A Ricardo Yang Wei y Rodney Dixon Johnson.—San José,

Over the past four decades, much child language research has focused on the acquisition of the passive voice, motivated largely by the common delay in the acquisition of passive constructions across languages, both in terms of production and comprehension (Deen, 2011:155). Overall late acquisition of the passive voice has been documented for many European languages – e.g. English at four to five years (cf. Maratsos, Fox, Becker, & Chalkley, 1985) and German at five years (cf. de Villiers, 1984) – as well as Hebrew at eight

years (cf. Berman, 1985; Mills, 1985).57 In the case of certain types of passive constructions,

56 _{According to Demuth et al. (2010:240), this is also the case in Sesotho as neuter-containing constructions in} this Bantu language differ morphologically, syntactically and semantically from passive constructions in several ways. First, the form of the neuter morpheme, i.e. either -eh- or -ahal-, is distinct from that of the passive morpheme -w- or -uw-. Second, the neuter morpheme is infixed at the end of the verb stem before the perfective marker -il-, whereas the passive morpheme is infixed after -il-. Third, as in isiXhosa, the neuter-containing construction cannot include a copular noun phrase specifying an AGENT argument, unlike long passives. And fourth, whereas passive constructions have only an event reading, the neuter has an adjectival/state-like reading corresponding to the English interpretation expressed by the suffix “-able”; for instance, -ratwa means “be loved” and -rateha means “be lovable”. As mentioned above, this is also one of the possible interpretations of the isiXhosa neutro-passive.

57 _{Note that the majority of studies do not provide a clear description of what exactly having “acquired” the} passive entails, neither in terms of a minimum accuracy score on a passives test nor in terms of the skill type (for example comprehension versus production) that has purportedly been “acquired”. As such, all references in this dissertation to specific ages at acquisition of the passive are quoted in line with the specific authors’

115 ages as late as nine and eleven years have been cited (Horgan, 1978; Maratsos, Kuczaj, Fox, Becker, & Chalkley, 1979). For example, in a large-scale study by Horgan (1978) that employed 234 children aged between two to fourteen years, instrumental non-reversible passives (e.g. The window was broken by the boys) were reportedly only produced spontaneously after the age of nine years.

Whereas early research suggests that the passive voice is generally only fully acquired after the age of five years, more recent research indicates that this general estimate may be exaggerated and that children perhaps have knowledge of the passive voice at a “significantly earlier” stage (Deen, 2011:184). In some languages, knowledge of the passive appears as early as three years or younger (Alcock et al., 2011:459). Regardless of the exact age at which the passive is acquired in specific languages, however, it may still be considered a typically later developing construction across languages (Southwood & Van Dulm, 2012:48). So widespread is the scholarly interest in this general delay in acquisition that Deen (2011:155) claims the passive voice to be “arguably the most well-studied phenomenon in all of child language [research – AP]”.

The primary focus in most of these studies has been on potential reasons for the delay in the acquisition of the passive voice, resulting in various theories to account for this general phenomenon. For a detailed overview and appraisal of such theories, cf. Crawford (2012) and Deen (2011); for some of the explanations offered for the comparatively early acquisition of the passive voice in some languages, cf. Alcock et al. (2011). An inquiry into the possible factors that may determine the comparatively late or early acquisition of the passive voice in various languages falls outside the scope of the current study. However, one attempt at explaining the late acquisition of passives, namely that of Seymour, Roeper and de Villiers (2005), will be touched on here, as it is the one that was drawn on in the design of the passive construction section of the REALt (the instrument used to assess knowledge of passives in the present study). These researchers propose that passive constructions are typically late- acquired because of the following “hidden properties” that children must decipher: (i) there is always an AGENT argument, even if unspecified in the case of short passives; (ii) there is a

conceptualisations (albeit sometimes vague or arbitrary) of what constitutes the completed acquisition of passive constructions.

116 difference between activity and consequence (i.e. the activity described in The paper is torn leads to the consequential state described in The paper is torn); and (iii) there is a difference between an agentive by-phrase and a prepositional by-phrase (i.e. The cars were driven by the

men describes an action performed by the men, whereas The cars were driving by the men

describes an unspecified agent’s action of driving the car past the men).

Not only typically developing children, but also those with language impairment (the specific population targeted in the design of the REALt) often find both the comprehension and production of passive constructions difficult (Southwood & Van Dulm, 2012:49). Leonard, Wong, Deevy, Stokes and Fletcher (2006) highlight three possible reasons for this phenomenon. Firstly, passive sentences have an atypical word order in that the AGENT argument does not occur in the typical sentence-initial, pre-verbal position found in most active sentences, but occurs post-verbally as part of a prepositional phrase. This means that a passive sentence cannot accurately be interpreted in a typical linear fashion where the first noun phrase is considered representative of the AGENT argument and the second noun phrase representative of the THEME; such interpretations are, however, common among children with language impairment and among young typically developing children (Southwood & Van Dulm, 2012:48). Secondly, active and passive sentences differ in terms of verb morphology: in the active sentence Christine baked the cheesecake, the verb baked expresses the past tense, but in the passive equivalent The cheesecake was baked by

Christine, the past tense is expressed not by the main verb but by the auxiliary was, with the

main verb now serving to express the passive voice (Southwood & Van Dulm, 2012:49). A third reason offered by Leonard et al. (2006) in explanation of the trouble children with language impairment have with comprehending and producing passive constructions is the fact that the generation of a passive construction requires a relatively complex syntactic computation (cf. Leonard et al., 2006, for an accessible account of this computation).

As regards the different types of passive constructions, there appears to be a specific order of acquisition (Israel, Johnson, & Brooks, 2000). Both corpus studies and empirical studies that investigate the comprehension and production of passive constructions report that the acquisition of short passives precedes that of long passives, and that the acquisition of actional passives (i.e. passive constructions in which a physical action is expressed) precedes

117 that of non-actional passives (i.e. passive constructions containing perceptual verbs or psychological verbs) (Alcock et al., 2011:461; Crawford, 2012:5; Southwood & Van Dulm,

2012a:47).58 The tendency among monolingual English learners to produce predominantly

short as opposed to long passives has led researchers such as Horgan (1978) to suggest that

these learners’ short passives are in fact adjectival rather than verbal in nature.59

Recall that short passives allow both an adjectival and verbal reading in English, whereas long passives are unambiguously verbal. According to Alcock et al. (2011:459), true verbal passives seem to be fully acquired only by age six in the case of English monolinguals. If verbal passives prove more challenging to acquire for whatever reason, a preference for adjectival passives thus seems motivated, thereby explaining the predominance of short passives in early production.

As regards the order in which actional versus non-actional passives are acquired, Southwood and Van Dulm (2012) found evidence of a clear sensitivity to the distinction between psychological and perceptual verbs on the one hand and actional verbs on the other in the piloting of their REALt instrument among 57 typically developing monolingual English children (ranging from four to eight years in age) and 29 typically developing monolingual Afrikaans children (ranging from four to nine years in age). Southwood and Van Dulm (2012:53) report that in the case of both languages, those comprehension items that elicited incorrect responses contained either a perceptual or psychological (as opposed to actional) verb, or a “counter-intuitive AGENT-THEME relationship” such as the one in Thandi was

dressed by Bubbles, where Thandi is a girl and Bubbles a dog (the items with perceptual and

psychological verbs forming part of the reversible passives subsets – cf. below). Southwood and Van Dulm’s analysis of their data on the reversible passives production subset revealed a similar trend: across age groups, children performed worse on this subset than on the subsets targeting the production of “straightforward” actional passives, with the items including

58 _{Babyonyshev, Hart, and Grigorenko (2005) report a further distinction in the case of children with Specific} Language Impairment (SLI), namely that the acquisition of actional passives is followed first by the acquisition of passives containing psychological verbs and then by that of passives containing perceptual verbs. They state, however, that typically-developing children’s acquisition does not seem to be influenced by this fine distinction between verb types.

59 _{Borer and Wexler (1987) suggest that the argument chain underlying verbal passives matures relatively late,} causing adjectival passives to be acquired earlier in comparison. Their explanation is in line with a broader maturational account for the delayed acquisition of the passive voice according to which the “cognitive architecture” of passive constructions matures later than that of other constructions (Alcock et al., 2011:460).

118 perceptual and psychological passives proving the most difficult. For example, the responses to two items using the verbs upset and thought of / onthou (“remembered”) remained incorrect in the case of 35% of the English and 34% of the Afrikaans participants, even after the administrator’s follow-up stimulus (Southwood & Van Dulm, 2012:55).

As for reversible passives, this type of construction too has been shown to be especially vulnerable to late acquisition. Such passives contain two animate noun phrases (denoting the AGENT argument and the THEME argument, respectively) that are interchangeable, even if such an alteration renders the interpretation somewhat improbable, e.g. The cat was chased

by the dog versus The dog was chased by the cat. The creators of the REALt decided to target

reversible passive constructions in both comprehension and production as there has been agreement in the literature, over the past decades, that children with language impairment find this type of construction particularly challenging (cf. Bishop, 1979, as cited in Southwood & Van Dulm, 2012). This challenge is reportedly increased in the case of improbable scenarios, limited contextual cues and/or unfamiliar verbs (Van der Lely, 1994; Van der Lely & Dewart, 1986). Southwood and Van Dulm (2012:51) argue that a child who has truly mastered the comprehension and production of passive sentences will necessarily be able to comprehend and produce reversible passives, however improbable the scenario denoted by the sentence. For this reason, both probable and improbable scenarios were included in the choice of reversible passive sentences for this instrument.

The results of the piloting of the REALt instrument suggest that difficulty with reversible passives is not limited to children with language impairment. Southwood and Van Dulm (2012:51,55) found that among typically-developing children, performance on the subsets targeting the comprehension and production of reversible passives (including passive sentences with counter-intuitive AGENT-THEME relationships) was rather lower across age groups than on the subsets targeting “straightforward” actional passives (all displaying the expected/traditional AGENT-THEME relationship), with an apparent increase in accuracy with age. More specifically, in the case of English, the scores for the comprehension of reversible passives improved from 81% to 94% between the ages of four and eight, whereas the scores for the comprehension of “straightforward” actional passives improved from 90% to 98% over time. A similar pattern is found among the Afrikaans monolinguals: their scores

119 for the comprehension of reversible passives improved from 53% to 84% between four and nine years, whilst their scores for the comprehension of “straightforward” actional passives improved from 62% to 96%. In terms of production, too, reversible passives proved more difficult than “straightforward” actional passives for the English monolinguals: their scores for reversible passives improved from 67% to 83% between four and eight years, whilst their scores for “straightforward” actional passives improved from 75% to 96%. (Among the Afrikaans monolinguals, however, the production scores for reversible passives are not consistently lower than the scores for “straightforward” actional passives.)

Turning to monolingual English children specifically, research suggests that these learners generally take up to five years or longer to fully acquire the rules relating to passive constructions (Demuth et al., 2010:238). An early study by de Villiers and de Villiers (1973) tested English monolingual children’s comprehension of the passive using an act-out methodology in which participants were given both active and passive sentence prompts to act out using toys. The oldest group of participants, aged between 32 and 37.5 months, exhibited a correct response to 87.8% of the active sentence prompts, but to only 34.4% of the passive sentence prompts. In another early study, that by Baldie (1976), English monolingual children were shown capable of imitating passive constructions before five years, but only capable of comprehending them around the age of six years. The ability to produce this type of construction reportedly occurred as late as 7;6 in this sample of participants. According to Vasilyeva, Huttenlocher and Waterfall (2006:170), the fact that English monolinguals have been shown to be able to, at the age of four years, produce at least some full passives and on grounds of their responses in conversations with adults, comprehend at least some such sentences, one may conclude, “on the basis of the analysis of spontaneous speech, that sometime around the age of 4 years the basic elements of the passive construction have been mastered” by monolingual English children.

As regards the acquisition of passive constructions by monolingual Afrikaans children, there is, to my knowledge, no available literature providing normative data indicating the age at which the passive voice is typically acquired. Some indication may, however, come from Dutch (the language from which Afrikaans was largely derived) – in this language, “hardly any” uses of the passive have been noted in the speech of monolingual children of pre-school

120 age, i.e. of four years and younger, although there may be “precursors of the passive” at this age (Gillis & De Houwer, 1998:28,35). Southwood and Van Dulm’s initial testing of the REALt also provides some information (cf. above for participant numbers). Despite all their participants having a mid-SES background (contrary to the low SES background of the participants in the present study) and the sample sizes of their two groups differing largely, the results of these pilot tests provide an indication of the relative difficulty of English versus Afrikaans passives for typically developing children.

In the case of all items testing comprehension (i.e. the comprehension of long “straightforward” actional passives, short “straightforward” actional passives and long reversible passives, the latter including a mixture of actional, psychological and perceptual verbs) as well as the items testing the production of long or short actional passives, the Afrikaans monolinguals lagged behind their English peers across all age groups (Southwood & Van Dulm, 2012b:51-55). However, like the English monolinguals, the Afrikaans monolinguals showed an improvement on all four of the above measures over time and, in the case of items testing the comprehension of short actional passives, caught up to their English peers by the age of eight years. Also, in the case of items testing the production of reversible passives, the Afrikaans group managed to outperform the English group by the age of seven years, despite the younger Afrikaans groups consistently having fared worse than the younger English groups on this measure. On average, these REALt data thus seem to indicate that the passive is acquired later in Afrikaans than in English, although this difference may be overcome from the ages of seven to eight years onwards.

Let us now turn to the acquisition of passive constructions by monolingual child learners of the Southern Bantu language isiXhosa. The context-appropriate spontaneous use of the passive voice in the speech of children as young as three years has been reported for languages belonging to several different families, including (i) languages of the North American Inuit family (Allen & Crago, 1996); (ii) various Mayan languages (Pye & Quixtan Poz, 1988); (iii) Eastern Bantu languages, e.g. Kiswahili and Kigiriama (Alcock et al., 2011); and (iii) importantly, Southern Bantu languages, e.g. isiZulu (Suzman, 1985, 1987, 1990) and Sesotho (Demuth, 1989, 1990). With regard to verb types, non-actional passives (a typically rare construction in European languages) are reportedly produced at as early an age as 2;1 in

121 the case of Kiswahili and Kigiriama (Alcock et al., 2011:459). Three-year-old monolingual Sesotho learners are also capable of matching both actional and non-actional reversible passives to pictures, and of successfully generalising passive syntax to novel verbs without priming, as reported in Demuth et al. (2010).

The latter study consisted of three experiments: a picture-based comprehension task (Experiment 1), an elicited production picture description task (Experiment 2), and a novel verb generalisation / syntactic priming task (Experiment 3). Three different groups of 16 three-year-old lower to lower-middle class monolingual Sesotho children were employed, all of them having had little experience with looking at picture books (Demuth et al., 2010:241). In Experiment 1, 12 sets of two pictures each were employed, each picture containing a boy, a girl and a mother, half of the verb stimuli being actional verbs and the other half non- actional verbs. Each of the 12 picture sets was presented twice, once with an active prompt, and once with a passive prompt. Results revealed participants to have fared significantly better at comprehending pictures of active verbs (82%) than passive verbs (73%), as was expected, but that the use of actional versus non-actional verbs did not result in a significant difference in scores, despite the scores for actional verbs being higher (Demuth et al., 2010:242)

In Experiment 2, the goal was to test whether the participants could, upon being presented with a picture of an actional verb with patient-focused prompt (e.g. What’s happening to the

boy?), produce a (long) passive. The stimuli included 12 pictures depicting actional verbs,

each picture again containing a boy, a girl and a mother. Each verb was tested with both an AGENT- and PATIENT-focused question in order to elicit, respectively, active and passive answers. Of the agent-focused prompts, 95% elicited the production of an active verb, whilst 98% of the patient-focused prompts elicited a passive verb, there being no significant difference between these results (Demuth et al., 2010:245). At 25%, the proportion of the participants’ passives that constitutes long passives aligned closely with the 21% reported for spontaneous speech by Kline and Demuth (2008), verb type not affecting truncation rates. However, in a second round of testing, upon being encouraged to tell the complete depicted story to the experimenter who could not see what was happening in the picture, participants’

122 production of long passives increased to 91% in the case of the active recasts and 71% in the case of the passives recasts (Demuth et al., 2010:245).

Lastly, in Experiment 3, participants were presented with two novel Sesotho verbs, each verb being paired with a novel toy and two bean-bag dolls (a boy and a girl). The goal was to test