Acquisition of French Causatives: Parallels to English Passives

Abstract

Guasti (2016) notes similarities between English get- and be-passives, and Romance causatives of the faire-par and faire-infinitif types, respectively. On this basis she conjectures that faire-infinitif will show an acquisitional delay similar to that found for English be-passives, which are not mastered until sometime after the age of four. Here, this prediction is tested and supported for French faire-infinitif causatives of transitive verbs. To explain the delay, the Universal Freezing Hypothesis (UFH) of Snyder and Hyams (2015) is extended to this type of causative: a restriction on movement is recast as a restriction on AGREE. A novel prediction, that faire causatives involving unergative or unaccusative verbs will be acquired much earlier, is also tested and supported. Finally, English get-passives and French “reflexive causative passives” are examined in light of the fact that both are acquired substantially earlier than age four.

1. Introduction

Children acquiring English often start using get-passives before they turn three, but most studies find that children have difficulties producing and understanding be-passives until much later ages. The details vary by study but, in general, children do not reliably succeed on the be-passives of eventive verbs (especially if a by-phrase is present) until around the fourth birthday, and they continue to have substantial difficulties with the be-passives of non-eventive verbs (e.g., see, like) until age five or six. Guasti (2016) conjectures that a similar acquisitional time course might exist for certain causative constructions found in Romance languages like French. Her idea is that the delay seen for English be-passives might also exist for French causatives of the faire-infinitif type.

In this paper we will first test, and strongly support, Guasti’s (2016) conjecture as it applies to French causatives of transitive verbs. Next, we will propose that the Universal Freezing Hypothesis (UFH) of Snyder and Hyams (2015), which seeks to explain the delay in English be-passives, can be extended in a natural way to explain the observed delay in this type of causative. The proposed extension leads to novel predictions for the acquisition of causatives involving intransitive verbs, and we will provide evidence supporting these predictions. Finally, we will examine two early-acquired structures, English get-passives and French reflexive causative passives. The key point will be that, in both cases, the external argument associated with the lower verb does not have to be present in the syntactic structure and, as a consequence, children’s early successes are fully consistent with the UFH.

2. French Causatives and Guasti’s (2016) Conjecture

2.1. Causatives and Passives

Within the Generative tradition there exists an extensive literature on French faire-causatives, dating to Kayne’s (1975) seminal analysis dividing them into two main types, faire-infinitif and faire-par. In faire-infinitif causatives, if the causativized verb is transitive, the causee is obligatorily realized as either a dative-marked Determiner Phrase (DP) following the embedded object, as in (1a), or as a dative clitic preceding faire, as in (1b):

1.

a.

Jean

a

fait

laver

la voiture

à

Paul

John

has

made

to-wash

the car

DAT

Paul

‘John made Paul wash the car.’

b.

Jean

lui

a

fait

laver

la voiture

John

him.DAT

has

made

to-wash

the car

‘John made him wash the car.’

c.

Jean

a

fait

laver

la voiture

(par Paul)

John

has

made

to-wash

the car

(by Paul)

‘John had the car washed (by Paul).’

The causee in the faire-par causative, however, appears in an optional par-phrase, which is located after the object of the lower verb, as shown in (1c).

In the present paper we will employ a somewhat larger, five-way classificatory scheme, as indicated in (2):

2.	a.	Faire-Dative (FD) = faire + Vtransitive with a dative causee
		Jean a fait laver la voiture à Paul.	Jean lui a fait laver la voiture.
	b.	Faire-Par (FP) = faire + Vtransitive with a par-phrase
		Jean a fait laver la voiture par Paul.
	c.	Faire-Null (FN) = faire + Vtransitive with a null (i.e., omitted) causee
		Jean a fait laver la voiture.
	d.	Faire-Unaccusative (FuA) = faire + Vunaccusative (and an accusative causee)
		Jean a fait partir Paul.	Jean l’a fait partir.
		John has made leave Paul	John him.ACC has made leave
	e.	Faire-Unergative (FuE) = faire+Vunergative (and an accusative causee)
		Jean a fait danser Paul.	Jean l’a fait danser.
		John has made dance Paul	John him.ACC has made dance

Thus, the faire-infinitif sentences in (1a,b) will both be referred to as FDs. The faire-par sentence in (1c) will be referred to as an FP if the par-phrase is pronounced, or an FN if the par-phrase is omitted.

Alongside the differences in the causee phrases, there exist a number of semantic differences across the various subtypes of causatives. Crucially, Guasti (2016) notes that certain semantic restrictions found in FP/FNs are also found in English get-passives: First, stative verbs (loved, aimer in 3a,b) are rejected in both get-passives (e.g., Hirsch and Wexler 2004) and FP/FNs:

3.	a.	?* John got loved
	b.	?* Ils	ont	fait	aimer	Jean		(par Marie)
		they	have	made	love	John		(by Mary)
		(?*) ‘They had John loved (by Mary).’

Second, Guasti (2016) notes that both get-passives and FP/FNs require the logical object to be [+affected]. For example, in (4a,b) the answer/la solution is [-affected]:

4.	a.	?* The answer got found
	b.	?* Ils	ont	fait	trouver	la solution		(par le chercheur)
		they	have	made	find	the answer		(by the researcher)
		(?*) ‘They had the answer found (by the researcher).’

These restrictions are absent in be-passives (5a,b), and in FDs (6a,b):

5.	a.	John was loved
	b.	The answer was found

6.	a.	Ils	ont	fait	aimer	Jean	à		Marie
		they	have	made	to-love	John	DAT		Mary
		‘They made Mary love John.’
	b.	Ils	ont	fait	trouver	la solution		au chercheur
		they	have	made	find	the answer		DAT+the researcher
		‘They made the researcher find the answer.’

On the basis of these parallels, Guasti (2016, p. 185) conjectures that the acquisitional time course of French causatives will mirror that of English get- and be-passives: just as be-passives are delayed relative to get-passives, FDs will be delayed relative to FP/FNs.1

2.2. Testing Guasti’s (2016) Conjecture

In order to assess Guasti’s (2016) conjecture we examined 11 longitudinal corpora from CHILDES (MacWhinney 2000) for children acquiring French in France: Anaïs, Marie, Marilyn, Nathan, Théotime (Demuth and Tremblay 2008); Anaé, Antoine, Léonard, Madeleine, Théophile (Morgenstern and Parisse 2007); and Anne (Plunkett 2002). The ages covered for the children in question are listed in

Table 1. Age Ranges of Corpora.

Child	Ages
Anaé	1;04,20–5;01,21
Antoine	1;00,24–6;03,08
Leonard	0;11,19–5;03,19
Madeleine	1;00,05–6;11,27
Théophile	1;00,09–4;11,11
Anais	1;00,23–3;00,15
Marie	1;00,02–4;00,05
Marilyn	1;06,13–2;11,14
Nathan	1;00,12–3;00,03
Théotime	0;11,17–3;00,03
Anne	1;10,12–3;05,04

, in Y;MM,DD format.

For our analysis we first used a computer search to identify all child utterances containing any form of the verb faire, and then we manually removed all non-causative uses. Next we classified each of the causatives as either “Clear FD”, or “Other”, according to the criteria in (7) and (8) (these criteria are based directly on the discussion in Guasti 2016).

7.	Clear FD: A causative containing an overt dative argument, and/or a transitive verb that is semantically incompatible with FP/FN (i.e., a non-eventive verb, or a verb with a direct object that is [-affected]).

8.	Other: A causative containing an overt par-phrase, or a lower verb that is either intransitive or semantically compatible with being an FN.

Note that children often omit material that is obligatory for adult speakers. Hence, our criteria do not rely exclusively on the form of the causee. An utterance like (9), where the causee is omitted, would nonetheless be classified as a clear FD, since the verb trouver ‘to find’ does not satisfy the semantic constraints on FP/FNs:2

9.	Ils ont fait trouver la solution (…)

Our prediction is that no instances of a clear FD will appear in the children’s utterances prior to the late age-range associated with the appearance of be-passives: any child using clear-cut FDs prior to age four will be a counterexample.3

2.3. Results

Our examination of the 11 children’s data found no use of a clear FD prior to age four. As expected, shortly after the age of four some of the children did begin using clear-cut FDs, as seen in (10). Additionally, as expected, other types of causatives appeared much earlier. In fact, the data included two examples of FPs (11a,b), and both were produced well before the age of three:4

10.	Je	faisais	faire	quoi	à	mes trois enfants?
	I	made	do	what	DAT	my three children?
	‘What did I make my three children do?’							(Madeleine 4;01,27)

11.

a.

Il va

[se]

faire

gronder

par

sa maman et papa

he goes

[self]

make

scold

by

his mom and dad

‘He’s going to get (himself) scolded by his mom and dad.’

(Antoine 2;09,16)

b.

Elle

se

fait

tirer

par

la boulle

comme

ça

fait

du bruit

she

self

makes

pull

by

the ball

as

that

makes

of-the noise

‘She is getting (herself) pulled (=becoming irritated) by the ball since it makes noise.’

(Madeleine 2;05,12)

Table 2. Results by Child.

	Onset of Non-FDs	#Non-FDs before Age 4	Onset of FD	#FDs (to End of Corpus)
Anaé	2;00	12	>5;01	0
Anaïs	2;08	5	>3;00	0
Anne	2;04	14	>3;05	0
Antoine	2;06	11	4;02	1
Léonard	2;01	2	>3;02	0
Madeleine	2;01	7	4;02	3 (see Figure 1)
Marie	2;05	18	>4;00	0
Marilyn	2;09	4	>2;11	0
Nathan5	>3;00	NA	>3;00	NA
Théophile	2;10	11	>4;11	0
Théotime	2;03	13	>3;00	0

FD = Faire-Dative: faire-causative + transitive V with dative causee.

provides the results by child. For example, the child Madeleine produced her first example of a (non-FD) faire-causative at the age of two years, one month (2;01), and produced an additional six such utterances before her fourth birthday. Her first recorded use of an FD was at the age of 4;02, and was followed by two additional FDs before the end of her corpus. In Figure 1 we have graphed Madeleine’s use of causatives as a function of her age.

To check for statistical significance of the observed delay in FDs, we employed an absolute-frequency binomial test, as described in Snyder (2007, chp. 5): as indicated in (12–14), we first calculated the total number (U) of child utterances in our sample that were produced after the given child was already producing both datives and (non-FD) causatives, but before the child had reached the age of four years:

12.	U = 33,244 utterances

To estimate the per-utterance frequency of FDs in speakers whose grammar allows them, we analyzed all parental utterances in our 11 corpora, calculated each parent’s per-utterance frequency of FDs, and then took the median, F:

13.	F = (7 FDs)/(22,778 utterances by the given parent)

To calculate the probability that U would be as high as observed (or even higher), under the null hypothesis that FDs were always available to children as soon as they had both non-FD causatives and dative arguments, we calculated the binomial probability, p = (1−F)^U:

14.	p = 0.00003651 < 0.0001

As seen in (14), the age gap between the onset of non-FD causatives and dative-marked arguments, on the one hand, versus the first clear FD, was robustly significant. Therefore, Guasti’s (2016) conjecture (at least as it applies to FDs) is strongly supported by our longitudinal data. In the next section we look at the connection between these results and the UFH, before moving on to more fine-grained predictions regarding the acquisition of causativized intransitives.

3. Explaining the Acquisitional Delay for Faire-Datives

In this section we will first review Snyder and Hyams’ (2015) account of the acquisitional delay found in English be-passives. Then we will show that their account, the UFH, can be extended in a simple way to account for the observed delay in FDs.

3.1. English Be-Passives and the Universal Freezing Hypothesis

Snyder and Hyams (2015) propose that the lateness of English be-passives is due to a freezing effect. For the syntax of the passive, they adopt the analysis of Collins (2005a), which is based on a strict interpretation of UTAH (i.e., the Uniformity of Theta-Assignment Hypothesis; Baker 1988, p. 46; 1997, p. 74). Under Collins’ (2005a) interpretation, a verb’s external θ-role must be assigned in exactly the same way in active and passive sentences, namely to the specifier of vP. In a passive the external argument is realized either as a PRO, in the case of a short passive; or as an overt DP (preceded by the overt voice-head by), in the case of a long passive.

Consider the be-passive in (15). Given the presence of the external argument John in Spec-vP, simple argument-movement of the object DP the book into Spec-IP would violate relativized minimality (RM; Rizzi 2001, 2004). The solution proposed by Collins (2005a) is that the object is “smuggled” past the external argument, via movement of a larger phrase, after which it raises to subject position without a minimality violation, as shown in (16).

15.	The book was written by John.

16.

Cf. Collins 2005a, pp. 90, 95

In (16), the V write raises to the head of PartP and forms the participle. Smuggling occurs when the entire PartP, including the object, raises past the external argument in Spec-vP and lands in the specifier of the passive voice phrase. From here the object DP can undergo cyclic movement into the subject position in a fashion compatible with minimality.

Collins (2005b, p. 292) defines “smuggling” in terms of the illustration in (17): The constituent XP cannot be directly related to Z, due to an intervening element W. Yet, after movement of the larger constituent YP to a position from which it c-commands W, XP can be related to Z. In this case, YP “smuggles” XP past W.6

17.	Smuggling (Collins 2005b, p. 292, ex. 6)

In the cases of smuggling discussed by Collins (2005a, 2005b), the relation between XP and Z is always one of movement: the relation created by moving XP to position Z.

The entire concept of smuggling hinges on the idea that the Freezing Principle (which was proposed independently in Ross (1967, 1986) and Wexler and Culicover (1980)) allows for some exceptions. One formulation of the principle, based on Müller (1998), is given in (18).

18.	Freezing Principle: In the following configuration, no operation (such as Move) may relate X and Z: * Z … [Y … X …] … <Y>

According to this principle, the direct object inside the PartP in (16), which corresponds to the X inside Y in (18), should be frozen to further syntactic operations, and therefore unable to undergo raising into subject position. Hence, Collins (2005a, 2005b) concludes that there must exist situations in which the Freezing Principle fails to apply, although he does not attempt to spell these out in a general theory.

Finally, the role of smuggling in the late acquisition of the be-passive, according to Snyder and Hyams (2015), is connected to a difference in the capacity of adults and children under four to make the required exception to the Freezing Principle. In children that young, the Freezing Principle blocks smuggling, and thereby blocks the be-passive: without smuggling, the underlying object cannot get past the intervener that separates it from subject position. Snyder and Hyams (2015) formulate this hypothesis as the UFH, as stated in (19).

19.	The Universal Freezing Hypothesis: For the immature child (until about age four), the Freezing Principle always applies. No subpart of a moved phrase can ever be extracted.

The UFH thus accounts for the observed delay in children’s mastery of be-passives, which depend on smuggling to circumvent a Minimality violation.

3.2. The Universal Freezing Hypothesis and the Locus of Maturational Change

As formulated above, the UFH describes a change in the availability of exceptions to the Freezing Principle, but it does not address why such a change would occur. The answer presumably depends on the precise nature of the maturational change that the UFH describes. For example, one possibility is that there are essentially two Freezing Principles provided by Universal Grammar (UG). Under this view, a “maximally restrictive” version of the Freezing Principle, which is active in the grammar of very young children, is replaced by a “selective” version sometime around the age of four.

Yet, this is not the approach that we advocate. For one thing, there are reasons to doubt that the Freezing Principle itself (whether selective or restrictive) is actually a primitive element of UG. For example, Uriagereka (1999) argues that freezing effects can be made to follow as a deductive consequence from mechanisms that are independently needed for cyclic spell-out. Alternatively, Keine (2016) argues that they can be derived from limitations on the search space of probes.

Furthermore, in Borga and Snyder (forthcoming) we review recent experimental evidence suggesting that an absolutely restrictive version of the Freezing Principle would be too strong, even for children younger than four. As an alternative, we propose that the syntactic structures resulting from a smuggling-based derivation impose excessive demands on the child’s computational resources for language processing. This proposal accommodates recent evidence suggesting that children’s ability to produce and comprehend be-passives improves in experiments that either (i) eliminate the need for the use of a smuggling derivation in the first place, for example by adding a feature like [+wh] or [+Topic] to the derived subject (cf. Rizzi 2004); or (ii) take steps to reduce the processing load that a smuggling-based structure creates, for example by providing a structural prime for the be-passive (e.g., Messenger 2010). In the case of (i), information structure is manipulated via context provided in the experimental task, resulting in a derived subject which is either discourse old ([+Topic]) or which undergoes wh-movement, avoiding Minimality violations and the need for smuggling in either case (cf. footnote 8). As regards (ii), one analysis of syntactic priming studies is that the comprehension of a sentence can aid in activating its underlying syntactic representation, which can then be re-used in further production and comprehension, at a reduced processing cost (Guasti 2016, p. 203). The point which follows from both observations is that the change taking place around the age of four is not a change in the grammar itself, but rather in the level of computational complexity (and, hence, grammatical complexity) that the child’s language processing system can handle.7

3.3. Extending the Universal Freezing Hypothesis to Faire Datives

Despite the existence of parallels between English passives and Romance causatives, it is not immediately clear that Snyder and Hyams’ (2015) freezing-based approach to passives can account for the acquisitional timing of causatives. The maturational delay described in the UFH appears to be tied to DP movement, which does not play any role in standard analyses of faire-causatives (e.g., Kayne 1975; Rouveret and Vergnaud 1980; Burzio 1986). Yet, there is reason to think a form of smuggling might also be needed in certain causatives.

A number of authors, going back to Kayne (1975), have proposed that a derivation along the general lines of (20) (which is based on a discussion in (Belletti and Rizzi 2012)) is found in French and Italian FDs.

20.	FD	Cf. Belletti and Rizzi 2012

In (20) we have labeled as ‘smuggling’ the step of the derivation in which the VP headed by laver ‘wash’ is moved past the external argument, à Paul ‘to Paul’. (In the version of this analysis sketched by Belletti and Rizzi (2012, p. 135), the VP undergoes feature-driven movement to the specifier of a functional head, X, which the authors, suggest is the head of a “(small) clausal complement” to the causative verb).

As emphasized by Belletti and Rizzi (2012), this is an instance of smuggling, even though nothing moves out of the moved constituent. The VP containing the lower verb’s direct object moves past the intervening dative causee, into a position where the direct object can be case-valued by v_cause.8 To repeat the definition from Section 3.1, stated in terms of the diagram in (17) (repeated as (21), with the elements from the FD inserted), smuggling is the movement of a constituent YP (here the VP containing the lower object) into a position where XP (the lower object) can be “related to” Z (the higher functional v_CAUSE head).

21.	Smuggling in the faire dative:

In the cases of smuggling discussed in Collins (2005a, 2005b), the relation between XP and Z was always one of movement, but if we take movement to require the establishment of an AGREE relation, then the crucial relation between XP and Z can be seen as one of AGREE. Correspondingly, in (20) the crucial relation is an instance of AGREE that simply values a case feature of XP, with no resulting movement.

Hence, to extend the UFH to FDs we really only need to make one change: freezing effects are not fundamentally about movement, but rather about AGREE. As a first approximation we can state this view in the form of a “Modified Freezing Principle”, as in (22):

22.	Modified Freezing Principle: In the following configuration, an AGREE relation (usually) cannot be established between X and Z: * Z … [Y … X …] … <Y>

The formulation in (22) includes the qualification “(usually)” as an explicit indication that certain exceptions will be possible for adults, but—precisely as before—the UFH will mean that these exceptions are unavailable to children younger than four. Hence, the UFH now works well for FDs: children will have substantial difficulties until the age of about four, because the lower verb’s direct object cannot be case-valued without making an exception to the Modified Freezing Principle.

Indeed, potential support for an interaction between freezing and AGREE can be found in recent work by Keine (2016, 2018), who notes a type of freezing effect on long distance agreement in Hindi.9 Agreement between a matrix verb and the object of a nonfinite embedded clause is usually optional in Hindi, as in (23), but is substantially degraded in instances where the lower clause has undergone extraposition, as in (24):

23.	shiksakõ-ne	[ raam-ko	kitaab	parhne]	d-ii
	teachers-ERG	Ram-DAT	book.F	read.INF	let.PFV.F.SG
	‘The teachers let Ram read a book.’

24.	?? shiksakõ-ne	ti	d-ii	[ raam-ko	kitaab	parhne ]i
	teachers-ERG		let.PFV.F.SG	Ram-DAT	book.F	read.INF
	‘The teachers let Ram read a book.’

Thus, it appears that Hindi may provide support for the idea that freezing effects can be found in the domain of AGREE, per se, even when extraction from the moved constituent is not at issue.

In order to interpret the UFH in terms of the computational demands of language processing, as described in Section 3.2, and in terms of a maturationally-timed change in children’s computational abilities, we will need one more adjustment to our assumptions. In terms of movement operations, a smuggling-based derivation requires the language processing systems to represent a chain whose “tail” is properly contained within the “head” of another chain (for example: The book was [written <the book>] by John <written the book>.) This configuration is plausibly responsible for a substantial increase in computational complexity. Yet, in order to subsume FDs under this account, we will need to assume that this configuration is a special case of a more general source of computational complexity: structures involving AGREE into a moved constituent. In other words, regardless of whether any phrase moves out of the moved constituent, simple AGREE into a moved constituent is something that is both necessary to represent, and difficult to represent, during the computations that subserve language production and comprehension.

Finally, note that our account crucially relies on the idea (shared with Snyder and Hyams (2015)) that get-passives can, at least in some cases, be derived without recourse to smuggling. We will discuss this point in detail in Section 5. First, however, we will derive and test a novel prediction of the proposals that we have made in the current section.

4. The Universal Freezing Hypothesis and Causatives of Intransitive Verbs

4.1. New Predictions for Causativized Intransitives

Further examination of the syntax literature on Romance causatives has led us to a novel prediction: the causative of an intransitive verb, whether unaccusative (FuA) or unergative (FuE), will be within the grammatical capabilities of a two- to three-year-old child. Specifically, we note that Guasti (1996) and Folli and Harley (2007) have argued that, in terms of Kayne’s (1975) distinction between faire-infinitif and faire-par, FuEs necessarily belong to the faire-infinitif type, with a derivation along the lines of (25):10

25.	Faire + unergative = ‘faire-infinitif’

In (25), the assumption is that FuEs, like FDs, require movement of the lower VP, but there is nothing to intervene between v_cause and the accusative-marked causee, Paul. Note, too, that in terms of the Modified Freezing Principle, the UFH does not create any problems for the child under four. This is because there is no AGREE operation into the VP that underwent movement.

In the case of FuAs, Folli and Harley (2007) and Guasti (1996) argue that faire-causatives without an external argument are at least potentially instances of Kayne’s (1975) faire-par causative (recall that, in present terms, these correspond to FP or FN causatives, depending on whether the optional par-phrase is present). For a typical FP, (26) shows a version of the analyses proposed by Folli and Harley (2007) and Guasti (1996): the complement of faire is not a vP but rather a nominalized VP.11 The par-phrase is a PP adjunct similar to the by-phrase in English derived nominals (Folli and Harley 2007) and does not count as an intervener in either movement or case valuation. Accordingly, neither smuggling nor the Modified Freezing Principle plays any role in FP/FN causatives. Similarly, when the causativized verb is an unaccusative, as in (27), there is nothing to intervene between v_cause and the accusative-marked causee, Paul. No smuggling occurs, and the Modified Freezing Principle is respected. Hence, the prediction of the UFH is that children under four will succeed.

26.	Faire-par (FP):	Cf. Guasti 1996; Folli and Harley 2007

27.	Faire + unaccusative = ‘faire-par’

4.2. Testing the Predictions

Note that Guasti’s (1996) acquisitional conjecture, which she expressed in terms of Kayne’s (1975) faire-infinitif/faire-par distinction, linked causatives of the faire-infinitif type with English be-passives. In Section 2, when we tested her conjecture, we focused on FDs, which are faire-infnitif causatives in which the causativized verb is transitive. Yet, as discussed in the preceding section, FuEs also seem to belong to the faire-infinitif type of causative. If so, one might expect them to show the same acquisitional delay seen for FDs in Section 2.3. Yet, as indicated above, the UFH actually leads to the opposite prediction: FuEs (as well as FuAs) are expected not to show any particular delay.

For the corpus results reported in Section 2.3, we simply distinguished between FDs and other causatives, but to evaluate the new predictions for causativized intransitives we will now examine utterances in the “Other” category more closely. In brief, both types of causative are attested at early ages, although at all ages FuAs are produced much more frequently than FuEs. Ten of the 11 children produced at least one example of an FuA in their recorded speech, and in all ten cases the first use was prior to the age of three. For FuEs, 6 of the 11 children produced at least one example, and in four cases the first use was prior to the age of three. Some representative examples of FuEs are provided in (28), and details for each child are provided in

Table 3. Causativized intransitives, by Child.

	First Use of FuA (Age)	#FuA (Total)	First Use of FuE (Age)	#FuE (Total)
Anaé	2;00	7	NA	0
Anaïs	2;08	3	NA	0
Anne	2;04	10	2;05	2
Antoine	2;06	11	4;02	1
Léonard	2;01	2	NA	0
Madeleine	2;01	10	NA	0
Marie	2;06	12	2;05	1
Marilyn	2;09	3	2;11	1
Nathan	NA	0	NA	0
Théophile	2;10	8	3;07	3
Théotime	2;03	9	2;05	4

.

28.	a.	Il	me	fait	pleurer
		he	me	makes	cry
		‘He is making me cry.’					(Marilyn 2;11,14)
	b.	Ça	me	fait	rire
		that	me	makes	laugh
		‘That makes me laugh.’					(Théotime 2;7,28)

Thus, our evidence from the acquisition of French faire causatives with intransitive verbs, both FuAs and FuEs, is fully in line with the predictions of the UFH.

5. Get-Passives and Reflexive Causative Passives

In this section we return to two structures that are acquired substantially earlier than the age of four, get-passives and reflexive causative passives, to clarify how they differ from be-passives and FDs, respectively.

5.1. Get-Passives

In contrast to be-passives, English get-passives are produced and understood by the age of two or three years (Turner and Rommetveit 1967a, 1967b; Harris and Flora 1982; Crain and Fodor 1993; Slobin 1994). While the underlying structure of get-passives is controversial, one influential view is that there exist multiple types of get-passives, each with its own structure (Reed 2011, p. 42). Snyder and Hyams (2015) contend that the types of get-passive that are produced and understood by children younger than four do not contain an external argument.

Tests for the syntactic presence of a phonetically null, external agent argument include compatibility with a purpose clause, as in (29), and compatibility with the adverbial modifier on purpose, as in (30) (these examples are based on Fox and Grodzinsky 1998, p. 327).

29.	a.	The ship was sunk [PRO to collect the insurance money].
	b.	*The ship sank [PRO to collect the insurance money].
	c.	*??The ship got sunk [PRO to collect the insurance money].

30.	a.	The book was torn on purpose.
	b.	*The book tore on purpose.
	c.	*??The book got torn on purpose.

The be-passives in (29a) and (30a) clearly have an understood agent in their syntactic structures, because the agent is available to control the PRO subject in a purpose clause (29a), or to license the use of the agent-oriented adverbial on purpose in (30a). The anticausative counterparts in (29b) and (30b), which uncontroversially lack any external argument, are sharply ungrammatical (at least on the interpretations available in 29a and 30a). Crucially, the get-passives in (29c) and (30c) appear to pattern with the anticausatives, which indicates that these particular examples of the get-passive are actually incompatible with the syntactic presence of an understood Agent.

Yet, while the native-speaker judgments indicated in (29) and (30) are the ones that are usually reported, there does exist some inter-speaker variation for the (c) examples.12 Indeed, Reed (2011) and Alexiadou (2012) report that get-passives are sometimes judged to be fully compatible with a by-phrase, or with other diagnostics for the syntactic presence of an Agent, especially if contextual support is provided. According to Alexiadou’s (2012) account, English actually has two different non-actional voice heads, corresponding to passive voice and middle voice, that are possible in a get-passive (in contrast, the be-passive always contains the ‘passive’ voice head). Crucially, passive voice assigns an external argument, but middle voice does not.

For Alexiadou (2012), two factors contribute to the structural ambiguity of get-passives. First, English middle voice is not associated with any overt morphology.13 Second, get is a semantically bleached aspectual verb introducing a change-of-state event, but lacking its own argument structure. Hence, the verbs that typically form get-passives are semantically unspecified for a distinction between external and internal causation. In contexts that strongly favor an agentive interpretation of the event associated with the participle, an underlying passive-voice construal of the get-passive is, at least, marginally possible for some speakers, and this yields the less-than-straightforward judgments for examples like (29c) and (30c).

For present purposes, the essential part of Alexiadou’s (2012) account is simply that the “passivized” verb in an English get-passive routinely lacks the external argument that would be present in a be-passive. If there is no external argument to intervene, the promotion of the logical object into subject position will not require smuggling. Hence the Freezing Principle will not be violated, and the UFH is fully compatible with children’s early success on get-passives.

5.2. Reflexive Causative Passives

As noted above, the earliest examples of FPs that we found were actually reflexive causatives; these were already present in the speech of two-year-olds. As discussed by Belletti (2016a) for Italian, these “reflexive causative passives” are, intuitively, an extremely complex grammatical construction, yet people’s intuitive ideas about grammatical complexity are a poor predictor of acquisitional timing, and in Italian (much as in French), children produce Reflexive Causative Passives (RCPs) remarkably early.

For RCPs (see 31), we again follow Guasti (1996) and Folli and Harley (2007) in their analyses of FPs. For the object-to-subject analysis of “formally reflexive” clitic constructions in French, we follow Sportiche (2010). Snyder and Hyams (2015) have provided extensive evidence for early mastery of reflexive and non-reflexive clitics in children acquiring French and Italian. Moreover, despite the passive-like interpretation, there is no passive morpheme in reflexive FPs, and thus—once again—no need for smuggling. Hence, children’s early success on RCPs is entirely as expected.

31.	Reflexive Causative Passive (RCP):

Belletti (2016a, 2016b), however, offers an alternative analysis of RCPs, in which the reflexive morpheme represents the external argument of the causative verb, and the lower argument does have to be smuggled past it into surface subject position. Belletti (2016b) further notes that the da-phrase in the Italian reflexive si-fare is compatible with non-agent arguments (32a), which are illicit in the regular, non-reflexive fare-da (32b).

32.	a.	Maria	si		fa	capire		da tutti
		Maria	SI		makes	understand		by everybody
		‘Maria makes herself understood by everybody.’							(Belletti 2016b, p. 5)
	b.	*Questa		spiegazione		ha	fatto	capire	il problema	da tutti
		this		explanation		has	made	understand	the problem	by everybody
		‘This explanation has made the problem understood by everybody.’

Belletti (2016b) takes this latter observation to indicate that the reflexive FP causative, at least in Italian, is not always as closely related to the non-reflexive FP as the presence of the da-phrase would suggest, and indeed she leaves open the possibility that the construction is potentially ambiguous between a more complex, semantically reflexive construction requiring smuggling, and a simpler derivation along the lines of (31). Hence, one possibility is that Belletti’s (2016b) analysis (with smuggling) is the correct one when a true, semantic reflexive is present (as it often is, in adult speech), while the structure is simpler (and smuggling is not involved) when the construction is only “formally” (not semantically) reflexive.

There are several reasons to assume that a simple structure of this kind is indeed available to children who have not yet mastered smuggling. As regards the reflexive morpheme, Snyder and Hyams (2015) provide detailed evidence for two-year-old French and Italian children’s mastery of formally (but not semantically) reflexive constructions with se. In addition, Labelle (2013) observes that the reflexive morpheme in the RCP (se-faire) construction in some cases must not be associated with the external argument of the causative verb, as the surface subject cannot be construed as playing a role in bringing about the caused event (33).

33.	Les habitants	se sont	fait	sorprendre	pendant	leur sommeil	par
	the inhabitants	SE AUX	made	surprise	during	their sleep	by
	l’eruption	du volcan
	the eruption	of-the volcano
	‘The inhabitants were taken by surprise during their sleep by the eruption of the volcano.’					(Labelle 2013, p. 238)

Hence, at least in some cases, se clearly is not the external argument of the causative verb. In our view it is quite reasonable to assume that the RCPs in early child speech are not semantically reflexive, that se is not an intervener for purposes of RM, and that no smuggling is required.

Finally, we would like to note that Labelle (2013) posits that se in RCPs is actually the overt realization of a non-active voice head which prevents the merge of an external argument, resulting in the promotion of the internal argument from the embedded predicate and the absence of a semantically reflexive interpretation. The causative verb itself is described as a semantically underspecified lexical realization of a verb which brings about an event. In light of the acquisitional evidence presented in this paper, we believe the parallels between these ideas and the structure Alexiadou (2012) formulates for the get-passive in English definitely warrant further exploration.

6. Conclusions

In this acquisitional study of French causatives, we first tested and found support for Guasti’s (2016) conjecture regarding the acquisition of faire-par versus faire-infinitif causatives, at least as it applies to causativized transitives: there was strong evidence for a delay in FDs. Next, drawing on the syntax literature for Italian, and updating the Freezing Principle to take AGREE relations into account, we showed that the UFH can explain the observed delay for FDs. We then derived a novel prediction: even though faire-causatives with unergative verbs require an analysis as faire-infinitif causatives, they will not be delayed like the FDs, since the unergatives have no internal argument requiring case valuation. This prediction was tested and supported. Finally, we examined certain parallels between RCPs and get-passives with respect to non-assignment of an external theta-role, which help to explain why these structures are not affected by the UFH.

The above findings raise several questions for future research. As described by Hirsch and Wexler (2006), children at least as old as age 5 struggle with accurate comprehension of the by-phrase in English nominals. One potential take-away from this observation is that children have difficulties acquiring the default Agent interpretation conveyed by the by-phrase, outside of those actional passives where young children do succeed, given adequate context or a structural prime. The par-phrase in both the FP and RCP constructions has been analyzed as similarly conveying a default Agent role (Guasti 1996; Folli and Harley 2007; Labelle 2013). As Guasti (2016) notes, this would seem to predict that children acquiring these constructions might experience difficulties in the accurate comprehension of the par-phrase. As several instances of par-phrases did appear in the spontaneous speech data prior to age 3, this prediction deserves further review. In light of Belletti’s (2016b) observation regarding the differences between da-phrases in the Italian FP and RCP constructions (the former involving a default Agent role assignment, the latter transmission of a thematic role from the verb), future research might test whether children acquiring French display asymmetries in the comprehension/production of these two constructions.

Another area in which the UFH makes predictions concerns the inverse/specificational copular construction in English. Specificational copular sentences (34a) appear to invert the typical order of arguments in a predicational copular construction (34b).

34.	a.	The doctor is Maria
	b.	Maria is the doctor

Under several analyses of specificational copular sentences including that developed in Mikkelsen (2005), the predicational argument (the doctor in (34)) is Merged in a structurally lower position than the referential argument (Maria) prior to being promoted to subject position, resulting in what appears to be a Minimality violation. If this violation must be circumvented via a more complex smuggling derivation, then a delay in the acquisition of specificational copular sentences is predicted under the UFH.

In conclusion, evidence from the acquisition of Romance causatives provides further evidence for the UFH, and contributes to a developing understanding of the role of freezing effects in acquisition. The similarities between get-passives and RCPs merits further acquisitional and theoretical research, particularly given the very early appearance of the latter in spontaneous speech data.