Hostname: page-component-7b9c58cd5d-7g5wt Total loading time: 0 Render date: 2025-03-14T23:03:22.974Z Has data issue: false hasContentIssue false

The roles of language use and vocabulary size in the emergence of word-combining in children with complex neurodevelopmental disabilities

Published online by Cambridge University Press:  28 May 2020

Susan FOSTER-COHEN*
Affiliation:
University of Canterbury, New Zealand Institute of Language, Brain and Behaviour The Champion Centre, Christchurch, New Zealand
Anne van BYSTERVELDT
Affiliation:
University of Canterbury, New Zealand Institute of Language, Brain and Behaviour
Viktoria PAPP
Affiliation:
University of Canterbury, New Zealand Institute of Language, Brain and Behaviour
*
Address for correspondence: Susan Foster-Cohen, New Zealand Institute of Language, Brain and Behaviour, University of Canterbury, Christchurch, New Zealand. Email: susan.foster-cohen@canterbury.ac.nz
Rights & Permissions [Opens in a new window]

Abstract

Parent report data on 82 preschool children with complex neurodevelopmental disabilities including Down syndrome, dyspraxia, autism, and global developmental delay suggests communicative language use must reach a threshold level before vocabulary size becomes the best predictor of word combining. Using the Language Use Inventory and the MacArthur-Bates CDI (with sign vocabulary option), statistical modelling using regression trees and random forests suggests that, despite high linear correlations between variables, (1) pragmatic ability, particularly children's emerging ability to talk about things, themselves and others is a significantly better predictor of the earliest word combining than vocabulary size; and (2) vocabulary size becomes a better predictor of later word combining, once this pragmatic base has been established.

Type
Brief Research Report
Copyright
Copyright © Cambridge University Press 2020

Introduction

The emergence of grammar is argued to be strongly linked to the emergence and growth of vocabulary (Bates & Goodman, Reference Bates and Goodman1997; Braginsky, Yurovsky, Marchman & Frank, Reference Braginsky, Yurovsky, Marchman and Frank2015; Marchman & Bates, Reference Marchman and Bates1994). However, this research, which shows tight relationships between the trajectories of lexical growth and grammatical development over time in the early years, is often provided as evidence that vocabulary size predicts “the emergence” of early grammar with the implication that it is a certain vocabulary size which actually predicts the beginnings of word combining (Bates & Goodman, Reference Bates and Goodman1997; Moyle, Weismer, Evans & Lindstrom, Reference Moyle, Weismer, Evans and Lindstrom2007; Ramírez, Lieberman & Mayberry, Reference Ramírez, Lieberman and Mayberry2013).

Typically developing children are argued to start combining words with a vocabulary somewhere around 50–100 words (Bates, Bretherton & Snyder, Reference Bates, Bretherton and Snyder1988; Crais, Reference Crais and Roush2001; Haynes, Moran & Pindzola, Reference Haynes, Moran and Pindzola2006; Paul & Norbury, Reference Paul and Norbury2012; Shore, Reference Shore1994) and to have passed into multiword speech by the time they have 200 words (Bates & Goodman, Reference Bates and Goodman1997, p. 519). A reason to question whether these relationships are inviolable is, however, seen in research aimed at understanding variability among children. Patterson (Reference Patterson1998) found that bilingual children between 21 and 27 months acquiring Spanish and English who were not combining words had vocabularies ranging from 14 to 194 and some children with fewer than 50 words were combining. In the clinical literature, Bates and Goodman (Reference Bates and Goodman1997) found that children with Down syndrome buck the trend of the tight correlation between vocabulary size and early grammar growth, although Buckley (Reference Buckley1993, Reference Buckley, Rondal and Buckley2003) argues they start combining around the 100 word mark, i.e., the upper end of the typical range. Thal and Bates (Reference Thal and Bates1990) and Moyle et al. (Reference Moyle, Weismer, Evans and Lindstrom2007) suggest that late talkers have larger vocabularies than so-called typically developing children before they combine words. And in a study of two-year-old preterm and full term children, Stolt et al. (Stolt, Matomäki, Haataja, Lapinleimu, Lehtonen & Group, PIPARI Study. Reference Stolt, Matomäki, Haataja, Lapinleimu and Lehtonen2013) suggested that the exclusion of children with neurological impairment resulted in the expected, highly predictive relationship between vocabulary size and word combining in both preterm and full-term groups. This suggests that the relationship between lexicon size and word combining in children who do have neurological impairments, but are routinely excluded from studies, bears closer examination.

Given the above suggestions that the link between vocabulary size and the emergence of word combining in these atypical populations may be weaker than in typical populations, we wanted to explore what role communicative language use might play in the emergence of word combining and how it might relate to vocabulary size. Using an aetiologically heterogeneous sample of children with complex neurodevelopmental disabilities, we wished to explore whether developments in vocabulary size or in language use better predicts the emergence of combinatorial expressive language.

The role of pragmatics in driving grammar development has been explored in a number of respects over the years, with arguments for communicative intentions as a driving force in the emergence of words (Bloom, Reference Bloom, Golinkoff, Hirsh-Pasek, Bloom, Smith, Woodward, Akhtar, Tomasello and Hollich2000) and the development of grammar (Tomasello, Reference Tomasello2003). Little attention, however, has been paid to capturing the specific role communication ability plays in the emergence of grammar or to the relative impact of that ability in relation to vocabulary size on word-combining. One reason for this lack of attention has been the lack of measures of language use that can be compared with parent reports of vocabulary as standardly collected through the use of the MacArthur-Bates Communicative Development Inventory – Words and Sentences (CDI) (Fenson, Marchman, Thal, Dale & Reznick, Reference Fenson, Marchman, Thal, Dale and Reznick2007). However, the Language Use Inventory (LUI) (O'Neill, Reference O'Neill2009; Pesco & O'Neill, Reference Pesco and O'Neill2012) provides such a measure that can be easily completed by parents alongside the CDI (Foster-Cohen & van Bysterveldt, Reference Foster-Cohen and van Bysterveldt2016). The LUI is normed for 18-47 month olds, but is appropriate for use with older children with language delays. We therefore used these measures to ask two questions:

Question 1: Does communicative/pragmatic development or vocabulary size better predict word combining in preschool children with complex neurodevelopmental disabilities?

Question 2: What is the relationship between pragmatic development and vocabulary growth in the emergence of word combining in children with complex neurodevelopmental disabilities?

Participants

The participants were a sample of 82 children with language delay in the context of a range of neurodevelopmental disorders aged 24–66 months (Table 1) who were attending the same centre-based family-centred early intervention programme for children birth to primary school age (i.e., up to the age of six) with delays in multiple areas of development. All children were assessed with varying degrees and types of language delay by clinical staff and, by definition, none of the children had a diagnosis of SLI.

Table 1. Participants by diagnosis and Adaptive Behavior Assessment System II (ABASII) scores

Note: The General Adaptive Composite of the ABASII can range from 40 to 160, with 100 as the mean in a typical distribution and SD of 15.

The largest group of children were those with Down syndrome, partly because the early intervention programme has specialised in this population over a number of years resulting in a large cohort, and partly because their parents were particularly keen to be approached for inclusion in the study. Children with Down syndrome present with a broad spectrum of challenges and are highly variable one from another in terms of long-term outcomes for language and pragmatics, and parents attending early intervention are keen to contribute to work that might help our understanding of how to get the best outcomes for children. The variability inherent within Down syndrome makes their status as the largest group in this study appropriate. Unusual for studies of this kind we have also included children with other diagnoses that share characteristics with Down syndrome including motor speech disorders (e.g., cerebral palsy) and challenges of social engagement (e.g. possible autism).

All the children were attending early intervention with their parents on a regular (usually weekly) basis and were receiving coordinated family-centred early intervention from a multi-disciplinary team that included a speech and language therapist, physiotherapist or occupational therapist, and an early intervention teacher. The early intervention programme is one in which therapists ask parents to contribute their observations of their children on a regular and ongoing basis so that the children's programmes can be individualised, be appropriate for their daily living contexts, and be delivered through a partnership between therapists and parents. The parents and therapists were blind to the study's questions and hypotheses.

At the start of the study, parents were invited to participate irrespective of the age of their children at the time and new children were added to the cohort gradually as children joined the early intervention service. As a result, the smallest number of data collection points per parent was from those whose children were close to transitioning to school at the start of the study, and the largest number from parents who entered the study when their children were younger and went on to provide multiple reports across the time period. Parents were asked to complete the measures at six monthly intervals, coinciding with their child's birthday and half-birthday, and return them in their own time within two weeks. The mean number of completions per parent was 2.6 (SD = 1.55). The data to be analysed here, consisting of 209 completions, results from a pooling of the responses across all completions and all children in the sample. We acknowledge that pooling the data risks hiding the effects of repeated completions on parents’ responses such as those suggested by Bates and Goodman (Reference Bates and Goodman1997) who found elevated vocabulary scores in a longitudinal sample compared to a cross-sectional one. However, we believe we have adjusted for this by treating participant identity as a random effect in regression trees and a fixed effect in random forests (see below).

As a means of comparing these children to typically developing children of the same age, we present in Table 1 their General Adaptive Composite Scores on the parent form of the Adaptive Behavior Assessment System (ABASII) (Harrison & Oakland, Reference Harrison and Oakland2000) collected at the same time as the LUI and CDI data. The ABASII is a well-established developmental questionnaire covering a wide range of adaptive behaviours and has a Standard Deviation of 15 and a mean of 100 based on a neurotypical population. The minimum possible score is 40 and scores less than 70 indicate significant deviation from expected norms. As can be seen from Table 1, in every diagnostic category there is considerable variation, both within and between diagnoses, in overall functioning as assessed with this measure; which is why, although diagnosis can be a useful predictor for some purposes, we suggest that for the current purpose, the pooling of data across diagnostic categories is appropriate.

The distribution of the data collection points by age at time of collection are shown in Table 2.

Table 2. Distribution of data collection points by child age

This table shows the number of data collections at each six-month point from 24 to 66 months. All data were collected within a two-week period on either side of each of the ages shown.

Procedures

Parents of the 82 participants completed the New Zealand version of the Bates-MacArthur Communicative Development Inventory: Words and Sentences (CDI) (Fenson et al., Reference Fenson, Marchman, Thal, Dale and Reznick2007; Reese & Read, Reference Reese and Read2000) and the Language Use Inventory (LUI) (O'Neill, Reference O'Neill2009). These test materials have already been demonstrated to have good levels of consistency and validity with this population of parents (Foster-Cohen & van Bysterveldt, Reference Foster-Cohen and van Bysterveldt2016). The New Zealand version of the CDI developed by Reese and Reed (Reference Reese and Read2000) has 41 words in the vocabulary section adjusted for the New Zealand dialect; otherwise it is identical to the original American English version. For this study, the CDI was modified further (with permission from the CDI board) to allow parents to select a sign option for each of the vocabulary items. This was particularly relevant for children with Down syndrome, and no requirements were placed on parents for the sign to be generally recognisable. They can therefore be assumed to be a mixture of the New Zealand Sign Language signs taught in the early intervention programme and home signs developed within each family. The result is a total expressive (conceptual) vocabulary – here named CDITOTAL – for each child at each completion point across both modalities.

The Language Use Inventory (O'Neill, Reference O'Neill2009) contains 14 sections of which 10 contribute to a possible maximum score of 161. Questions in those ten sections are as follows: Section C (21 items) about types of words used (e.g., ‘food items’, ‘animals’, ‘up, down, open or close’; Section D (7 items) requests for help (e.g., ‘Does your child ask for help to play a game?’); Section F (6 items) using words to get someone to notice something (e.g., ‘Does your child ever try to get your attention by naming something he/she is interested in?’); Section G (9 items) questions and comments about things (e.g., ‘When talking about things like toys does your child ever talk about or ask about why something happened’); Section H (36 items) asking questions and making comments about things, self and others (e.g., whether parents have heard their child talk or ask about what something is, how something tastes, what he/she wants, how someone else is feeling); Section I (14 items) using words in activities with others (e.g., asking an adult to show him/her how to do something); Section J (5 items) engaging in verbal or non-verbal teasing and humour (e.g., saying the wrong things in a teasing way); Section K (12 items) showing an interest in words and language (e.g., playing with the pronunciation of words); Section M (15 items) adapting conversations to other people (e.g., trying to answer when asked to tell someone about something); and Section N (36 items) building longer sentences and stories (e.g., will use utterances containing ‘forgot’, ‘could’ or ‘possibly’). It is worth noting that because the LUI asks questions in terms of what children ‘say’ it may under-represent the communicative skills of those who make significant use of sign or whose communication is supported by assistive technologies.

Three outcome measures of word combining were parents’ responses to one question on the CDI and two questions on the LUI. The CDI asks parents to respond to the question: “Has your child begun to combine words yet, such as “Nother cracker” or “doggie bite”? with the options to record “Not yet”, “Sometimes”, or “Often”. In what follows this ordinal variable is called WORDCOMBCDI and is entered into the analysis as 0 (“Not yet”), 1 (“Sometimes”) and 2 (“Often”). The Language Use Inventory asks “Has your child begun to use sentences of more than 2 words?” and “Has your child begun to use sentences of more than 4 words?” –each with the options of reply of “Never”, “Rarely”, “Sometimes”, or “Often”. These variables are labelled TWOPLUS and FOURPLUS respectively in what follows, in each case entered into the analysis as 0 (“Never”), 1 (“Rarely”), 2 (“Sometimes”), and 3 (“Often”). The three questions can be ordered developmentally as (1) WORDCOMBCDI; (2) TWOPLUS; (3) FOURPLUS. None of these questions contribute to either the CDITOTAL or LUI Total Scores.

Before investigating relationships among CDITOTAL (i.e. the total vocabulary score on the CDI), LUI Total Scores and the three outcome measures, we created a more conservative variable from the LUI by excluding two types of questions. Firstly, all of Section C and most of Section N were excluded as these consist of questions about the use of vocabulary items and therefore would increase the correlation between the CDITOTAL and the LUI scores. Secondly, we excluded all questions where parents were offered only multiple word examples as in “Does your child ever try to get your attention by saying “You know what?” or “Guess what?”, for example, as these might increase the correlation between responses on the LUI and responses to the word combining questions. Fifty-two items from the LUI were retained as part of the new variable (PRAGLUI) and included “Does your child ask for your help to play a game”, “Does your child tell another person to stop doing something (e.g., ‘Don't do that; Stop!’)”, “Does your child try to make others laugh by saying wrong things in a teasing way (e.g., giving the wrong name for something even though you know he/she knows the right name for it)” and “Have you noticed that your child answers questions that you ask while reading books?”). The 52 items are: D1, D2, D3-D7, F1, F2, G6-G8, H13, H15, H23, H33, H34, I1, I2, I6, I7, I11-I14, J1, J3-J5, K1-K7, K10-K12, M1, M2, M4-M7, M9, M10, M12, N32-N35. For those unfamiliar with the LUI, a sample of the complete measure can be requested from https://languageuseinventory.com

The more conservative pragmatic variable (PRAGLUI) was then used as a predictor in a regression tree analysis to tease out the non-linear relationships between children's scores on PRAGLUI and CDITOTAL predicting the three outcome measures: WORDCOMBCDI, TWOPLUS and FOURPLUS. In the R statistical language (R Core Team, 2014) an untrained partitioning algorithm (Hothorn, Hornik & Zeileis, Reference Hothorn, Hornik and Zeileis2006) was used to recursively sort the scores of the PRAGLUI and CDITOTAL in relation to the three outcomes measures. The resulting binary decision trees illustrate the diagnostic breaking points that matched responses to each of the word-combining questions. During the tree generation, we considered only the breaking points that reached the p < 0.001 criterion.

Finally, in order to explore which sections of the LUI (using the LUI Total Score rather than the more restricted PRAGLUI) had more impact on the three outcome measures than others, the raw scores within each of the sections were scaled by conversion to a percentage of the possible maximum score for each section (to control for the different sizes of possible section totals) and a random regression forest appropriate for non-linear multiple regression was fitted.

Results

A correlation matrix was created as an initial data visualization of predictors and outcomes (Table 3). The matrix summarises the Pearson's r correlations between each of the outcome variables (FOURPLUS (i.e., parent response to whether child uses utterances of 4 words or more), TWOPLUS (i.e., parent response to whether child uses utterances of 2 words or more) and WORDCOMBCDI (i.e. parent response to whether child is reported to combine words) and the three predictors (CDITOTAL (i.e., vocabulary size), PRAGLUI (i.e., non word-combining biased pragmatic items on the LUI) and LUI Total Score).

Table 3. Correlation matrix for outcomes and predictors

Notes: CDITOTAL = Vocabulary Total; PRAGLUI = Pragmatic subset of LUI TotalScore; WORDCOMBCDI = Word combining question from CDI; TWOPLUS and FOURPLUS = Word combining questions from LUI.

As Table 3 shows, the PRAGLUI variable correlates at .80 or above with all three outcome measures (WORDCOMBCDI, TWOPLUS and FOURPLUS). CDITOTAL also correlates highly with each outcome measure. In a standard linear model, these correlations would be interpreted as too high to enter the variables as independent predictors. However, a high correlation can come about despite systematic deviations from close correlation at one or more points along the line. Regression trees, however, can reveal the complexity and differential predictability, even within highly correlated items as we have in this case. It is thus a useful statistical tool in cases where relationships risk being smothered by the overall generality of simple correlation (Tomaschek, Hendrix & Baayen, Reference Tomaschek, Hendrix and Baayen2018).

Figure 1 shows the regression trees for WORDCOMBCDI, TWOPLUS and FOURPLUS. The nodes represent statistically significant breaking points in the data (p < 0.001), and the terminal nodes (or leaves) are labelled for the number of observations captured by the boxplot. The y-axis on the terminal boxplots represents the scale 0 = “not yet”, 1 = “sometimes”, and 2 = “often” for WORDCOMBCDI and 0 = “Never”, 1 = “Rarely”, 2 = “Sometimes”, and 3 = “Often” for TWOPLUS and FOURPLUS. The trees support the suggestion hinted at by the linear correlations that for WORDCOMBCDI and TWOPLUS a level of pragmatic capacity is necessary before CDITOTAL becomes a better predictor. For FOURPLUS, vocabulary size overwhelms the role of pragmatics. However, in all three trees the size of vocabulary becomes a useful predictor when it reaches around the 300 word mark, even for the emergence of word combining as measured by WORDCOMBCDI; considerably larger than some authors have suggested for other cohorts. Thus the trees in Figure 1 reflect a picture in which vocabulary size has better explanatory power only once a threshold of language use development has been reached, a picture that can be summarised as Table 4 using the strongest parent response “often” for the establishment of word combining in each case.

Figure 1. Regression tree models for three outcome measures

Table 4 Relationships between PRAGLUI, CDITOTAL and “Often” responses to word combining questions

Table 4 suggests that when a child has a PRAGLUI score greater than 18 and a CDITOTAL score greater than 322, then parents will record 2 (“Often”) on the word combining question on the CDI. They will record 3 (“Often”) on the LUI to the TWOPLUS questions when the PRAGLUI score is greater than 23 and the CDITOTAL is greater than 285. The answer to the FOURPLUS question on the LUI is only predicted by the vocabulary size (greater than 454) and not by the score on the PRAGLUI, in accordance with our suggestion that as vocabulary size increases, the predictive role of pragmatics disappears.

The results of random forests of (scaled) sections of all sections and questions on the LUI are shown in Figure 2.

Figure 2. Variable importance plots from random forest models predicting outcomes measures from scaled LUI sections (all questions included)

The permutation-based variable importance measures of the random forest models predicting the outcomes from the scaled sections of the LUI suggest sections C and H are the best predictors of all three outcomes. Section C is a minimal measure of vocabulary development that was removed from the PRAGLUI variable. Section H ‘Your child's questions and comments about self and others’ is the most predictive of WORDCOMBCDI, is less so of TWOPLUS and less so again of FOURPLUS. The three forest models thus mirror the increasing importance of vocabulary as word combining expands. For all three outcome measures, Section G ‘Your child's questions and comments about things’ is the next most influential after the two sections just discussed, with Section N, which contains words that require a level of syntactic ability, not surprisingly rising up the variable importance rankings as children move to using expressions of four or more words (FOURPLUS).

Discussion

The aim of this study was to determine whether parents’ answers to questions about their children's word combining were better predicted by the same parents’ answers on the LUI or by their report of vocabulary on the CDI. Our results suggest that communicative ability as measured by the LUI is a better predictor of when children with multi-system disabilities start to combine words and in the earliest stages expand the number of words in their utterances beyond two-word utterances. Moreover, their abilities to talk about things, themselves and others, the key components of any theory of pragmatics, are the key components that best predict word combining. Vocabulary size does appear to play a role; but it does so only after communicative abilities have reached a certain threshold level. Importantly, those vocabulary sizes appear to be considerably larger than found in much of the literature, before they predict word combining. The vocabulary sizes we found, however, do fit the data found by Bates and Goodman (Reference Bates and Goodman1997) who suggested that “children with vocabularies under 300 words have very restricted grammatical abilities: some combinations, a few function words in the right places, the occasional bound morpheme, but little evidence of productive control over morphology or syntax” (p.545). They suggest that 300 words reflects a ‘critical mass’ necessary for grammar. Our data indicate the same conclusion and suggest that atypical populations are perhaps not as atypical in developmental path as is sometimes supposed (Paterson, Parish-Morris, Hirsh-Pasek & Golinkoff, Reference Paterson, Parish-Morris, Hirsh-Pasek and Golinkoff2016).

Despite this, Bates and Goodman (Reference Bates and Goodman1997) found their cohort of children with Down syndrome were the only ones to buck the trend of what they describe as “approach[ing] the status of a psychological law” (p. 542) governing the relationship between vocabulary and grammar development. Consequently, it is important to consider whether our results are due to 39% of the children in this study having Down syndrome. Singer Harris, Bellugi, Bates, Jones, and Rossen (Reference Singer Harris, Bellugi, Bates, Jones and Rossen1997) similarly found that the children with Down syndrome had grammatical development that lagged behind vocabulary growth, but unfortunately failed to capture the moment of word combining because they tied the decision about which version of the CDI that parents completed (Words and Gestures or Words and Sentences) to a combination of vocabulary size and word combining (those reporting more than 50 words or combining words completed the Words and Sentences form), thereby pre-empting the outcome measure of interest here. We suggest that the large number of children with Down syndrome in our sample may well have amplified the degree of non-significance of vocabulary size in our study; but given the match with typically developing children found by Bates and Goodman and identified above, it is not clear that it has skewed the picture as much as might be imagined.

Another issue calling for further exploration is the nature of the vocabulary acquired by the children. The children with Down syndrome, in particular, were reported to use signs as well as spoken words. In this study CDITOTAL is made up of the children's total conceptual vocabulary, irrespective of modality, on the assumption that at the pre-grammatical stage of development words and signs can be treated as symbols that code concepts equivalently, at least in a language like English where uninflected base forms can stand alone. This assumption requires further exploration, however, and is an issue we are currently exploring.

Another question revolves around what parents understand by the notion of “word combining”. As documented by several researchers (Bates et al., Reference Bates, Bretherton and Snyder1988; Peters, Reference Peters1977; Tomasello, Reference Tomasello2003 among others), there are (at least) two routes to word combining, often referred to as “analytical” and “gestalt/holistic”, the latter having the danger of leading to an over-inflation of reported word combinations as a result of formulaic and other fixed utterances that, to the child, are essentially single words being interpreted as productive combinations. We have no way of knowing how well attuned parents are to this distinction. Our informal analysis of parents’ responses to the “three longest utterances” question suggests that a fairly large proportion of these reported utterances may be formulaic (something suggested also by Bates and Goodman (Reference Bates and Goodman1997: 531) and Newport (Reference Newport1990) for late talkers); but this may just be because such utterances are memorable to the parent and may bear no relation to the basis on which parents answer affirmatively to the word combining questions. However, we are continuing to explore this issue as it bears importantly on clinical interventions aimed at increasing children's mean length of utterance (MLU).

The clinical importance of understanding what the precursors are for effective word combining lies in the contribution it can make to determinations of when to encourage the move from one- to two-word utterances in children with developmental disabilities. If there is merit in our results, then developing a vocabulary to 50-100 words is not a sufficient basis for encouraging the move to grammar. Rather a larger basis of active intentional communication through non-verbal gestures and one-word/sign utterances, (including unanalysed multi-syllable expressions) may need to be established before word combinations are actively encouraged.

Footnotes

We are grateful to the parents who made this research possible; to the staff of The Champion Centre who understood its importance; and to the action editor and reviewers at the Journal of Child Language for incisive feedback that improved the quality of the work.

References

Bates, E., Bretherton, I., & Snyder, L. (1988). From first words to grammar. Cambridge: Cambridge University Press.Google Scholar
Bates, E., & Goodman, J. C. (1997). On the inseparability of grammar and the lexicon: Evidence from acquisition, aphasia and real-time processing. Language and cognitive Processes, 12(5-6), 507584.Google Scholar
Bloom, L. (2000). The intentionality model of word learning: How to learn a word, any word. In Golinkoff, R., Hirsh-Pasek, K., Bloom, L., Smith, L. B., Woodward, A. L., Akhtar, N., Tomasello, M. & Hollich, G. (Eds.), Becoming a word learner: A debate on lexical acquisition (pp. 1950). Oxford: Oxford University Press.CrossRefGoogle Scholar
Braginsky, M., Yurovsky, D., Marchman, V. A, & Frank, M. C. (2015). Developmental Changes in the Relationship Between Grammar and the Lexicon. Paper presented at the CogSci. 2015. langcog.stanford.eduGoogle Scholar
Buckley, S. (1993). Language development in children with Down syndrome-Reasons for optimism. Down syndrome research and practice, 1(1), 39.CrossRefGoogle Scholar
Buckley, S. (2003). Literacy and Language. In Rondal, J. A. & Buckley, S. (Ed.), Speech and Language Intervention in Down Syndrome (pp. 132153). London: Whurr Publishers Ltd.Google Scholar
Crais, E. (2001). Developmental Milestones for Speech, Language and Communication. In Roush, J. (Ed.), Screening for Hearing Loss and Otitis Media in Children (p. 96). San Diego, CA. : Singular.Google Scholar
Fenson, L., Marchman, V. A, Thal, D. J., Dale, P. S., & Reznick, J. S. (2007). MacArthur-Bates communicative development inventories: User's guide and technical manual: Baltimore, MD: Brookes.Google Scholar
Foster-Cohen, S.H., & van Bysterveldt, A. K. (2016). Assessing the communication development of children with language delay through parent multi-questionnaire reporting. Speech, Language and Hearing, 1-8. doi: http://dx.doi.org/10.1080/2050571X.2015.1108067CrossRefGoogle Scholar
Harrison, P.L., & Oakland, T. (2000) ABAS, adaptive behavior assessment system : manual. San Antonio, Tex.: Psychological Corporation.Google Scholar
Haynes, W. O, Moran, M. J, & Pindzola, R. H. (2006). Communication disorders in the classroom: An introduction for professionals in school settings: Jones & Bartlett Learning.Google Scholar
Hothorn, T., Hornik, K., & Zeileis, A. (2006). Unbiased recursive partitioning: A conditional inference framework. Journal of Computational and Graphical statistics, 15(3), 651674.CrossRefGoogle Scholar
Marchman, V. A, & Bates, E. (1994). Continuity in lexical and morphological development: A test of the critical mass hypothesis. Journal of child language, 21(2), 339366.CrossRefGoogle ScholarPubMed
Moyle, M. J., Weismer, S. E., Evans, J. L., & Lindstrom, M. J. (2007). Longitudinal relationships between lexical and grammatical development in typical and late-talking children. Journal of Speech, Language, and Hearing Research, 50(2), 508528.CrossRefGoogle ScholarPubMed
Newport, E. L. (1990). Maturational constraints on language learning. Cognitive science, 14(1), 1128.CrossRefGoogle Scholar
O'Neill, D. K. (2009). Language Use Inventory, An assessment of young children's pragmatic language development for 18- to 47-month-old children [manual]. Waterloo, Ontario, Canada: Knwoledge in Development.Google Scholar
Paterson, S. J, Parish-Morris, J., Hirsh-Pasek, K., & Golinkoff, R. M. (2016). Considering development in developmental disorders. Journal of Cognition and Development, 17(4), 568583.CrossRefGoogle Scholar
Patterson, J. L. (1998). Expressive vocabulary development and word combinations of Spanish-English bilingual toddlers. American journal of speech-language pathology, 7(4), 4656.CrossRefGoogle Scholar
Paul, R., & Norbury, C. (2012). Language disorders from infancy through adolescence: Listening, speaking, reading, writing, and communicating: Elsevier Health Sciences.Google Scholar
Pesco, D., & O'Neill, D. K. (2012). Predicting later language outcomes from the Language Use Inventory. Journal of Speech, Language, and Hearing Research, 55(2), 421434.CrossRefGoogle ScholarPubMed
Peters, A. M. (1977). Language learning strategies: does the whole equal the sum of the parts? Language, 560573.CrossRefGoogle Scholar
R Core Team. (2014). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. 2013.Google Scholar
Ramírez, N. F., Lieberman, A. M., & Mayberry, R. I. (2013). The initial stages of first-language acquisition begun in adolescence: When late looks early. Journal of child language, 40(2), 391414.CrossRefGoogle ScholarPubMed
Reese, E., & Read, S. (2000). Predictive validity of the New Zealand MacArthur communicative development inventory: Words and sentences. Journal of Child Language, 27(02), 255266.CrossRefGoogle ScholarPubMed
Shore, C. M. (1994). Individual differences in language development (Vol. 7): Sage Publications.Google Scholar
Singer Harris, N. G, Bellugi, U., Bates, E., Jones, W., & Rossen, M. (1997). Contrasting profiles of language development in children with Williams and Down syndromes. Developmental Neuropsychology, 13(3), 345370.CrossRefGoogle Scholar
Stolt, S., Matomäki, J., Haataja, L., Lapinleimu, H., Lehtonen, L., & Group, PIPARI Study. (2013). The emergence of grammar in very-low-birth-weight Finnish children at two years of age. Journal of child language, 40(2), 336357.CrossRefGoogle Scholar
Thal, D., & Bates, E. (1990). Continuity and variation in early language development. Individual differences in infancy: Reliability, stability, prediction, 359-383.Google Scholar
Tomaschek, F., Hendrix, P., & Baayen, R. H. (2018). Strategies for addressing collinearity in multivariate linguistic data. Journal of Phonetics, 71, 249267.CrossRefGoogle Scholar
Tomasello, M. (2003). Constructing a language: A usage-based approach to child language acquisition. Cambridge (MA).Google Scholar
Figure 0

Table 1. Participants by diagnosis and Adaptive Behavior Assessment System II (ABASII) scores

Figure 1

Table 2. Distribution of data collection points by child age

Figure 2

Table 3. Correlation matrix for outcomes and predictors

Figure 3

Figure 1. Regression tree models for three outcome measures

Figure 4

Table 4 Relationships between PRAGLUI, CDITOTAL and “Often” responses to word combining questions

Figure 5

Figure 2. Variable importance plots from random forest models predicting outcomes measures from scaled LUI sections (all questions included)