988:42, 248) shows that there is agreement on P in Fijian only if its NP complement is extracted (by topicalization, in this case).
a. ‘Eimami saa qaaqaa a ‘ai-Boumaa [i-na drano].
we ASP victorious ART native-place about-ART lake
‘We, the natives of Boumaa, were victorious concerning the lake.’
b. A drano ‘eimami saa qaaqaa [‘i-na --] a ‘ai-Boumaa.
ART lake we ASP victorious about+3.SGART native-place
‘The lake, we the Boumaa people were victorious concerning it.’
Agreement does not appear on as wide a range of functional heads in Fijian as in Kinande. In Kinande, agreement also appears on certain quantifiers, on complementizers, on the focus particle, and on a special particle found inside the verb phrase. For all of these heads, it can be shown that agreement depends on the agreed with NP asymmetrically c-commanding the agreeing head. In Fijian, however, quantifiers and complementizers do not undergo agreement at all, so the setting of the DAP is moot for those categories.2 2.2 The Case Dependence of Agreement Parameter Next consider the kinds of data that motivate the CDAP in (2), and how it applies in Fijian.
In most IE languages, it is transparently true that T agrees with a phrase X only if T assigns NP nominative case to that phrase. For example, the finite verb agrees with a nominative subject in Icelandic, but not with a dative subject; if anything, the finite verb agrees with the nominative object in a dative subject construction in this language:
Henni leiddust Þeir. (Icelandic)
‘She was bored with them.’
Similarly, in Hindi the finite verb agrees with its nominative subject in imperfective clauses like (15a), but not with a subject marked with ergative case in perfective clauses like (15b).
‘Nina lifted up the boy.’
There is arguably no such requirement in Bantu languages. Claims about case assignment in the Bantu languages are necessarily somewhat abstract, since there is no morphological case marking in these languages. But recall that in locative inversion and object fronting structures, T agrees with the fronted XP in the Bantu languages, as shown again in (16).
a. Oko-mesakw-a-hir-aw-a ehilanga. (Locative Inversion)
LOC.17-table 17S-T-put-PASS-FV peanuts.19
‘On the table were put peanuts.’
b. Olukwi si-lu-li-seny-a bakali (omo-mbasa). (Object Fronting)
wood.11 NEG-11S-PRES-chop-FV women.2 LOC.18-axe.9
‘WOMEN do not chop wood (with an axe).’
Why is this kind of agreement possible in the Bantu languages, but not in the IE languages, from the point of view of case theory? The IE side of the question is straightforward: T cannot agree with the fronted phrase because it does not assign it nominative case; rather the object has accusative case in (4b) from Yiddish, and the PP has no case at all in (4a) from English. Now how does case work in Bantu? There are two plausible options. The first is that case in Bantu works the same way that it does in IE languages: the postverbal NPs in (16a) and (16b) have nominative case (what other case could they have?), the preverbal NP in (16b) has accusative case (compare (4b)), and the locative expression in (16a) has no case at all. Then it is clear that T agreeing with a phrase does not depend on T giving that phrase nominative case in Kinande. The other plausible option is that there simply is no morphological case assignment in Bantu languages. Then a fortiori the phrases in Spec, TP do not get case in (16a,b), and T does not assign them case. Nevertheless, T agrees with them. This assumption too points to the conclusion that the CDAP must be set negatively in the Bantu languages. There are other conceivable approaches to this issue—approaches that strive to maintain a tight relationship between Case and agreement even in Bantu—but they all lead to reasonably well-known difficulties (see Ndayiragije 1999 for some discussion).
It is widely assumed that these considerations apply to v as well, that v can only agree with the NP that it assigns accusative case to in an IE language. Certainly that is true for the Spanish example in (9), where the verb agrees with the plural direct object in accusative case, and not with the singular dative expression. However, true object agreement (as opposed to object clitics) is rare enough in IE languages that it is hard to show a compelling range of evidence for this assumption. On the other side of the question, given that (2) is set negatively for Bantu, we might expect that v could agree with an argument in oblique case rather than an accusative object in Bantu. This certainly happens in some other languages (see (27) below), but for Bantu the issue is again moot because there is no accusative/oblique case distinction.
Given that Bantu languages don’t have overt case marking, the clearest consequence of the CDAP being set negatively is the possibility of multiple full agreement in auxiliary constructions. Only a single finite verb can agree with a given argument in IE languages. As a result, complex tenses are built out of a finite auxiliary and an uninflected participle, not out of two agreeing verbs, as shown in (17) from English.3
a. Chris is coming.
b. *Chris is comes.
Why is (17b) ruled out? The ungrammaticality of such examples can be derived from the CDAP. The lower T agrees with the thematic subject, by hypothesis. Given a positive setting of the CDAP, this is only possible if the lower T values the case of the agreed with NP as nominative. Now case assignment is unique; the case feature of an NP can only be valued once. Once the lower T values the case feature of the subject NP, the higher T node cannot do so. The positive setting of the CDAP then implies that the higher T cannot agree with the subject NP. Either the higher T could agree, or the lower one, but not both, given the positive setting of the CDAP and the uniqueness of case assignment. But structures similar to (17b) are perfectly possible in many Bantu languages. (18) gives two examples in which the same kind of agreement with the same NP appears on both the main verb and the auxiliary verb in Kinande.
a. Abakaliba-bya ba-ka-gul-a amatunda.
women.2 2S-were 2S-PCPL-buy-FV fruits.6
‘The women were buying fruits.’
b. Tú-lwé tú-ká-ly-a.
‘We were eating.’
This leads to the conclusion that the CDAP is set negatively in the Bantu languages.
Consider now Fijian. Like Kinande, Fijian does not have overt morphological case. It does not even have inversion constructions comparable to (16), as far as I know. Hence we cannot expect there to be a wide range of evidence as to how the CDAP is set. However, Fijian does have auxiliary-main verb constructions, and in some of these there is agreement on both verbs, much as there is in Kinande:
Era dodunu me+ra la’o (Dixon 1988:280)
3pS must C-3pS go
‘They must go.’ (similarly with bese ‘not want’)
This suggests that Case assignment must not be a requirement for agreement in Fijian, any more than it is in Bantu.4
I conclude, then, that Fijian has the same parameter settings as Kinande does, and the kinds of data that motivate those parameter settings are similar, apart from a few obvious surface differences (like the fact that Fijian has VOS word order and Kinande is SVO). This is an important replication of the parameters in (1) and (2), confirming that they are not just artifacts of comparing NC and IE languages, but have value in the analysis of other languages as well—as one would expect from a proposal concerning parameters that are embedded in a theory of Universal Grammar. The evidence for these parameters is admittedly not as rich for Fijian as it is for Kinande. For example, Fijian does not have the same range of inversion constructions that Kinande has, where different kinds of NP can move to Spec, TP. Fijian also does not have the same rich system of gender that Kinande does, so agreement on v and P is only agreement for number, and hence it is easy to overlook and may be open to other kinds of analysis. Finally, not as many functional heads bear agreement in Fijian as in Kinande, and there is little or no case marking in Fijian. Nevertheless, the evidence for these parameter settings in Fijian is far from trivial, especially for the positive setting of the DAP. The three categories that bear agreement in Fijian (T, v, and P) all behave in a consistent way, in that each kind of agreement is contingent on some kind of movement taking the agreed with NP to a high enough position (NP raising to Spec, TP; object shift out of VP; topicalization out of PP). The various functional heads thus behave in a consistent way in Fijian, just as they do in Kinande. This supports my claim that (1) and (2) are rightly thought of as true parameters in the syntax of agreement, not simply as properties stipulated in the lexical entries of the functional heads, which could differ from one functional head to the next internal to the same language.