A-Movement in Language Development
Abstract and Keywords
This chapter provides an overview of the literature on children’s acquisition of constructions involving A(rgument)-movement: passive, unaccusative verbs, raising-to-subject, and raising-to-object. Considering A-movement within a derivational theoretical framework (GB/Minimalism), we provide some historical and theoretical context for treating these constructions under the same operation. In all cases, the surface position of an NP is incongruous with its syntactic configuration for receiving its thematic role. For each construction we discuss empirical evidence concerning children’s knowledge of the construction (including, where available, cross-linguistic data), and the major theoretical debates that have arisen around them, notably Maturation. We suggest that variability in experimental outcomes, both within and across constructions, can be linked to methodological choices and not likely to lack of linguistic knowledge.
Decoding the argument structure of a verb involves interpreting what roles are played by the NPs that appear with the verb. In some cases, this is a fairly straightfoward task, given the fact that a single verb will tend to consistently assign the same thematic roles to the same (local) positions. These patterns are particularly transparent in agentive-transitive verbs, where the subject serves as the agent of the action, and the object serves as the Patient/Theme of the action, as in (1).
Verbs come out of the lexicon with idiosyncratic patterns of θ-assignment, and in accordance with the θ-Criterion, these thematic requirements must be met at some initial stage in the construction of the syntactic representation (at D-structure, or in newer terminology, at Merge). Sentences like (1) retain a fairly transparent mapping between a verb and its semantic arguments.
However, some sentences do not retain this local configuration; in cases of A(rgument)-movement such as passives (2) or raising (3) constructions, this basic (p. 231) configuration may be deformed, resulting in non-local, nontransparent mappings between a verb’s idiosyncratic pattern of theta assignment and the final utterance.
It seems reasonable to assume that constructions like (2) and (3) would be problematic for children, for several reasons. First, as just noted, these utterances deform the canonical semantics–syntax mapping, by “moving” NPs away from where their thematic roles are assigned. Second, the patterns of θ-assignment themselves may be unusual; unlike typical verbs, raising verbs like tend and seem do not assign an agent or experiencer role (or any other role) to their subject and, as a result, one of the argument positions is left “empty,” semantically speaking. Thus, in (3), no thematic role is assigned to the matrix subject position. Meanwhile, in passives, a thematic role that is usually assigned—here, again, that of agent, usually assigned to subject—is lost.
In short, the number and arrangement of (overt) NP arguments is deformed in A-movement structures. Given the unusual status of this state of affairs with regard to the general patterns seen in agentive-transitive verbs, this may result in difficulty or delays for language learners. This is especially likely, given the fact that learners appear to exploit syntactic bootstrapping in learning (e.g. Gleitman 1990; Naigles 1990, 1996; Fisher et al. 1991; Gleitman et al. 2005). In order to use the number and configuration of arguments to deduce the meaning of a novel verb, learners must assume that thematic relations are local to the main predicate, and that the roles assigned to arguments are predictable based on general principles of thematic alignment (Perlmutter and Postal 1984; Baker 1988). As a result, the acquisition of A-movement is of particular interest to researchers working in child language acquisition.
In this chapter, we will discuss the acquisition of A-movement in first language, including the acquisition of passives, unaccusatives, and subject raising and raising-to-object constructions. Before we discuss these individual constructions in adult and child language, we will turn to the issue of A-movement at large, and consider what linguistic features and representations such movement entails.
12.2 A-Movement in Adult Grammar
Argument movement, or A-movement, is a term used to describe the movement of an argument (i.e. an NP/DP1 subject or object) from its base-generated (initial) position to some other c-commanding position in the syntactic structure.
(p. 232) The term “A-movement” subsumes several distinct processes and predicate types, including passives, unaccusatives, and raising verbs. To understand how A-movement generally works, we will need to make crucial reference to the notions of arguments and movement (including the motivations behind such movement), and so it is to these issues that we now turn.
Jaeggli (1986) notes that the term argument is used ambiguously in Chomsky (1981a). In one sense, an argument of a verb is any DP that holds a particular thematic/grammatical role for the verb: for example subject, direct object, etc. In another sense, arguments are simply referential expressions (NPs) that may hold thematic roles.
However, regardless of the sense in which the term is used, the notion of argumenthood (and by extension, cases of A-movement) crucially relies on ideas about thematic assignment and θ-roles—in particular, the θ-Criterion (discussed in greater detail in section 12.2.3). Jaeggli (1986: 590) in fact operationalizes the term by suggesting the following formulation: “X = NP is an argument of Y iff X is assigned a θ-role listed in the lexical entry of Y by Y or by a projection of Y.”
In the following sections we will present the most widely accepted movement-based analyses2 of these constructions; this discussion will focus primarily on the analyses developed in Government and Binding (GB) Theory (e.g. Chomsky 1981a, 1986) and the Minimalist Program (Chomsky 1995, 1998).
12.2.2 Movement in Government-Binding Theory
With the advent of the Revised Extended Standard Theory (Chomsky 1973), and carrying over into early Government-Binding theory (GB; Chomsky 1981a, 1986), the syntactic movements of elements as varied as NPs, PPs, and wh-elements were reconceptualized as instantiations of the broader rule Move-α. As a result, transformations (p. 233) like “passivization” and “raising,” which had previously been conceptualized as separate rules, were now subsumed under the same rubric, and subject to essentially the same syntactic considerations (as we shall discuss in the following two subsections).
GB (and by extention, Minimalist) theories of A-movement require a number of key assumptions about elements/modules that the grammar contains, including Move-α, the θ-Criterion, the projection principle, the Case Filter, and multiple levels of syntactic representation. These assumptions combine to form the basis of how GB conceptualizes the process of A-movement, and extend to comprise claims about the basic grammatical constructs required for the acquisition of linguistic representations involving such movement.
220.127.116.11 Movement: Levels of Representation, C-Command, Traces, and Chains
Government-Binding theory suggests that the grammar’s syntactic component compiles items from the lexicon, in a manner which satisfies their unique lexical requirements. This process forms “D(eep)-structures,” on which transformations apply to construct abstract “S(urface)-structures,” which are then given phonetic and logical/semantic representations at PF (phonetic form) and at LF (logical form), respectively (PF and LF being the “interface levels”). Thus, the GB model minimally requires the assumption of rules which can generate S-structures from D-structures, and those which can map S-structures to PF and LF (Chomsky 1981a).
Syntactic movement—what GB calls Move-α (Chomsky 1981a)—is thought to apply when some element would not be “interpretable” at one of the interface levels, if it were to remain in its current location. This situation occurs as a result of abstract features not being “checked” on the lexical item; the element must move to have its features checked, usually by some functional projection (e.g. TP, DP) elsewhere in the structure.
It is not logically required that movement be to c-commanding positions, but it appears that all applications of Move-α do result in such movement. This movement forms a “chain” between the moved element (in the cases under consideration here, the DP) and the “trace” it leaves behind in its base-generated position (this chain is usually represented via coindexation of the DP and the trace). A new (coindexed) trace is generated with each successive application of the movement rule, such that a single moved DP may leave behind multiple traces (if there are any intermediate landing spots). The movement chain (here, an A-chain) is thought to represent not only a kind of “locational history” of the moved element, but also a current relationship between that element and its previous locations; the moved DP retains access to the features and thematic assignment related to those positions.
The rule Move-α performs a wide range of functions, and subsumes previous notions of individual rules which moved only wh-elements, only NPs or PPs, etc. (Chomsky 1986). Because Move-α applies to so many constituents, it is clearly a heavily used rule. However, its application is not entirely lawless, and several general properties (p. 234) characterize it. For our considerations here, the most important constraint is that the antecedent (i.e. the moved DP) must originate in a θ-position, but it may not move into one; this stricture arises as a result of the θ-Criterion and the projection principle.3
12.2.3 The θ-Criterion and the Projection Principle
Movement is made possible (and constrained) given the notion of the θ-Criterion (“Each argument bears one and only one θ-role, and each θ-role is assigned to one and only one argument”4 Chomsky 1981a: 36). Because lexical requirements, including thematic assignment, must be satisfied at D-structure,5 any arguments will (or better put, must) be licensed to appear in the structure by bearing a θ-role before transformations (including Move-α) begin to apply. As a result, to avoid violating the θ-Criterion, DPs will always originate in a θ-position, and they may only move into positions to which no θ-role is assigned.
The issue of θ-assignment remains important even after D-structure (i.e. after movement transformations have begun to apply) due to the projection principle, which states that lexical requirements are maintained at every level of the derivation—that is, they are “projected” up through S-structure and LF.
Now that we have sketched a picture of where DPs might move from and to (that is, the beginning and end points of A-movement), we turn to the issue of why A-movement might occur.
12.2.4 Case Theory
Movement of DPs is constrained by θ-theory, but it is made necessary given Case theory. The Case Filter demands that every DP with phonetic content must have (abstract, if not morphological) Case. Only phonetically “null” DPs, like trace, escape this filter (Chomsky 1981a).6
(p. 235) DPs from the lexicon must be inserted into θ-positions at D-structure, but θ-positions are not always Case positions. When a particular θ-position is not a Case position, the DP must move to get Case. Given the θ-Criterion and Case theory, we end up with a system in which A-movement usually (if not always) occurs from a θ-position into a Case position.7
In short, every lexical (i.e. overt) DP must have one (and only one) θ-role, as well as abstract Case. θ-assignment happens at an initial stage in the derivation, and A-movement may apply to avoid Case Filter violations, when the DP’s D-structure position is not a Case position. This is precisely what happens in situations of passives, unaccusatives, and raising verbs. In the following sections we will discuss each of these phenomena in turn.
But first, let us turn our attention to more recent formulations of A-movement within the Minimalist Program, and how (and to what extent) these differ from the GB approach.
12.2.5 The Minimalist Program
Over the years, GB theory moved from having highly specific rules for individual phenomena (e.g. passivization, subject-to-subject raising) towards a system in which individual phenomena fall out from more general rules (e.g. “Move-α”). The Minimalist Program (MP; e.g. Chomsky 1995) constitutes a further step in the attempt to streamline or remove any parts of the theory which relate only to particular constructions or modules, leaving behind only the “conceptually necessary” components of the grammar (specifically, interface levels, and the elements appearing at each of these levels), and the computations needed to construct them (Chomsky 1995: 169). These refinements result in several theoretical notions which distinguish the GB and MP approaches to A-movement.
Perhaps the most important theoretical distinction between early GB accounts8 of movement and more current MP accounts relates to the circumstances under which movement may occur.9 Movement operations were initially conceptualized in the GB (p. 236) framework as applying essentially free of cost; there were no limitations on when and where movement could apply (see the formulation of “Move-α” in section 18.104.22.168).
However, later GB accounts, as well as accounts of movement in MP, have been formulated in a much stricter way. Especially in MP, derivations are hypothesized to be economical, and movement is considered to be a “Last Resort”: it occurs only when there is no other way to make a derivation converge (that is, be “fully interpretable” or grammatical). Moreover, movement is only licensed for a lexical item which has some uninterpretable abstract or overt feature (such as Case or agreement features) that must be checked (or “erased”), in order for the derivation or structure to be licit; in MP, this notion that elements move only to satisfy their own needs is referred to as the principle of “Greed.”
Most of our discussion of the empirical results on children’s acquisition of A-movement will be couched in GB terms, as this framework was the dominant paradigm at the time the research presented was undertaken. However, these results are generally also interpretable within MP, modulo (1) the assumption that movement must be motivated by Greed (and as a Last Resort), and (2) the basic terminological differences between the two frameworks.
In the GB framework, passive sentences such as (4a) below are generally assumed to derive from the form in (4b), where the subject position is underlyingly empty.10
Chomsky (1981a) proposes that passives have two crucial characteristics. First, [NP, S]11 does not receive a θ-role; thus, in accordance with the θ-Criterion, no NP may be present here at D-structure (in MP terminology, at Merge). Secondly, the passive suffix -en (p. 237) (e.g. writt-en, driv-en) absorbs12 the Case that the verb would otherwise assign to its complement. As a result, [NP,VP] cannot receive Case13 within the VP, and the complement undergoes A-movement to a Case position in order not to violate the Case Filter. These two characteristics result in a conceptualization of the passive in which the verbal morphology -en carries features usually assigned to NPs: namely, it receives a θ-role and Case.
Note that the passive suffix is being claimed to carry some semantic weight. First, the passive in (4b) may appear with an oblique by-phrase (by Mira) which expresses the agent of the action; these are sometimes called “full passives.” In such an oblique, the agent θ-role is transmitted to the by-phrase by the passive suffix, which may optionally subcategorize for this PP. Jaeggli (1986) claims that when the agent by-phrase does not appear, as in a so-called “short passive” (the peas were eaten), the verb’s external θ-role is carried by the suffix itself, resulting in an “implicit argument” (specifically, that denoting the agent of the action).
Second, the passive suffix –en is assigned the object Case of the verb, so that movement of [NP,VP] is forced by Case theory: the object NP must raise to receive its Case from the tensed INFL/T. This movement also allows the passivized subject to check the EPP (Extended Projection Principle) feature of INFL/T. Promotion of the lower NP into that position is possible with respect to the θ-Criterion, since [NP,S] is no longer a θ-position. This means that the structure of (4a) is that in (5).
Jaeggli’s account makes several other verifiable predictions. First, it correctly predicts that clausal complements, which do not require Case in English, may appear in their unraised position (with a concomitant expletive subject) in passives, as in (6).
And second, this account predicts that in some languages, passives may appear with post-verbal subjects, assuming that two conditions are met. First, the language may not be an EPP language: it must not require an overt NP subject.14 Second, the verbal complement must be able to receive (inherent nominative) Case in this post-verbal position (since the structural objective Case is absorbed by the passive suffix). While these conditions are not met in English, they both obtain in Spanish, and as a result, post-verbal subjects do appear in passives, as seen in (7) (from Jaeggli 1986).
12.3.1 Typology of Passives
Keenan (1985) notes that passives play a more essential role in some grammars than in others, and that still other languages lack passives entirely; furthermore, given that a language does use passives, there appears to be a markedness hierarchy of what kind of passives appear. The most widespread are basic passives (e.g. The peas were eaten): these have no agent (i.e. by-) phrase and are formed from a transitive main verb which (in the active) takes an agent subject and patient object. Keenan suggests that if a language has any passives at all, they will include basic passives. Indeed, some languages—including Latvian and some styles of Persian and Classical Arabic (Comrie 1977), Urdu, Amharic, Igbo, Shoshoni, and others (Siewierska 1984)—allow only these basic short passives; agents may not be expressed. Thus, although it has been suggested that other languages (e.g. Kota, Palauan, and Indonesian) require an agent phrase in the passive (Siewierska 1984), Keenan claims that the following entailment relationships obtain: if a language has passives with agent phrases, they will have them without; if it has passives of stative verbs, it will have passives of activity verbs; and if it has passives of intransitive verbs, it will have passives of transitive verbs.
Complex (i.e., non-basic) passives include passives of internally complex transitive verb phrases (8), passives of non-transitive verbs (including the subjectless “impersonal passives,” (9)), passives of ditransitive verb phrases (in which the recipient or the location is passived, (10)), and passives with non-Patient subjects (11) (these examples from Keenan 1985).
(p. 239) 12.3.2 Passives in Child Language: Empirical Results
In this section we review the literature on children’s comprehension and production of the passive construction. We begin with an overview of some older work in this area (on child English), then turn to more recent experimental work on this construction. After discussing several of the major theoretical analyses of children’s performance on the passive, we finally turn to developments based on cross-linguistic work. As we will see, the main issues addressed by work in this area concern (a) whether passives are in fact difficult (i.e. late acquired) for children, and (b) if so, why they are difficult.
22.214.171.124 Older Work: Passives are Difficult
One of the earliest investigations of children’s knowledge of the passive within modern linguistic theory was conducted by Fraser et al. (1963), who investigated children’s ability to comprehend, imitate, and produce a range of grammatical constructions. Among the ten constructions investigated was the passive. Setting the stage for much work to follow, Fraser et al. found that children aged 3–4 years (mean age of 40 months) displayed a great deal of difficulty with the passive construction. Out of 24 target items, only 12 (50 percent) imitations of the passive were correct, only 7 (29 percent) comprehension (picture identification) responses were correct, and only 2 (17 percent) productions of the passive were correct.
Slobin (1966) took investigation of the passive in a slightly different direction, both by examining knowledge of this construction in older children (kindergarteners through to 6th-graders) and by measuring children’s reaction times (RT) and errors in a truth-value judgment (TVJ) task. Children were shown a static picture and were presented with a sentence that was one of four types: affirmative active (what Slobin called “kernel”; (12a)), negative active (12b), affirmative passive (12c), or negative passive (12d).
In addition, Slobin constructed half of the sentences to be reversible, like those in (12) (in which both NPs are possible agents), and half to be nonreversible (in which only one NP is a possible agent), such as The girl is watering the flowers. Nonreversible kernel sentences, when passivized, become anomalous (#The girl is being watered by the flowers).
Slobin (1966) found that the youngest children made the most errors overall, but this rate was still fairly low (18.2 percent), showing that even these young children were not simply guessing in their responses. Slobin did not report the error rate according to sentence type, but RT measurements showed a significant advantage of nonreversibilility on responses to the passive sentences: subjects were significantly faster in responding to nonreversible passives than to reversible passives, and they made fewer errors on (p. 240) the nonreversible passives. In fact, response times (across subjects) to the nonreversible (affirmative) passives were slightly shorter (1.39 seconds) than response times to the nonreversible (affirmative) actives (1.45 seconds).
In the context of the syntactic theory being considered at the time (namely, Chomsky’s Syntactic Structures; 1957), Slobin was interested in evaluating whether the number of transformations within a structure had an effect on the time needed to comprehend the respective sentence, and the resulting accuracy (in line with previous psycholinguistic work by Mehler, 1963, and others). Since there is a syntactic transformation associated with formation of the passive, but none related to reversibility, the strong effect of reversibility could not be fully accounted for under that syntactic framework (neither could Slobin’s finding, consonant with others’ results, that negative stimuli were slower and more likely to lead to error than affirmative passives). Therefore, Slobin suggested that the facilitative effect of nonreversibility on passives stemmed from semantic or psychological effects, though he did not provide a more detailed analysis. He pointed out that some of the discrepancy between reversible and nonreversible sentences might be accounted for in terms of anomaly, but, interestingly, this cannot be the whole explanation. In particular, anomalous sentences (e.g. #The girl is being watered by the flowers / #The flowers are watering the girl) induced twice as many errors as the normal, non-anomalous nonreversible sentences (The flowers are being watered by the girl / The girl is watering the flowers). Slobin’s assessment was that “anomaly tended to confuse and slow down responding” (1966: 227). Therefore, some semantic effect other than anomaly appears to be a facilitating factor in the nonreversible sentences.
Returning attention to younger children and to reversible passives, both Bever (1970) and Maratsos (1974a) found that while children at age 3 (3;0–3;3) perform poorly in an act-out comprehension task (29 percent correct in Maratsos’s study), children around age 3;6 show considerable improvement (72 percent correct). But that improvement is short-lived. In the months that follow, children’s performance dips again (Maratsos’s children aged 3;8–3;11 were only 35 percent correct) before gradually rising back up (4;4–4;7-year-olds were 81 percent correct). This replicates the pattern reported by Bever (1970). The cause of this temporary dip in performance is not well explained and has not been discussed in more recent work. Maratsos suggested that as children grow, they become influenced by statistically more frequent constructions (i.e. the NVN active structure) and overgeneralize this interpretation to other constructions. In this way, the child’s expectation of the more frequent active construction interferes with interpretation of the passive word order (see Braine and Brooks 1995; Ambridge et al. 2008, among others, for a modern account along these lines). Finally, children overcome this frequency-driven bias and are able to correctly interpret passive sentences.
Horgan (1978) also reported a late development of the full passive construction. Horgan performed an elicitation task in which children aged 2–11 (and college-aged controls) were shown pictures and asked to describe them. Some of the pictures contained an agent or instrument, such as a boy or a ball next to a broken lamp, while others contained only the broken lamp. She found that the youngest children produced very few full passives but frequently produced truncated (i.e. basic/short) passives (which, (p. 241) Horgan argued, should not be analyzed as passives for the children, since they could be stative or adjectival). Among those children who produced full passives, some produced only reversible passives and others produced only nonreversible passives; children did not produce both types of passives until age 11.
Interestingly, Horgan’s argument is not that very young children are necessarily incapable of producing full passives (since some of the 2-year-old children she studied did so), but rather that the semantics of the passive takes a long time to acquire. Young children (aged 2–4) who produced only nonreversible passives produced constructions with an instrument as the logical subject (the lamp was broken by the ball), not an agent; these children did not produce nonreversible passives with an agent by-phrase until age 9. Thus, the passive construction is not completely productive from a semantic point of view. Horgan suggests that for children younger than 11, the passive is restricted to a non-agentive causation meaning, and might be analyzed by children on analogy to other constructions such as reflexives (I was hammering and I got hit = I hit myself with the hammer).
In summary, the early picture of the acquisition of the passive in English is that 3–4-year-olds have difficulty comprehending and producing the passive construction in general, children older than kindergarteners can comprehend it but are helped when the construction is “nonreversible,” and children produce very few full passives until age 10 or 11.
126.96.36.199 More Recent Work: Not All Passives Are Difficult
In the study by Maratsos et al. (1985), research on children’s knowledge of the passive turned to finer distinctions among verbs that are passivized (this study replicates findings by Maratsos et al. 1979). Specifically, Maratsos et al. (1985) contrasted children’s comprehension of actional versus non-actional (mental verb) passives.
This distinction is important, because it points to the fact that the external argument of a passive can be an experiencer rather than an agent, a fact that becomes relevant in later work on the passive (e.g. in Fox and Grodzinsky 1998).
In the first task, children were given sentences about Sesame Street characters and were asked “Who did it?” (E.g. Experimenter: “Grover is held by Ernie. Who did it?”) Answers were indicated verbally or by pointing to a finger puppet character. Although all children (4- and 5-year-olds) performed well on the active sentences containing both types of verbs (90 percent) and passive sentences containing actional verbs (67 percent—this number is clearly not at ceiling, but it is reportedly above chance), children performed poorly on the passive sentences containing mental verbs (40 percent, not significantly different from chance). There were no significant differences between the age groups, although, interestingly, the 5-year-olds were slightly worse on the passive mental verbs (35 percent) than the 4-year-olds (47 percent).
(p. 242) The second task was a picture-selection task with children aged 4–11 years. For the mental verbs, pictures included thought balloons, and for perception verbs (see, hear), one character in the picture was blindfolded or wearing earmuffs, while the other character had unobstructed eyes or ears. Maratsos et al. (1985) found that, similar to the previous experiment, while children in all age groups were statistically above chance for the active sentences and the actional passives, the 4-year-olds were below chance (34 percent correct) on mental passives. Children in older age groups were at or above chance on these items, but only children in the oldest group (10–11 years) were at ceiling.
Maratsos et al. (1985) reject the explanations that children either do not have the formal categories of subject and object, or that they do not hear enough instances of mental verbs used in the passive voice (although they do note the infrequency of this construction in child-directed speech). Even the 7-year-olds in the second experiment were at chance for non-actional passives, and Maratsos et al. estimate that neither of these explanations are plausible for children as old as 7. As an alternative, the authors suggest that the failure to comprehend mental passives has to do with a “general semantic constraint or condition” (Maratsos et al. 1985: 186). They mention semantic “transitivity” as a possible candidate, where a predicate that assigns an agent and a patient thematic role, such as break, would be high in transitivity, and a predicate that assigns an experiencer subject or object, such as scare, would be lower. Transitivity might correlate with passivization, such that children expect passives to be more likely (and therefore comprehend them better) with verbs that take an agent subject and a patient object, than with verbs that take an experiencer subject and theme object.
Borer and Wexler (1987) suggest an alternative explanation for Maratsos et al.’s asymmetry between actional and non-actional passives. Namely, they point out that action verbs, appearing in the passive without a by-phrase, are ambiguous between a verbal and an adjectival reading.
Hebrew distinguishes morphologically between adjectival and verbal passives, and Borer and Wexler note that children acquiring Hebrew acquire the adjectival form of the passive significantly earlier than the verbal passive (Berman and Sagi 1981; Berman 1986). Thus, they suggest, English-acquiring children likewise acquire the adjectival passive first—we just cannot distinguish the underlying difference on the surface. Therefore, when an English-speaking child responds correctly to a passive sentence such as (15), she is in fact analyzing it as an adjectival (or lexical) passive, which requires no NP-movement. Constructions requiring NP-movement (i.e. verbal passives) are acquired much later due to constraints on the biological maturation of these structures (see section 188.8.131.52 below).
While the foregoing research emphasizes children’s difficulty with the passive construction (notably in comprehension tasks), there is evidence that children in fact (p. 243) produce passives in their own speech from quite young ages, starting around age 3 to 3;6. Relevant here are the studies by Pinker et al. (1987) and Crain et al. (1987).
Pinker et al. (1987) report findings from both naturalistic speech (from CHILDES files) and experimental results involving elicited novel forms. The authors sought to distinguish between three possible explanations of how the passive is acquired: (a) children are conservative learners, producing and comprehending only those forms already encountered in the input (see Gordon and Chafetz 1990), (b) children are fully productive learners, allowing any transitive verb to be used in the passive, or (c) children are constrained productive learners, allowing only certain transitive verbs (those that take a patient object) to be used in the passive. The problem with option (b), the authors note, is that the passive is not in fact fully productive in adult grammar. Certain transitive verbs, such as fit, cannot be passivized (The suit fits John; *John was fit by the suit). Given that children do not receive negative evidence (Chomsky 1959; Brown and Hanlon 1970; Morgan and Travis 1989; Marcus 1993), if children were to adopt strategy (b), which generates a superset of the target grammar, it would be impossible for them to correct this assumption on the basis of positive evidence only: hence, it raises a learnability conundrum.
In their spontaneous speech sample (Brown 1973), Pinker et al. (1987) found 72 potential passives (counted liberally—i.e. including likely adjectival forms like named and crowded) in Adam’s data, and 32 potential passives in Sarah’s data. Of these, 25 percent of Adam’s and 22 percent of Sarah’s were “productive” in the sense that they were forms that were ungrammatical in the adult grammar and therefore could not have been modeled for the children, as in (16).
The fact that children produced passives they could not have heard in the input suggests that children are not conservative learners in the sense suggested in strategy (a) above: this construction is at least partially productive for them. Moreover, in their controlled experiments (see the following discussion), Pinker et al. (1987) found that children were able to produce passive forms of novel verbs which had been previously taught to the children only in the active voice, further supporting their claim that the passive is at least partially productive for children.
The authors conducted four controlled studies of both comprehension and production of novel verbs in order to see whether children are fully productive in their formation of the passive (allowing all transitive verbs in this construction) or if their production of the passive is limited to certain classes of verbs, for example actional verbs. Across the four experiments, Pinker et al. (1987) manipulated either the lexical meaning of the novel non-actional verbs (perception meaning or spatial meaning) or the θ-roles assigned by the novel verb (agent/location subject and theme object, or theme subject and agent/location object).15 They presented novel verbs to children in (p. 244) either active or passive voice, and then tested comprehension (by an act-out task) and production (by asking children to describe events). They elicited both the form the child had already heard (active, if the verb was presented in active voice) and the form the child had not heard (passive, if the verb was presented in active voice). Here we will focus on the production aspect of this study.
In eliciting the passive, the researchers were careful to establish the intended patient/theme as a topic in the discourse (e.g. “Here’s the elephant. Nothing’s happening to the elephant. Now something’s going to happen to the elephant. I want you to tell me what’s happening”; Pinker et al. 1987: 209).
Their main results can be summarized as follows:
• In the first experiment, which compared actional and perception verbs,16 children were more likely to produce passives for verbs that were modeled in the passive than for those modeled in the active voice (62% versus 59% for action verbs, and 94% versus 69% for perception verbs), a difference which Gordon and Chafetz (1990) note supports at least some verb-based learning of the passive. However, children produced novel productive passives with reasonable frequency (59% action verbs, 69% perception verbs), which Pinker et al. take to indicate the existence of some kind of productive rule.
• The difference between (productive) actional and non-actional passives (59% versus 69%) was non-significant and clearly did not involve a categorical ban on non-actional passives.
• In experiments 3 and 4, which examined children’s willingness to passivize verbs according to their thematic role assignment, results showed that children were much less willing to passivize “anticanonical” verbs (i.e. those verbs that assign a theme subject and an agent object role in the active voice17) than canonical verbs (agent subject, theme object). The researchers elicited 69% passives from active-taught canonical verbs, but only 28% passives from active-taught anticanonical verbs.
• Children easily passivize novel actional verbs that assign an agent subject role and a theme object role (canonical verbs; experiment 1) and are highly resistant to passivizing novel actional verbs that assign a theme subject and agent object role (anticanonical verbs; experiment 3); for verbs that assign experiencer/location and theme roles, children were somewhat more likely to passivize those that assigned their theme argument to the logical object, but the asymmetry was not as strong as for the actional verbs (experiments 2 and 4).
(p. 245) These findings suggest that if children apply a semantic constraint to the application of the passive rule, it is not one that limits the passive to verbs that take only an agent subject and patient object. That is, children readily produced passives for verbs that assigned an experiencer subject and theme object (i.e. the perception verbs in experiment 1). However, children were overall less willing to passivize certain types of verbs. Children were moderately less likely to passivize non-actional verbs than actional verbs, but not significantly so. Children were significantly less willing to passivize verbs that are considered “anticanonical,” that is, those that assigned a theme subject and an agent or location object. And children were somewhat less willing to passivize verbs that have a spatial meaning, as compared with canonical actional verbs, instead assigning a locative role to one of its arguments.
Strong resistance to passivizing anticanonical verbs follows if children’s rule for producing the passive hinges on the Thematic Hierarchy (Jackendoff 1972), which orders arguments within a sentence in terms of decreasing animacy. That is to say, the argument that is highest on the Thematic Hierarchy should appear as the subject of the sentence, while arguments that appear lower on the Hierarchy should surface as verbal and prepositional complements, etc. The argument is that the surface subject of a passive (i.e. the logical object) must be lower on the Thematic Hierarchy than the argument in the by-phrase (the logical subject), which would result in preferentially allowing animate agent-obliques, and inanimate theme-subjects, in passives. The milder but still notable resistance to passivizing spatial verbs observed by Pinker et al. suggests that children’s rule for passive formation is not one that applies only to actional versus non-actional classes of verbs, nor one that categorically excludes verbs that assign certain θ--roles. Rather, Pinker et al. (1987) conclude that children develop a productive rule for forming the passive that approximates a Thematic Hierarchy-driven preference for passivizing canonical verbs and a gradual dispreference for passivizing less canonical verbs. Verbs that assign an agent/experiencer subject and theme object are most easily passivized; verbs that assign a location subject and theme object are less so; verbs that assign a theme subject and an agent/location object are the least passivizable.
Another study that revealed early production of the passive in English-speaking children is that by Crain et al. (1987), who found that even quite young (3- and 4-year-old) English-speaking children could produce full passives in an elicited production task. As an example, the experimenter posed the following scenario:
The fact that the child uses a novel form (crashen) is evidence that the child is producing the construction based on a productive rule, rather than reproducing something modeled in the input.
Crain et al. (1987) argue that the act-out methodology masks children’s true early knowledge of the passive. As a comparison, Crain et al. gave an act-out and a picture (p. 246) verification task to their subjects, and found much worse performance on the act-out task (70 percent) than on the picture verification task (91 percent). They do not provide quantitative results from their elicited production task, but they note that 29 of the 32 children they interviewed produced at least one full passive, and 24 of the children produced three or more full passives. Thus, the authors argue that that by manipulating the pragmatics of the scenario, full passives can be elicited from children as young as 3;4, and that, consequently, the ability to form verbal passives cannot be lacking in young children due to maturational factors (see section 12.3.3 below).
It bears mentioning, however, that (a) all of the predicates used in Crain et al.’s study are actional verbs, not mental verbs (thus, they were the type of verb that younger children performed well on in the Maratsos et al. 1985 study), and (b) a high proportion of children’s passives were get-passives (as in (17) above) rather than be-passives (please see also Budwig 1990). As a result, we next turn to Fox and Grodzinsky’s (1998) analysis of get and be passives.
Fox and Grodzinsky (1998) argue against Borer and Wexler’s (1987) claim (see section 184.108.40.206 below) that children have difficulty with passives due to an inability to form A-chains. They note the finding by Crain et al. (1987), discussed above, that 3- and 4-year-olds produce passives in certain elicitation tasks, and further observe that many of these children’s passives were get-passives. Fox and Grodzinsky’s evidence that get-passives, like be-passives, involve an A-chain is based on the facts that get can intervene in an idiom chunk (18), and that it can take an expletive subject (19), two traditional tests for verbs that allow NP-movement (A-movement).
Thus, Fox and Grodzinsky (1998) argue, if children produce get-passives, those constructions involve A-movement and therefore A-movement cannot be subject to (late) maturation. Rather, Fox and Grodzinsky suggest that young children’s difficulty with passives has to do specifically with the by-phrase. They point out that children perform well in both comprehension and production tasks on either short (truncated) passives (John was kissed/seen), or full passives that have an actional verb (John was kissed by Mary). The problem case is full non-actional passives (John was seen by Bill). Fox and Grodzinsky conducted an experiment that tested all of these types of constructions.
Eight of the 13 children in their study showed the expected pattern: perfect performance on actives, on actional get- and be-passives, and on short non-actional passives. On the non-actional full passives, however, these children were at chance. (Of the remaining children in their study, two showed adultlike performance, and three were at or below chance on both truncated and full non-actional passives; the latter group is problematic for Fox and Grodzinsky’s account.)
Fox and Grodzinsky’s (1998) explanation of the results of their main group of subjects is that the by-phrase is problematic for children because of a process they call θ-transmission—a process not used in active voice sentences. In the derivation of the (p. 247) passive from the active form, the verb’s external argument is “suppressed” (since it is not obligatorily spelled out in the passive; Jaeggli 1986, and see the beginning of section 12.3). Fox and Grodzinsky (1998) argue that in the case of non-actional verbs, the external argument—which is in an adjunct position, and thus not governed by the verb—cannot get a θ-role without the mechanism of θ-transmission. Theta-transmission involves the by-phrase, which carries the θ-role belonging to the logical subject, transmitting that θ-role to the NP with which it appears. However, it is precisely this mechanism which is problematic for young children. In the case of actional verbs, the external argument is able to receive an agent (or affector) role directly, from the homophonous (non-passive) preposition by, thus rendering θ-transmission unnecessary. Fox and Grodzinsky (1998) conclude that this is why children selectively have trouble comprehending full (nontruncated) non-actional passives—in those constructions, the by-phrase argument requires an experiencer role, which cannot be assigned directly by the non-passive by, and therefore requires θ-transmission (please see Chapter 10 by Guasti in this volume for further discussion).
The studies discussed so far have focused on children’s grammatical ability to comprehend and produce passive constructions. But an experiment by O’Brien et al. (2006) manipulated the pragmatic conditions under which a by-phrase might be warranted. The authors point out that in a scene in which there is only one possible agent/experiencer argument, it is somewhat infelicitous to spell out the agent in a by-phrase; rather, a short passive is sufficient to accurately describe the scene. But if there are multiple potential agents, it is most felicitous to indicate which of those characters is the agent. For example, if Mickey is chasing Donald, and Pluto—another potential chaser—is also standing nearby, it is felicitous to describe the scene as “Donald is being chased by Mickey.” This method was employed in the Crain et al. (1987) study described earlier in this subsection; one of the chief contributions of the O’Brien et al. study is that they looked at nonactional verbs as well as actional verbs.18
The result of O’Brien et al.’s (2006) first experiment was that children (ages 4–4;10) were significantly above chance on all types of stimuli: long and short passives, with both actional and non-actional verbs. The authors do not state whether the difference in performance between long non-actional (82 percent) and short non-actional passives (100 percent) was significant.
In a second experiment, slightly younger children (mean age 3;6) were given long actional and non-actional passives in two conditions. In one of the conditions, only a single potential agent/experiencer was present, and in the other condition an additional agent/experiencer was present. By manipulating this factor (the number of potential agents or experiencers) within the experiment, the researchers could test for this effect directly. They found that the children in this experiment performed significantly above chance on both actional and non-actional passives in the condition with the extra agent/ (p. 248) experiencer, but not above chance in the other condition (single agent/experiencer), confirming their prediction.19
In summary, many of these studies indicate that passives are not problematic for children across the board. Instead, some show that certain types of verbs are more readily passivized than others (Maratsos et al. 1985; Pinker et al. 1987), or that by manipulating pragmatic conditions, children’s performance on passives (either in production or comprehension) can be improved (Crain et al. 1987; O’Brien et al. 2006). Nevertheless, pace Crain et al. and O’Brien et al., the received wisdom has been that before age 5, children are not uniformly correct in interpreting all types of passives, and therefore, in some sense, their grammatical knowledge of the passive is not quite adultlike. Thus, various theoretical accounts of children’s knowledge of the passive have been put forth. We turn to these accounts now.
12.3.3 Accounting for Passive Acquisition: Theoretical Approaches
Researchers have struggled to account for children’s acquisition of the passive, since this has appeared to be delayed with respect to that of the active. The major division within the generative literature has been between those researchers who believe passive acquisition relates to linguistic maturation, and those who do not. In the following sections we discuss the maturation account and alternative approaches.20
220.127.116.11 Maturation of A-Chains
The relatively late acquisition of the passive documented in languages such as English, German, and Hebrew prompted Borer and Wexler (1987, 1992) to propose the linguistic maturation hypothesis;21 according to this hypothesis, certain constructions in Universal Grammar (UG) (including those involving A-chains, such as the passive) are not immediately available to the child, but rather mature over time, just as do secondary sex characteristics. Structures relevant to the passive are assumed to mature around the age of 4;0.22 Before this age, the A-Chain Deficit Hypothesis posits that A-chains are (p. 249) ungrammatical for the child, and predicts that passives will therefore not appear in spontaneous speech, and will not be comprehended by children. The naturalistic and experimental findings of other researchers (e.g. Horgan 1978; Pierce 1992a) have been interpreted as providing support for the maturation hypothesis (Borer and Wexler 1987; Pierce 1992a).
However, other observations have cast doubt on this claim, requiring that Wexler and his colleagues reformulate their hypothesis. For instance, Borer and Wexler (1987) initially argued that pre-mature children have trouble forming A-chains of any kind. But the widely accepted VP-internal subject hypothesis (among others, Koopman and Sportiche 1991)—according to which all DP subjects are generated in SpecVP and then raise to SpecTP, thereby forming an A-chain—proved problematic for this conceptualization of the maturation hypothesis, since children seem to have no trouble with this type of subject raising (see also Fox and Grodzinsky 1998; Köppe 1994). Borer and Wexler (1992) relax their approach somewhat by suggesting that the only problematic A-chains are those relating two potential θ-positions: the so-called “(subject, object)” (Babyonyshev et al. 2001) or “non-trivial” (Chomsky 1995; Guasti 2002) A-chains.
A number of distinct approaches (discussed in the next section) have taken issue with the notion that components of UG may not be available to children, instead suggesting that some other issue—either grammatical or pragmatic—lies behind children’s “problems” with the passive. And more recently, work examining children’s acquisition of other types of A-movement (including subject raising and raising-to-object; see sections 12.5 and 12.6, respectively) support these non-maturational accounts by providing evidence that children have very little trouble with either trivial or non-trivial A-chains.
18.104.22.168 Non-Maturational Approaches
As we will see in section 12.3.4 below, a significant amount of cross-linguistic data has been presented which appears to disconfirm the maturation approach. But even before considering these results, researchers have suggested alternative accounts, which do not appeal to linguistic maturation, for the late acquisition of the passive in languages such as English.
For example, Weinberg (Berwick and Weinberg 1984; Weinberg 1987) argued for an account of children’s passive acquisition based on Markedness Theory. In a similar vein, several researchers have proposed that the “late” acquisitional status of the passive may only be apparent; specifically, naturalistic data may have been incorrectly interpreted. For instance, Pinker et al. (1987) and Crain and Fodor (1993) have pointed out that the corresponding scarcity of full passives in naturalistic adult speech is never interpreted as a lack of grammatical knowledge, but instead as evidence that the passive is simply a marked construction. The same could reasonably hold true of children’s speech.
Likewise, some experimental studies indicating late access to passives have suffered from methodological flaws, including—but certainly not limited to—the type of pragmatic infelicity addressed by O’Brien et al. (2006). Due to such experimental flaws, Crain and Fodor (1993) suggest that in many cases, the actual cause for children’s errors on experimental tasks is not a lack of linguistic maturity, but rather the result of (p. 250) nonlinguistic cognitive demands, including sentence parsing, the planning of responses, and pragmatic presuppositions. This nonlinguistic maturation hypothesis proposes that experimental linguistic performance improves over time due to the maturation of these extralinguistic cognitive abilities. Indeed, results from a number of other experiments have indicated at least partial, if not full, mastery of the passive by children younger than age 4 (seen in sections 22.214.171.124 above and 12.3.4 below). These results come from experimental designs that minimize nonlinguistic cognitive demands, provide felicitous pragmatic contexts for use of the passive, or even examine a language with a passive distinct in its characteristics from the English construction.
For instance, Borer and Wexler (1987) had specifically claimed that any passives which did appear in English-acquiring children’s speech at this young age were not verbal (syntactic) passives, but instead adjectival (lexical) passives; recall that the two are homophonous in English, but the latter involve no A-chain. Grimm (1973) found evidence that seemed to support this proposal. Although German verbal passives differ from adjectival passives in their choice of auxiliary (werden ‘become’ versus sein ‘be,’ respectively) and are thus not entirely homophonous, Grimm found that a common error in a repetition experiment was to replace werden with sein, which suggests that children perceive the two as similar, if not identical; this could be taken as support for Borer and Wexler’s adjectival-passive hypothesis. However, as detailed more fully in section 12.3.4 below, Demuth (1989) reports acquisition of the passive by age 2;8 in Sesotho, a Bantu language in the Niger-Congo language family, which has verbal—but no adjectival—passives. Moreover, Eisenbeiss (1993) reports for German that in a picture identification task, children aged 2;0 and older chose the correct picture 90 percent of the time for verbal passives, and there was even a strong tendency (70 percent) for 2-year-olds to incorrectly interpret adjectival passives as verbal passives. Likewise, in an elicited production task in the same study, even children younger than 4 produced verbal passives. In light of these facts, it seems unlikely that all early passives are lexical rather than syntactic, and it may prove that experimental design is to blame for the discrepancy between Eisenbeiss’s study and earlier claims about late acquisition of the passive in German. One possibility is that a sentence-imitation task (used by Grimm) may introduce nonlinguistic cognitive demands of the sort discussed by Crain and Fodor, while a picture-identification task (used by Eisenbeiss) reduces the cognitive load. In short, extra-linguistic cognitive demands may have masked German-speaking (and other) children’s true linguistic competence. Eisenbeiss’s data, however, do indicate that the distinction between sein passives and werden passives is not fully adultlike at this young age, as the 2-year-olds showed a tendency to misinterpret adjectival passives as verbal passives in the picture-ID task. This result is especially noteworthy as it directly contradicts Borer and Wexler’s claims that early passives are adjectival rather than verbal.23
(p. 251) Such evidence that children presented with pragmatically felicitous and experimentally controlled circumstances can comprehend and use the passive earlier than age 4;0 seems to provide clear evidence against the linguistic maturation hypothesis. Likewise, much of the aforementioned data has come from languages in the Indo-European family, but a growing body of evidence from children acquiring non-Indo-European languages indicates that much of the previously documented performance on passives is not a result of maturation (linguistic or nonlinguistic), but is rather dependent on language-specific factors. In section 12.3.4, we turn to this cross-linguistic data.
12.3.4 Cross-linguistic Findings on Acquisition of the Passive
Our understanding of children’s ability to produce and comprehend the passive has been enhanced by cross-linguistic work. Some cross-linguistic evidence has indicated a late acquisition of the passive similar to that claimed for English. For instance, Mills (1985) reports that German-acquiring children produced no full passives before age 6;0, that sein (adjectival) passives are more frequent in children’s productions than werden (verbal) passives, and that in comprehension tasks, German-speaking children (incorrectly) apply an active construal to passive sentences until about age 7 (that is, they interpret the first NP as an agent/experiencer). Berman (1986) also reports a very late acquisition of the verbal passive in child Hebrew (age 8), but an earlier acquisition of the adjectival passive, which is morphologically distinct from the verbal passive. Terzi and Wexler (2002) report that the verbal passive in Greek is acquired much later than the adjectival passive in that language.
On the other hand, several studies of non-Indo-European (and non-Semitic) languages have suggested a much earlier facility with this construction. In this section, we discuss some of these findings in detail.
Demuth (1989) presents data from children acquiring Sesotho (a Bantu language) showing that as early as age 2;8, children use the passive in spontaneous (non-imitative) production, and argues against Borer and Wexler’s maturation account of the English passive. Her argument is based on two claims: (a) Sesotho children produce this construction spontaneously at very early ages, and (b) Sesotho has no adjectival passive construction; all of these children’s passives are verbal, and therefore involve NP-movement to subject position (A-movement). Similarly, Suzman (1985) found that Zulu-acquiring children begin using the passive productively around age 2;6 to 3 years. (Zulu, like Sesotho, is a Bantu language; its passive construction has similar properties to those of the Sesotho passive, and it is similarly frequent in both adult-directed and child-directed speech.) The passive utterances Suzman observed at that age were non-imitative, although she notes that the contexts in which children uttered passives were semantically and pragmatically limited.
(p. 252) Demuth’s data include spontaneous utterances obtained as part of a longitudinal study of four children over a 2-year period. The overall proportion of passives is admittedly low, ranging from 0.4 percent of all utterances to 2.1 percent for the children (compared to adult caregivers, who produced passives in 6 percent of their spontaneous utterances in this sample), and some of the earliest passive productions may have been rote-learned forms (since at the earliest stage as many as 41 percent of the children’s passives were modeled on forms that were frequent in the discourse). However, Demuth notes that by age 2;8, children are producing passives quite productively (i.e. using non-rote forms), including full reversible passives, impersonal passives, and passives in both declaratives and interrogatives. Thus, Demuth argues, Sesotho-speaking children can form A-chains by age 2;8, considerably younger than the age proposed by Borer and Wexler.
It appears that very young Sesotho-speaking children produce verbal passives spontaneously. The implication of this finding is that if Sesotho children have A-chains, then perhaps at least some of English-speaking children’s passives are verbal passives too, and not simply adjectival passives. That is to say, even if the ability to form A-chains is something that matures in young speakers, if some Sesotho children give evidence of having reached this maturational stage by age 2;8, then presumably English-speaking children reach this stage at roughly the same age.
In more recent work by Demuth et al. (2010), Sesotho-speaking 3-year-olds were given a picture identification task (to test comprehension), as well as elicited production and novel verb generalization tasks (to test production). In the comprehension task, the children were better in responding to active than passive prompts (82 percent correct on actives, 73 percent correct on passives), though their performance on the passive prompts was still quite high compared to results reported for English-speaking children. The children also showed a slight (but non-significant) advantage of actional passives (77 percent) compared to non-actional passives (69 percent). Interestingly, adult controls performed significantly worse on non-actional than actional passives (89 percent versus 99 percent), but showed equal performance on actives versus passives overall. Demuth et al. suggest that worse performance on non-actional verbs, for both children and adults, stems from difficulty in achieving an accurate mental depiction of these predicates.
In the elicited production task, children were shown a picture of an action, asked to point to each of the characters depicted (e.g. a girl, a boy, and a mother), and were given a verbal description of the event (“This is kissing”) and then asked about the patient (“What is happening to the boy?”). In each picture a third character was depicted who observed the event but did not take part.24 This task resulted in at-ceiling performance: in 95 percent of agent prompts, children produced an active sentence, and in 98 percent of patient prompts, children produced a correct passive construction (although only 25 percent of their passives included a by-phrase).
(p. 253) In the novel verb generalization experiment, the child was taught two novel verbs by being shown a novel action (e.g. boy and girl dolls riding on a see-saw, so that one character slid down or fell off the see-saw). One of the verbs was modeled in the active voice and the other was modeled in the passive voice. For active-modeled verbs, the passive was then elicited by asking “What is being done to the boy/girl?” For passive-modeled verbs, an active form was elicited (“What is the boy/girl doing?”). The result was that all children successfully produced novel passives when asked patient-focused questions; 95 percent of these included correct passive morphology and 65 percent had a by-phrase.
The passive construction in Sesotho and Zulu (as well as in Quiche Mayan and Inuktitut, discussed in the following paragraphs) is formally akin to that in English and other Indo-European languages; that is, each involves A-movement operating under similar morphosyntactic demands. As a result, it cannot be argued that Sesotho- or Zulu-speaking children acquire the passive earlier due to the relative simplicity of the construction in their language (for instance, if there were no A-chain involved). Moreover, if children acquiring these languages show mastery of these forms at such a young age, it cannot be the case that A-movement only biologically matures after the age of 4.
What must be explained, then, is why English-speaking children appear to acquire the passive later than, for example, Sesotho-speaking children. One possibility is that they do not in fact acquire it later, and their early “adjectival” passives are in fact true verbal passives. Another possibility is that there is a true cross-linguistic asymmetry in the development of this construction, and it is linked either to typological differences between the target grammars or to effects of input frequency. Passives are used more frequently in Sesotho caregivers’ speech (6 percent) than in English-speaking caregivers’ speech (0 percent, according to Brown 1973; 0.036 percent according to Gordon and Chafetz 1990). In a study based on a more extensive language sample (98 hours), Kline and Demuth (2010) report that parents use passives in 2.7 percent of their utterances to children, and that 60 percent of those passives are full passives. Parents of English-speaking children produce by-phrases in only about 4 percent of their passives (Gordon and Chafetz 1990).
In addition to the frequency of the construction in speech to children, Demuth et al. (2010) suggest that another advantage for Sesotho-speaking children over English-speaking children is that the passive in Sesotho is morphologically unambiguous. That is, while the English short (i.e. agentless) passive is ambiguous between a verbal and adjectival structure, in Sesotho the two constructions are morphologically distinct, and this may provide cues for learners about distinguishing the true passive construction. On the other hand, it is worth noting that the verbal and adjectival passives in Hebrew are, as in Sesotho, morphologically distinct, and yet Hebrew-speaking children acquire the verbal passive at quite a late age, around age 8 (Berman 1986; Borer and Wexler 1987).
Allen and Crago (1996) discuss spontaneous speech data from four Inuit children acquiring Inuktitut natively. As in Bantu languages, Inuktitut’s verbal and adjectival passives are morphologically distinct. The children, aged 2;0 to 2;10 at the start of the nine-month study, produced verbal passives quite frequently, averaging about 2.8 passives (p. 254) per hour of recorded speech, or about 2.55 percent of their verbal utterances. (Based on Pinker et al.’s 1987 study of English from the CHILDES database, Allen and Crago report that English-speaking children produced about 0.4 passives per hour of recorded speech.)
Although the majority of the children’s passives are short forms (which include no overt agent) and involve an actional verb, Allen and Crago (1996) point out that (a) since Inuktitut is a null argument language, leaving out the agent is extremely natural in speech, and (b) children do produce more complex passive forms, such as full passives,25 passives with experiencer verbs (see, want), the “habitual” passive, passives containing a non-patient internal argument (i.e. subject), and passivized causative structures.
To explain the earlier acquisition of the passive in Inuktitut as compared to English, Allen and Crago point to the more frequent occurrence of the passive in adult caregiver speech: 7.8 passives per hour in adult Inuktitut versus 1.1 passives per hour in adult English (calculated based on data reported in Gordon and Chafetz 1990).26
In addition to frequency, Allen and Crago (1996) note other aspects of the grammar of Inuktitut that might contribute to their early acquisition. One is that, due to the highly inflected nature of the language, speakers tend to use strategies to avoid producing two-argument sentences, since the verb would then have to agree with both arguments. One way to avoid a two-argument structure is to use a passive, which requires agreement with only one argument; this is a strategy that may be used by both child and adult speakers of Inuktitut.
Allen and Crago (1996) suggest that another possible explanation for the early productive passive in child Inuktitut is that Inuktitut employs far more NP-movement than a language like English. Inuktitut is an ergative language, and according to some syntacticians, the object of a transitive verb must undergo raising in order to receive Case (Bittner 1994). Under this assumption, and given the assumption of the VP-internal subject hypothesis, both subject and object must raise to get case. In English—and given GB-theoretic notions of syntactic structure and movement—only the subject must raise from SpecVP to SpecIP to get Case. It should be noted, however, that under more current MP assumptions (e.g. Chomsky 1995), both subject and object DPs are hypothesized to move for Case reasons: subjects to AgrSP, and objects to AgrOP. If this syntactic (p. 255) analysis is on track, Allen and Crago’s second possible explanation could not account for the cross-linguistic differences in acquisition.
Pye and Poz (1988) report similar early acquisition of the passive in Quiche Mayan. In spontaneous speech samples, they found that children use the passive construction productively in their speech at roughly eight times the frequency of English-speaking children, starting from the age of 2;1. In a comprehension (picture-selection) task, they also found that their 4- and 5-year-old Quiche-speaking subjects were significantly above chance on actional passives, marginally above chance on non-actional passives, and (unexpectedly, given the English data) at chance on actional and non-actional actives. Pye and Poz note, however, that Quiche actives containing two third-person referents are ambiguous with respect to their theta-role assignment (i.e. agent versus patient), and that this may have impeded the children’s performance on those items.
While the evidence from Bantu languages, Inuktitut, and Mayan appears to point rather uniformly to early comprehension and production of the passive, the evidence from Japanese seems to indicate a delay in the acquisition of this construction. However, the cause of the delay remains a point of disagreement. Some authors have argued in favor of A-chain maturation as the explanation (Sugisaki 1999; Minai 2000; Machida et al. 2004), while others have argued against this account (Sano 2000; Sano et al. 2001; Okabe and Sano 2002).
Based on Miyagawa (1989) and Kubo (1990), Sugisaki (1999) argues for (at least) two types of passives in Japanese: gapped passives (also called “direct passives”), which involve an A-chain (shown in (20)), and gapless (or “adversity”) passives, which do not (shown in (21)).
Sugisaki (1999) provides experimental evidence that children acquiring Japanese (ages 3 to 5) may have acquired both types of passives or neither type, but if they have acquired only one, it will be the gapless passive. He concludes that the gapless passive is acquired earlier than the gapped passive, which involves A-movement, in support of the A-chain maturation hypothesis.
Since both full passives and full unaccusatives involve A-movement, Sano et al. (2001) argue that children’s difficulty with the passive is not due to problems forming A-chains per se, but rather to problems with the passive suffix -rare.
All these cross-linguistic findings have important implications for explanations of English-speaking children’s apparent difficulty with the passive. If children acquiring Language X have difficulty with a construction, while children acquiring a different Language Y produce and comprehend the construction much earlier, the apparent difficulty in Language X cannot be due to biological reasons. Babyonyshev et al. (2001) point out, however, that many of these studies assume, rather than demonstrate, that the passive constructions in these languages follow an English-like structural derivation. Thus, there may be room for debate over whether the earlier acquisition of passives in languages like Inuktitut and Sesotho constitute evidence against claims such as the A-chain maturation hypothesis.
Indeed, a number of arguments have been made that the cross-linguistic data discussed here do not necessarily speak against the A-chain maturation hypothesis. As pointed out by Hyams et al. (2006), Crawford’s (2004) reanalysis of Demuth’s (1989) Sesotho data argues for an “adversity passive” analysis of the children’s constructions, involving an applicative structure and no A-movement. Similarly, Johns (1992) argues that passives in Inuktitut involve a kind of nominalization and do not involve A-movement. Hyams et al. suggest, instead, that A-movement may in fact be problematic for children—but only when the application of this movement results in a non-canonical ordering of thematic arguments (their Canonical Alignment Hypothesis, or CAH). Specifically, they argue that children expect that a verb’s external argument (usually an agent) will map to subject position (SpecIP or SpecTP), and that children will experience difficulty when this alignment does not obtain; this is precisely what happens in verbal passive constructions, though not in unaccusatives or raising-to-subject constructions (discussed in section 12.5). In their own data on Malagasy, Hyams et al. (2006) report findings from three children studied longitudinally (age 1;6 to 2;8) indicating that the children produce a passive-like construction quite early. However, on the basis of other properties of child and adult Malagasy, Hyams et al. also argue that these constructions, in which a theme argument gets promoted to a subject-like position, actually involve topicalization, and therefore A′-movement instead of A-movement. The CAH will again become important when we discuss children’s acquisition of raising-to-object, in section 12.6 below.
12.3.5 Theoretical Approaches to Non-Passive A-Movement
While we have presented these maturational and non-maturational approaches in the context of children’s acquisition of the passive, similar accounts have also been (p. 257) proposed for each of the other types of A-movement that we will discuss in section 12.4. Given our discussion of A-movement in adult grammar, it should be clear that—just as the existence of a rule “Move-α” would suggest—all the types of A-movement make use of similar linguistic features and representations. As a result, it seems reasonable to assume that the acquisition of these grammatical elements should underlie the acquisition of each of these constructions. That is to say, there should be some “acquisition of A-movement” independent of, for example, acquisition of the passive, acquisition of unaccusatives, etc. However, such an assumption does not equate to the prediction that once one of these types of A-movement appears in the child’s grammar, all the other types should, as well. In some cases, other issues may be at play: for instance, MLU or working memory (and whatever other abilities these issues relate to) may be strongly implicated in the cases of longer construction types, such as raising-to-object.28
In general, though, the accounts of (non-passive) A-movement in L1A that have been suggested in the literature tend to fall into these two main subcategories (specifically, linguistic and non-linguistic maturation), and the reader is encouraged to keep these approaches in mind as they examine the empirical findings presented in this chapter. In the coming sections, we will return to accounts of the acquisition of other types of A-movement in more detail, especially in the cases where researchers have detailed new analyses that do not fit so neatly into one of these two general camps.
While a rich literature is devoted to children’s acquisition of the passive, less attention has been paid to other constructions involving A-movement, including unaccusative verbs, raising-to-subject, and raising-to-object constructions. We now discuss each of these constructions in turn.
12.4 Unaccusatives: Inherently Passive Verbs
Unaccusative verbs are those intransitive verbs which assign only a single internal θ-role; that is, semantically, they take only a patient/theme argument (Perlmutter 1978). According to the GB account of unaccusatives, due to Burzio (1986), if a verb does not assign an external θ-role it will also fail to assign Accusative Case (this is known as Burzio’s Generalization). As a result, the argument generated within the VP of an unaccusative verb must raise—for Case reasons—to SpecTP, to be assigned nominative Case. In this position, it acts as the subject of a sentence, including triggering agreement with the verb. Importantly, though, this promotion (p. 258) of the complement NP to SpecTP involves the same type of A-chain as that found in passives.
Thus, a sentence like (23a) has a representation as in (23b) (see also Guasti 2002). Unaccusative verbs in English include members like arrive, come, go, remain, descend, climb, run, and fall29 (Burzio 1986).
Unaccusatives contrast with the class of unergative verbs, a second type of intransitive. Unergatives, which include such agentive verbs as shout, talk, hide, hate, and think (Burzio 1986), differ from unaccusatives in that they project only a single external θ-role and—crucially, for our interest here—involve no A-chain (Pierce 1992b). This difference in underlying structures between unaccusatives and unergatives can be observed in the contrast in (24)–(25). Because the unergative argument is base-generated in subject position (rather than raising into this position, as is the case for the unaccusative), it may never appear post-verbally; however, the unaccusative argument may appear post-verbally (i.e. unraised) with an expletive subject.
(p. 259) 12.4.1 Unaccusatives in Child Language: Empirical Results
The studies we describe in relation to unaccusatives offer conflicting perspectives on children’s ability to represent A-movement. Snyder et al. (1995) argue that children can in fact represent A-movement, based on their correct production of auxiliary selection with past participles; supporting this claim are data from Snyder and Stromswold (1997), who present evidence that children use unaccusative verbs with preverbal subjects in spontaneous speech. Babyonyshev et al. (2001), on the other hand, argue that children cannot represent A-movement. Their data come from Russian-speaking children’s errors in applying the so-called “genitive of negation.”
Snyder and Stromswold (1997) used CHILDES files to examine 12 English-speaking children’s spontaneous productions of the verbs break, come, fall, go, grow, and leave, in unaccusative contexts. The authors report that children’s production of unaccusative verbs begins around age 2;1 (range: 1;6–2;7), and that all the children acquired unaccusatives before passives (with the gap ranging from 0–14 months, with a mean of 3.6 months, and a median of 2.7 months). The question is whether children correctly represent these verbs as unaccusative, or incorrectly represent them as unergative.
Evidence from Pierce (1989) suggests that English-speaking children do distinguish unaccusative from unergative verbs: when children produce intransitive utterances with a post-verbal subject, it is almost always with an unaccusative verb; unergative verbs are consistently used with preverbal subjects. More recent investigations of child Italian and Hebrew support this view: Lorusso et al. (2005) demonstrate that Italian children’s (age 2–3) subjects are more frequently overt and in post-verbal position when the verb is unaccusative than when it is unergative, and Friedmann (2007) shows that Hebrew-speaking children by the age of 2 years consistently allow unaccusative verbs to appear with both SV and VS orders, but limit unergative verbs to SV order. If these children correctly represent unaccusative verbs as having an underlying internal argument, then their productions of unaccusatives with a preverbal (i.e. raised) subject are instances of A-movement.
Further support for the claim that children can represent A-chains before age 4 comes from Snyder et al. (1995), who investigated children’s application of the auxiliary selection rule in Italian and French. In these Romance languages, past participles occur with auxiliary avoir/avere ‘have’ if they are transitive or unergative (see (27), (28), and (29)) and with être/essere ‘be’ if the verb is unaccusative (26). Although transitive verbs typically take have as their auxiliary, as in (28)–(29), when used as reflexives with an object clitic, these verbs take the auxiliary be instead of have (30). After Marantz (1984), the reflexive clitic construction is analyzed as involving an A-chain.
(p. 260) (27)
According to Snyder et al. (1995), children acquiring French and Italian make almost no errors on such auxiliary selection.30 Snyder et al. examined the auxiliary selection by one French-speaking child and three Italian-speaking children; Table 12.1 presents the results from Philippe (French, ages 2;1–3;3).
Although the Italian-speaking children Snyder et al. studied had fewer utterances overall, these children showed a similar facility correctly applying auxiliary selection with reflexive clitics in past participles. One of the children made three errors overall, two in which she used have with an unaccusative verb that had no clitic (out of 61 such utterances), and one in which she used have with a reflexive that had an object clitic (out of 34 such utterances). The other children made no errors.
Babyonyshev et al. (2001) present a different view of children’s acquisition of A-movement, by investigating the “genitive of negation” construction in Russian. Babyonyshev et al. point out that children’s spontaneous production of both (short) passives and unaccusatives is not unambiguous evidence that children have A-chains. They instead argue that children represent these constructions as alternative structures that involve no NP-movement, which they call S(yntactic)-homophones. In the case of passives, the child’s S-homophone construction is actually a lexical or adjectival passive, as argued by Borer and Wexler (1987). In the case of unaccusatives, Babyonyshev et al. suggest that children incorrectly represent these as unergative structures. Note that if children are representing unaccusatives incorrectly as unergatives, they are representing a theme argument (an underlying object) in an external argument (subject) position. This requires an assumption that children either violate, or do not have, Baker’s (1988) Uniformity of Theta Assignment Hypothesis (UTAH), or any similar universal restriction on the assignment of thematic roles to particular structural positions. (p. 261)
Table 12.1 Use of Aux be versus have with reflexive versus non-reflexive verbs
French; from Snyder et al. (1995)
The genitive of negation is a Russian construction in which indefinite objects of transitive verbs or (indefinite) internal arguments of unaccusative verbs are marked with genitive case instead of accusative case, when in the scope of negation. Some examples from Babyonyshev et al. (2001) are given below:
(p. 262) (38)
As illustrated in the examples, genitive marking is restricted to indefinite internal arguments within the scope of negation (including internal arguments in passives); definite or specific objects are marked with accusative case, and arguments of unergative verbs are marked with nominative. Babyonyshev et al. (2001) further point out that this construction is, in many ways, ideal for investigating the claims of the A-chain maturation hypothesis. One reason is that the internal argument of the unaccusative verb remains in the complement of the verb on the surface. The A-movement of that argument to subject position is covert—it occurs after Spell-Out.31 As a result, there is no S-homophone of this construction, so the A-chain maturation view predicts that this construction will simply be problematic for children—there is no other way for children to analyze the structure if they cannot represent A-chains. The second reason this construction is useful is that it is reportedly extremely frequent in adult speech (though Babyonyshev et al. do not provide frequency data). Thus, if children do have difficulty with this construction, their difficulty cannot be explained by a lack of exposure to the construction (as is sometimes argued for passive; see section 12.3.4).
Babyonyshev et al. (2001) predict that if Russian-speaking children have acquired the genitive of negation construction, but cannot form A-chains, they will allow genitive case on the negated object of a transitive verb and disallow genitive on subjects of unergative verbs, but—crucially—they will fail to allow genitive case on the argument of a negated unaccusative verb.
Using a sentence-completion task, the experimenters presented Russian children (ages 3;0–6;6) with brief stories involving a character who performed some action on various objects. After the story, a puppet summarized what happened but left out the object, which the child was then prompted to provide. Multiple object referents were provided in the story context in order to make indefinite reference pragmatically felicitous. For example, in one story a cat has some houses and bicycles that he will paint. He paints two houses but does not paint any bicycles. At the end the puppet says, “I know what happened. The cat painted two houses and didn’t paint …” The child should then complete the sentence with an expression referring to the object. Since Russian allows variable word order, subjects of unergative verbs (as well as objects of transitive verbs and arguments of unaccusative verbs) may occur post-verbally.
The results of this study are reproduced in Table 12.2.
The difference in the responses to unaccusatives versus nonspecific object of transitives (73 percent versus 45 percent) was significant. The fact that children consistently marked the nonspecific object of transitive verbs with genitive but never did so with (p. 263) subjects of unergatives (and virtually never with specific objects of transitives) suggests that the children studied do know this construction and the semantic and syntactic constraints on its application. Thus, the fact that the children failed to consistently mark the argument of unaccusative verbs with genitive case is consistent with Babyonyshev et al.’s prediction, and their explanation that children cannot represent A-chains. Instead, the authors conclude, children misrepresent unaccusatives as unergative structures. Babyonyshev et al. (2001) note further that if their child subjects are divided into an older group (mean age 5;4) and a younger group (mean age 4;0), the older children do mark the unaccusative arguments with genitive more frequently than the younger children do.
Table 12.2 Results of Babyonyshev et al. (2001)
Percentage of genitive responses
Nonspecific object of transitive
Specific object of transitive
Subject of unergative
Argument of unaccusative
We have seen three studies of children’s production of unaccusative structures with apparently conflicting results. Snyder and Stromswold (1997) provide evidence from spontaneous speech indicating that English-acquiring children have productive use of unaccusatives like break, come, and leave around age 2;1. Similarly, Snyder et al. (1995) show that children acquiring French and Italian produce correct auxiliary selection, indicating a correct analysis of verbs with reflexive clitics as unaccusatives; recall that the A-chain maturation hypothesis predicts that these children should overgeneralize have as the auxiliary with these past participles. On the other hand, Babyonyshev et al. (2001) show that Russian-speaking children fail to mark objects of unaccusatives in the scope of negation with genitive case, as is required by the adult grammar. This suggests that children incorrectly analyze these structures as unergative. How can these two sets of findings be reconciled?32
Babyonyshev et al. (2001) suggest an alternative interpretation of the A-chain maturation account—which they call the External Argument Requirement Hypothesis (EARH)—which would be consistent with both the Russian data and the French/Italian data. According to the EARH, the problem with A-chains is not the movement per se, (p. 264) but the absence of an external argument in the underlying form of passive and unaccusative constructions. If children required there to be an external argument projected at all levels of representation, they would misanalyze unaccusative constructions as unergative structures.
The EARH is consistent with Snyder et al.’s (1995) findings because unlike other unaccusative constructions, Romance reflexive clitic constructions do in fact project an external argument. The EARH is also consistent with the Russian genitive of negation findings, since these constructions do not involve an external argument. Therefore, if the problem for children lies in requiring an external argument to be projected, rather than forming the A-chain itself, they should perform poorly on genitive of negation in Russian but well on auxiliary selection with Romance reflexive clitics. However, if English-acquiring children actually do have an adultlike unaccusative analysis of the verbs examined by Snyder and Stromswold (1997), the EARH will not suffice to account for the facts.
We should point out that Snyder and Stromswold (1997: 294, fn. 18) explicitly argue against the hypothesis that children initially apply an unergative analysis to unaccusative verbs. They note that at least in English, the primary linguistic input is unlikely to correct such a faulty analysis, and suggest that children must exploit “the interaction of UG linking rules” and semantics to arrive at the correct unaccusative analysis. The evidence from auxiliary selection in Snyder et al. (1995) also suggests that children are analyzing these verbs in an adultlike way. Thus, the contrast between the Russian data and the English/Romance data may be unresolved.
In the next section we will see that the EARH is not compatible with Hirsch and Wexler’s (2007) account of raising-to-subject constructions in English. It is to these raising constructions that we now turn.
Constructions like that in (39a) are variably called raising-to-subject, (subject-to-)subject raising, or simply raising constructions. Because these sentences alternate with “unraised” counterparts like (39b), even early (pre-GB) analyses assumed that the NP John in (39a) is base-generated in the embedded clause, but “raises” out of that clause to become the matrix subject at later stages in the derivation (Rosenbaum 1967; Postal 1974).
Chomsky (1981a) suggests that, due to a lexical idiosyncracy, predicates like seem and likely assign no external θ-role to their subjects. As a result, given the strictures of the (p. 265) θ-Criterion, no NP will be generated here at D-structure, and this subject position will be available as a landing site for movement.
All NPs require Case, and subject NPs usually get Case from tensed INFL/T in their clause. However, if the embedded INFL is [–Tense] (i.e. infinitival T, as in (39a)), it will be unable to assign (nominative) Case to its subject; in this situation, the subject NP in the embedded clause will be forced to raise to matrix subject position for Case, thereby satisfying the Case filter (and, in a language like English, also checking the matrix EPP feature). (In the unraised version, (39b), the embedded INFL is [+Tense]—finite—and thus able to assign Case to the embedded subject. Under these circumstances, an expletive is inserted in matrix subject position to check an EPP feature.)
It should be explicitly noted, however, that just as in other cases of A-movement (such as passives and unaccusatives), the moved NP receives its θ-role in one position (its D-structure position, or—under MP assumptions—its position at Merge), and moves to receive its Case feature in another position (its S-structure, or Spell-Out, position).
12.5.1 Raising-to-Subject in Child Language: Empirical Results
Both Wexler (2004) and Hirsch and Wexler (2007) provide evidence that the raising-to-subject construction, like that in (40), develops late—around age 7. They attribute this late development to the late biological maturation of A-chains. On the other hand, Becker (2006, 2009) argues that children as young as age 4 can correctly interpret raising constructions. We will examine each of these claims in this section.
Wexler (2004) reevaluated the A-chain Deficit Hypothesis (ACDH; Borer and Wexler 1987, 1992) in terms of phasal theory in Minimalism (Chomsky 2001). Rather than defining the constraint on child grammar in terms of the grammaticality of forming an A-chain, the Universal Phase Requirement (UPR) proposes that in child grammar, υ (the light verb) always defines a phase, regardless of whether υ is “defective” or not.
In Minimalism, phases are thought to be the smallest units of syntactic processing; they are “semantically complete” units33 containing a verb, its arguments, and C “as the domain of force plus proposition” (Wexler 2004: 164). Chomsky (2001) suggests that the derivation of any given expression proceeds phase by phase. The idea here is that each phase is placed individually in active/working memory, where features are deleted as necessary; afterwards, it is routed to the phonological component, where deleted features are removed from the narrow syntax so the derivation can converge at LF. Importantly, though, once a phase has been through the derivation cycle, no further (p. 266) changes can occur within the phase; only the “edges” of the phase (usually specifiers) are available to other operations higher in the syntactic structure.34
Normal (nondefective) υ assigns an external argument. In the case of a normal/nondefective υ, the complement to υ (i.e. VP and its complements) is not visible to operations at the higher phase, which is headed by C. This means that while the specifier of υ may move up into the higher phase (as happens to the subject, according to the VP-Internal Subject Hypothesis), internal arguments of the verb cannot. However, in the case of a defective υ phase—that is, one in which no external argument is assigned—the object (complement of V) is free to move up into the higher phase. Such is exactly the case with passive and unaccusative verbs. Wexler argues that the same applies to constructions involving raising-to-subject verbs, like seem.
According to UPR, however, children (unlike adults) take υ to operate as the head of a phase regardless of the type of verb involved (i.e. defective or not). The result is that internal arguments within the VP complement of that υ will never be available for movement into the higher phase. Thus, the UPR predicts that children should be able to represent constructions with an expletive subject and a raising verb, as in (41a), but should not be able to represent constructions in which the subject of the lower clause has raised to the matrix clause, as in (41b).
Recall from the discussion in section 12.4.1 that the EARH, which Babyonyshev et al. (2001) invoke to reconcile their results with those of Snyder et al. (1995), predicts that children should have difficulty with both constructions.
To evaluate children’s comprehension of these constructions, Hirsch and Wexler (2007) conducted a picture selection task with children aged 3–9. Children were given sentences such as (41a) and (41b) above, plus the two types of sentences in (42) and (43).
In the cartoon pictures, thought bubbles were used to depict both “thinking” and “seeming.” The non-matching picture in the think and seem conditions differed from the correct picture in one of a number of ways:
• Matrix-reversal (MR) pictures depicted an incorrect character doing the “thinking.” For example, this foil for sentence (43) above would show Bart both playing an instrument and thinking about Lisa.
(p. 267) • Embedded-reversal (ER) pictures depicted an incorrect character performing the action of the embedded predicate. This foil for (43) would show Lisa playing an instrument and thinking about Bart.
• Double-reversal (DR) pictures reversed the characters doing the thinking and the acting. This foil for (43) would show Bart thinking about Lisa, and Lisa (in the thought bubble) playing an instrument.
Hirsch and Wexler report that while all children were at ceiling on the active control sentences like (42), and performed very well on the think sentences (43) and unraised (41a) sentences, children younger than age 7 were at chance on the raised sentences (41b). In fact, only the 9-year-olds scored above 90 percent for these constructions. Hirsch and Wexler’s (2007) combined results are reproduced in Table 12.3.
Breaking down children’s errors according to foil type, Hirsch and Wexler (2007) found that with the think and unraised seem items, children had the hardest time rejecting matrix-reversal (MR) foils (those in which the agent of the embedded clause is also—incorrectly—represented as the agent/experiencer of the matrix clause). It was mainly the youngest children who had this difficulty, and the authors suggest that an undeveloped Theory of Mind, or a failure to understand the lexical meaning of the verbs think or seem, may play a role in these errors.
For the raised seem items, however, children showed a different error pattern. With these items, children made the most errors with double-reversal (DR) foils: they chose the picture in which the experiencer of the matrix clause was performing the embedded clause action, and the agent of the embedded clause was doing the thinking. In the DR conditions, children are not just at chance; rather, they are significantly below chance, suggesting a strong preference for the DR foil over the correct picture. (With the other foil types, children are not below chance.)
Table 12.3 Results from Hirsch and Wexler (2007)
(p. 268) Hirsch and Wexler (2007) interpret this result as meaning that because children’s grammar cannot represent a raising structure, children interpret raised seem sentences as if they were think sentences. That is, they interpret a sentence like (41b) above, repeated here as (44a), as if it meant (44b).
Becker’s (2006, 2009) experimental work suggests a somewhat different picture of children’s facility with raising constructions.35 In Becker (2006), 3- to 5-year-old children were shown a cartoon picture that depicted characters and normal inanimate objects (i.e. inanimate objects were not made “more animate” by adding eyes, etc.), and a puppet uttered a description of the picture. The child’s task was to say whether the puppet’s sentence was “OK” or “silly” with respect to the picture. The puppet’s sentences were constructed to contain an inanimate subject, either a raising or a control verb (seem, appear, want, try), and an embedded predicate that was either semantically compatible or incompatible with the matrix subject. (See Table 12.4 for examples.)
This design was driven by the fact that control verbs select for animate subjects while raising verbs do not project any subjects; therefore, raising verbs, but not control verbs, are free to occur with an inanimate matrix subject. For both classes of verbs, however, the matrix subject is semantically linked to the embedded predicate, and therefore this predicate must be semantically compatible with the matrix subject. The prediction was that if children distinguish the classes of raising and control verbs correctly, they should reject all sentences containing a control verb since there is a semantic clash between the control verb and the inanimate matrix subject; conversely, children should accept the sentences containing raising verbs, as long as the embedded predicate was semantically compatible with the matrix subject. If children do not distinguish the verb classes and interpret all verbs as control verbs, they should reject all items; if they interpret all verbs as raising verbs, they should accept sentences with a compatible lower predicate and reject sentences with an incompatible lower predicate. Example items and their types are given in Table 12.4.
The older and younger children in the study behaved differently. Five-year-olds consistently rejected sentences containing a control verb, and accepted only those raising sentences that contained a compatible lower predicate, indicating an adultlike distinction between the two verb classes. The 3-year-olds, however, did not appear to discriminate the two verb classes. In both cases, they accepted or rejected the sentence on the basis of the compatibility of the embedded predicate. Thus, these children accepted (p. 269) sentences such as The flower wants to be pink/The hay seems to be on the ground since flowers can be pink and hay can be on the ground, but they rejected The flower wants to fly away/The hay seems to be excited since flowers cannot fly away and hay cannot be excited. Four-year-olds showed an intermediate performance between the younger and older children.
Table 12.4 Test items in Becker’s (2006) experiment
The flower wants to be pink
The flower wants to fly away
The hay seems to be on the ground
The hay seems to be excited
(*) compatible = subject and lower predicate are semantically compatible
(**) incompatible = subject and lower predicate are semantically incompatible
Becker’s interpretation of these results is that younger children correctly interpret raising structures as involving a semantic relationship between the matrix subject and embedded predicate, but not between the matrix subject and matrix verb; however, they make the error of allowing control verbs to have this same type of interpretation. A follow-up experiment showed that when children are given raising or control sentences with an animate subject, they correctly interpret both types of verbs. In later work, Becker (2009) argues explicitly that it is the animacy of the matrix subject that drives children’s interpretation of the sentence as involving a raising (A-movement) or control structure.
The results of this study leave open the possibility that the younger children do analyze control verbs as selecting the matrix subject, but they do not yet place a semantic restriction on that subject requiring it to be animate. In Becker (2009), children’s comprehension of raising and control verbs with expletive subjects (weather it) was tested. Expletive subjects differ from inanimate referential subjects in that they do not refer to anything. Therefore, they can occur only with predicates that do not assign a θ-role to that position (in this case, subject position). That is, a sentence like (45) must involve a raising predicate.
Three- and 4-year-old children were shown pictures of weather events, and a puppet commented about each picture using either a raising or a control verb. The child was asked to say whether the puppet’s sentence was “OK” or “silly.” Example prompts are given in (46)–(47). (p. 270)
Table 12.5 Responses in Becker’s (2009) experiment (percent judged acceptable)
The mean percentage of “OK” responses is given in Table 12.5.
Both age groups were significantly above chance in accepting the raising weather sentences like (46a), suggesting an adultlike interpretation of these sentences. In their responses to the control weather sentences like (46b), 3-year-olds, but not 4-year-olds, were significantly above chance in their OK responses, suggesting that on average, they considered control verbs to be compatible with an expletive subject. Moreover, 4-year-olds, but not 3-year-olds, showed a significant difference between their proportion of OK responses to raising versus control items, suggesting that 4-year-olds are beginning to distinguish these two verb classes.
Let us now consider how the differing claims of Hirsch and Wexler (2007) and Becker (2006, 2009) might be reconciled. Hirsch and Wexler point out that children in the Becker (2006) study may have performed as they did because they do not yet have an adultlike concept of animacy, such that they allow inanimate objects to want or try to do things. Although Becker (2009) provides arguments against this possibility, young children’s conception of the world—and the effect this might have on their interpretations of sentences—is important to consider. It is possible that children’s grammaticality judgments are influenced by alternative ontological possibilities, and accordingly that they respond to these tasks differently than adults would.
On the other hand, there are various methodological differences between the Hirsch and Wexler (2007) and Becker (2006) studies that could account for their different results. Hirsch and Wexler employed raising sentences that contained an overt experiencer argument (to Lisa in Bart seems to Lisa to be playing an instrument), while Becker’s test sentences lacked this phrase. Although Hirsch and Wexler argue that the inclusion of the experiencer argument does not make the sentence harder to parse (based on processing data from adults), it is possible that this phrase, when overt, does make the processing load high enough for children that they cannot correctly parse the sentence. To (p. 271) our knowledge, this has not been tested within a single experiment, though we do know that increasing the number of animate NP referents in a sentence may significantly increase the processing load (Goodluck and Tavakolian 1982; see also Orfitelli 2012).
Another way to interpret Hirsch and Wexler’s results is to consider the relative ordering of thematic roles in a raising sentence with an overt experiencer argument:
While it is not problematic, in principle, to have a Theme subject (as is the case in a regular unaccusative sentence, like The rock fell), when more than one argument is present, the Thematic Hierarchy requires the argument higher on the hierarchy (in (48), the experiencer) to appear in subject position (Jackendoff 1972). However, in (48) it is the argument that is lower on the Thematic Hierarchy (the theme) that is in subject position. Along the lines of the claims made by Hyams et al. (2006; see discussion in section 12.3.4 above), it may be that when the canonical ordering of arguments is disrupted, the sentence becomes difficult, if not impossible, for young children to parse. These possibilities must be explored in future research.
Like raising-to-subject constructions, raising-to-object (R-to-O) constructions are also biclausal. However, the latter constructions have an overt embedded subject that is distinct from the matrix subject.
Whereas the raising analysis for subject raising constructions like (39a) has been fairly uncontroversial, two major approaches to what we will call R-to-O constructions have appeared in the literature, only one of which actually involves movement (for a more detailed discussion, see Kirby et al. 2010 and the references therein).
The R-to-O analysis (Postal 1974; and more recently, Lasnik and Saito 1991) parallels the subject raising account detailed in section 12.5 above; if the embedded clause is untensed, it will not be able to assign nominative Case, and so the embedded subject (in (49), Mary) will have to move to avoid violating the Case Filter. Because the matrix verb assigns no θ-role to its complement (i.e. in (49), there is no “expectee/believee”), matrix object position is available as a landing site (in accordance with the θ-Criterion), resulting in a structure like that in (50).
(p. 272) In later GB models, as well as some more current MP models, “object” position is SpecAgrOP, where Case features are assigned (or checked) for objects (51). Linear order is achieved by the matrix verb raising to a functional head above this projection—possibly TP or υ P—and the matrix subject raising out of its base-generated, VP-internal position and up to SpecAgrSP, also for Case reasons.
That the embedded subject indeed receives accusative (i.e. object) Case becomes obvious when a pronoun occurs in this position (52). As a result, proponents of this approach sometimes call this construction subject-to-object raising, for obvious reasons.
Anaphors may appear as the embedded subject in such constructions, and a pronominal embedded subject may not be coindexed with the matrix subject (53). Given Principles A and B, this fact is taken by proponents of the raising analysis to indicate that the embedded subject does indeed move to the matrix clause. (In (53), subscripts represent binding indices, rather than co-indexation based on movement).
(p. 273) Likewise, the fact that matrix material can appear between the embedded subject and its predicate, as in (54), has also been used to support the raising analysis. (See Lasnik and Saito, 1991, for details on these and other arguments in favor of a raising analysis.)
However, an alternative approach to such “raising-to-object” constructions instead suggests that no movement whatsoever is involved. Chomsky (1981a) argues for an “exceptional case-marking” (ECM) account, on the basis of his assumption that all subcategorized positions (including objects) are θ-marked. Given the θ-Criterion and the projection principle, θ-marking of verbal complements effectively precludes any NP moving into object position, even if that position is not filled at D-structure. In short, movement into θ-positions is impossible, and the NP subjects of infinitival complements must receive their Case through other means.
This ECM account (Chomsky 1981a, 1986; Chomsky and Lasnik 1993) suggests that structures like (49) arise through the application of a marked rule of S′(CP)-deletion, governed by certain verbs like expect and believe (55a–b). Here, the embedded subject remains in situ and is “exceptionally case-marked” as an object by the matrix verb (exceptionally, since Case-marking does not usually take place across a clausal boundary). While we will use the term “raising-to-object” to describe these constructions, see Kirby (2009) for some evidence—from patterns in child language—which suggests that something like the ECM analysis may instead be correct.
Regardless of the details of the particular syntactic account, several patterns appear in the behavior of R-to-O verbs which should be noted explicitly here, and all of which stem from the lack of θ-assignment between the matrix verb and the embedded subject NP. First, R-to-O utterances with embedded active clauses are synonymous with similar utterances with embedded passive clauses (i.e. (56a) = (56b)).
Next, because R-to-O verbs assign no θ-role to the matrix object position (i.e. the overt subject of the embedded clause), it is grammatical for an expletive subject to appear in this position.
And finally, because R-to-O verbs place no semantic selectional restrictions (including animacy) on the subject of the embedded clause, these verbs may embed any clause that is itself internally semantically felicitous.
12.6.1 Raising-to-Object/Exceptional Case Marking in Child Language: Empirical Results
Almost no research in child language has focused on the R-to-O construction. A study by Goro (2004b) using spontaneous speech in CHILDES (data from Adam, Sarah, and Abe) turned up no spontaneous productions of R-to-O constructions of the type presented in the first part of section 12.6. Kirby’s (2009) work covered a wider range of spontaneous data and included experimental work. We describe this work here.
Kirby’s (2009) examination of spontaneous speech included all American English corpora available in the CHILDES database, and looked at both adult and child utterances. All instances of the R-to-O verbs want and need were searched, and then examined by hand so that only those instances occurring in actual R-to-O contexts were included. In children’s speech, 689 R-to-O occurrences of want and 19 R-to-O occurrences of need were found. These constructions included a variety of syntactic features, and to a large extent, children’s R-to-O constructions resembled those spoken by caregivers, with some minor exceptions. For example:
• The majority of R-to-O utterances are in present tense (90–100 percent for children and adults).
• The majority of R-to-O utterances contain a pronoun subject in the matrix clause.
• For want utterances, about 15–20 percent contain matrix negation; for need, this number is around 40 percent.
• 1 percent of adults’ want constructions, and 1.6 percent of children’s, contain an embedded passive (e.g. I didn’t want my hand to be holded; Abe, 2;11).
• Children produced a small number of embedded expletives (0.4 percent of want constructions; e.g. I want it to be warm; Nina 3;2.4).
• Adults produce more questions than children (71.4 percent of their want-utterances, versus 27.7 percent for children).
(p. 275) Thus, there is some productive use of R-to-O constructions in these data. It is important to remember, however, that these data come from a large number of children (over 50) and that the search spanned a wide range of ages (from the youngest files available, through late childhood38). Nevertheless, many of these utterances, including ones with embedded passives, come from quite young ages: the earliest is from Abe, at age 2;11.30.
In addition to examining spontaneous productions of R-to-O, Kirby also conducted a number of comprehension studies with children aged 4 and 5 years. In all of Kirby’s (2009) experiments, she compared children’s performance on R-to-O constructions with object control (John asked/told Mary to wash the car). Here, however, we will focus only on the R-to-O results.
The first experiment tested active R-to-O constructions using a truth-value judgment (TVJ) task. Children were read a brief story; then the experimenter posed a question about the story to a puppet, and the puppet answered.
Children heard either three items with want (two of which had target-true answers, one of which had a target-false answer) or three items with need (one target-true and two target-false answers). This experiment was included for several reasons: (1) to check whether children are familiar with the lexical meanings of the verbs want and need; (2) to see if 4- and 5-year-olds know the basic argument structure of R-to-O verbs (e.g. where in the sentence the wanter/needer must appear); and (3) to check whether children have the processing ability required to parse biclausal sentences with three animate NPs.
Children in both age groups were significantly above chance in responding correctly to these items (4-year-olds: 87.5 percent; 5-year-olds: 91.2 percent).
Another experiment tested children’s comprehension of embedded passives. As noted earlier in section 12.6, one of the properties of the R-to-O construction is that an embedded passive is truth-functionally equivalent to its embedded active counterpart.
As in the first experiment, this was tested by giving children a brief story and then having them rate the truth value of the puppet’s comment about what happened in the story.
(p. 276) (63)
Again, children heard either three items with want or three items with need. Each set of test items had two target-true answers and one target-false answer.
Although their performance was not at ceiling, both 4- and 5-year-olds were still significantly above chance in correctly responding to these items (4-year-olds: 75 percent; 5-year-olds: 79.2 percent). Interestingly, these same children were tested on matrix passives, and although the 5-year-olds gave the same performance (also 79.2 percent correct, significantly better than chance), the 4-year-olds were not significantly better than chance on the matrix passives (64.6 percent correct). Thus, in this experiment, the 4-year-old children performed better on passives embedded under R-to-O verbs than they did on matrix passive constructions.39
Kirby (2009) notes that this asymmetry in performance calls into question the A-chain maturation hypothesis. Not only do active R-to-O constructions contain a non-trivial A-chain (from the subject of the embedded clause to the object of the matrix clause) but R-to-O constructions with an embedded passive contain two A-chains: one within the embedded clause (passivization), and a second one raising the NP into the matrix clause (R-to-O). If children perform better on constructions involving two A-chains than constructions involving one A-chain, it is unlikely that their difficulty with matrix passives has to do with A-chain formation per se.
Kirby (2009) suggests two distinct (but related) viewpoints which may account for children’s relatively high performance on R-to-O embedded passive constructions. First, Hyams et al.’s (2006) account of children’s late acquisition of the passive relies crucially on a default alignment of semantic and grammatical roles. Hyams et al. suggest that children only have difficulty with A-chains when these obscure the canonical pairing of agent with subject, patient with object. Kirby suggests that due to the double A-chains in R-to-O utterances, semantic objects (which receive their θ-role in the embedded clause) surface as (p. 277) syntactic objects (in the matrix clause), and that it may be precisely this alignment which supports children’s interpretation of R-to-O embedded passives. Kirby suggests that the CAH is a piece of what she calls “semantic scaffolding”: a strategy which children use at early (non-adultlike) stages of grammatical knowledge, in which they take recourse to canonical semantics, in a number of ways, to support their syntactic interpretations.
Second, Kirby (2009) argues that reconceiving biclausal constructions in terms of phases may provide a way to understand children’s facility with R-to-O. Specifically, because there is no θ-assignment between the matrix verb and the subject of the embedded clause, R-to-O utterances may allow for discrete phasal processing. This would allow children to parse the matrix frame (e.g. John wanted X) before parsing the embedded clause (X = Mary to call Bill), in turn lowering the processing resources required to understand such utterances, and leading to an early facility with R-to-O utterances (active or passive).40
Kirby also directly tested children’s knowledge of the lack of θ-marking between matrix R-to-O verbs and their embedded subjects, by eliciting semantic anomaly and grammaticality judgments from the same children. These judgments tested whether children would accept inanimate (64) and expletive (65) subjects embedded under R-to-O verbs. Children were shown pictures and heard a puppet make a comment about each picture. The child’s task was to say whether what the puppet said was “okay” or “silly,” compared to “what we normally say.”41
Results indicated that neither age group was significantly above chance on correctly accepting embedded inanimate subjects (4-year-olds: 56.3 percent; 5-year-olds: 58.3 percent), and that only 5-year-olds were above chance on correctly accepting embedded expletives (4s: 62.5 per cent; 5s: 77.1 per cent). However, based on children’s justifications for their rejections, Kirby concluded that children were (1) attending to truth values (rather than grammaticality or felicity), and (2) not parsing the R-to-O utterances as a whole, but rather selectively attending to the embedded clause.42 Kirby (2009) suggests that the demands of the metalinguistic (p. 278) judgment task itself, combined with the processing load inherent to biclausal sentences, may have overloaded children’s processing abilities, resulting in this effect.43
Further research on children’s acquisition of R-to-O is clearly necessary, to determine at what point children have the full adultlike representation of these utterances.
In this chapter, we have discussed children’s acquisition of A-movement in its major forms, including passives, unaccusatives, and raising verbs (both subject raising and raising-to-object).
Children’s acquisition of these constructions seems to be staggered, with some constructions (e.g. unaccusatives) appearing quite early in spontaneous production (around 2;0) and other constructions (passives, R-to-O utterances) appearing much later. But even given a single construction—for instance, the passive—children may display varying levels of “adultlike” performance, depending on the semantics of the utterance in question, the particular thematic roles assigned (and the positions to which they are assigned), the pragmatic felicity of the utterance context, and the processing load associated with the comprehension task (including sentence-internal factors, such as the number of animate NPs, and sentence-external factors, like the experimental method).
In short, researchers have determined that all of these issues are at play in children’s performance. When these factors are stacked in children’s favor, children may show comprehension at much earlier ages than the literature has previously claimed—even if they do not yet use the constructions in their spontaneous speech. As a result, one of the major theoretical claims about the acquisition of A-movement—namely, that A-chains are subject to biological maturation—is weakened. As we find more sensitive metrics to test children’s underlying representations of these constructions, it is conceivable that researchers could determine A-movement to be something made available quite early by UG. In short, it may ultimately prove less helpful for researchers to talk about the “acquisition of A-movement,” and more helpful to focus on the surrounding linguistic and extra-linguistic limitations that prevent children from demonstrating what they have known all along.
We would like to thank the editors of this volume and an anonymous reviewer for very helpful suggestions and comments. All errors and shortcomings are due to us.
(1) In the following discussion, we will be somewhat lax in using the terms NP and DP more or less interchangeably. Though we do subscribe to a DP analysis of all NPs, the essence of the points made should hold, regardless of the theoretical stance. Furthermore, using the term NP here will, in many cases, allow consistency with much of the GB literature referenced.
(2) A number of non-derivational approaches assume that no movement is involved in these constructions. Instead, “structure sharing” analyses like LFG (Bresnan 1982a), GPSG (Gazdar et al. 1985), and HPSG (Pollard and Sag 1994) assume that a single argument occupies a position in both the matrix and embedded clauses. On the most abstract level, this proposal overlaps significantly with transformational approaches, perhaps especially given current construals of movement as a process of copying (i.e. multiple Merge) plus deletion (Chomsky 1995), and the fact that movement-based analyses still, on some level, require a single element to be linked to multiple positions, semantically. In this way, the learnability considerations with respect to semantic linking and construal would be quite similar, regardless of the particular theoretic approach.
However, since structure-sharing approaches do not necessitate that the grammar contain any process of movement (or copy-deletion) and/or multiple levels of derivation, they essentially represent a proper subset of the requirements that must be met by a mental grammar, given a movement-based account. As a result, we will not consider these non-derivational approaches in any detail here.
(3) The other two constraints are that (1) applications of the rule must respect subjacency, and (2) the trace must be “governed in some sense” (Chomsky 1981: 56). As these constraints are less crucial to the child data discussed below, we will not consider them in any detail here.
(4) This formulation of the θ-Criterion contrasts with semantically based (non-movement) analyses of so-called “raising” constructions (e.g. Jackendoff 1972), as well as with more recent notions of control as movement (e.g. Hornstein 1999), in which a single argument may bear multiple thematic roles. We will not consider these proposals in any detail here, although Hornstein’s proposal does bear on the broader acquisition findings presented in Kirby (2009). See the latter citation for more detail.
(5) Chomsky (1981: 37) assumes that “if α subcategorizes the position β, then α θ-marks β … We will say that α θ-marks the category β if α θ-marks the posisiton occupied by β or a trace of β. Note that α subcategorizes a position but θ-marks both a position and a category.”
(6) Another null DP is the silent PRO involved in control constructions. PRO was originally hypothesized to not need Case (Chomsky 1981), but later this assumption was revised (Chomsky and Lasnik 1993). We will not explore these hypotheses—or the phenomenon of control—in any detail in this chapter; however, both Becker (2006, 2009) and Kirby (2009) have made crucial use of the fact that certain raising and control constructions form surface-identical pairs.
See those citations (and below) for more detail.
(7) Chomsky (1986: 94) formulates these notions in a slightly different way, essentially reversing the order of operations: “let us assume that an element is visible for θ-marking only if it is assigned Case…. A lexical argument must have Case, or it will not receive a θ-role and will not be licensed.” However, the end result is the same: every overt DP must have a θ-role and Case.
(8) It should be noted that later GB accounts resembled MP accounts to a much larger extent.
(9) Where (i.e. when) movement occurs in the derivation is another point which distinguishes the two theoretical approaches, and their answers arise to a great extent out of notions of the interface levels involved. GB proposes that items are removed from the lexicon and assembled into a D-structure, after which point movements occur. In contrast, MP proposes that the assembly of elements from the lexicon (“Merge”) and the movements of these elements proceed simultaneously; this allows MP to dispense with the notion of D-structure. The competing claims about interface levels are not crucial to the results presented below, and as such, we will not consider them in any detail here. Interested readers should consult and compare Chomsky (1981) and Chomsky (1995).
(10) The following discussion relates to verbal passives, which are in some cases homophonous in English with adjectival passives (cf. The peas were eaten (by Mira)/were gone (*by Mira); adjectival passives are sometimes also called lexical passives). However, the latter are not formed by A-movement but instead stored (as adjectives) in the lexicon. This homophony is, as we will see below in section 12.3.2, important in interpreting some of the data on children’s acquisition of the English passive.
(11) S is equivalent to TP in more current terminology. [NP,S] is the syntactic position of the semantic subject (and/or the subject NP itself) of the sentence, and [NP,VP] is the verbal complement.
(12) Jaeggli (1986: 591) defines the notion of “absorption” in the following way: a “passive suffix ‘absorbs’ the external θ-role of a predicate simply by being assigned that θ-role. Nothing more is involved.”
(13) In MP: have its Case feature checked/erased.
(15) The ages of the children varied across the four experiments. Experiment 1 tested only 4-year-olds, experiment 2 tested 4- and 5-year-olds, and experiments 3 and 4 tested children aged 5–8.
(16) These verbs corresponded to activities which are not lexicalized in English, and had meanings like “to rub the back of the neck of” (actional) and “to see through a binoculars-like instrument” (perception).
(17) None of these anticanonical action verbs exist in English, nor do Pinker et al. (1987) mention any language in which they do exist. An example for this category would be The dog floosed the giraffe, meaning ‘the giraffe leapfrogged over the dog.’
(18) In their first experiment they report results from only one non-actional verb, see. They had tested hear as well but found very poor performance on that particular verb. In their second experiment O’Brien et al. (2006) tested both see and like as non-actional verbs.
(19) A reviewer points out that children may also be good at correctly accepting long passives in a “match” condition even when there is no extra potential agent in the situational context, but they may have difficulty correctly rejecting a long passive in a “mismatch” condition. Since O’Brien at al. (2006) do not break down their results according to match versus mismatch items, it is not possible to tell whether this manipulation led to a significant difference in performance.
(20) For reasons of space we limit ourselves to nativist theoretical approaches. Please see Chapter 9 by Viau and Bunger in this volume and Chapter 10 by Guasti in this volume for some discussion of usage-based accounts of the development of argument structure and passive.
(22) It should be noted that the age of maturation was later revised to about 7 years (Wexler 2004; Hirsch and Wexler 2007). The A-Chain Deficit Hypothesis has also been revised in this later work to be formulated in terms of phasal syntax (Chomsky 2001); see section 12.5.1.
(23) As a side note, Borer and Wexler (1987) did not conduct an experiment of their own, but rather reported on naturalistic production (e.g. Berman and Sagi 1981, cited therein, for Hebrew) and experimental data collected by other researchers (e.g. for English, Horgan 1978; Maratsos et al. 1985).
(24) The inclusion of a third character was motivated by O’Brien et al.’s (2006) finding that English-speaking children performed better on non-actional full passives when the presence of an additional referent made a full passive statement about the patient more felicitous.
(25) Most of the full passives are produced by one subject, Juupi, who has a higher MLU at earlier ages than the other children.
(26) We do not necessarily advocate a frequency-based explanation of cross-linguistic asymmetries in the acquisition of the passive. Some studies in which children received extra exposure to the passive construction report an improvement in performance on passive constructions (Gordon and Chafetz 1990) while others report no such advantage (Maratsos et al. 1985). Hyams et al. (2006) also point out that children produce many forms that are not attested in their input (e.g. Root Infinitives) and fail to produce large proportions of constructions that are extremely frequent in parental speech (e.g. imperatives). However, it is noteworthy that in the languages in which children are reported to acquire the passive early, this construction is also more frequent in caregiver speech in comparison to languages in which children are reported to acquire the passive late. Thus, the role of input frequency in the acquisition of this construction remains an open question.
(29) Burzio labels these verbs ergatives, to distinguish them from non-ergatives (for more discussion on which, see the following discussion).
(31) For specific arguments for covert A-movement in this construction, see Babyonyshev et al. (2001); but also see Potsdam and Polinsky (2011) for arguments against the covert movement analysis of this construction.
(32) It is worth noting that a similar disagreement arises in the case of Japanese: Machida et al. (2004) argue that children acquiring Japanese perform well on unaccusatives (cf. Sano et al. 2001) because they incorrectly analyze them as unergatives. Shimada and Sano (2007), in turn, argue that children do correctly distinguish unaccusative from unergative constructions in Japanese, based on te-iru constructions.
(34) Wexler (2004: 164) summarizes this Phase Impenetrability Condition in the following way: “When working at a phase, only the edge (the head and spec(s)) of the next lower phase are available for analysis, and nothing lower than the edge. In particular the complement isn’t available.”
(35) It is worth noting that Becker sought to answer a different type of question than Hirsch and Wexler (2007). Whereas Hirsch and Wexler were asking whether children could represent A-chains, so as to interpret constructions involving that operation, Becker was interested in how children come to distinguish the class of raising verbs from the class of control verbs. These two classes of verbs differ in their thematic structure, yet they overlap partially in their surface distribution. This was the primary point of interest in Becker’s research.
(36) The use of the progressive or plain present tense on the main verb is determined by whether the verb itself is eventive (must be progressive) or stative (must not be progressive) and is unrelated to whether the verb is a raising or a control verb.
(37) Instead of CP complements, expect/believe may instead take NP complements, as in the sentences John has been expecting a delivery or John believes the story. In these cases, expect/believe clearly must assign a θ-role to the following NP, as there is no other verb which might do this work.
However, the existence of such sentences should not be taken as evidence that the these verbs always θ-mark a following NP. The fact that R-to-O verbs may either take an NP complement (and assign it a θ-role) or a CP complement (in which case there is no θ-relation between the matrix verb and the embedded subject) should be no more controversial than claiming that verbs like eat or sing can be optionally transitive (Julia has already eaten (dinner); Marguerite sings (opera) beautifully).
(38) The exact ages are not detailed, but Kirby included all available CHILDES files in her search. Taken together, these files represent children spanning from around 1-year-old to 13 years and older.
(39) There are two ways in which this comparison may not be a fair one. First, in the matrix passive test, children heard three items: one with target-true and two with target-false answers. This item set is not an ideal match for the embedded passive task (which had one target-false and two target-true items), as it is widely accepted that children have a “yes”-bias, and may perform more poorly on target-false than on target-true answers.
And second, both matrix passive items with target-false answers had agent-subjects and patient-objects (just like in active sentences); if children interpret passives as actives, they should accept these sentences. In contrast, only one of the target-false R-to-O embedded passives had a similar (i.e. active-like) θ-role alignment; the other was false for basic truth value reasons, given the preceding story.
Ideally, the R-to-O embedded passive and matrix passive items would have equal numbers of target-true and target-false answers (both within and across tasks), and all the R-to-O items would be specifically crafted to check whether children interpret embedded passives as active clauses. Kirby (2011) reports that a follow-up study is in progress which makes both of these changes.
(40) This would contrast notably with object control (OC) utterances, in which the embedded clause must be interpreted with respect to the matrix object.
Thus, although R-to-O and OC utterances have a similar surface string, R-to-O utterances may correspond to a lower processing load.
(41) In Kirby’s experiment, these grammatical R-to-O sentences were contrasted with anomalous/ungrammatical foils containing object control verbs, e.g. #The girl asked the trees to be tall, *The woman told there to be flowers on the table.
(42) For instance, one child rejected the sentence The teacher needed the books to weigh less because “they’re so heavy” (MB, 4;1.15). Another child rejected Big Bird wanted there to be crayons in the box, saying “It’s silly because there’s no crayons” (MC, 4;2.16). Kirby argues that such justifications indicate that children are not parsing the matrix frame, which can change the truth value of the overall sentence.
(43) Recall that children were successfully able to parse similar biclausal R-to-O utterances in the first (basic, active) TVJ task (see example (61)). Children also performed at ceiling on monoclausal filler items in the felicity/grammaticality judgment tasks. However, the metalinguistic task combined with the load of more clauses (and more NPs) seemed to be more than 4- and 5-year-olds could handle.