Jump to ContentJump to Main Navigation
Merging FeaturesComputation, Interpretation, and Acquisition$

José M. Brucart, Anna Gavarró, and Jaume Solà

Print publication date: 2009

Print ISBN-13: 9780199553266

Published to Oxford Scholarship Online: May 2009

DOI: 10.1093/acprof:oso/9780199553266.001.0001

Show Summary Details
Page of

PRINTED FROM OXFORD SCHOLARSHIP ONLINE (oxford.universitypressscholarship.com). (c) Copyright Oxford University Press, 2020. All Rights Reserved. An individual user may print out a PDF of a single chapter of a monograph in OSO for personal use.  Subscriber: null; date: 13 August 2020

Number agreement in the acquisition of English and Xhosa

Number agreement in the acquisition of English and Xhosa

(p.104) 6 Number agreement in the acquisition of English and Xhosa
Merging Features

Jill De Villiers

Sandile Gxilishe

Oxford University Press

Abstract and Keywords

The chapter considers the acquisition of number agreement in English, revealing asymmetries in production and comprehension. The possibility is explored that agreement is a post-syntactic phenomenon such that agreement on the target is inaccessible to interpretation. The consequences are explored for a radically different agreement paradigm, namely Xhosa.

Keywords:   production, comprehension, acquistion, interpretation, testing

6.1 Overview

How do children come to understand number agreement, that is, how does an inflection on the verb carry information about subject number? The puzzling fact that emerges is that children acquiring English control number agreement in production quite early, while failing to use the information from the verb in comprehension tasks until several years later. Two proposals are compared to account for this asymmetry: a formal one from modern linguistics accounts that predicts that features on the target of morphological agreement should not be accessible to interpretation, and a conceptual one that verbs, unlike pronouns, do not carry notional number.

English is compared with other languages in which agreement on the verb is not redundant, for example in pro‐drop languages. Xhosa, one of the Bantu languages, is taken as a test case: it has a rich noun class system and correspondingly rich subject agreement on the front of the verb. The status of subject agreement in Bantu languages is a topic of several decades of debate, in particular, whether it should be treated as a clitic pronoun or as an agreement marker, and whether the different members of this family of languages differ along the continuum of possibilities. A proposal is made in which data from children's comprehension might contribute to this discussion.

6.2 Basic number agreement

Agreement has been considered in several different ways under different theories. The example to be considered here is from the agreement between the subject and the verb in number, though there are many languages that have number agreement also with adjectives, determiners and so forth. How (p.105) does it happen when the subject and verb agree in number in Standard (or Mainstream) English?

(1) He goes to the store.

(2) They go to the store.

The classical account is that the number agreement on the subject noun is copied to the verb, in a unidirectional fashion; that is, the verb's number is dictated by the number of the subject noun. Person, number, and gender are known as phi‐features, and in a variety of languages they enter into agreement dependencies with elements in the clause distant from their source. The mechanism by which the source and target are connected varies with the linguistic theory. In modern generative accounts, the verb may raise in the syntactic structure to check number in some node called Agreement, which may or may not coincide with Tense (Pollock, 1989). In other accounts such as HPSG (Pollard and Sag, 1994), the copying is non‐directional and initiates from a “referential index” in the world that dictates the number both to the subject and the verb (see also Murphy, 1997, on Bantu languages). A constraint then operates to ensure that both elements have the same number.

In the copying model, the subject is the controller that has the semantics attached (e.g. number). In generative syntax, the subject raises out of the VP‐ merged position into a higher node, which triggers the verb to move to a subject agreement node to check its phi‐features. In the constraints model, the “reference” properties are dictated to both subject and verb, so neither has priority. It is possible that data on child languages could be used to differentiate these approaches, a question to which we will return in considering languages with very rich agreement systems.

6.3 Number agreement in AAE (African American English) and MAE (Mainstream American English)

How do children acquire number agreement on verbs? Although it has been established for decades now that young children speaking Standard English reliably produce agreement (3rd person /s/ on verbs) at about age three and a half years, less was known about comprehension of the form. The process of pursuing comprehension of agreement led us to some surprising findings.

Our work began with an innocent question about AAE, which has “optional” number agreement (Green, 2002; Labov, 1969; Myhill and Harris, 1986).

(3) He go to the store.

(4) They go to the store.

(p.106) As part of a large project to investigate the normative course of development of AAE (Seymour, Roeper, and de Villiers,2003a, b), it became necessary to understand whether children acquiring that dialect might be missing the 3rd person /s/ on the verb for purely phonological reasons. It is well established (Labov, 1969; Seymour and Pearson, 2004) that AAE has a different set of phonotactic rules for the ends of words than other dialects of English, especially in final consonant clusters (producing “tes” for “test” etc.), and the question arose about the child's sensitivity to the final /s/ in comprehension. If it could be shown that AAE speakers could understand the information carried by the /s/ even though they did not say it for phonotactic reasons, then the prospect was good for using a comprehension test rather than the usual production test in assessment of children suspected of language disorders. A chronic problem recognized for many years is that the inventory of morphemes used for language assessment on standardized tests is at variance with the inventory of morphemes in AAE, potentially leading to misdiagnoses of language disorder in AAE‐speaking children (Seymour, Bland‐Stewart, and Green, 1998).

In previous work on comprehension of the information conveyed by the number agreement on the verb (e.g. Fraser, Bellugi, and Brown, 1963), the trick has been used of making the subject an irregular plural with no overt marking so all the information on number is carried by the verb ending:

(5) The deer run in the park.

(6) The deer runs in the park.

Results suggested that this was difficult even for Standard‐English‐speaking children (e.g. Fellbaum, Miller, Curtiss, and Tallal, 1995), but it is also well known that children aged four to six have problems with irregular plurals such as “deer” and “sheep” and “feet” (Brown, 1973). In our pilot testing, four‐year‐ olds often asked us “Do you mean the deers?” suggesting that this was a source of confusion. As an alternative, one can disguise the plurality of the noun by ensuring that the following verb begins with /s/, in which case if the sentences are pronounced as in running speech, the existence or not of the plural /s/ on /cat/ in (7) and (8) is indeterminate:

(7) The cat sleeps on the bed.

(8) The cats sleep on the bed.

Johnson (2005) devised a set of stimuli of this sort and used a picture comprehension test of the kind shown in Figure 6.1 to ask whether AAE‐speaking children aged four to six years could use the number agreement on the verb to identify the number of the subject noun. Since no other information was (p.107)

Number agreement in the acquisition of English and Xhosa

Figure 6.1 Sample stimulus for the recorded sentence: /therabbitsnifftheflowers/

Source: From studies of Johnson (2005) and Johnson, de Villiers, and Seymour (2005)

available from the referential context, the assumption was that a child who had mastered the grammar of subject number agreement would be able to use the 3rd person /s/ to determine the number of the subject even if AAE phonological constraints meant that it was not produced in their own speech. In fact, AAE‐speaking children showed no sensitivity to the information in 3rd person /s/ by age six. Johnson concluded that the 3rd person /s/ is not present as number agreement in the grammar of AAE‐speaking children at this age, and maybe not in adult AAE either (Green, 2002; Myhill and Harris, 1986).

The complication with the story is that Johnson also had MAE‐speaking children as participants, originally considered a “control” for the AAE speakers (Johnson, de Villiers, and Seymour, 2005). It was expected that these children, who have full control over 3rd person /s/ in their speech from about three years of age (Brown, 1973), would have no difficulty detecting the /s/ as a clue to subject number in comprehension. However, three‐, four‐, and even many five‐year‐old MAE‐speaking children failed to use the 3rd /s/ as a clue to subject number. Several alternative explanations were explored, for example, that something about the pictured stimuli might have been responsible. Perhaps spontaneous speech offers more clues than these simple pictures, clues that somehow support the 3rd person /s/ production. As a precaution, we ran a study with children of three and four using single pictures, and asking them to describe what they showed. To ensure the use of 3rd person /s/ rather than (p.108) past tense or progressive, we said the picture showed “what the animals do everyday”, and we taught the children to start every sentence with “Every day” to legitimize the use of the generic tense that 3rd person also marks. After providing a sample of both morphological forms on different verbs, e.g.

(9) Every day the raccoon washes in the pool.

(10) Every day the pigs roll in the mud.

the children were then presented with the series of pictures again, in which the number of raccoons or pigs sometimes varied from the original. The preschoolers had no difficulty producing matched number agreement in this study. The same subjects were tested on the comprehension test used in Johnson (2005) and Johnson et al. (2005), but with the words “Every day” inserted before test sentences like (7) and (8). Despite these methodological improvements, the results were the same, namely the children showed no discrimination of subject number based on the cue from the 3rd person /s/.

Thus, even MAE speakers show a two‐year gap in performance between production and comprehension, with comprehension lagging behind production. The question becomes, why? Why is it that the information in the agreement feature on the verb is “bleached” of its numeric content? Before offering some theoretical alternatives, consider another question: why does anyone over six years old succeed? The speculation is that six‐year‐olds might be capable of comparing multiple representations, that is, they may succeed by comparing the output of their own production given the scene with the test sentence. This process would entail the ability to hold the sentences in working memory and to compare multiple representations. Others have proposed that this capacity for comparing representations is late‐developing and may be responsible for the delay of Principle B effect, in which ambiguity in Principle B interpretation is also resolved only after age six or so (Reinhart, 2004b).

6.4 Alternative theories

Here we compare two basic alternatives for why agreement features on the target, i.e. the verb, may be inaccessible in interpretation without special effort (of the sort that requires comparing representations). The first is more linguistic and the second more conceptual.

In explaining the problem in interpretation of 3rd /s/ on the verb, Johnson et al. (2005) borrowed a concept from the discussion in Chomsky's Minimalism (1995), namely that, once the number is checked off in agreement, it no longer survives as semantic information at LF. This would allow automatic procedures to arrange number agreement in production, but in (p.109) comprehension the agreement marker on the verb would carry no semantic information. Bobaljik (2006) makes a more specific proposal that agreement is a late operation, part of the post‐syntactic morphological component. He uses data from a variety of languages to address the prediction that it should be possible for an NP to control agreement on a predicate, even if it bears no syntactic relationship to that predicate other than being “close enough” (for which the technical details are not needed here). The conclusion he reaches then forces a second prediction, namely that “agreement features on the target of agreement do not contribute to interpretation”.

But, if there is something right about the failure in interpretation of agreement targets, then it should not just be a feature of child language but should also appear in adult processing of language. No direct test of this has been carried out, though there is a different literature on how adults resolve number agreement with ambiguous nouns such as collectives.

For example, in a study by Bock, Nicol, and Cutting (1999), adult subjects participated in a production task in which they either had to produce verb agreement or pronoun agreement with a subject noun. The subject nouns were of different types but included forms such as “committee”, which is a collective noun that is notionally plural, but grammatically singular:

(11) The committee meets on Tuesdays.

(12) *The committee meet on Tuesdays.

Adults responded differently when asked to provide verb agreement or pronoun agreement. The verb agreed with grammatical number:

(13) The committee meets. …

but pronouns agreed notionally with the subject:

(14) and they said. …

Bock et al. contend that when an agreement “controller” (namely the subject) carries a grammatical number that is not the same as its notional number, verb agreement targets generally match the grammatical number and pronoun agreement targets instead generally match the notional number.

However, these authors also draw a larger conclusion about why nouns and verbs may behave differently with respect to number, which brings us to the second, more conceptual account. “Verbs denote things whose number properties are at best slippery. As a property of states and events, number is abstract (Shipley & Shepperson, 1990; Wynn, 1990) and often indeterminate. Is hand‐shaking singular or plural? Is kissing singular or plural? Is football playing singular or plural? It may well be that the syntactic work of indicating (p.110) what goes with what in a string of words is more readily accomplished by using the number features of the subject to mark the verb, especially since English verbs usually occur with a morphologically explicit subject” (1999, p. 341).

The implication is that the information on the verb is secondary, or derived, for a very good conceptual reason. In more recent work, Eberhard, Cutting, and Bock (2005) provide a synthesis of several experiments in this vein to argue that, in speech production, the status of number on pronouns and verbs is derived from different sources, and repeat the general conclusion from their model that pronouns match the notional number of the subject more readily than verbs. Nonetheless, it remains to be investigated whether or how the verb information on sentences such as (7) and (8) might be accessed during online processing in adults. Furthermore, the conceptual account provides a different angle on the phenomena but is not incompatible with the generative account.

These authors raise interesting historical questions about how pronouns and agreement relate. Eberhard et al. (2005) bring in evidence that verb agreement in English arose by a process in which antecedent—pronoun number agreement was linked to subject—verb number agreement. Historically, there is evidence that topicalizing constructions move the subject into an initial topic position, often introducing a pronoun repeat of the subject (Givón, 1976):

(15) The girl, she like candy.

The argument is that in the earliest Germanic origins of English, the topical‐ ization results in a post‐posed subject:

(16) The girl, like‐she candy.

Phonological reduction and assimilation processes over time then reduce the pronoun to a verb inflection such as the one in Mainstream English dialects:

(17) The girl likes candy.

Once the form has grammaticized, it becomes an obligatory verb inflection insensitive to discourse requirements such as those that give rise to topi‐ calization. Eberhard et al. (2005) argue that verb inflections and pronouns share a common sensitivity to number historically, but in present‐day English, “singular and plural verb forms are comparatively numb to number meaning” (p. 538). In their model pronouns achieve their number via concord, a semantic process of coreference with the subject noun. But verbs get their number via syntactic agreement. Eberhard et al. admit that their model so far is meant to account for agreement‐lean languages such as English, and they make no claims for generality to languages that might bear richer agreement.

The developing picture is one in which pronouns carry agreement features by a different process than does verb agreement, and that will become an (p.111) interesting issue for us when dealing with a class of languages where the distinction between pronoun and agreement is not so evident.

6.5 Number agreement in languages other than English

English is a difficult language from which to reach a broad conclusion. Agreement in general is very weak: there is no marking of case or gender on nouns or verbs, and the verb number agreement on regular verbs is only for third person subjects, and only in the case of the so‐called and misnamed (Sauer‐ land, 2002) “present tense”. Furthermore, the circumstances are rare in which the notional plurality of the subject number is disguised, as with abstract collectives (“committee”) or when the following verb starts with an /s/ and there are no other contextual or linguistic (e.g. pronoun) clues. Writing about this problem, Brown (1973) argued that the clue from number agreement was not salient to children because it is rare in English to have to rely on it.

It is very important to consider data from languages in which the verb provides a more consistent and important cue to number, namely pro‐drop languages. If the subject is not there, then the only clue to its number (and/or gender, etc.) comes from the morphology on the verb. Looking at children speaking Dominican Spanish, Pérez‐Leroux (2006) found strikingly parallel results to those described above for English (Johnson et al., 2005). When the children were exposed to sentences in which the agreement morphology on the verb was the only cue to detecting which picture to choose (singular or plural subjects), the three‐, four‐, and five‐year‐old Spanish speakers were no better than the English‐speaking preschoolers at the task. The verb agreement morphology was not used as a clue to the number of the subject even though in Spanish the subject must frequently be absent because of pro‐ drop. At least for children, it seems that verb agreement is not more salient as a marker of notional number even when its “cue validity” is increased by pro‐drop.

In the chapter by Arosio, Adani, and Guasti (this volume), a similar result is found in Italian with more complex structures involving relative clause interpretation. In that case it is not until children are well advanced in years, perhaps nine years of age, before they will use the verb agreement properties to differentiate whether something is a subject or object relative clause. The control case demonstrates that they know the relative clause structures, in that they can use information from structural position to make the right interpretation, but not from agreement. The authors argue that, in parsing, agreement is a post‐syntactic operation, a position potentially compatible with the formal grammar model by Bobaljik (2006) discussed earlier.

(p.112) 6.6 The case of Xhosa

Our goal is to enlarge the discourse even further by considering a very rich agreement language, Xhosa, and to consider the ramifications for how children learn number agreement in that language. One of nine national African languages of South Africa and one of the Nguni group of Bantu languages, Xhosa is primarily agglutinative, with morphology accumulating on the verb stem. There are nine positions on the verb into which a grammatical morpheme might slot, and they include markers of agreement with both subject and object noun class. Like many Bantu languages, Xhosa has numerous noun classes that are relatively arbitrary, but may have historically semantic roots. For example, noun class 1 is mostly names for humans. However, in the present‐day language there are many exceptions and overlaps (e.g. names for humans also appear in seven other noun classes) so the semantics of the referent are only weakly associated with noun class membership. A partial ordering of morphemes on the verb in Xhosa is as follows:


  • U ‐ya ‐m ‐fund‐is ‐a.

  • SM‐1a ‐ T ‐ OM‐1a ‐ Verb ‐ CAUS ‐ M1

  • She present hi learn cause indicative

  • She causes him to learn

  • ‘She teaches him.’

There are arguments that the verb in Xhosa is not a complex head (See Buell (2005) on the closely related Nguni language, Zulu), but that each morpheme in fact has its own head in the hierarchy (Du Plessis, 1997; Deen, 2005a) and the full morphologically complex verb is only created at Spell‐Out (Julien, 2000, cited in Deen, 2005a). Xhosa has SVO word order but other variations of this order occur frequently. The subject noun can be dropped (pro‐drop), leaving only the subject agreement on the verb appropriate to the class of the absent subject noun. The basic sentence form is thus:


  • I‐si‐lumko si‐thanda iincwadi.

  • 7‐genius 7‐SM‐likes 9‐books

  • ‘The genius likes books.’

but post‐verbal subjects occur, as in:


  • Si‐thanda ii‐ncwadi isilumko.

  • 7‐SM‐like 9‐books  7‐genius

  • ‘Likes books the genius.’

(p.113) or with pro‐drop:


  • Si‐thanda ii‐ncwadi.

  • 7‐SM‐like 9‐books

  • ‘Likes books.’

What about number agreement? Number is not associated with a single morpheme but instead the form changes by noun class. Of the 15 noun classes, eight are singular and seven are plural; however, the formation of the plural is not straightforward in morphology. For example, in the following examples, the change from singular to plural is different for each noun class, unlike languages such as English, which have only tiny irregularity in plural formation (man/men, child/children, foot/feet).




Class 1: um‐ntwana 1a: u‐tata

Class 2: aba‐ntwana 2a: oo‐tata

Class 7: isi‐lumko

Class 8: izi‐lumko

The assumption can be made that the plural morphology on the noun arises, as with noun class, in the lexicon.

When it comes to subject and object verb agreement with the noun class, once again it is not a straightforward copy of an agreeing prefix, rather the plural form of agreement varies with class:


  • Oosisi ba‐hlala phezu kwesofa.

  • 2a‐sisters 2a‐SM‐sit on top sofa

  • ‘The sisters sit on the sofa.’


  • Izinja zi‐hlala phezu kwesofa.

  • 10‐dogs 10‐SM‐sit on top sofa

  • ‘The dogs sit on the sofa.’

How does a child acquire such a system, and does the child learn it in a piecemeal fashion, verb by verb and morpheme by morpheme?

6.7 Acquisition of subject and number agreement in Xhosa

The initial data to answer this question come from a sample of children studied longitudinally by the second author. This group consists of six children growing up with Xhosa as their first language in the township of Gugulethu outside of Cape Town, South Africa. Beginning at age two they were audio‐ taped approximately once every two months in natural interaction in their homes with family members and with a native Xhosa‐speaking researcher who (p.114)

Table 6.1 Number of utterances and number of samples ( ) by age band










80 (3)

152 (3)

142 (4)

45 (3)

149 (4)

72 (4)

640 (21)


124 (3)

132 (3)

56 (2)

75 (3)

86 (3)

54 (2)

527 (16)


69 (2)

92 (3)

104 (3)

50 (2)

315 (10)


273 (8)

376 (9)

198 (6)

120 (6)

339 (10)

176 (8)

1,482 (47)

also transcribed the tapes. The transcriptions were checked by a second native Xhosa speaker. Table 6.1 shows the number of samples from each child and the number of utterances collected for each sample over the period between approximately 24 to 39 months of age for each child. Samples were combined into three age bands to provide enough utterances in each age band for reliable developmental analysis. The transcripts consist of the child's utterance, a gloss of the intended utterance as it would be produced by an adult speaker, and an English gloss. The research assistant's speech is also transcribed and provided with an English gloss.

These data and also data from an even younger group of Xhosa speakers collected in the same manner reveal that subject agreement was well established by age two years (Gxilishe, de Villiers, and de Villiers, 2007). The group of children aged two to three years used subject agreement appropriately, with practically no substitution errors. The finding of omission but no substitution errors has been reported commonly for Bantu language acquisition (Deen, 2005a; Demuth, 2003; Suzmann, 1982). That is, children do omit the subject agreement at age two years but almost never use the wrong form, despite the complexity of the agreement paradigms. It does not seem to be the case that the children are using many rote utterances, in fact quite the contrary: like children everywhere, they are using novel utterances. As an index of productivity, the number of noun classes per transcript were tallied, and they averaged three to five different noun classes per transcript. This means that children were not restricting their talk to one or two familiar noun classes and hence achieving success by limited productivity. Neither is it the case that the children use only a few verb roots to achieve success: the different verb roots on which subject agreement is reliably provided vary from five to 15 per transcript. There is thus ample opportunity for errors that nevertheless do not occur.

The question that is significant for the present chapter is, how well do children mark number agreement from the subject to the verb? In the current data, plurals represent only 13% of the potential cases of subject agreement (p.115)

Number agreement in the acquisition of English and Xhosa

Figure 6.2 Data on plural and singular subject agreement from two‐ to three‐year‐old Xhosa speakers

from the children aged 24 to 39 months. Nevertheless, they are very well supplied. Figure 6.2 shows the graph of subject agreement averaged across the six children by age, and it is clear that plural agreement is better supplied than singular subject agreement. Most of the plural agreements are from noun classes 2 and 10, and most of the singulars are from the corresponding noun classes 1a and 9. There is nothing particularly transparent about the plural/singular marking for these classes. Again, omission is the only source of error, not substitution.

Again the question can be asked, what if the subject is present versus absent? Does it make a difference in the likelihood that the children will produce the number agreement? Unfortunately, the data are limited, given that the number of obligatory contexts for plural agreement is only 13% of the total subject agreement opportunities. The plural subject noun was present in only six cases, and verb agreement was appropriately supplied in five out of those six (83.3%). The plural subject noun was absent in 21 cases, with plural subject agreement provided in 19 of those 21 instances (90.9%). It appears to be the case that the subject does not have to be overt in the sentence for the child to supply correct number agreement on the verb. This would not be surprising (p.116) in a familiar language like English, where the form of irregularly pluralized lexical nouns has no influence on the form of number agreement on the verb:

(25) The man dances.

(26) The men dance.

But in Xhosa the form of agreement is dictated by the noun class of the subject. If the missing subject noun had a noun class that was semantically transparent (say human) then the referent properties could dictate the form of number agreement. But in Xhosa, noun classes are not so semantically transparent, and noun class is provided in the lexicon, not directly in the world of reference. It seems to be an inescapable conclusion that the subject noun had to be present to dictate the subject agreement on the verb before undergoing deletion. The one qualification necessary to this conclusion is that the predominant noun classes used at this age are classes 1/2 and 9/10, which are mostly humans and artifacts, so it is possible that semantic properties could assist the child at the beginning. But since both humans and artifacts occur in other noun classes, such a hypothesis will soon lead to substitution errors, which, as we have seen, are virtually nonexistent.

The puzzle arises when one considers not just production but comprehension, which must be the route by which the system is acquired. The adults around the child do not necessarily restrict themselves to four of the 15 noun classes, so the input will provide evidence counter to a simple semantic mapping. In order to make sense of the input, the child must recover any deleted subject nouns in the input to figure out the relation of noun class and number agreement marker. Little is yet known about the naturalistic input from care‐ givers to very young children acquiring Xhosa, but it would be interesting to see if “motherese” somehow makes the subject nouns more accessible, either by less deletion, or by having adjacent utterances with and without the subject noun. This question remains a subject for further research.

6.8 The nature of subject agreement in Xhosa

There is considerable work on the issue of whether subject agreement in Bantu languages should be treated as a kind of pronominal clitic attached to the verb, or an affix like English /s/ or Italian verb endings. As discussed, there is often a historical move from free pronoun, to clitic, to agreement morpheme, and it must be borne in mind that different languages in this group could be at different points in this progression.

The classic work on this question is by Bresnan and Mchombo (1987), who raise the question about subject agreement in Chichewa, another Bantu (p.117) language. In their typology the status of agreement markers is based on the co‐occurrence possibilities of person markers and their controllers in the same construction rather than on the morphophonological form of the agreement markers (Turunen, 2007). A syntactic agreement marker cannot occur without an overt controller in the same sentence; however, a pronominal marker cannot occur with an overt local controller or it would violate Principle A. For that reason, they classify Chichewa subject agreement as pronominal in type when an overt subject is absent (pro‐drop), and as agreement when an overt subject is present. However, object agreement in Chichewa obligatorily occurs when the object is dropped or displaced beyond the phrase, and so the OM is classified as pronominal in form.

Even for Chichewa, however, there are proponents of the view that the SM is also a pronominal clitic. Baker (2001, 2005) raises the possibility that Bantu languages like Chichewa have a parameter setting of “Optional Polysynthesis” in which crucial parts of the event are incorporated into the verb. In particular, he argues that, in Bantu languages such as Chichewa, the overt subject must be moved outside the clause (before or after) when there is subject agreement. That is, the language may have a grammar like the historical stage of early English verb agreement discussed in Eberhard et al. (2005):

(27) The girl, she like candy.

In such an analysis, the subject agreement marker occupies the subject position, namely Spec‐AgrS, and behaves more like a clitic pronoun. Baker argued that the subject in Chichewa is displaced outside the phrase by the presence of SM, i.e. SM occupies the subject position. In that way, there would no longer be a Principle A violation with an overt subject, since it is displaced. One of the convincing rationales for a subject displacement into topic is that wh‐ questions can never be asked directly in Chichewa, but only using a cleft or passive construction. This is because topics cannot be directly replaced with wh‐questions. On a view such as this, perhaps both SM and OM markers, being pronominal clitics, lead to dislocation of the corresponding arguments outside the clause, rescuing Principle A.

Zeller (2008) puts forward a complex proposal that in Zulu, closely related to Xhosa, the SM is a pronominal clitic that forms a constituent with the subject DP, and that “agreement” in SVO constructions is a form of clitic‐ doubling like that found in Northern Italian dialects.

In contrast to the pronominal view, Buell (2005) puts forth evidence that Zulu has SM markers that do not behave like pronouns. One of his pieces of evidence is that Zulu has a range of compound tenses with a lexical verb embedded under a variety of modals, auxiliaries, and aspect markers. All of (p.118) these auxiliary forms are marked also with subject agreement, in positions that pronouns would not usually occupy. In Buell's view, subjects are in the specifer of Agr‐S (but null in pro‐drop) and contribute their features to the verb when it raises to the head of Agr‐S, as in generative accounts of English, hence are not displaced.

Buell also adduces evidence that the OM in Zulu is an agreement marker, departing from other linguists of Bantu who argue that OM is pronominal (Bresnan and Mchombo, 1987). His argument is an interesting one that raises more questions about possible interpretive differences between agreement markers and pronouns, but it is not clear that it is decisive. In Zulu, the second conjunct of a coordinated sentence must take OM such as:


  • a. Ngi‐dl‐e a‐mahhabula a‐mabili no Sipho u‐ wa‐

  • 1S. eat PERF 6‐apple 6.REL and 1‐Sipho 1.SM ‐ 6.OM

  • dl‐ ile.

  • eat‐ PERF

  • (lit. ‘I ate two apples, and Sipho ate them, too’) (Buell, ex 82, 2005, p. 52)

Buell points out that this does not mean Sipho ate the same two apples, in fact, it means Sipho ate his own. Compare this with the overt English pronoun:

(29) I ate two apples and Sipho ate them too.

Comparing the status of subject agreement markers in Chichewa and Nairobi Swahili, Deen (2006) argues that, unlike in Chichewa, subject questions are possible in Nairobi Swahili, along with other diagnostic differences between the two languages. Deen concludes that the SA marker in Nairobi Swahili is not a pronominal clitic, though it may be ambiguous in other dialects of Swahili (Keach, 1995). There are many complex arguments about the nature of SM in Bantu languages, therefore it is impossible to do justice to them here. However, it is clear that the matter is not settled for even one language at this point, and the languages may indeed differ.

The literature suggests at least three possible mechanisms of SM for Xhosa, each with their strengths and problems. One mechanism, proposed by Du Plessis and Visser (1998), is that the subject noun is merged VP internally, and the noun class marker is provided in the lexicon. The subject noun then moves to the Specifier of AgrS. The verb moves from its base position into Tense and then to AgrS, where subject agreement is then dictated by the noun class of the subject noun (see Figure 6.3). In keeping with that proposal are the facts in Xhosa about subject agreement on compound verbs, as in Zulu in Buell (2005).


Number agreement in the acquisition of English and Xhosa

Figure 6.3 Tree diagram of derivation of subject agreement in Xhosa

In contrast, a pronominal account like that given for Chichewa might be that the SM itself occupies Spec of Agr‐S, displacing the subject noun into a topic position. That would be compatible with the lack of ordinary subject wh‐questions in Xhosa (Du Plessis and Visser, 1998; Zeller, 2008).

In a very recent paper, Zeller (2008) proposes that the SM is part of a “big DP”, in which the SM is the head of the preverbal subject, and takes the subject DP as its complement (which could be null). The SM then incorporates with the verb in T, as a pronominal clitic. But without agreement, what motivates subject movement? Zeller claims that SM is an antifocus marker in order to rationalize its move from VP. This account attempts to explain a variety of complex facts about Bantu languages, for Zulu in particular and by extension, Xhosa.

(p.120) As can be seen from this condensed review, the matter of the nature of subject agreement as pronominal or agreement marker is far from clear in Xhosa. Can child data help in distinguishing the alternatives?

6.9 What can child language tell us?

In Gxilishe et al. (2007) a question is explored about the nature of subject agreement in Xhosa child language. Du Plessis and Visser (1998) argue that the morpheme in Xhosa is a subject agreement marker with either an explicit or deleted subject. The children's data are certainly compatible with the latter position, in that there is no difference in the likelihood of supplying the subject agreement on the verb as a function of whether the subject is overt or deleted. Deen (2006) draws a similar conclusion for the status of the agreement prefix in child data from the Nairobi dialect of Swahili.

Yet with respect to subject wh‐questions, Xhosa behaves like Chichewa, and it remains possible that the status of SM may be pronominal in character. If so, it may more readily carry notional number, in concord with the subject noun, perhaps by virtue of its position in Spec‐Agr (see also Zeller, 2008). In contrast, the suffixes in a language like Italian or Spanish cannot raise to Spec‐Agr, and can only carry grammatical number by syntactic agreement. Furthermore, Bobaljik's claim about the post‐syntactic nature of morphological agreement is assumed not to apply to pronominal clitics. Nothing in the interesting work of Eberhard, Cutting, and Bock connects their production model to particular alternative grammatical configurations in a generative framework, but a bridge may be possible.

Recall that young English children cannot seem to use the agreement marker on the verb to establish subject number. The possibility that this is because the information is usually redundantly marked is contradicted by data from Pérez‐Leroux (2006) on Spanish‐speaking children, who also cannot use the verb marking even when the subject noun is pro‐dropped (see also Arosio et al., this volume). But it is as yet unexplored whether children speaking Xhosa could use the subject agreement marking on the verb to determine subject number. It is a much more difficult problem in the case of Xhosa, given the variety of forms and their dependence on noun class. However, if the forms behave more like pronouns, perhaps children will be able to judge number from these forms.

We have begun pilot research with stimuli very like those used in Johnson et al. (2005) and Pérez‐Leroux (2006). One such study is designed as picture choice as in the study in English by Johnson (2005). All stimuli are common enough nouns and verbs that three‐ to six‐year‐old Xhosa speakers should (p.121)

Table 6.2 Pilot studies of subject number agreement comprehension in Xhosa

Xhosa sentence

English gloss

Choices: correct in bold

1. a‐dlala kuswingi

‘they swing on the swing’

2 girls

1 girl

2. ba‐thetha efonini

‘they talk on the phone’

2 women

1 woman

3. i‐dlala emanzini

‘he plays in the water’

2 boys

1 boy

4. li‐tshaya estratweni

‘he smokes in the street’

2 police

1 police

5. zi‐lala ebhedini

‘they sleep on the bed’

2 cats

1 cat

6. u‐nukisa amablomu

‘it sniffs at the flowers’

2 bunnies

1 bunny

know (see Table 6.2). The question is, will young Xhosa‐speaking children be able to retrieve the subject number in keeping with their production?

If they behave like English speakers and show a production/comprehension asymmetry, then perhaps that would count as evidence that the markers are indeed agreement affixes, and possibly2 their targets therefore inaccessible to interpretation of number. If children can retrieve number from the markers, despite the complexity of form mapping, then perhaps it can add to the arguments on behalf of SM as a pronominal clitic in some Bantu languages.

We have also begun to explore the potential contrast with object agreement, for which there is much more consensus that the OM is pronominal in form. For instance, the object in a transitive sentence is displaced outside the prosodic envelope of the verb phrase when OM is present (see Van der Spuy, 1993; Buell, 2006; also Gxilishe, de Villiers, and de Villiers, 2007 on Xhosa children's language). If children can retrieve number information about the object from OM, but not from SM, that will give support to other linguistic arguments that the two forms are different in type. The task is very similar. The child sees for instance two pictures, one in which a woman is watering a single flower, and one in which she is watering three flowers. After saying about the pair of pictures:


  • Jonga … Oomama, … amablomu

  • See 2a‐women, 6‐flowers

  • ‘See? Women, flowers.’

the child is asked to “show the picture where”:


  • Umama u‐ya‐wa ‐nkcenkceshel‐a

  • 1a‐Woman 1a‐SM‐TNS‐ 6‐OM waters ‐ M

  • ‘The woman waters them.’

(p.122) versus


  • Umama u‐ya‐li‐nkcenkceshel‐a

  • 1a‐Woman 1a‐SM‐TNS‐ 5‐OM waters ‐ M

  • ‘The woman waters it.’

Since the lexical object is dropped, the only clue to object number is contained in the object agreement marker (Class 5 or 6) in preverbal position. Will it behave like a pronoun and allow number to be accessed?

We3 have tested eight children aged four and five years, all native Xhosa speakers, in a small day care center in a township in the Western Cape Province of South Africa, three girls and five boys. Each received six test examples of subject agreement and six of object agreement, after checking that they would respond by pointing to pairs of simple pictures. In no case did any child show mastery of the number properties of subject marking, that is, they did not use the marker to recover the subject number (average score 3/6 correct, no discrimination between singular and plurals as in Table 6.2). However, neither did they use the object marking as a cue to object number (average score 3.36/6) on examples like (31) and (32). It is premature to draw a firm conclusion at this stage, especially about the status of object agreement, as there are no data on when object agreement enters Xhosa children's speech production. However, it seems likely that young Xhosa speakers, like their English, Spanish, and Italian counterparts, may produce subject agreement successfully at age three but fail to interpret it as a cue to number of the subject for several years thereafter.

Such a finding might then give credence to arguments that the subject marker in Xhosa is indeed a post‐syntactic, morphological agreement, and not a pronominal. Although child language data is rarely used to arbitrate between different theories of the adult language, it is the continual hope of child language researchers that data from children may play some useful role in theoretical accounts.


(1) Noun classes and agreement are marked with numbers, according to convention. TNS = Tense, SM = Subject Marker, OM = Object Marker, REL = Relative Marker, PERF = Perfective.

(2) Only possibly, because, for example, Buell (2005) would have subject markers heading their own projections in the syntax, and therefore not being generated post‐syntactically as in Bobaljik (2006).

(3) Many thanks are due to Dr. Rose Mantoa Smouse, Thabisa Xhalisa, and Nolubabalo Tyam of the University of Cape Town, Clara Feldmanstern of Smith College, and the staff and children of the Kaya Mandi crèche. A full study is under way.