We present evidence that individual variation in grammatical ability can be predicted by individual variation in inhibitory control. We tested 81 5-year-olds using two classic tests from linguistics and psychology (Past Tense and the Stroop). Inhibitory control was a better predicator of grammatical ability than either vocabulary or age. Our explanation is that giving the correct response in both tests requires using a common cognitive capacity to inhibit unwanted competition. The implications are that understanding the developmental trajectory of language acquisition can benefit from integrating the developmental trajectory of non-linguistic faculties, such as executive control.
Citation: Ibbotson P, Kearvell-White J (2015) Inhibitory Control Predicts Grammatical Ability. PLoS ONE 10(12): e0145030. doi:10.1371/journal.pone.0145030
Editor: Sonja Kotz, Max Planck Institute for Human Cognitive and Brain Sciences, GERMANY
Received: February 3, 2015; Accepted: November 28, 2015; Published: December 14, 2015
Copyright: © 2015 Ibbotson, Kearvell-White. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited
Data Availability: All relevant data are within the paper and its Supporting Information files.
Funding: The authors have no support or funding to report.
Competing interests: The authors have declared that no competing interests exist.
Innate mental representations of grammar were initially proposed because it was thought that general learning mechanisms were not powerful or subtle enough to explain adult language competence . The original claim was made when ‘learning mechanisms’ more-or-less meant the associationist framework of Behaviourism. Decades of work since then has shown the child has a much richer and more sophisticated set of cognitive and social resources to bring to bear on the language acquisition process than the Behaviourist conception of learning [2,3,4,5,6,7,8]. Despite advances, much painstaking work still remains to be done to show exactly how aspects of cognition (e.g., memory, attention, executive function) interact with social reasoning (e.g., intention-reading, cultural intelligence) in a way that guides the child to an adult grammar. One branch of linguistics that has taken up this challenge is Cognitive Linguistics. It seeks to render accounts of language “consonant with aspects of cognition which are well documented or self-evident, or at least highly plausible, and which may well be manifested in non-linguistic activities” ( p.9). Contrast this with the Chomskian position which stresses language as separate from the rest of cognition: “It would surprising indeed if we were to find that the principles governing [linguistic] phenomena are operative in other cognitive systems…there is good reason to suppose that the functioning of the language faculty is guided by special principles specific to this domain …” ( p. 44).
Here we take forward the Cognitive Linguistics enterprise and explore a specific conjecture about the relationship between language and cognition. The hypothesis is this: grammatical ability to produce an irregular past tense form, for example, fly → flew, depends on the ability to inhibit a temping but incorrect response, flyed. The idea is that the correct form flew is facing unwanted competition from analogous patterns such as tie → tied, die → died and lie → lied. So, to put it very simply, if children are to learn language they must learn patterns and they must learn exceptions to those patterns. Giving the correct linguistic response involves suppressing this competition by using a cognitive faculty that is independent of language—inhibition.
The prediction is that those participants who are good at avoiding the overgeneralisation error (e.g., flyed) should also be good at inhibition. The implications of this are (1) it provides evidence that performance on a linguistic and non-linguistic test are recruiting the common cognitive faculty of inhibition, strengthening the case that language is deeply integrated with the rest of cognition (2) it provides new insights into the process of language acquisition. These overgeneralisation errors have traditionally received a very linguistic, domain-internal analysis [11,12,13,14,15]. A positive result would open the door to more cognitive-based explanations of the phenomena, for example, the retreat from overgeneralisation errors witnessed in child development could be the result of maturing inhibitory control (3) it identifies a source of individual variation in language ability which may in turn have implications for linguistic interventions, particularly for those at the far end of the spectrum of language ability like those with Specific Language Impairment.
As a test of grammatical ability we use a past tense elicitation task where participants hear a standard frame “…every day I fly, yesterday I….” and have to complete the sentence. As a test of inhibition we use the Sun-Moon Stroop task which involves participants responding “sun” to a picture of a moon and vice versa. There is evidence to suggest bilinguals are better than monolinguals at Stroop tests as they are well practiced in the skills of cognitive control and conflict resolution which switching between languages requires [16,17] (but also see ). Here we test whether the variation within monolinguals is also related to their ability to inhibit. To test this we use 5-year-old participants because adult monolinguals are at ceiling performance on the past tense elicitation task, meaning that there would be no between-participant variation to investigate. It is worth noting however, that these errors are not entirely absent from adult speakers, especially when the speaker is tired or under some communicative stress suggesting these too could be elicited under different experimental conditions than the ones we use here.
It could be that those children who are good at the Stroop test simply know more words or are a little older and this makes them better at the grammatical task. For this reason we also recorded each participant’s vocabulary ability and age in months. All three variables will be assessed to see what extent they predict grammatical ability. The main hypothesis is schematically summarised in Fig 1.
A total of 81 five-year-olds participated in this study (M = 5;6, SD = 3.54), 44 male, 37 female. All children were recruited from schools located in the Milton Keynes, Market Harborough and Henlow areas of the UK. All research conducted adhered to the British Psychological Society code of Ethics. The research was approved by the Open University Ethics Board. The purpose of the experiment was communicated to parents who chose to opt-in their children by a consent letter. Children were free to discontinue the experiment by assent.
Materials and procedures
Each child completed the Past Tense Task-20 , the English Peabody Picture Vocabulary Test  and the Sun-Moon Stroop test . All tasks were presented individually to each child in a quiet room of their school. The presentation of the tasks was counterbalanced.
Past Tense Elicitation Task
The PPT-20 was developed by Conti-Ramsden et al  as an adaption of the well-known Marchman et al  past tense task. This version is shorter, provides appropriate UK norms and is easily administered to children of this age group. For each item, children are presented with a set of three consecutive sentences, with an accompanying picture depicting a scene where an action is taking place. The third sentence has a missing word at the end and the child is asked to fill in the gap to describe ‘what happened yesterday’. Rising intonation suggests that the sentence is incomplete. The sentences all follow the format:
Three practice items are administered first (‘walk’, ‘fish’, ‘catch’), followed by the test items. When more than one response was provided (that is, self-corrections), coding was based on the final production. Both ‘spelled’ and ‘spelt’ were accepted as correct responses for the item containing the verb ‘spell’ as both forms are generated with equal frequency by speakers of British English. Although the PPT-20 records various types of error (e.g., non responses, responses with non-target verbs, no change of verb stem) we were interested in suffixation errors which comprised the addition of a stem-final suffix of vowel change plus suffix, for example fly → flyed. This is because our hypothesis is concerned with inhibiting the competitors that this type of overgeneralisation involves. A list of verbs used in this study appears in S1 File.
Following the procedure of Archibald and Kerns , the Sun—Moon Stroop consists of two conditions. In condition A, children are shown a single page consisting of 30 sun and moon pictures, which are randomly arranged into equal rows and columns. Children are then instructed to say ‘sun’ to the pictures of the suns, and ‘moon’ to the pictures of the moons, as fast as they can (within a 45-s time limit) and to correct themselves if they made a mistake before moving on. The experimenter (E) points to each picture as it is named, and moves across the rows. If a child makes an error on a picture, E leaves his finger on this picture until the self-correction. Children are told that if they name all the pictures on the page within the given time limit, they start from the top again. As a practice trial, children are asked to name the first four pictures. In condition B, the stimuli were arranged differently but consisted of the same number of pictures. Children are now instructed to say ‘moon’ to a picture of a sun, and ‘sun’ to a picture of a moon. We were specifically interested in the number of incorrect responses in Condition B because participants have to inhibit the tempting response to say “moon” to a picture of the moon. This measure gives us the closest analogy to the past tense task where participants are also scored on number of incorrect responses.
The English Peabody Picture Vocabulary Test  is a receptive vocabulary test in which children are shown an array of four coloured pictures. The experimenter reads out the target word to them, and they are required to select the picture that matches the target word. The test becomes progressively more difficult; once children fail eight items within one block (of 12 items) the test is terminated.
First we wanted to check whether our participant group was broadly in line with UK norms on grammatical ability and vocabulary (Table 1).
The mean scores for our participant were slightly lower than the average for the PPT-20 and slightly higher on the BPVS. Perhaps more importantly than the means, the standard deviations in our sample are broadly comparable to those of the wider population. This is important as it is this individual variation we are trying to account for and shows the amount of variation we have to work with in our sample is representative of the variance in the population. We now turn to our main research question. Our hypothesis is that there should be a positive correlation between inhibition and grammatical performance. In order to test whether the Stroop score is a predictor of verb errors we fitted a series of mixed effects logistic regression models using the glmer function in the R package lme4 (R script in S1 File; Raw data used in this analysis in S2 File). The glmer function estimates the parameters of the model using a maximum likelihood method.
In the models, the binary response variable denotes the presence or absence of a verb error. There are two grouping variables that we considered as random effects: participant ID (since each participating child has multiple responses) and verb (since each verb has a response from every participant). The fixed effects we considered were Stroop score, age and vocabulary score. The age and vocabulary variables are on a scale that is very different to Stroop score so, in order to help with the numerical stability of the model-fitting algorithm, these variables were each normalised by subtracting their mean and dividing by their standard deviation.
In order to determine the best form for the model, we began by fitting the most complex model we believed could be supported by the data. This model has all three main effects, random participant and verb intercepts, and random verb slopes for Stroop score, age and vocabulary score. We could not include random participant slopes because there is no variation of Stroop score, age or vocabulary score within participants. Nor could we include interactions between the main effects because including them caused the fitting algorithm to fail to converge, indicating that the sample size is too small to support this level of complexity. In R notation, the full model (Model 1) is:
After fitting the full model we performed backwards selection on the random effects whilst keeping the fixed effects structure constant. We used AIC as the criterion for selecting terms to drop out of the model. The resulting model (Model 2) has random participant and verb intercepts and a random verb slope for Stroop score. In R notation Model 2 is:
The reduced complexity of the random effects structure in Model 2 allowed us to then consider interactions between the main effects without running into convergence issues in the fitting algorithm. We therefore supplemented Model 2 with all two-way interactions between the main fixed effects and then performed backwards selection on the fixed effects whilst holding the random effects structure constant. Again, we used AIC as the criterion for selecting terms to drop out of the model. The backwards selection dropped age, vocabulary score and all interactions from the model, so the final model (Model 3) retains just the Stroop score as a fixed effect. In R notation Model 3 is:
We took Model 3 as our final and best model in the analysis. The fact that Stroop score was retained as a fixed effect in the final model indicates that it is a predictor of verb errors. We went further and used the ANOVA function in R to perform a formal likelihood ratio test between Models 3 and 4, a reduced version that excludes Stroop score as a fixed effect. In R notation Model 4 is:
The p-value for the likelihood ratio test is 0.021, indicating that the Stroop score effect is statistically significant at the 5% level.
Table 2 gives the estimated fixed effects coefficients from Model 3 along with their standard errors and 95% confidence intervals. In a logistic regression model, the exponential of the estimated coefficient for a numerical predictor such as Stroop score gives an estimate of the multiplicative effect on the odds of the outcome of a unit increase in the predictor. This means that a unit increase in Stroop score is associated with the odds of a verb error increasing by a factor of 1.25 with a 95% confidence interval given by (1.04, 1.55). This result is an average over participants and verbs. To give an example, the estimated error probability for an average verb in an average participant with Stroop score 5 is 0.76, compared to 0.80 for an average participant with Stroop score 6.
In summary, Stroop score is found to be a statistically significant predictor of verb errors. A unit increase in Stroop score is associated with the odds of a verb error increasing by a factor of 1.25 with a 95% confidence interval given by (1.04, 1.55).
We have found evidence here that individual variation in grammatical ability can be predicted by individual variation in inhibitory control and perhaps it does so in way schematically shown in Fig 1. From other work we know grammatical categories such as the transitive can behave in similar ways to non-linguistic categories . We also know speakers can use the eye-gaze of the speaker to work out the meaning of novel verbs  and grammatical constructions . We perhaps can add past tense formation to that list of core linguistic abilities that are integrated at deep level with the rest of cognition. If the developmental trajectory of grammatical ability is in some sense dependent on the developmental trajectory of inhibition then the question is whether it makes sense to study these topics independently. For some phenomena it might not be productive to draw a sharp line between what is language and what is cognition. This is not, as Chomsky might respond, a “non-existence” argument—the idea that language does not exist as a serious independent topic of study. Rather, it emphasises the need to look at how deeply and complexly different areas of the mind and brain are integrated if we hope to explain these developmental trajectories. This integrated view of language and cognition is predicted by an evolutionary perspective on the development of language. For instance, language developed in our species perhaps as recently as 2–300,000 years ago. It is very likely then that language must have used and continues to use pre-existing cognitive mechanisms. As Evans and Levinson put it “The null hypothesis here is that all needed brain mechanisms, outside the vocal-tract adaptation for speech, were co-opted from pre-existing adaptations not specific to language” ( p. 447).
These results also bear on the processes of language acquisition. Overgeneralisation errors have received a lot of attention from developmental psycholinguists as they are thought to be a window in the processes that underlie linguistic creativity: Adults do not say ‘flyed’ so children must have created that form for themselves. One particular concern is how children ‘retreat’ from these errors if they never receive explicit feedback that they are incorrect. Currently, the three main solutions to this problem—pre-emption, entrenchment and semantic class—are all born out of a very linguistic analysis. Our result does not negate the need for these explanations (which offer more fine grained predictions than inhibition could) but it does suggest they should be considered along side more cognitive based explanations. We know inhibitory control is maturing throughout the years that children’s overgeneralisation errors are reducing . It is possible therefore that inhibitory control accounts for some of the individual variation in past tense performance that linguistic accounts do not (and vice versa).
Finally, we know children with SLI struggle with the grammar of past tense formation , almost by definition. We also know that children with SLI have impaired performance on tasks involving inhibition . Our result suggests further research on interventions designed to target both linguistic and non-linguistic components, with the hope that by raising the performance of one domain it might have a positive effect on the other.
The measure of inhibition we used in this study did involve participants recruiting language—they had to say the word ‘moon’ or ‘sun’. The main point is that ‘sun’ and ‘moon’ do not carry with them any prepotent linguistic response other than the one primed by the visual stimuli (pictures of suns and moons). This is a different situation to grammatical test where the tempting generalisation is generated by the linguistic system itself (analogy to other past tense forms). On reflection, an age appropriate non-verbal measure of Stroop would have been a cleaner measure and it would be interesting to see whether using such a test replicates our findings.
In summary, grammatical errors involving past tense formation and errors of performance on the Stroop test are significantly related. What appears to be the most parsimonious explanation for this relationship is that giving the correct response in both tests requires using a common cognitive capacity to inhibit unwanted competition. That performance on a linguistic and non-linguistic test are recruiting a common cognitive faculty of inhibition, strengthens the case that the complexity of language emerges through the interaction of cognition and language use over time [2,3,4,5,6,7,8].
S1 File. Verbs used in this study (Figure A).
R script (Figure B).
S2 File. Raw data.
Thank you to the parents, staff and children of Kibworth Primary, Great Bowden Primary School, Market Harborough CofE Primary School, Wavendon Gate School and Raynsford Academy. Thank you to Gina Conti-Ramsden who promptly replied to questions about the PTT-20.
Conceived and designed the experiments: PI. Performed the experiments: JK. Analyzed the data: PI. Contributed reagents/materials/analysis tools: PI JK. Wrote the paper: PI.
- 1. Chomsky N. (1957). Syntactic Structures. The Hague: Mouton.
- 2. Langacker R. (1987). Foundations of Cognitive Grammar, Vol. I. Stanford: Stanford University Press.
- 3. Langacker R. (1991). Foundations of Cognitive Grammar, Vol. II. Stanford: Stanford University Press.
- 4. Croft W. (1991). Syntactic Categories and Grammatical Relations: The Cognitive Organization of Information. Chicago: University of Chicago Press.
- 5. Givón T. (1995). Functionalism and Grammar. Amsterdam: Benjamins.
- 6. Tomasello M. (2003). Constructing a Language: A Usage-Based Theory of Language Acquisition. Cambridge: Harvard University Press.
- 7. Goldberg A. (2006). Constructions at Work: The Nature of Generalisations in Language. Oxford: Oxford University Press.
- 8. Bybee J. (2010). Language, Usage and Cognition. Cambridge: Cambridge University Press
- 9. Taylor J. (2002) Cognitive Grammar. OUP.
- 10. Chomsky N. (1965). Aspects of the Theory of Syntax. MIT Press.
- 11. Pinker S. (1989). Learnability and Cognition: The Acquisition of Argument Structure. Cambridge, MA: MIT Press.
- 12. Ambridge B, Pine JM, Rowland CF. (2012). Semantics versus statistics in the retreat from locative overgeneralization errors. Cognition 123, 260–279. doi: 10.1016/j.cognition.2012.01.002. pmid:22325040
- 13. Ambridge B, Pine JM, Rowland CF, Chang F. (2012a). The roles of verb semantics, entrenchment and morphophonology in the retreat from dative argument structure overgeneralization errors. Language 88, 45–81. doi: 10.1353/lan.2012.0000
- 14. Marchman VA. (1997). Children’s productivity in the English past tense: The role of frequency, phonology and neighborhood structure. Cognitive Science, 21, 283–304. doi: 10.1207/s15516709cog2103_2
- 15. Plunkett K, Marchman V. (1993). From rote learning to system building: acquiring verb morphology in children and connectionist nets. Cognition, 48, 21–69. pmid:8403834 doi: 10.1016/0010-0277(93)90057-3
- 16. Bialystok E, Craik FIM, Grady C, Chau W, Ishii R, Gunji A, Pantev C. (2005). Effect of bilingualism on cognitive control in the Simon task: Evidence from MEG. NeuroImage, 24, 40–49. pmid:15588595 doi: 10.1016/j.neuroimage.2004.09.044
- 17. Van Heuven WJB, Conklin K, Coderre E, Guo T, Dijkstra T. (2011). The influence of cross-language similarity on within- and between-language Stroop effects in trilinguals. Frontiers in Psychology, 2 (374), 1–15. doi: 10.3389/fpsyg.2011.00374
- 18. de Bruin A, Treccani B, Della Sala S. (2014). Cognitive advantage in bilingualism: An example of publication bias? Psychological Science, 1–9. doi: 10.1177/0956797614557866
- 19. Conti-Ramsden G, Durkin K, Simkin Z, Lum JAG, Marchman V. (2011). Short Report: The PTT-20: UK normative data for 5- to 11-year-olds on a 20-item past-tense task. International Journal of Language & Communication Disorders, 46(2), 243–248. doi: 10.3109/13682822.2010.487884
- 20. Dunn M, Dunn LM. (2007). Peabody Picture Vocabulary Test—4. Circle Pines, MN: AGS.
- 21. Archibald SJ, Kerns KA. (1999). Identification and description of new tests of executive functioning in children. Child Neuropsychology, 5(2), 115–129. doi: 10.1076/chin.18.104.22.16867
- 22. Marchman VA, Wulfeck B, Ellis Weismer S. (1999). Morphological productivity in children with normal language and SLI: A study of the English past tense. Journal of Speech, Language and Hearing Research, 42, 206–219. doi: 10.1044/jslhr.4201.206
- 23. Ibbotson P, Theakston A, Lieven E, Tomasello M. (2012). Prototypical transitive semantics: developmental comparisons. Cogn. Sci. 36, 1–21. doi: 10.1111/j.1551-6709.2012.01249.x
- 24. Nappa R, Wessell A, McEldoon KL, Gleitman LR, Trueswell JC. (2009). Use of Speaker’s aze and Syntax in Verb Learning. Language Learning and Development, 5(4), 203–234. doi: 10.1080/15475440903167528
- 25. Ibbotson P, Lieven E, Tomasello M. (2013). The attention-grammar interface: eye-gaze cues structural choice in children and adults. Cogn. Linguist. 24(3):457–481. doi: 10.1515/cog-2013-0020
- 26. Evans N, Levinson S. (2009). The myth of language universals: Language diversity and its importance for cognitive science. Behavioral and Brain Sciences 32(5) doi: 10.1017/s0140525x0999094x
- 27. Leon-Carrion J, Garcia-Orza J, Perez-Santamaria FJ. (2004). Development of the inhibitory component of the executive functions in children and adolescents. International Journal of Neuroscience, 114, 1291–1311. pmid:15370187 doi: 10.1080/00207450490476066
- 28. Rice ML, Tomblin JB, Hoffman L, Richman WA, Marquis J. (2004). Grammatical tense deficits in children with SLI and nonspecific language impairment: Relationships with nonverbal IQ over time. Journal of Speech, Language, and Hearing Research, 47, 816–834. pmid:15324288 doi: 10.1044/1092-4388(2004/061)
- 29. Henry LA, Messer D, Nash G. (2012) Executive functioning in children and young people with specific language impairment. Journal of Child Psychiatry and Psychology 50/1: doi: 10.1111/j.1469-7610.2011.02430.x