Ferid Chekili (Tunis, Tunesia)
The Position of the Postverbal Subject and Agreement Asymmetries in Arabic
The Position of the Postverbal Subject and Agreement Asymmetries in Arabic
Concentrating on the position of the subject NP in the clause structure, most work on Standard Arabic (SA) up to the early 1990's, has assumed, whether explicitly or implicitly, and as a result of the VP-Internal Subject Hypothesis, that the difference between SVO order and VSO order is to do with the position of the subject: in VSO, it is located in [Spec, VP], in SVO, in [Spec, IP].
This has been correlated with the different agreement possibilities inherent to each word order. For example, the difference between (1) illustrating full agreement, i.e. agreement in Number and Gender, and (2) illustrating partial agreement, i.e. agreement in Gender only has been accounted for by generative linguists by exploiting the distinction between the two subject positions:
In (1), the subject in [Spec, IP] correlates with full agreement on the verb; in (2), the subject in [Spec, VP] correlates with partial agreement: i.e. there is an agreement asymmetry.
Agreement asymmetries in SA have been accounted for within different frameworks and using different components of the grammar: within the Chomskyan framework, two different positions, representing different analyses of agreement asymmetries, have generally been proposed in the literature: a syntactic account exploiting the distinction between [Spec, VP] and [Spec, IP], and a morphological account where the subject in both VSO and SVO is assumed to be located in [Spec, IP] and thus, the agreement asymmetries are accounted for in the PF component.
In this squib, I present some data from Tunisian Arabic (TA) which may indicate that the [Spec, VP] position of the subject (in VSO) and the resulting 'lack of' or 'partial' agreement which appears to be intuitively correct, may still be considered as a possible location for postverbal subjects and therefore, a syntactic analysis of agreement asymmetries exploiting the difference between [Spec, VP] and [Spec, IP], may still be a valid one.
1 Syntactic analysis of agreement asymmetries in Arabic
Prior to Minimalism (e.g. Koopman and Sportiche (1991), Benmamoun (1992), Bahloul and Harbert (1993), and others)1 these agreement asymmetries were sometimes analysed by exploiting the distinction between government and spec-head: in SVO (1 a., b.), there is a Spec-Head agreement relation between the subject in [Spec, IP] and the verb in I; in VSO (2 a., b.), on the other hand, a relationship of government between the moved verb and the subject left in situ. In Minimalism, following Chomsky (1992), we find the claim that in VSO, the subject remains in the lexical projection [Spec, VP] whereas in SVO, it moves to [Spec, IP] (or in a more articulated split-I framework, [Spec, AGRSP]) before spell-out. On the assumption that [Spec, VP] is not a checking position, the agreement asymmetries are accounted for2.
1.1 Agreement asymmetries in other languages
Crosslinguistically, there is also an agreement asymmetry between SVO and VSO which is sometimes attributed to the location of the subject: if we assume the VP- Internal Subject Hypothesis to be correct, there will be no agreement or 'partial' agreement if the subject remains within VP but full agreement if it moves from [Spec, VP] to [Spec, IP]: There appears to be, at least, a tendency for partial agreement when the lexical subject is in its canonical position within VP, but full agreement when it is outside VP. This agreement asymmetry is found in the Celtic languages (see, for example, Huybregts (1991), Kaplan (1991), and Taraldsen (1993). For example, in Breton, we find the following contrast:
In (3b) there is partial agreement because the postverbal subject is presumably within VP3.
Similarly, in French (4) examples from Fassi (1993) , the postverbal nonexpletive lexical subject is in [Spec, VP]. Hence, partial agreement is expected:
In certain varieties of English, we find the contrast in (5) examples from Bolotin (1995):
Although (5 b.) is not found in all varieties of English generally, in Standard English, agreement is with the associate it is suggestive: three candidates is the associate of there and therefore, it would be located in [Spec, VP] hence the partial agreement.
I am not claiming here, that every instance of a postverbal subject that is located in [Spec, VP], will trigger partial agreement (nor am I saying see discussion of example (15) that every postverbal subject is located in [Spec, VP]): the optionality of (5 b.) shows that things are more complicated than this. All I am saying is that there seems to be a tendency for this in other languages, but I am primarily concerned with the postverbal subject in SA and how the TA data (see below) seem to confirm the VP-location of this subject.
2 PF analysis of agreement asymmetries in Arabic
The second view after the 1990's holds that in (SA) VSO like SVO the postverbal subject is in [Spec, IP] with the verb occupying a position outside IP e.g. C or F (cf. Aoun et al. (1994)). For example, Benmamoun (1997, 1998), offers an account of the agreement asymmetries in SA employing postsyntactic (PF) merger of the verb and the postverbal subject: this account claims that in the VSO order, the verb and the subject merge postsyntactically in the morphological component.
As a result, "the inherent features on the noun spell out the relational agreement features on the verb. This in turn, makes the number suffix redundant" (Aoun and Benmamoun 1999: 181). In order to introduce their morphological analysis, Benmamoun (1997, 1998), and Aoun and Benmamoun (1999) have to show that a syntactic analysis in terms of the location of the subject either in VP (VSO) or IP (SVO) is incorrect since they take the subject to be in [Spec, IP] in both orders and therefore checking of full agreement is possible in both orders. This led them to conclude that a better analysis was a PF one.
One type of argument against the [Spec, VP] location of the subject, is based on the behaviour of sentential negation in Moroccan Arabic (MA) (see Benmamoun (1997) for a discussion) which, incidentally, many hold to be SVO rather than VSO4. Also, even if the subject is indeed in [Spec, IP] in MA (which might well be the case), it does not necessarily mean that it is also the case in SA given the differences in agreement between the two languages.
Some data from TA which, uncharacteristically, involve partial agreement5, suggest that the postverbal subject in SA (which is associated with partial agreement) may indeed be positioned in [Spec, VP] and that this may be the factor determining agreement asymmetries.
Notice that, as for MA, the subject is not necessarily in the same position in TA and SA , given the differences in agreement between the two languages. However, where TA becomes relevant i.e. may be used to argue for the location of the subject in SA is when the TA subject can be shown to be located in [Spec, VP], triggering partial agreement and suggesting that this position may be associated with partial agreement in SA and possibly in other languages as well.
The [Spec, VP] analysis of the postverbal subject would provide a natural account of structures like (6) where two verbal elements precede the subject.
Assuming the subject to be in [Spec, VP], allows for the two verbal slots to be easily accounted for: One verbal element would be internal to IP, the other external (e.g. C or F (see above)). Whereas if the postverbal subject were located in IP, then it would be hard to account for such structures.
To show that the correlation between the location of the postverbal subject in [Spec, VP] and partial agreement may be plausible, I argue that, when the subject whether preverbal or postverbal is in IP e.g. SVO and VSO in the modern dialects (cf. Benmamoun's analysis of negation in MA), there will be full agreement; when the postverbal subject is in [Spec, VP] in the same dialects, there will be partial agreement. This could argue for the position of the postverbal subject in SA which is associated with partial agreement.
The TA data to be considered involve Expletive-Argument chains where the argument subject is clearly in [Spec, VP]. On analogy with 'there-sentences', we might refer to these as 'famma-sentences'.
3 The famma-construction
Before considering the behaviour of 'famma-sentences', I shall first try and establish the expletive nature of famma, firstly, by comparing it with English there, and secondly by considering its potential thematic possibilities:
Famma, like there, may have a locative meaning as in (7):
On analogy with there which can occur in existential constructions as an expletive element, it is not unreasonable to treat famma in existential sentences as an expletive element forming an Expletive-argument chain with the argument NP.
Like there, famma does not inflect for phi-features, i.e. Person, Number, and Gender:
Like there (cf. Chomsky 1995: 155), famma must have an associate which licenses it. E.g. (9):
In 'famma-sentences' like in 'there-sentences' (cf. Chomsky (1995: 155) there may be agreement with the associate of famma. E.g. (10):
In existential sentences, famma has no intrinsic reference. For example, its reference cannot be questioned:
In view of the similarity between 'there-sentences' and 'famma-sentences', it is not unreasonable to suggest that both the expletive element famma and the argument NP-associate occupy A-positions: [Spec, IP] and [Spec, VP] respectively:
Having established the expletive nature of famma6, consider the following data from TA: TA shows full agreement in both SV(O) and VS(O) orders. Compare, for instance, (13) and (14) where, in both, the subject is presumably located either in [Spec, IP] (cf. Benmamoun's analysis of sentential negation in MA which is closely related to TA) or in an adjoined or dislocated position (Chekili (to appear)):
If the correlation between the location of the postverbal subject in [Spec, VP] and partial agreement is maintained and if the representation in (12) is correct, then we predict that VSO order with the associate in [Spec, VP] as in (12) will show partial agreement, and this is indeed the case as shown by the following contrast between 'famma-sentences' (15) and non-expletive ones (16):
In (15a) the argument subject NP barsha naas in [Spec, VP] is plural but the auxiliary verb kaan singular. (15b) shows that the same holds in embedded contexts: a correlation can be made between the location of the subject in VP and partial agreement. In (16) the postverbal subject is presumably outside VP: either in [Spec, IP] or an adjoined or dislocated position.(cf. (13) and (14) above). No such correlation can be made here, because the postverbal subject is not in VP.
One possibility is that kaan in (15) agrees, not with naas, but with famma7. This is unlikely, however,because of the possibility of (10) repeated as (18) which shows that kaan agrees directly with the postverbal subject: i.e. in (15) full agreement is optional:
(15) shows that not every instance of a postverbal subject is located in [Spec, VP] because, if the associate is in VP, then the expletive cannot be in VP, unless these constructions are analysed on a par with English Unaccusative structures within the light verb analysis (Chomsky 1995). The correlation between the location of the postverbal subject in VP and partial agreement which is made in this paper does not make this implication; it only implies that when the postverbal subject is in VP, there will be partial agreement, or conversely, partial agreement seems to be determined by a subject in VP: some postverbal subjects are clearly not in VP, e.g. the postverbal subject in TA VSO nonexpletive constructions, and postverbal expletives as in (15) or in comparable constructions in SA (cf. Benmamoun 2000): e.g.
What these sets of data have shown is that there is some initial plausibility in the correlation between the location of the postverbal subject in VP and partial agreement, and therefore, this lends credibility to the location in SA of the postverbal subject in VP, since VSO constructions in SA are also associated with partial agreement. This position will be strengthened if it can be demonstrated that partial agreement in TA like in SA is determined by all classes of subject NP's located in VP and not just by indefinite subjects as in the case of all the examples above. To see this, consider the data in (1921) (adapted from Halila (1992: 253255):
(19 a., b.) illustrate the expletive construction in TA and English respectively. (20 a.) is an alternative to (19 b.); (20 b.) shows that a comparable construction is not allowed in TA. (21 a.) shows that in English the associate must be indefinite; (21 b.) on the other hand, shows that there is no such restriction in TA; (21 c.) shows that partial agreement also occurs with postverbal definite subjects in [Spec, VP].
4 Partial Gender agreement
Nothing has been said as yet about partial agreement (as opposed to total loss/ lack of agreement) in SA: it is a well-known fact that SA displays agreement in Gender in VS contexts, e.g. (2 a., b.) above repeated as (22):
Aoun et al. (1994), drawing on an insight in Bahloul and Harbert (1993) propose that agreement is not always retained under head movement. Only 'intrinsic' and not grammatical features are retained: 'intrinsic' refers to features which are arbitrarily determined and hence not predictable from the meaning of the words. In (22 a., b.) only Gender (an intrinsic feature) is retained; Plural, being predictable and not intrinsic, is not.
For the insight proposed in this paper which relies on a unified treatment of agreement in SA and TA to be plausible, it must, therefore, predict that the same pattern of agreement i.e. agreement in Gender will be displayed in TA: in other words, the occurrence of sentences like (23) is predicted:
As indicated by the asterisk, this does not occur. Rather, it must be as in (24):
However, this is expected in light of the fact that subject-verb agreement in TA shows no Gender distinction in the Plural, cf. (25):
Consequently, the TA data is compatible with the analysis advocated by Aoun et al. for SA since, in (24), when the verb/auxiliary undergoes head-movement, the reason the intrinsic Gender feature is not retained is that contrary to SA it has not been there in the first place. (24) is derived from (26):
TA shows a Gender distinction in the Singular, therefore we predict that Gender will be retained under head-movement: (27) demonstrates that in TA as in SA agreement is not lost totally, only partially, since Gender IS retained under head-movement when it is present underlyingly, as would be predicted by Aoun et al.'s analysis:
(27) is presumably derived from (28):
5 Concluding remarks
Finally, although the objective here, is not to propose solutions to theoretical problems,8 it would be interesting to speculate briefly on the nature of the correlation between the location of the postverbal subject in VP and partial agreement: The SA data seem to suggest that, what makes the difference in agreement between SVO and VSO, is whether the subject remains in its canonical position (29 b.) or whether it is raised to [Spec, IP] (29 a.) in other words, whether [Spec, VP] contains an overt subject or a null one:
Under the assumption that lexical items enter the syntactic derivation with their features fully specified (Chomsky 1995) and that, in case subjects are overt, some of their features may become redundant (Chekili (to appear)), partial agreement is predicted to obtain in (29b) since the external argument position [Spec, VP] is occupied by a lexical NP but not in (29a) as there is a null subject in VP.
In conclusion, the correlation in SA and possibly other languages between the position of the postverbal subject in VP and partial agreement, has been shown to be plausible on the basis of data from TA which seem to indicate that when the subject is in IP i.e. SVO and VSO orders in non-expletive contexts there is full agreement; when the subject is in VP e.g. in Expletive-Argument constructions there is partial agreement. This pattern suggests that partial agreement is indeed determined by the position of the postverbal subject in VP.
Aoun, Joseph, Elabbes Benmamoun and Dominique Sportiche (1994): "Agreement and Conjunction in Some Varieties of Arabic", in: Linguistic Inquiry 25, 195220.
Aoun, Joseph and Elabbes Benmamoun (1999): "Gapping, PF Merger, and Patterns of Partial Agreement", in: Fragments: Studies in Ellipsis and Gapping. Ed. S. Lappin and E. Benmamoun. New York: Oxford University Press, 175192.
Bahloul, Maher and Wayne Harbert (1993): "Agreement Asymmetries in Arabic", in: Proceedings of WCCFL 11, 1531.
Benmamoun, Elabbes (1992): Functional and Inflectional Morphology: Problems of Projection, Representation and Derivation. Doctoral dissertation. USC, Los Angeles, CA.
Benmamoun, Elabbes (1997): "Agreement in Arabic and the PF Interface", in: WCCFL 15, 3347.
Benmamoun, Elabbes (1998): Agreement Asymmetries and the PF Interface. [available: http://www.Usc.Edu/dept/LAS/Linguistics/Semitic/index.html (November, 26, 1999)].
Benmamoun, Elabbes (2000): The Feature Structure of Functional Categories: a Comparative Study of Arabic Dialects. Oxford: Oxford University Press.
Bolotin, Naomi (1995): "Arabic and Parametric VSO Agreement", in: Perspectives in Arabic Linguistics 7, 727.
Borer, Hagit (1995): "The Ups and Downs of Hebrew Verb Movement", in: NLLT 13, 527606.
Chekili, Ferid (to appear): "Agreement Asymmetries and the Lexical/ Null Subject Parameter", in: Al'Arabiyya.
Chomsky, Noam (1992): "A Minimalist Program for Linguistic Theory". MIT Occasional Papers in Linguistics 1. Cambridge: MIT.
Chomsky, Noam (1995): The Minimalist Program. Cambridge: MIT Press.
Doron, Edit (1996): "The Predicate in Arabic", in: Studies in Afroasiatic Grammar. Ed. J. Lecarme, J. Lowenstam, Ur Shlonsky. Leiden: Holland Academic Graphics.
Eid, Mushira (1987): "Arabic Theoretical Linguistics: The Seventies and Beyond", in: Al'Arabiyya. Journal of the American Association of Teachers of Arabic 20, 1&2.
Fassi Fehri, Abdelkader (1993): Issues in the Structure of Arabic Clauses and Words.. Holland: Kluwer Academic.
Halila, Hafedh (1992): Subject Specificity Effects in Tunisian Arabic. Unpublished Ph.D. Dissertation. USC.
Huybregts, Riny (1991): "Allosteric Agreement in VSO Languages", in: Linguistics in the Netherlands. Ed. F. Drijkoningen and A. Van Kemenade. Amsterdam: John Benjamins.
Kaplan, Tami (1991): "A Classification of VSO Languages", in: ESCOL 191, 198209.
Koopman, Hilda and Dominique Sportiche (1991): "The Position of Subjects", in: Lingua 85, 211258.
McCloskey, J. (1996): "Subjects and Subject Position in Irish", in: The Syntax of the Celtic Languages. Ed. J. Borsley and I. Roberts.Cambridge: Cambridge University Press, 241283.
Mohammad, Mohammad (1989): The Sentence Structure of Arabic. Doctoral Dissertation. Los Angeles. USC.
Roberts, Ian and Ur Shlonsky (1996): "Pronominal Enclisis in VSO Languages", in: The Syntax of the Celtic Languages. Ed. R. Borsley and I. Roberts. CUP, 171199.
Taraldsen, Knut Tarald (1993): "Subject/ Verb Agreement and Word Order in Celtic and Romance", in: NELS 23, (2), 495504.
Van Gelderen, Elly (1996): "Parametrising Agreement Features in Arabic, Bantu Languages, and Varieties of English", in: Linguistics 34, 753767.
Van Gelderen, Elly (1997): Verbal Agreement and the Grammar behind its Breakdown. Tübingen: Max Niemeyer Verlag.
2 There are actually a number of variations on this general theme, based on a distinction between spell-out and LF and a distinction between strong and weak features (Bolotin (1995), Doron (1996), Roberts and Shlonsky (1996), Van Gelderen (1996, 1997 etc.)).
5 TA displays full agreement in both SVO and VSO orders (cf. examples (13) and (14)).
7 Another view see, for instance, Halila (1992) would be that kaan agrees with a null expletive in its spec.
8 For a possible theoretical explanation, see Chekili (to appear).