Cognitive inhibition explains children’s production of medial wh-phrases

ABSTRACT

Non-adult-like linguistic behavior in children is sometimes taken as evidence for endogenous factors that drive selection of grammatical features from the child’s hypothesis space of possible grammars. Analyses of English-acquiring children’s productions of medial wh-phrases exemplify this trend in particular. We provide an alternative account of these productions as performance errors arising from underdeveloped cognitive inhibition. We offer experimental evidence in favor of our failure of inhibition account. The results argue against treating these errors as reflecting incomplete or non-target acquisition of grammatical features. Instead, the results support a theory of how these errors arise and are subsequently purged from children’s productions that reduces to a theory of how cognitive inhibition develops during childhood.

KEYWORDS:

• Syntax
• movement phenomena
• wh-movement
• executive function
• inhibition control

Acknowledgments

We would like to thank all of the children who participated in this study as well as their parents/guardians. For helpful suggestions and feedback, we are grateful to the UMD acquisition lab as well as audiences at the 42nd Boston University Conference of Language Development and the 2019 Michigan State Undergraduate Linguistics Conference.

Funding

The first author was partially supported by an NSF-NRT grant (NSF: #1449815) while working on this project, and the second author was partially funded by the Brazilian National Council for Scientific and Technological Development (CNPq), grant #306462/2020-6. We are grateful for the support of both of these agencies.

Disclosure Statement

No potential conflict of interest was reported by the authors.

Notes

¹ Similar findings are reported for Dutch learners (van Kampen 1997, Jakubowicz & Strik 2008), Spanish learners (Gutiérrez Mangado 2006), and French learners (Oiry 2006, Demirdache & Oiry 2008, Jakubowicz & Strik 2008). In this article, we focus on English-acquiring children, but we expect that our account of English-acquiring children’s production of medial wh-phrases would apply to children acquiring other languages where these questions are also ungrammatical in the target language of acquisition.

² Thornton (1990:ii) described only the productions like (1a), with the same wh-word in both positions, as “medial-Wh questions”; she described productions like (1b) as partial movement questions. Throughout this article, we refer to both types of questions, at least as produced by English-acquiring children, as medial wh-questions. While this conflicts with the terminology originally put forward by Thornton (1990), it more accurately reflects our account of these non-target productions, which holds that they arise via the same mechanism.

³ For the sake of expository convenience, we will talk in terms of copies of wh-phrases throughout this article. However, our account does not depend on the copy theory of movement being true. In fact, all that matters for our purposes is that some form of the wh-phrase is reactivated in the mind of the producer at places in the structure where copies are posited to be, due to the successive cyclic nature of wh-movement. There is a good amount of psycholinguistic evidence that supports this view, which we will discuss in $\S$ 2.1. As long as one believes the psycholinguistic findings, our account goes through, and it doesn’t particularly matter which syntactic theory underpins these psycholinguistic facts.

⁴ Scontras et al. (2015) also show that it takes longer to begin a sentence with a longer dependency, suggesting that there is a planning cost as well.

⁵ For additional psycholinguistic evidence that the production of linguistic dependencies, not necessarily just A-bar- (i.e., wh-)dependencies, involves active maintenance of the relevant element in question in memory, see also Badecker & Kuminiak (2007) and Franck et al. (2010).

⁶ A reviewer asks whether our account also predicts that wh-phrases should be pronounced in other locations along the movement path. We return to this in more detail in $\mathcal{\S }$ 3.5.3. However, in brief, such productions are, in principle, expected given our account. That being said, we expect them to be extremely rare. Even when cognitive inhibition fails, we would expect the failure to still be constrained by aspects of the grammar, including the statistical distribution of the utterances that the grammar gives rise to. Specifically, wh-phrases are only ever pronounced in base positions in very highly constrained contexts in English (echo questions and multiple questions), while, on the other hand, wh-phrases occur more frequently at the clause boundary in English, including in embedded interrogatives (I wonder what time it is) and embedded polar interrogatives (do you know what time it is). So, because the child would only hear wh-phrases in base positions on an extremely rare basis, we would expect that children would exhibit many fewer instances of failed cognitive inhibition in such cases, compared to at the clausal boundary. If anything, children might instead be more likely to produce the corresponding DP that gives the answer in the base position, especially since they know the answer given the setup of the task and since it is licit (and common) to pronounce DPs in the complement to V position in the adult target language. This is similar to an aspect of Dell‘s (1986:291–292) theory which holds that speech errors will follow the categorical constraints of the grammar.

⁷ Lutken, Legendre & Omaki (2020) propose an account of medial wh-phrase productions that is similar to our account. We will return to this in $\mathcal{\S }$ 3.5.3. However, it is worth noting that they claim our account, as proposed in Grolla & Lidz (2018), cannot account for medial wh-phrase productions where the wh-phrases are distinct (Lutken, Legendre & Omaki 2020:43, note 3). As we’ve just discussed, this is not the case. Our account does offer an explanation of medial wh-phrase productions with distinct wh-phrases. And, in fact, they themselves adopt two distinct accounts of the two different types of productions, with the productions involving distinct wh-phrases hypothesized to be restart errors (Lutken, Legendre & Omaki 2020:38–39).

⁸ It is of course possible that executive functions impact the speed of language acquisition, but, absent a detailed and testable proposal to this effect, non-target grammar accounts predict there to be no relation between cognitive inhibition and medial wh-phrase production. Moreover, given the issues for non-target grammar accounts that we discuss immediately following as well as converging evidence from other experiments like the ones reported in Lutken, Legendre & Omaki (2020), we think that non-target grammar accounts, even if they were supplemented with a developmental proposal in the vein suggested here, are likely wrong.

⁹ One exception to this is the non-target grammar account in Thornton (1990), where children’s productions with non-identical wh-phrases are treated as distinct from children’s productions with identical wh-phrases; however, no concrete analysis of the former sorts of productions is given. McDaniel, Chiu & Maxfield (1995:740) do also briefly acknowledge this problem, and they propose, in passing, an ad hoc solution to the issue, which is particular to the details of their account that we’ve glossed over.

¹⁰ The YouTube channel for Turma da Mônica can be found at the following link: https://www.youtube.com/channel/UCV4XcEqBswMCryorV_gNENw

¹¹ The data are available at https://osf.io/vp6mg/

¹² Note that this participant asked 12 subject questions even though we only tried to elicit eight; similarly, this participant asked five object questions, but we tried to elicit seven, and this participant asked four adjunct questions, but we tried to elicit six. Participants did not always ask the type of question that we tried to elicit. We return to this in $\mathcal{\S }$ 3.4.6 and $\mathcal{\S }$ 3.5.3.

¹³ For comparison, 22% of the utterances in Experiment 1 (30 children, 354 child utterances) from Lutken, Legendre & Omaki (2020) included medial, and 15.4% from their Experiment 3 (20 children, number of child utterances not reported) included medial wh-phrases.

¹⁴ A reviewer notes that our account would predict there to be all three of these types of medial wh-phrase productions, but the reviewer points out that the data sets from Thornton (1990) and Lutken, Legendre & Omaki (2020) predominantly include productions with matching wh-phrases and productions with what + target wh-phrase. The reviewer is correct to point out that our theory predicts a mix of production types, including non-matching medial wh-phrase productions that aren’t solely of the what + target wh-phrase type. This is indeed what we observe in our data set, as can be seen in Table 2.

Moreover, it’s not clear to us that these productions actually are absent from the data sets of Thornton (1990) and Lutken, Legendre & Omaki (2020). It is hard to ascertain the extent to which they were observed in Thornton (1990) since few summary statistics are given. That being said, Thornton (1990:237) does report some such productions, including Which Smurf do you think who has roller skates on?, Which animal do you think what really says “woof woof”?, Which one do the bear and the squirrel think who has two bears?, and Which guy did they say which had the orange marble?. On the other hand, Lutken, Legendre & Omaki (2020:16) do give the relevant summary statistics for their Experiment 1. In this experiment, a majority of the utterances with non-matching wh-phrases, 11 of 14, are not of the what + target wh-phrase type. This is consistent with our account. Furthermore, if these types of medial wh-phrase productions are more absent from other data sets than one would expect given our account, we suspect this is due to other studies involving fewer participants and fewer utterances than is the case for our study; that is to say, these other studies might just have happened to fail to observe a preponderance of various other non-matching types of medial wh-structures in virtue of smaller sample sizes.

¹⁵ Note that the fixed effect coding the error rate in the incongruent condition of the motor inhibition task was not in the best fitting model, despite the seeming difference between groups that can be seen in Figure 4, where we simply plotted some descriptive statistics about the measures from this task. This is consistent with the predictions of our account, as discussed in $\mathcal{\S }$ 3.1.

¹⁶ A reviewer points out that there is very little variability in our motor ability task measure and so, because of this, the measure may not have been predictive of medial wh-phrase production. This is indeed possible. As Hedge, Powell & Sumner (2018) point out, when between-subject variability in a measure is low, detecting a correlation with that measure becomes mathematically more unlikely. That being said, neither our account nor any other account predicted there to be a correlation between medial wh-phrase production and motor ability. We only included this task as a control task. If we had a motor ability measure with more between-subject variability, our account (and all other accounts) would still predict there to be no correlation with medial wh-phrase production.

¹⁷ The question in (12m-i) clearly involves argument resumption in the context of the video, which involved a girl popping balloons by spitting watermelon seeds at them. The “[inaudible]” portion of the transcription presumably involved the preposition with.

¹⁸ Note that this was not one of the 1,352 multiclausal utterances.

¹⁹ Note that this participant did not complete all four tasks and so their data is not analyzed in our main results. This was also not a multiclausal utterance.

²⁰ One author of this article has collected some preliminary data with Brazilian-Portuguese-acquiring children that tentatively suggests that the production of resumptive elements in relative clauses is also correlated with cognitive inhibition. Unfortunately, this data collection effort has been interrupted by the COVID-19 global pandemic.

²¹ A reviewer asks whether the account from McDaniel et al. (2015) could explain the subject asymmetry that we see in our data set. McDaniel et al. posit that there is a general production pressure such that subject extraction across a clausal boundary is more difficult than other types of extraction across a clausal boundary. They furthermore argue that the effects of this production pressure are grammaticalized (2015:421, fn. 7), in an attempt to explain certain Empty Category Principle contrasts and that-trace effects in English.

We do not think this account could explain the subject asymmetry that we see in our data. The core of their account still turns on the status of the grammar, even if the nature of grammar has been in part shaped by certain production pressures. We see no evidence in our data set that leads us to posit different grammars for different children at different times. That being said, perhaps the production pressures that disfavor subject extraction across a clausal boundary, which McDaniel et al. (2015) posit as the basis for some grammatical knowledge, could still explain the subject asymmetry in our data set, without appeal to the grammatical knowledge itself. We still, however, fail to see how this would explain the preponderance of medial wh-phrases with subject questions. It’s not clear why the general difficulty of subject extraction across a clausal boundary should result in a greater frequency of medial wh-phrases. Moreover, if their account is correct, it’s actually surprising that in 22 of the 26 cases in our data set where children asked a question with a medial wh-phrase while also switching the type of question to one other than the one that we tried to elicit, they switched an intended non-subject question to a subject question.

²² Lutken & Legendre (2020) and Lutken (2021:256–260) do find a negative correlation between a composite working memory score and the production of medial wh-phrases of $r=-0.2$ , which could be taken as further evidence in support of the account in Lutken, Legendre & Omaki (2020). However, although this correlation is significant, it is a relatively small correlation. Moreover, it could be that they see a correlation between working memory and medial wh-phrase production because the development of working memory is correlated with the development of cognitive inhibition (see, e.g., Carlson, Moses & Breton 2002). This would explain both our finding as well as theirs.

One possibility would be to redo our study but also include the working memory measure from Lutken & Legendre (2020) and Lutken (2021). If one still observed a correlation between cognitive inhibition and the production of medial wh-phrases in such a followup where working memory is controlled for in the statistical analysis, this would suggest that our account is correct and that the correlation with working memory arises because of an underlying correlation with cognitive inhibition. Another possibility for teasing these things apart is to upregulate cognitive inhibition on a trial-by-trial basis (cf. Hsu & Novick 2016); if one were to see fewer medial wh-phrases on trials where cognitive inhibition has been upregulated and more on trials where cognitive inhibition has not been upregulated, this would be consistent with our account but unexplained on the account from Lutken, Legendre & Omaki (2020).

²³ We do not consider subject questions here, since there would be no observable subject-auxiliary inversion in these questions anyway.