Acquisition of English adjectival resultatives: Support for the Compounding Parameter

ABSTRACT

Two distinctive types of complex predicates found in English are separable verb-particle combinations (“particles”) and adjectival resultatives (“ARs”). Snyder ties both to the positive setting of the Compounding Parameter (“TCP”). This predicts that during the acquisition of a [+TCP] language, any child who has acquired ARs or particles will also permit “creative” bare-stem, endocentric compounding. Existing support comes from children acquiring Japanese and English. Yet the same evidence introduces two new puzzles: (i) why is compounding acquired roughly a year earlier in English than in Japanese?; and (ii) in English, why is compounding always acquired at the same time as (and never substantially prior to) particles? Here, we argue that both puzzles can be explained if we allow the trigger for a single parameter-setting (e.g., [+TCP]) to be completely different for children acquiring different languages. Specifically, the trigger for [+TCP] (and hence, ARs) in English is proposed to be particles, which are unavailable in Japanese. Two novel predictions are tested and supported: (i) the frequency will be higher for particles than for any (other) potential trigger in child-directed English or Japanese; and (ii) children acquiring English (unlike Japanese) will have reliably adult-like comprehension of ARs by the age of 3 years.

Acknowledgments

We would like to thank the members of the University of Connecticut Language Acquisition Group (“Acqui-Lab”) for helpful discussion at all stages of this project. We are grateful to Pietro Cerrone, Emma Nguyen, Sarah Asinari, and Chantale Yunt, as well as UConn K.I.D.S., for their help with recruitment; to Nic Schrum, for recording our test sentences; and to Hiromune Oda, for help with the Japanese corpus. We also thank the audiences at GALA 14 and BUCLD 44 for their many helpful comments. Last but not least, we would like to express our gratitude for the insightful comments and feedback provided by three anonymous reviewers, Associate Editor Kristin Syrett, and Editor Jeff Lidz.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Declaration of interest

The authors report no conflict of interest. The authors alone are responsible for the content and writing of this article.

Data availability statement

The data that support the findings of this study are available at https://figshare.com/s/20813fe7e111fef58ecd.

Notes

¹ A third category, “spurious,” will be set aside here.

² As discussed by Carrier & Randall (1992:174, fn.1), there is variation across English speakers in their acceptance of specific instances of strong resultatives. For example, it seems that (3a) is fairly widely accepted, but (3b) is a bit more controversial.

³ Conceivably, the nine native speakers who accepted (5a) might have differed from other speakers in their lexical semantics for tatai-ta. If, for these nine, flatness of the patient actually was an outcome expected from their (personal) lexical semantics, the example would have been acceptable to them as a weak (not strong) resultative.

⁴ The cross-linguistic evidence from these surveys can be found in Snyder (2011:20, Table 1).

Table 1. Fixed effects, Experiment 1, all ages (N = 28)

⁵ The English particle in (9a) is prepositional; the Khmer particle in (9b) can also function as a verb. Yet in both cases, the verb and the particle function together to characterize an event of lifting the table and (thereby) causing it to move upward, and in both cases, the verb can be separated from the particle by a direct object, as shown in (9).

⁶ Note that the form of an AR in a [+TCP] language will surely vary somewhat as a result of other parameter settings, but the cross-linguistic findings behind Generalization I suggest that this will not be a problem. Despite formal differences in ARs, it seems that all [+TCP] grammars allow some type of complex predicate with the distinctive semantics of an AR. As long as this is true, the prediction of concurrent acquisition for ARs and compounding should be valid.

⁷ Age of acquisition was estimated as age of the child’s First of Regular Uses (or FRU) (Snyder 2007), the first clear use that was followed (within a month) by additional uses, with different lexical items. FRU as a diagnostic is based on “Grammatical Conservatism,” the idea that children do not start using a grammatical structure productively until they know how to form it correctly (Maratsos 1998; Snyder 2007).

⁸ Much the same pattern was found for German in Hanink & Snyder (2014).

⁹ Note that creative bare-stem compounding was tested using N-N compounds. In [+TCP] languages, N-N compounds are the type used the most frequently by adults and children.

¹⁰ Note that in (20) the result predicate aka-ku has been scrambled to a position before the direct object. This has the effect of minimizing the difference between (20) and (21). Given Otsu’s (1994) finding that 3-year-olds are reasonably comfortable with Japanese scrambling, the authors did not expect this use of scrambling to increase the difficulty. Also, note that it is highly controversial whether Japanese has asingle syntactic category of “Adjective” in the same way assumed for English. Nonetheless, for present purposes we will use “Adjective” as aconvenient label, and abstract away from the differences.

¹¹ This pattern plausibly follows from the fact that the V and the result AP form a complex predicate, and tense is already marked on the V.

¹² In addition to these 18 children, of whom six failed on the compounding task, there were two more (ages 4;08 and 4;11) who also failed on compounding. Yet the authors indicated that those two children may have been having difficulty attending to the task. We have therefore excluded them from the calculations in the text.

¹³ Alongside particles, one other [+TCP]-related structure that is available in English but not Japanese is the Goal-PP construction (cf. Beck & Snyder 2001). Yet Goal-PP constructions are quite problematic as a trigger for [+TCP]. Please see Xu & Snyder (2017) for data and discussion.

¹⁴ The age range is somewhat wide because we started with older children (to confirm that the task worked) and then moved progressively younger.

¹⁵ Some participants in Experiment I were bilingual. Among the 34 participants, seven used a language other than English at home (Bangla, Chinese, French, Rukiga, Russian, or Spanish). In all cases, parents reported that their child spoke English at least 90% of the time. Moreover, to anticipate the results, the children all performed at or near ceiling, regardless of whether they were mono- or bilingual.

¹⁶ As the associate editor (AE) noted, the current study cannot test whether children know that ARs pick out a maximal state since we did not contrast the degree of color in the study. Although it would be interesting and important to test this, it will be left for future research. In fact, we had reason to show that the painting event had just started and for the result color to be only a small part of the entity. In particular, we wanted to show the original color to children so that the contrast between that and the result color would be clear to them. Otherwise, children might get confused. For instance, if Lisa is painting the green table purple and the painting event is almost accomplished (i.e., the table is almost completely purple), children might accept the test sentence “Lisa is painting the table purple” for the wrong reason. In other words, children might wrongly interpret the adjective as attributive (i.e., “Lisa is painting the purple table”) and accept the sentence because the color of the table now is almost completely purple.

¹⁷ All the materials (including the PowerPoint and the stories for Experiment I and Experiment II) can be found at https://figshare.com/s/9d560a43e348e40ff810.

¹⁸ Of the six children who were excluded, five showed either a “yes” bias or a “no” bias; these children consistently answered “yes” (or “no”) to both practice/filler items and test items. One child, the youngest we tested (2;11), appeared to be answering randomly.

¹⁹ Note that WSR was chosen over a paired t-test because there were only five possible values of the DV (i.e., from 0 to 4 “Yes” responses) for each level (True, False); hence, the paired t-test’s assumptions about normality were unlikely to be met. WSR requires only that the pairwise difference score for each participant be on a scale that is at least ordinal; this will hold true when the difference scores result from subtraction performed on values along a scale that is at least interval. In our case, the values are the numbers of “Yes” responses to “True” versus “False” items. Given that the test items were extremely similar to one another (aside from their truth value), and given that we detected no outliers among the items, these values lie on a scale that is at least interval. Hence, as best we can determine, our data conform fully to the WSR assumptions.

²⁰ The sample size was limited due to the aforementioned pandemic. Fortunately Experiment I yielded a large effect size (Cohen’s standardized difference score d_z = 5.43), and it was therefore deemed reasonable to proceed with Experiment II even if the sample size would have to be smaller.

²¹ Regarding the excluded two children (3;08 and 3;04), one (3;08) showed a “yes” bias for both test items and control items; the other child kept changing her answers (e.g., initially answering “right” but then immediately saying “no”). Note, however, that even the “yes”-biased child produced clear ARs in the justifications. It was interesting to see a child fail on the TVJT but succeed at producing ARs.

²² As noted by a reviewer, some maternal ARs were discontinuous. In (36b), the result AP appears before the verb. There is an interesting question of how these discontinuous examples compare to continuous examples, in terms of their usefulness to the learner. In any case, even when discontinuous examples are included, ARs were quite infrequent. All the AR utterances that were identified in maternal speech can be found here: https://figshare.com/s/5ee048df8fe5e864fce1. Recursive compounds (both English and Japanese), English particles, and Japanese ARs can also be found at this link.

²³ Note that this is a conservative way to calculate the frequency of ARs in maternal speech because we did not include the corpora that did not contain any ARs. In other words, if we included those corpora, the average frequency of ARs would be much lower.

Additional information

Funding

This material is based in part on work supported by the program in Science of Learning & Art of Communication at the University of Connecticut, which is supported by the National Science Foundation under Grant DGE-1747486. (Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation.)