On the Acquisition of English Complex Predicates and Complex Word Formation: Revisiting the Parametric Approach

Abstract

Languages vary in their availability of productive endocentric bare-stem compounds (e.g., flower hat) and a range of complex predicates (separable verb-particles, double object datives, adjectival resultatives, put-locatives, make-causatives, and perceptual reports). To account for these cross-linguistic variations, two parameters have been proposed: the Compounding Parameter (TCP), which governs the formation of bare-stem compounds, separable verb-particles, and adjectival resultatives, and the Small Clause Parameter (SCP), which determines whether a verb can take a small clause complement. These parameters make testable predictions about children’s acquisition. If TCP and SCP are on the right track, we would expect correlations in the acquisition of structures governed by each parameter. This study examines these predictions by analyzing longitudinal corpora from 23 English-speaking children, assessing both the correlation between the acquisition of different structures and their acquisitional ordering. Our findings support both TCP and SCP, confirming that the acquisition of bare-stem compounds is closely associated with that of separable verb-particles, while the acquisition of (some) complex predicates is related. In addition, our results offer new insights into the potential triggers that children use to set each parameter. These findings contribute to our understanding of language variation and the role of parameter setting in first language acquisition.

1. Introduction

A central question in linguistic theory concerns the nature of cross-linguistic variation and how children make the correct choices for their target language in the process of language acquisition. In this study, we investigate one such case regarding English complex word formation and complex predicates.

A prominent proposal in this domain is the Compounding Parameter (TCP) (Snyder, 1995, 2001, et seq.), which posits that languages vary systematically in the availability of productive, endocentric, bare-stem noun-noun compounds (N-N compounds, e.g., cookie chair). Crucially, TCP has been argued to underlie not only compounding creativity but also the availability of complex predicates such as V-NP-particles (e.g., lift the table up, also referred to as separable verb-particles in the literature)1 and adjectival resultatives (e.g., wipe the table clean), as these constructions depend on a shared mechanism that combines the main verb and the secondary predicate to form a unit akin to a morphological compound at the point of semantic interpretation. Consequently, there is systematic association in terms of their availability: cross-linguistic typological surveys revealed that languages with a V-NP-particle construction consistently allow productive endocentric compounding (Snyder, 2007), and resultatives of the English type are found only in languages with productive endocentric compounding (Snyder, 2001).

Further extending this framework, Snyder (2012) introduces the Small Clause Parameter (SCP), which governs the availability of small clause complements, to explain why some languages (e.g., Japanese) allow endocentric N-N compounds but lack V-NP-particles and some type of adjectival resultatives. This parameter is hypothesized to explain the availability of a broader set of complex predicates, including V-NP-particles, (at least some) adjectival resultatives, double object (DO) datives, put-locatives, make-causatives, and perceptual reports, as illustrated in (1).

(1)	a. They lifted the table up.	(V-NP-particle)
	b. I hammered the metal flat.	(Adjectival resultative)
	c. Mary gave me a book.	(DO dative)
	d. I put the apple on the table.	(Put-locative)
	e. Jenny made the kid cry.	(Make-causative)
	f. John saw the boy leave.	(Perceptual report)

Together, TCP and SCP generate two key acquisition predictions: (i) in their acquisition, V-NP-particles, resultatives, and novel N-N compounds are predicted to be related with each other, reflecting their shared reliance on TCP, and (ii) in their acquisition, small-clause-dependent predicates may associate with each other, reflecting their shared reliance on SCP.

Empirical support for these predictions has emerged from acquisition studies in English, Japanese, and German (Snyder & Stromswold, 1997; Sugisaki & Isobe, 2000; Snyder, 2001, 2007; Wang et al., 2020, 2022). However, there remain some open questions. While correlations have been observed both among complex predicates and between complex predicates and N-N compounding (Snyder & Stromswold, 1997; Snyder, 2001, 2007), some prior research did not control for potential confounding factors, raising the question of whether these constructions are acquired together because they share some linguistic prerequisites or simply because they are part of the “grammar explosion”. Some studies, while controlling for the general trend of language development, examined only a subset of the relevant structures, leaving open the question of whether the acquisition of the remaining structures is correlated.

This study revisits these questions by examining the longitudinal corpora of 23 English-acquiring children. Employing statistical controls to disentangle parameter-driven acquisition from broader developmental factors, we investigate potential association between the acquisition of productive N-N compounds and complex predicates (V-NP-particles, put-locatives, DO datives, make-causatives, and perceptual reports). As a preview of the results, our findings confirm and extend some key findings in Snyder’s series of work, namely that a child’s acquisition of English V-NP-particles is closely connected to the acquisition of N-N compounds and that the acquisition of (some) complex predicates is interrelated.

By addressing some of the remaining questions in prior work, this study aims to provide more empirical evidence for the parametric approach of language acquisition. We hope the findings will shed light on discussions about the role of macroparameters in linguistic theory and the extent to which abstract parameters guide language development.

The rest of this paper is structured as follows: Section 2 reviews the relevant background of TCP and SCP, alongside prior acquisitional studies of these two parameters. Section 3 presents, in detail, the methodology, data analyses, and results of our corpus study. Section 4 summarizes the major findings, discusses their broader implications, identifies unresolved questions, and outlines directions for future research. Section 5 concludes the paper.

2. Background

2.1. On the Nature of Variations in Root Compounding and Complex Predicates: TCP and SCP

This section reviews Snyder’s parametric accounts for cross-linguistic variations in N-N compounding and complex predicate formation. Snyder (1995, 2001) proposed that languages vary parametrically with respect to whether the grammar allows creative endocentric, bare-stem N-N compounding, termed the Compounding Parameter (TCP). The positive setting of this binary parameter makes it possible for speakers to create novel endocentric compounds from uninflected roots on the fly, and for listeners to readily interpret their meanings based on context. For example, in English, a [+TCP] language, the compound flower hat could refer to a hat with flowers on it or a hat shaped like a flower. Furthermore, N-N compounds can be recursively formed, as in flower hat store, which might denote either a store selling flower hats or a store decorated with them (but not necessarily selling them). In contrast, [−TCP] languages do not permit speakers to freely create and interpret novel bare-stem N-N compounds. While bare-stem N-N compounds may exist, they are typically lexicalized, meaning that their interpretations are fixed and stored in the lexicon rather than creatively generated. For example, in Spanish, a [−TCP] language, the compound hombre rana (literally “man frog”) has a rigid, conventionalized meaning: underwater diver. In contrast, the English counterpart frog man could mean underwater diver but could just as easily refer to any man associated with frogs—a man studying frogs, a man with a pet frog, or even a man dressed in a frog costume.

Crucially, Snyder argues that the availability of English V-NP-particles and adjectival resultatives, as exemplified in (1a) and (1b), depends on the positive setting of this parameter.2 This link between N-N compounds and these complex predicates has gained support from cross-linguistic surveys. For example, Snyder (2001) reported that if a language allows adjectival resultatives of the English-type, it also has productive N-N compounding (see also Son, 2007). In addition, languages with a V-NP-particle construction consistently allow productive endocentric compounding (Snyder, 2007).

To formally characterize whether a language permits novel N-N compounding and its association with the availability of V-NP-particles and resultatives, Snyder (2012) proposes that a single semantic interpretation mechanism, Generalized Modification (GM, as defined in (2)), is responsible for interpreting both novel compounds and these complex predicates. Accordingly, TCP is defined in terms of GM, as in (3).3

(2)	Generalized Modification (GM)
If α and β are syntactic sisters under the node γ, where α is the head of γ, and if α denotes a kind, then interpret γ semantically as a subtype of α’s kind that stands in a pragmatically suitable relation to the denotation of β.
	(Snyder, 2012, p. 285, ex. 11.5)

(3)	The Compounding Parameter (TCP)
	The language (does/does not) permit Generalized Modification.
	(Snyder, 2012, p. 285, ex. 11.4)

Under this formulation, a [+TCP] language (e.g., English) permits GM, which allows two root nouns (e.g., frog and man) to combine syntactically into a bare-stem N-N compound, where man serves as the head and denotes individuals. The resulting compound frog man is interpreted as a subtype of men that stands in a pragmatically suitable relation to frogs—whether this means a man who studies frogs, a man who owns a frog, or a man dressed in a frog costume.4 The notion of a pragmatically suitable relation is deliberately underspecified, enabling speakers to derive contextually appropriate interpretations on the fly.

GM can apply not only to individual-kind predicates (such as frog and man), but also to “eventuality-kind” predicates, thereby allowing the formation of V-NP-particles and adjectival resultatives. Consider an English V-NP-particle lift the table up. The verb lift combines with a small clause (4) (Kayne, 1985; Hoekstra, 1988; Den Dikken, 1995, a.o.). Extending Chierchia’s (1998) specific proposal of semantic kinds, Snyder assumes that the verb lift denotes a kind of activity, while the small clause complement denotes a kind of state. Applying GM, the entire phrase is interpreted as a subtype of the “lifting” event-kind that stands in a pragmatically suitable relation to the result state (the table being up).5

(4)	[VP lift [SC the table up]]

Consider an adjectival resultative like wipe the table clean, whose syntactic structure is given in (5). In this structure, the verb wipe combines with the AP clean, and the object the table occupies the specifier position of the VP (5a). The verb wipe undergoes head movement and adjoins to v, thereby surfacing to the left of the table (5b). Snyder assumes that verb wipe denotes (in effect) a kind of activity, while clean denotes a kind of state. Applying GM, the whole phrase is construed as a subtype of the “wiping” event-kind that stands in a pragmatically suitable relation to the result state (the “clean” kind of state). The result is a kind of accomplish event, with “wiping” as its development and “clean” as its culmination.

(5)	a. [VP [the table] [V’ wiped [AP clean]]]]	(before head movement)
	b. [vP [wiped v] [VP [the table] [V’ <wiped> [AP clean]]]]

While GM being available ([+TCP]) enables the formation of certain complex predicates, such as V-NP-particles and adjectival resultatives, it does not guarantee their availability. Even among [+TCP] languages, the availability of specific complex predicates may vary. For example, Japanese, like English, is [+TCP], allowing endocentric N-N compounds as a fully productive and creative process. However, it does not permit V-NP-particles or “strong” resultatives of the English type (6).6

(6)	Strong resultatives:
	a.	??	John-ga	kinzoku-o	petyanko-ni	tatai-ta.
			John-NOM	metal-ACC	flat	pound-PAST
			“John pounded the metal flat.”
			(Washio, 1997, p. 5, (16b))
	b.	*	boku-wa	zibun-o	kutakuta-ni	odot-ta.
			I-TOP	self-ACC	tired	dance-PAST
			“I danced myself tired.”
			(Washio, 1997, p. 20, (67c))

Snyder (2012) proposed that the licensing of small clause complements requires the positive setting of another parameter, which he called the SCP, defined in (7). Specifically, languages vary with respect to whether a verb can take a small clause complement. [+SCP] languages (e.g., English) allow small clause complementation whereas [−SCP] languages (e.g., Japanese) do not (8). It is worth noting that there exist multiple views and definitions of small clauses. In Snyder’s usage, a small clause is characterized primarily by two properties: (i) its subject receives structural accusative case from its selecting verb, and (ii) if the predicate of the small clause is verbal, it appears in the form of a bare infinitive.

(7)	The Small Clause Parameter (SCP)
	The language (does/does not) allows small clause complements to V.

(8)	Perceptual reports
	a. English: John saw [SC Mary walk].
	b. Japanese:
		*	John-ga	[SC Mary-o	aruku]	mita.
			John-NOM	Mary-ACC	walk	see-PAST
			“John saw Mary walk.”
			(Snyder, 2012, p. 296, 11.24)

According to Snyder, the formation of English complex predicates in (1), including V-NP-particles, strong adjectival resultatives, DO datives, put-locatives, make-causatives, and perceptual reports, requires the positive setting of the SCP, in the sense that the main verbs take small clauses as their complements.

Table 1. Constructions requiring the positive setting of TCP or SCP.

	Require Positive Setting of TCP?	Require Positive Setting of the SCP?
N-N compound	Yes
Weak adjectival resultative	Yes
Strong adjectival resultative	Yes	Yes
V-NP-particle	Yes	Yes
DO dative		Yes
Put-locative		Yes
Make-causative		Yes
Perceptual report		Yes

summarizes the constructions that require a positive setting of TCP or SCP.

2.2. Previous Acquisition Studies

TCP and SCP make distinct yet interrelated predictions for the acquisition of creative endocentric N-N compounds and complex predicates. TCP predicts that the acquisition of creative endocentric N-N compounds, adjectival resultatives, and V-NP-particles are inherently linked, as all three structures depend on the positive setting of the parameter. SCP predicts that the acquisition of the complex predicates in (1), excluding weak resultatives, should be interrelated, as they all require the positive setting of SCP.

Previous acquisition studies report findings consistent with these predictions. Snyder and Stromswold (1997) investigated the acquisition of English complex predicates by analyzing longitudinal corpora from 12 English-acquiring children in the CHILDES database. Their results showed that DO datives, V-NP-particles, put-locatives, make-causatives, and perceptual reports tend to emerge as a group, typically before age three. This finding is consistent with the predictions of SCP. However, it is possible that these constructions emerge together as part of the “grammar explosion”—a developmental phase in which children acquire multiple syntactic structures in rapid succession. This confound gives rise to a concern about the absence of control measures to rule out this alternative explanation.

Snyder (2001) investigated the relationship between creative N-N compounds and various complex predicates, including V-NP-particles, DO datives, put-locatives, make-causatives, and perceptual reports in terms of children’s acquisition. Analyzing 10 children’s longitudinal corpora, this study incorporated rigorous controls to eliminate confounding factors like the “grammar explosion”. A key finding was that there was an exceptionally strong correlation between V-NP-particles and N-N compounds (r = 0.98, t(8) = 12.9, p < 0.0001) and a near identity relation in children’s age of acquisition of these two constructions. These findings were later replicated in Snyder’s (2007) expanded analysis of a larger sample (19 children), despite the fact that he used a different measure to estimate when children acquire certain constructions. Similar findings were reported for German in Hanink and Snyder (2014). These findings provide strong support for TCP.

As one may notice, none of the aforementioned corpus studies examined adjectival resultatives. This exclusion is due to their infrequent occurrence in spontaneous speech, both in child and adult production, making corpus analyses less reliable for studying their acquisition. For instance, Wang et al. (2020) examined longitudinal corpora from 20 English-acquiring children but found that only 10 of them produced at least one clear instance of an adjectival resultative. Furthermore, Wang et al. (2022) analyzed maternal input and found that adjectival resultatives occurred at low frequencies, ranging from 2 to 568 per 100,000 utterances, with a mean frequency of 30. Given this low frequency in spontaneous production, corpus-based methods may be unreliable for determining their age of acquisition (AoA), and behavioral experiments provide a more reliable approach to studying the acquisition of resultatives. Wang et al. (2022) conducted a Truth Value Judgment task to examine English-acquiring children’s comprehension of adjectival resultatives. The study tested children as early as experimentally feasible—shortly after their third birthday. The results showed that even at this early age, children were able to interpret adjectival resultatives, aligning with previous findings on the acquisition of novel N-N compounds and other complex predicates. Another experimental study supporting TCP comes from Sugisaki and Isobe (2000), who investigated Japanese-speaking children’s acquisition of resultatives and novel N-N compounds. Using a comprehension task, they assessed children’s understanding of resultatives, while an elicitation task was used to test their production of novel N-N compounds. Their findings revealed a strong association between the acquisition of resultatives and N-N compounding, with both structures emerging around age 3;06. (For a discussion of factors contributing to AoA differences between Japanese and English, see Wang et al. (2022).)

2.3. Summary and Remaining Question

In sum, previous acquisitional studies revealed that the acquisition of V-NP-particles and adjectival resultatives is closely correlated with the acquisition of novel N-N compounds, which supports TCP.

Compared to the observed correlation between V-NP-particles and resultatives with N-N compounds, the question of whether different complex predicates are interrelated in the way predicted by the SCP remains unresolved. Although Snyder and Stromswold (1997) reported that complex predicates tend to be acquired as a group, consistent with what the SCP predicts, their findings are potentially confounded by general developmental factors. While Snyder (2001) controlled for this confound, he did not examine the correlations among complex predicts themselves.7 As a result, it remains unclear whether these constructions are genuinely linked in terms of their acquisition, particularly after controlling for general developmental factors (such as “grammatical explosion”). In the next section, we present a new corpus study examining the acquisition of novel N-N compounds alongside various complex predicates, including V-NP-particles, put-locatives, DO datives, make-causatives and perceptual reports. Our main goal is to assess potential relations between the acquisition of these structures.

In addition, some prior corpus studies have been limited to 10 or 12 English-speaking children, raising concerns about generalizability. To address this concern, we conducted a larger-scale investigation to determine whether previous findings can be replicated and extended to a larger sample.

3. Corpus Study

3.1. Method

We analyzed the spontaneous production data of 23 children, whose longitudinal corpora are available from the English-North America and English-UK repositories of the CHILDES database (MacWhinney, 2000). These children’s ages fall within the range of 1;0 to 3;0 at the time of the initial transcript recording (see

Table 2. Information of the corpora analyzed in the study (Braunwald, 1971; Brown, 1973; Demuth & McCullough, 2009; Jones & Rowland, 2017; Nelson, 1989; Suppes, 1974; Theakston et al., 2001).

Collection Name	Corpus	Child Name	Age Range (Year; Month)	No. of Transcripts	No. of Child Utterances
Eng-UK	Lara	Lara	1;09–3;03	120	60,441
Eng-UK	Manchester	Anne	1;10–3;08	35	22,687
Eng-UK	Manchester	Aran	1;11–2;10	33	18,293
Eng-UK	Manchester	Carl	1;08–2;08	33	26,280
Eng-UK	Manchester	Dominic	1;10–2;10	35	24,239
Eng-UK	Manchester	Gail	1;11–2;11	34	18,462
Eng-UK	Manchester	Joel	1;11–2;10	35	21,374
Eng-UK	Manchester	John	1;11–2;10	32	14,449
Eng-UK	Manchester	Liz	1;11–2;10	33	16,457
Eng-UK	Manchester	Ruth	1;11–2;11	33	21,433
Eng-UK	Manchester	Warren	1;10–2;09	36	17,813
Eng-NA	Braunwald	Laura	1;05–7;00	200	40,860
Eng-NA	Brown	Adam	2;03–5;02	55	46,743
Eng-NA	Brown	Eve	1;06–2;03	20	12,113
Eng-NA	Brown	Sarah	2;03–5;01	139	38,096
Eng-NA	Nelson	Emily	1;07–2;08	54	4893
Eng-NA	Suppes	Nina	1;11–3;03	52	33,181
Eng-NA	Providence	Alex	1;04–3;05	51	29,252
Eng-NA	Providence	Ethan	0;11–2;11	50	21,920
Eng-NA	Providence	Lily	1;01–4;00	80	40,027
Eng-NA	Providence	Naima	0;11–3;10	88	43,500
Eng-NA	Providence	Violet	1;02–3;11	51	17,296
Eng-NA	Providence	William	1;04–3;04	44	21,291

). This age range is determined based on the findings of Snyder and Stromswold (1997).

Our primary goal was to investigate potential links in children’s AoA across several constructions. Specifically, we examined (i) the association between novel N-N compounds and V-NP-particles, as predicted by TCP; and (ii) the interrelations among complex predicates (V-NP-particles, DO datives, put-locatives, make-causatives, and perceptual reports), as predicted by the SCP. (iii) Although put-locatives, DO datives, make-causatives, and perceptual reports (1c–f) are not directly tied to TCP, their acquisition may nonetheless correlate with that of N-N compounds if the triggers that English-acquiring children use to set TCP also contribute to setting SCP. Therefore, we also examined the relation between novel N-N compounds and these complex predicates.

To determine the AoA of a construction, we used the measure of First Regular Use (FRU), defined as the first use followed within a month by additional uses involving non-identical lexical items (Snyder, 2007). In cases where FRU could not be determined due to the low frequency of a construction in the corpus, we also noted the First Clear Use (FCU) for each child.

We first used a Python version 3.10 program together with Garber’s (2019) CHA file Python parser to identify, for each child, potential candidates for the following constructions: (a) novel endocentric N-N compounds; (b) V-NP-particles in lines containing a verb and a potential particle; (c) put-locatives in lines containing the verb put, place, or set along with a preposition; (d) DO datives; and (e) make-causatives in lines containing any form of the verb make followed by a bare infinitival verb with its NP “subject”; (f) perceptual reports in lines containing any form of the verbs hear, see, or watch followed by a bare infinitival verb with its NP “subject”. Next, manual searches were conducted. For constructions (b) through (f), two research assistants manually examined the Python output data and identified the first five occurrences of each construction. One of the authors then reviewed their annotations and determined the FRU/FCUs for each child. For (a) N-N compounds, the authors assessed each potential compound in the context of the transcript to determine whether it was truly novel based on two criteria: first, it could not be a lexicalized form (e.g., toothbrush, orange juice); second, the context supports the interpretation of the compound that the child created on the spot, meaning that it made sense within the context and had not been previously used by an adult in the corpus. The authors identified the first three to five clear uses of novel compounds before determining the FRU/FCUs for each child. Across all the constructions examined, we excluded imitations, routines, repetitions, and otherwise unclear utterances.

3.2. Data Analysis and Results

Table 3. Ages of FRU/FCU (in years) for each construction type and child. For N-N compounds, V-NP-particles, put-locatives, and DO datives, FRU was reported unless otherwise marked. An asterisk (*) indicates cases where FCU was used as an approximation for AoA. For make-causatives and perceptual reports, FRU could not be determined for most children due to their low frequency, so only FCU was reported. A slash (/) indicates that neither FRU nor FCU was available.

Child	N-N Compound	V-NP-Particle	Put-Locative	DO Dative	Causative *	Perceptual Report *
Lara	2.15	1.88	1.93	2.52	2.56	2.66
Anne	2.05	1.93	2.05	2.33	/	/
Aran	1.99	2.01	2.08	2.39	2.72	2.62
Carl	1.92	1.73	2.00	2.24	2.59	1.73
Dominic	2.21	2.17	2.25	2.34	2.67	/
Gail	2.12	1.99	2.11	2.18	2.11	2.95
Joel	2.09 *	2.07	2.39	2.57	2.32	/
John	2.00	1.96	2.24	2.05 *	/	/
Liz	2.04	1.96	2.04	2.88	2.32	2.86
Ruth	2.34 *	2.26	2.61	2.49	2.61	2.72
Warren	1.85	1.85	2.18	2.49	/	2.60
Laura	2.22	1.76	1.76	1.57	3.28	1.56
Emily	1.96	1.82	1.82	1.82	1.86	2.59
Nina	1.96	2.03	2.05	2.10	2.07	/
Alex	2.44	2.32	2.52	2.52	2.32	2.86
Ethan	1.29	1.46	1.55	1.94	1.82	/
Lily	1.78	1.99	2.21 *	2.38	2.51	2.53
Naima	1.38	1.55	1.72	1.77	/	1.91
Violet	2.42	1.82	1.91	2.26	2.28	/
William	2.30	2.27	2.42	2.73	/	2.61
Adam	2.26	2.26	2.26	2.26	/	2.34
Eve	1.92	1.75	1.75	1.83	/	1.67
Sarah	2.60 *	2.55	2.85	2.83	2.85	2.96

reports the ages of FRU/FCU (in years) for the constructions under investigation. As noted in Section 3.1, we used the age of FRU as a measure for AoA. However, when FRU could not be determined from the corpus for a given child, FCU was used instead. For make-causatives and perceptual reports, FRU could be determined for only 5 out of 23 children (less than one-fourth), due to the low frequency of these two constructions in the corpora. Moreover, for 4 of these 5 children, the FRU and FCU ages were identical. Therefore, only FCU was reported for make-causatives and perceptual reports.

Given the limited data for make-causatives and perceptual reports, we set these constructions aside for now and focus on the others. We will return to make-causatives and perceptual reports later in the paper.

3.2.1. N-N Compounds, V-NP-Particles, Put-Locatives, and DO Datives

TCP predicts that the AoA of V-NP-particles should correlate with that of N-N compounds. SCP predicts that the AoAs of complex predicates (V-NP-particles, put-locatives, and DO datives) should be correlated with one another. To test these predictions, we conducted a series of correlation analyses examining the relationship between different construction pairs.

Table 4. Correlation between different constructions pairs (n = 23).

Comparison	r	t	df	p
N-N compound vs. V-NP-particle	0.796	6.019	21	<0.001
N-N compound vs. put-locative	0.685	4.314	21	<0.001
N-N compound vs. DO dative	0.490	2.577	21	0.018
V-NP-particle vs. put-locative	0.931	11.693	21	<0.001
V-NP-particle vs. DO dative	0.674	4.185	21	<0.001
put-locative vs. DO dative	0.725	4.819	21	<0.001

summarizes the correlation coefficients (r), t-values, degrees of freedom (df), and p-values.

The results revealed significant associations across all construction pairs, with some pairs exhibiting particularly strong correlations, such as N-N compounds and V-NP-particles, V-NP-particles and put-locatives, as well as put-locatives and DO datives. These results support the idea that N-N compounds and V-NP-particles are closely related (both requiring GM). Furthermore, the results suggest that the acquisition of these complex predicates relies on shared prerequisite(s), such as allowing the verb to take a small clause complement (i.e., the positive setting of the SCP).

However, alternative explanations for these results exist. As pointed out by Snyder (2001, 2007), the co-emergence of these structures could stem from a broader “grammatical explosion” phase, during which children’s processing capacity crosses a critical threshold, enabling these constructions to naturally enter their spoken repertoire. In other words, the grammatical prerequisites for these constructions might be entirely irrelevant.

To examine this possibility, we followed Snyder’s (2007) approach, using morpheme-based mean length of utterance (MLUm) as an index of children’s processing capacity. Moreover, we took the age at which a child’s MLUm first reaches a threshold as an estimate of when they attained the corresponding level of processing ability. Specifically, we calculated the average MLUm from each child’s transcript that contained the FRU of V-NP-particles. This construction is frequent in the corpora, making its FRU reliably determinable for each child. The resulting threshold, 2.134, closely aligns with the value reported in Snyder (2007), 1.919, despite differences in sample selection. For each child, we then calculated the age that their MLUm first reached this threshold, which we subsequently controlled for in our analyses.

We conducted partial correlation tests for each construction pair, using the age at which a child’s MLUm first reached 2.134 as a covariate. As shown in

Table 5. Partial correlation between different constructions pairs (n = 23).

Comparison	r	t	df	p
N-N compound vs. V-NP-particle	0.726	4.720	20	<0.001
N-N compound vs. put-locative	0.551	2.950	20	0.008
N-N compound vs. DO dative	0.262	1.212	20	0.240
V-NP-particle vs. put-locative	0.871	7.916	20	<0.001
V-NP-particle vs. DO dative	0.412	2.021	20	0.057
put-locative vs. DO dative	0.516	2.693	20	0.014

, most correlations remained statistically significant even after controlling for this developmental measure, suggesting that the observed AoA correlations among these constructions cannot be fully explained by general developmental effects. More specifically, the significant partial correlation between N-N compounds and V-NP-particles supports the hypothesis that both constructions are linked to TCP. In addition, the significant partial correlations between V-NP-particles and put-locatives, as well as between put-locatives and DO datives, provided support for the hypothesis that their acquisition shares some common prerequisite, such as the positive setting of the SCP.

Two additional patterns do not follow directly from the predictions of TCP and the SCP. The first is the observed correlation between N-N compounds and put-locatives; the second is the lack of correlation between DO datives and V-NP-particles, which we discuss next.

First, although put-locatives are not hypothesized to depend on the positive setting of TCP, their AoA was significantly correlated with that of N-N compounds, a pattern also observed in Snyder (2001). One possible explanation for this finding is that N-N compounds and put-locatives are indirectly linked through their shared association with V-NP-particles, even though their acquisition is governed by distinct parameters (TCP for N-N compounds, SCP for put-locatives). Here we propose that for English, the same set of linguistic triggers may be responsible for setting the values of both TCP and SCP.8 Two plausible candidates for these triggers are V-NP-particles and strong adjectival resultatives, as these structures are subsumed under both TCP and SCP. Comparing the two candidates, strong adjectival resultatives are less likely to serve as triggers for these parameters, because they are infrequent in spontaneous speech, as discussed in Section 2.2. Wang et al. (2022) found that resultatives occur at a very low frequency in English maternal input, averaging only 30 instances per 100,000 utterances. In contrast, they found that V-NP-particles are much more frequent in English maternal input, with an average of 3570 instances per 100,000 utterances. Given this disparity in input frequency, V-NP-particles are a more plausible candidate for serving as the trigger for both TCP and SCP. Their associated structures, N-N compounds which are subsumed under TCP, and put-locatives, which are subsumed under the SCP, are thus indirectly correlated.

Second, DO datives exhibit no statistically significant correlation with N-N compounds, and their correlation with V-NP-particles is only marginally significant. The lack of correlation with N-N compounds, also reported in Snyder (2001), is unsurprising, given that the acquisition of DO datives is not predicted to depend on the positive setting of TCP. The latter pattern, however, is unexpected, given that both DO datives and V-NP-particles are hypothesized to require the positive setting of the SCP. We consider two possible explanations for this finding: First, it is possible that DO datives do not actually depend on the positive setting of the SCP. If this is the case, there would be no expectation that DO datives and V-NP-particles should be related to one another in their timing of acquisition. A second possibility is that DO datives do depend on the positive setting of the SCP, but also require some additional prerequisite, which is acquired late by at least some children. In this case, the positive setting of the SCP (evidenced by V-NP-particles becoming available in child’s speech) alone would be a relatively weak predictor of when DO datives became available in a child’s speech. This explanation further predicts that children’s ages of acquisition for DO datives will be either the same as, or later than, the corresponding ages for V-NP-particles. To test this, we conducted a pairwise t-test comparing children’s age of FRU for V-NP-particles and DO datives. The results showed that DOs are acquired significantly later than V-NP-particles (t(22) = 5.865, p < 0.001). Thus, despite the relative “noisiness” of the data for DO datives (and without yet identifying the proposed late-acquired prerequisite), the available evidence is still consistent with the hypothesis that DO has the positive setting of the SCP as one of its prerequisites.

Along similar lines of reasoning, if V-NP-particles serve as the trigger for TCP and the SCP, we make a further prediction that their associated constructions, N-N compounds and put-locatives, should not be acquired prior to V-NP-particles; rather they should emerge around the same time as or later than V-NP-particles. To test this prediction, we conducted pairwise t-tests for each construction pair to investigate potential ordering effects, including the previously reported ordering between DO datives and V-NP-particles. The results, summarized in

Table 6. Pairwise t-tests between different construction pairs (n = 23).

Comparison	t	df	p
N-N compound vs. V-NP-particle	2.143	22	0.043
N-N compound vs. Put-locative	−1.190	22	0.247
N-N compound vs. DO dative	−3.317	22	0.003
V-NP-particle vs. Put-locative	−5.844	22	<0.001
V-NP-particle vs. DO dative	−5.865	22	<0.001
Put-locative vs. DO dative	−3.279	22	0.003

, are consistent with our prediction: N-N compounds are acquired significantly later than V-NP-particles, and put-locatives are acquired significantly later than V-NP-particles. However, N-N compounds and put-locatives show no consistent ordering effect, because the acquisition of the two constructions is governed by different parameters. DO datives are acquired significantly later than other constructions subsumed under the SCP, V-NP-particle and put-locatives, suggesting an additional prerequisite for acquisition, which we will discuss in Section 4.

3.2.2. Make-Causative Constructions

We next examine make-causative constructions. First, we excluded children for whom the FRU or FCU of make-causatives was not found in the corpora, resulting in a dataset of 16 children. We then conducted a series of correlation tests to examine the relationship between make-causatives and the other constructions under investigation. If the acquisition of make-causatives requires the positive setting of the SCP, we predict that it should correlate with that of some other complex predicates subsumed under the same parameter. In contrast, we do not predict such a correlation with N-N compounds, as make-causatives do not depend on the positive setting of TCP.

As shown in

Table 7. Correlation between make-causatives and other constructions (n = 16).

Comparison	r	t	df	p
N-N compound vs. make-causative	0.491	2.110	14	0.053
DO vs. make-causative	0.098	0.370	14	0.717
V-NP-particle vs. make-causative	0.318	1.253	14	0.231
Put-locative vs. make-causative	0.308	1.210	14	0.246

, while all correlations were positive, none reached statistical significance. We propose two possible explanations for the lack of significance correlation. One possibility is that the acquisition of make-causatives does not require the positive setting of SCP, but instead depends on distinct linguistic prerequisites. Another possibility is that the low frequency of this construction inherently produces “noisier” data, making it harder to detect correlations in spontaneous production. This is supported by the lack of FRU in most children (18 out of 23) and the absence of both FRU and FCU in about one-third of children (7 out of 23) in our sample.

3.2.3. Perceptual Reports

In this section, we examine perceptual reports, another low-frequency construction, as evidenced by the absence of FRU/FCU in several children in our sample. Following the approach used for other constructions, we first excluded children for whom the FRU or FCU for perceptual reports was not present in the corpora, resulting in a dataset of 16 children. We then conducted a series of correlation tests to examine the relationship between perceptual reports and other constructions under investigation. If perceptual reports require the positive setting of the SCP, their acquisition should correlate with that of other complex predicates. On the other hand, since they do not require the positive setting of TCP, no such correlation is expected with N-N compounding.

The results, summarized in

Table 8. Correlation between perceptual reports and other constructions (n = 16).

Comparison	r	t	df	p
N-N compound vs. perceptual report	0.446	1.866	14	0.083
DO dative vs. perceptual report	0.743	4.158	14	0.001
V-NP-particle vs. perceptual report	0.665	3.332	14	0.005
put-locative vs. perceptual report	0.652	3.214	14	0.006

, showed that perceptual report constructions were correlated with other complex predicates but not with N-N compounds, suggesting that their acquisition requires the positive setting of the SCP but not TCP.

Next, we controlled for developmental factors using the same approach as in Section 3.2.1. Specifically, we calculated the average MLUm from transcripts where children’s FRU of V-NP-particles was observed, which was 2.219. We then calculated the age at which each child’s MLUm first exceeded this threshold. Partial correlation analyses were conducted while controlling for this covariate. The results are summarized in

Table 9. Partial correlation between perceptual reports and other constructions (n = 16).

Comparison	r	t	df	p
N-N compound vs. perceptual report	0.160	0.583	13	0.570
DO dative vs. perceptual report	0.594	2.661	13	0.020
V-NP-particle vs. perceptual report	0.459	1.862	13	0.085
put-locative vs. perceptual report	0.451	1.821	13	0.092

.

The results revealed a significant correlation between perceptual reports and DO datives, suggesting that their acquisition shares a common prerequisite, such as the positive setting of the SCP. However, only marginally significant correlations were observed between perceptual reports and V-NP-particles or put-locatives, and (as expected) no significant correlations were found between perceptual reports and N-N compounds.

Similar to our analysis of DO datives, we conducted pairwise t-tests to examine whether there is an ordering effect between perceptual reports and other complex predicates, particularly V-NP-particles, which may serve as a trigger for the SCP. This analysis allowed us to assess whether the acquisition of perceptual reports is truly independent from the acquisition of other complex predicates. The results are summarized in

Table 10. Pairwise t-tests between perceptual reports and other complex predicates (n = 16).

Comparison	t	df	p
DO dative vs. V-NP-particle	4.392	15	0.001
DO dative vs. put-locative	2.514	15	0.024
DO dative vs. perceptual report	−1.788	15	0.094
V-NP-particle vs. put-locative	−4.671	15	0.000
V-NP-particle vs. perceptual report	−5.105	15	0.000
put-locative vs. perceptual report	−3.451	15	0.004

, where we also included results on DO datives to verify previous results regarding the ordering relationship between DO datives, V-NP-particles, and put-locatives.

The results showed that, like DO datives, perceptual reports are acquired significantly later than V-NP-particles and put-locatives, suggesting that their acquisition may depend on the positive setting of SCP, along with additional prerequisites.9 At the same time, we emphasize that, similar to make-causatives, the low frequency of perceptual reports in children’s spontaneous speech may introduce noise, making it difficult to draw definitive conclusions about their relationship to other complex predicates. As such, the results should be interpreted with caution.

4. Discussion

In Section 3, we analyzed data from 23 children to explore potential associations among six constructions: novel N-N compounds, V-NP-particles, put-locatives, DO datives, make-causatives, and perceptual reports. We summarize the key findings below: First, children’s acquisition of V-NP-particles correlates with that of N-N compounds, supporting the hypothesis that both depend on the positive setting of TCP. Second, we found associations between V-NP-particles and put-locatives, suggesting that these constructions share common linguistic underpinnings, such as the positive setting of the SCP. Third, DO datives and perceptual reports showed no significant correlations with V-NP-particles but emerged later in children’s spontaneous speech, suggesting that they may involve not only the positive setting of the SCP but also additional prerequisites. In contrast, make-causatives did not show correlation with any other constructions, suggesting that their acquisition may not rely on either TCP or SCP. However, results for make-causatives and perceptual reports should be interpreted with caution due to their low frequency in children’s spontaneous speech. For such low-frequency constructions, behavioral experiments may provide a more reliable means for investigating children’s acquisitional trajectories, an avenue we leave for future research. Overall, our findings align with Snyder’s (1995 et seq.) parametric approach to complex word formation and complex predicates.

We next discuss the broader implications of these findings and outline some remaining questions for future research.

4.1. TCP and SCP Setting Across Languages: Focusing on English and Japanese

According to Snyder (2001), TCP has an initial, unmarked setting: “−“. This means that the child initially has a grammar that does not permit GM. At some point, if the child encounters evidence that her language allows structures requiring the application of GM (such as V-NP-particles), she will set the parameter to its marked value “+”.

In the same spirit, and assuming a system of “subset parameters” in the sense of Wexler and Manzini (1987), the SCP also has an initial, unmarked value “−”. That is, a child starts with a grammar that disallows verbs from taking small clause complements. If she later encounters structures that requires small clause complementation, she will switch the setting of SCP from “−” to “+”.

For children acquiring English, we have proposed that V-NP-particles may serve as the trigger for setting both TCP and SCP. This proposal is based on two findings: First, the acquisition of N-N compounds is significantly correlated with that of put-locatives, although the two constructions fall under different parameters; and second, V-NP-particles are acquired significantly earlier than the other constructions under discussion. These observations suggest that V-NP-particles may play a central role in setting both parameters.

A related question arises for languages like Japanese, which are [+TCP, −SCP] and lack V-NP-particle constructions: How do children acquiring such languages determine the parameter values?

Snyder (2016) and Wang et al. (2022) argue that children acquiring different languages may rely on different structures and traverse distinct paths to reach the same parametric setting for TCP. They propose two candidate structures that might enable Japanese-acquiring children to determine that their target language is [+TCP]. One candidate is (weak) resultatives. If learners can recognize a resultative construction encodes both a primary activity and a secondary change of state within a single VP, this could signal that the target language allows GM. Another candidate is recursive compounding,10 based on Namiki’s (1994) generalization that a language allows bare-stem endocentric compounding if and only if it also allows recursive compounding (e.g., peanut butter sandwich in English or gakusee eiga kenkyuukai “student film club” in Japanese). Exposure to a robust number of recursive endocentric compounds may allow learners to infer that the language permits endocentric compounding as a creative process, leading to a [+TCP] setting. Wang et al. (2022) further suggest that, for Japanese, recursive compounding is a more plausible trigger for [+TCP] than resultatives, given their frequency in child-directed speech: on average, resultatives occur only about 2 times per 100,000 utterances, whereas recursive compounds appear around 210 times per 100,000 utterances.11

Before ending this subsection, we add a brief note on the setting of SCP and the acquisition of resultatives. As discussed earlier, children acquiring Japanese are assumed to begin with the unmarked [−SCP] setting and to shift to [+SCP] only when they encounter supporting evidence. This raises the question of whether exposure to weak resultatives (despite their rarity) might lead Japanese-acquiring children to mistakenly identify their language as [+SCP], given that weak resultatives superficially resemble strong resultatives and could be analyzed as involving a small clause structure. The absence of small clause structures in adult Japanese (e.g., V-NP-particles, strong resultatives, perceptual reports) suggests two possible acquisition paths: One possibility is that children temporarily adopt the incorrect [+SCP] setting upon encountering weak resultatives, but later backtrack when they fail to find further evidence that would support the positive setting. Another possibility, which we find more plausible, is that children are “grammatically conservative” (see Snyder 2007, 2011, 2021 for other independent evidence): they retain the unmarked [−SCP] setting unless and until they encounter robust, unambiguous positive evidence for [+SCP]. Only then do they commit to the marked value. The main reason to favor this conservative acquisition path is that weak resultatives do not necessitate a small clause analysis. They are compatible with alternative syntactic structures that lack a small clause (e.g., (5)) and therefore fail to provide unambiguous triggering evidence for [+SCP]. This raises a further question: Given the surface similarity between strong and week resultatives, do English-acquiring children distinguish between them and, if so, how? We leave this question for future research.

4.2. Prerequisite for Acquiring DO Datives

In Section 3, we proposed that the acquisition of DO datives requires not only the positive setting of SCP, but also some additional prerequisite. Here, we speculate on what this prerequisite might be, in light of the delayed emergence of DO datives compared to V-NP-particles and put-locatives. One possibility is that children need to recognize the connection between DO datives and the syntax of possession. According to Harley (2002), a language will not permit surface structures in which a “possessor” NP c-commands a “possession” NP, such as DO datives, unless it has P_HAVE, a prepositional element expressing possession. In fact, researchers have argued that English DO datives involve a small clause headed by an element like P_HAVE (Green, 1974; a.o.). On this view, a sentence like John gave Mary the book expresses a caused-possession meaning: John’s giving of the book caused Mary to have the book. If this analysis is correct, acquiring DO datives requires more than recognizing the presence of a small clause. English-acquiring children must also posit an abstract functional head like P_HAVE, and determine that it heads the lower VP shell. This process may also involve understanding how possession is grammatically encoded in English, and adding P_HAVE to the inventory of abstract functional elements that are allowed to occur in a syntactic numeration. Developing these may take extra time, beyond what is needed to set the SCP.

4.3. On the Relation Between English To-Datives and Other Complex Predicates

Another open question concerns the relation between English to-datives (e.g., give a book to John) and the complex predicates discussed in this paper. According to Snyder and Stromswold (1997), to-datives share some features with other complex predicates, particularly DO datives, suggesting that their acquisition should be related.

Previous studies have consistently found that to-datives are acquired significantly later than DO datives (e.g., Campbell & Tomasello, 2001; Snyder & Stromswold, 1997; Viau, 2007). Campbell and Tomasello (2001) attributed this ordering effect to an asymmetry in linguistic input, arguing that the earlier acquisition of DO datives results from their greater frequency in child-directed speech. In contrast, Snyder and Stromswold (1997) proposed that while to-datives share some common property with DO datives (and other complex predicates), to-datives require an additional prerequisite that DO datives lack, which explains their later acquisition. Specifically, they suggested that this extra requirement could be children’s understanding of the dative use of to. Supporting this hypothesis, they found that children’s acquisition of to-datives was significantly correlated with their acquisition of dative to in dyadic constructions (e.g., This bag belongs to John; Something happened to Mary).

Following Snyder and Stromswold’s (1997) proposal, the acquisition of to-datives should be related with the acquisition of other complex predicates. Additionally, their account predicts a developmental ordering effect, where to-datives should be acquired later than other complex predicates. This issue requires further investigation, and future research may consider including to-datives as part of their evaluation of the SCP.

5. Conclusions

To conclude, in this study we conducted a larger-scale corpus analysis to examine the parametric accounts for the acquisition of complex predicates (V-NP-particles, put-locatives, DO datives, causatives, and perceptual reports) and complex word formation (creative endocentric N-N compounds). By examining the partial correlation between their acquisition and ordering effects, we aimed to determine whether there are genuine parametric dependencies among these constructions. The results provide supporting evidence for TCP and SCP, shedding light on the potential triggers (i.e., V-NP-particles) that English-acquiring children use to set the values of these two parameters. We hope this work advances our understanding of how parametric variation influences language acquisition and elucidates the developmental interplay among complex predicates and endocentric N-N compounding.