8 October 2009
Developmental pathways in language emergence from two to seven: Late starts and surprising arrivals
Abstract, slides and audio of presentation
Edited transcript
The following audio presentation is brought to you by the Australian Institute of Family Studies as part of our monthly seminar series in which we showcase national and international research related to the family.
The seminars are designed to promote a forum for discussion and debate. They are open to the public and free of charge.
Seminar facilitated & speaker introduced by Diana Smart.
Professor Stephen Zubrick:
Thank you very much and it's a pleasure to be here. The Institute Seminar Series is a well-known and respected opportunity and it's not one that's lightly declined. I jumped at it when offered the opportunity and hope that what I have to offer you today will be engaging and of interest. I've titled the talk Developmental Pathways into Language Emergence from Two to Seven, Late Starts and Surprising Arrivals.
The work is based upon work that my colleagues and I are engaged in - those colleagues are specifically Kate Taylor and Mabel Rice at the University of Kansas - and this is work that we've had in train through the National Institutes of Health. These are awards that are spanning the period from 2002 to 2012. It's funded research, about $12 million over this particular period of time, to cover the longitudinal work. Sample pools are variously large and comprise twins - or twin pairs rather - and singletons but there'll be a little more on that as we progress.
We're also supported in the work by a team in molecular genetics that's headed up by Professor Shelley Smith at the University of Nebraska. Hugh Catts provides us with reading consultancy in the research that attends the reading component of the work that we're doing. Javier Gayan at Oxford leads the behavioural genetics team and Lisa Hoffman handles some of the biostatistical pieces that fall in and around the work we're doing.
Not all of this is play in the talk that I have today and if this is successful I might tempt AIFS to offer another invitation in the future. It's going to depend on how well this works, I suspect.
I want to start the presentation today by asking you to think very carefully about the way that language emerges. That's "Steve-speak" and I either get through this hurdle with you or the talk fails. It's an important aspect of your consideration of what I'm going to talk about because today the word and the emphasis is on emergence, the way that language starts. We are so fundamentally shaped right now at this moment by language and the way that we use it, that it's impossible to imagine or think about how it was that it emerged.
I think it's fairly safe to say I haven't met anyone who is certain that they remember what it was like not to have language. I can make claims about what I am told about my first words, but I don't remember that "light", which happened to be the first word that I used - that's what's claimed at any rate by my mum and dad. But "light" was it and I reflected on my wife, Ann's first word; her father was listening a cricket game and somebody was out for a duck and Anne uttered "quack" as a first word.
But I want you to think about the issue of language emergence, the way that language gets started. Its features are quite striking.
It's early, around 18 to 24 months, is where the start-up period is generally agreed to occur. At least that's where we can observe it. It's certainly timed from the standpoint of the way that it emerges. One of its prevailing features is, for the child, a prodigious acquisition in the absence of any formal structured input.
Now parents often bring into this, "oh no we helped Fred acquire language", yet help really wasn't required. Parents often feel quite affronted by that. What you were doing was good, I'm not denying that at all, but the fact of the matter is that language acquisition occurs in all children without really the assistance of any formal structured input. They're out there imbibing, as it were, the language environment and acquiring without, as it were, the assistance of formal structured input for it.
The other feature about it is that it is relentlessly focused on the adult model. Acquisition progresses towards the adult model and that adult model is the adult model that's available to the child in the environment that they're in. So for example, you can hear in my dialect that I'm not Australian, okay? That's present in the sound of what I'm producing. The adult model of language that I achieved was the adult model that was around me and acquisition is focused on that. Boy, does it get it and I'll tell you what; you're extremely sensitive to when it's not perfect, when it's not there. You'll recognise it immediately. That's one of the features of language acquisition, its relentless developmental progression towards the adult model.
To the audience: Now I know there's been a discussion in here about questions or not, but it's really important that you put your hand up and ask questions as we go along, because there's some stopping points along the way where if you're not with me, I need to know that. This is an important stopping point. Early timing, prodigious and relentless. Okay? Good.
Again focussing on first language acquisition; how many of you speak more than one language? I want to focus on first language acquisition, or sometimes it's first languages because sometimes children acquire multiple languages right from the start, but we're focusing right at the first and early emergence of language. What are the fundamental features of that? First, one of the fundamental features of language acquisition is its universal capability. The child brings a universal capability.
The underlying mechanisms for acquisition have to be capable of responding to an arbitrary system. If you think of the child being parachuted to earth - I'm sure it's not quite done that way - parachuted to earth, and trying to work out language, there is nothing that would allow the child to predict cat, le chat, il gatto, mala, and mau (all words in different languages for "cat"). That exhausts the languages that I can do for cat. There's nothing that would allow the child to predict that depending on where they are. They have to have a universal capability to sort that out. It's any language, anywhere, spoken by anyone.
This commences at a period roughly 18 to 24 months and is apparently effortless. The other thing about first language acquisition is how robust it is. Enormous biological and environmental forces are required to prevent language from emerging and the sorts of things that prevent language from emerging really many times make you wonder if life's worth living. There can be great biological assaults, extremely confronting impoverishment of environment and the point here is that language is remarkably robust. It's a feature of being human.
The other thing about first language acquisition is that it's time sensitive. A second language that's acquired later is not acquired as effortlessly. The other thing is - and I think that we very quickly forget this - is that the neurobiological representation of the language in the cerebral cortex is a matter of great study and is extremely informative. We have that from laterality studies, brain imaging techniques and studies of course of adult traumatic brain damage. I had a former life as a speech pathologist and spent many years working with adults who had acquired aphasia, mostly through traumatic brain damage.
Nothing deepens your understanding of how language gets organised neurobiologically than being confronted by someone who has had an assault, a stroke or a brain tumour and had some of those systems impaired or compromised. The other thing that we have to observe are the conditions in which you lose a language and those occur in non-traumatic language loss where we observe mostly a decline in production of language ahead of comprehension (ie the understanding of language).
Language comprehension or language reception is something that's represented in the cerebral cortex much more broadly and much more widely. The distinction between expression and reception is a very particular distinction. Normally we see a diminishment of vocabulary and grammatical forms first and semantic loss, or the loss of meaning, occurs in very late stages where you're looking at non-traumatic language loss such as dementia.
The other thing about language is that language is a developmental facilitator. It is great to have it in your took kit. It's a really super thing to have. It's a fantastic substitution for the fist. It's better to negotiate - I like to think that it is - the world is a better place with language. It provides a tool for the acquisition of new knowledge and that knowledge can be vicarious rather than through direct experience. "Don't touch that, you'll get burnt." Often times though, direct experience does tend to follow.
It's a necessary entry ramp to reading and part of the language cycle. It's used to signal social class and dialect boundaries. We also have formal uses and registers such as standard Australian English for example is a spoken register and a written register in Australia, but we also have technical and professional language registers that go with this. It's a tool for life-long learning and it can provide increasing returns; not a minor thing. That's kind of a bit of a background on some of the foundations that make language an important developmental phenomenon.
I want to talk a little bit about now - still focusing on first language acquisition - some of the fundamental processes. I'm going to introduce the second staging post here. This is a piece that needs to be got - as it were - because I'm going to be using the word, "components", through this talk and this is the place where I talk about what are the components of language.
The component systems within language are sounds, grammar and semantics. There's a fourth system that developed in what would have been probably the late 1970s, called pragmatics, that deals with the social uses of language. I'm not bringing that into this aspect of the talk, I can go there if there are questions later, but the principle pieces that really fundamentally comprise the language system are the three components of the sound system, grammar and semantics.
Now it's important to appreciate that these systems develop over time on their own time course. They are synchronised but the sound system, for example, of the adult model is usually stable by the age of eight. Usually it's pretty much represented by the age of eight and we don't expect further growth or change in it, unless you're learning a different language. Normally the native or available language model is phonemically stable quite early.
Grammar or syntax is normally, that is the adult grammar, is achieved by the age of 12; we expect that to be a stable system. Yes, there's elaboration of grammar but that tends to be for special purposes and writing purposes. The basic grammatical or syntactic system of the adult model is achieved by around the age of 12. Semantics, or meanings, that's a life-long process. You're always adding to your semantic system; you're always learning. We see this particularly in - this is an example of it - life-long growth in vocabulary would be an example of it.
Very important, three component systems: the sounds, grammar and semantics. Any questions about this? Yes?
Question 1: In terms of the first two, which seems to have [inaudible], is it neurological, is it the brain that drives that to those points and what happens after that or is this a statistical representation?
Stephen Zubrick: That's what this talk is about. So you couldn't have a better question. Watch what happens here and then we'll come back to it if we've got a gap there.
Okay, so when we look at populations of children we largely see three developmental phenomenon and at a population level, 1) the vast majority of the population acquires the normal adult model of language. 2) Some children start talking late. 3) Some children never acquire the adult model of language. They go on to have specific language impairment or SLI. These are children with apparent normal motor and sensory function, they have at least average intelligence, no known syndrome or condition and nothing's compounded here by bi- or multi-lingualism. They just don't make the adult model or achieve it and I'll be highlighting a few pieces of what that looks like in a moment.
The question that our team has been working on through the research that we're funded to do is really a big question. How does language work? We really want to know, how does it work? There's this amazing thing happening. There's an obvious timed genetic underpinning, but where are the clocks that start it going? Where are the clocks; at 18 to 24 months, something gets going, for all children, everywhere. There's an obvious neurobiology.
We know the brain has to handle this and we've got plenty of evidence to show some of the mechanisms that are in play and there are obvious environmental inputs, you do learn to speak the language that you're hearing, from that standpoint. The reciprocal relationships between the genetic, biological and environmental sub-straights remain opaque, that's one of the fundamental tensions in the research; how does language acquisition work?
To be more specific in this talk, does starting late matter? Are there late talker effects on later language acquisition? Some children do start to speak late, does it matter that they do? Are there environmental risk effects on measures of grammar, vocabulary or composite language scores as indexed by the mother's education or SES? Is there something in the social and material circumstance of the child that might be essential or important, that will drive language emergence, thus making it late for some and not for others?
What does "late language emergence" - that's a Steveism , an LLE - what does late language emergence predict about later language development? The fact that you're observing it, does it predict anything? Now we're moving right down into this talk from the standpoint of what the focus of interest is. Why do this? Well we don't know why some toddlers start talking later than others and frankly we don't know if it predicts anything, or we don't know a lot about what it predicts.
There's some reasons why we have an interest in - and I'll go back to language emergence, because this again can be tricky - the speed of language growth and its responsiveness to environmental inputs quickly pose some big methodological challenges, really big challenges. There's an obvious covariance between neurobiology and environment. There's a potential confounding of causes and effects as the kid sweeps towards acquiring the language model. If you're trying to understand how it works, you're having to battle your way through larger and larger proportions of the environment that's shaping what the kid is doing. So the interplay of biological, genetic and environment mechanisms is most observable early, because you've got less of it, you've got less of the environment in the form of manifest language pouring out of the kid. The fact that language begets language, you can use language to acquire more.
There's a steady scientific debate about the role and primacy of genes and the environment, however when you look at the extant literature around language acquisition there is a predominant view that late language emergence is governed principally by the lack of environmental inputs. When you look at the broad literature that sits there, that's the story that's then narrated and you can go back to the original literature on mothers, because this was actually put at the doorstep of certain people over others from the standpoint of likely culprits.
Let's give you a quick overview then. Normal language, we expect to see emergence occur between 18 and 24 months. It follows an expected growth trajectory and the component systems, there's that word "component", are well synchronised; things sort of hum along. In late language, typically it's defined by a start that's past 24 months of age, it follows a slower trajectory in terms of acquisition and the component systems, sounds, syntax, semantics are not well synchronised. What you tend to see is the semantic system outstrips some of the other components. Yes?
Question 2: Could you just give an example of that [inaudible]?
Stephen Zubrick: Coming. They'll beat it. Yes, we'll get that for you. So let's see what we've got here. In a normal language your first words might occur at 12 months, you'll hear mummy; teddy; drink. Two to three word combinations by 24 months, so here you see an emergent grammar, mummy splash, teddy drink, but none of the grammatical marking occurring. We can get a noun phrase, we get a verb for you, not quite ready to inflect verbs, but we're beginning to line up the grammatical pieces.
Then what you tend to see a little later are verb inflections at 36 months and that's called a morphosyntax because the shape of the word changes so you get "mummy splashed me", "teddy wants a drink", - better verb marking. In late language again a late start, a slower trajectory and the component systems are not well synchronised, so you see vocabulary growth - and this answers Andrea's question - vocabulary growth outstrips morphosyntax.
So all the vocabulary is there but it's missing pieces of the morphosyntax. It's not there and that's a feature that we're particularly interested in. Where language is on time, we expect at 24 months a vocabulary of at least 50 words and we expect the child to be combining words. Frequent use of two or three words together, "no bed", "me up", "teddy sleep', "want mummy' - and that's reasonable - okay Fred looks to be on time. In late language emergence, the definition of the phenotype, as it's increasingly being called, is a vocabulary of less than 50 words and the use of single words. No word combinations, quite a diminished presentation.
There's some early warnings here in the literature about late language emergence. Donna Thal and her group published a work on a thousand single-born children that they had studied starting at 10 months and finishing at 36 months and Thal concluded in her work that we've consistently failed to predict language delay from any age using a single language variable. We haven't been able to do it. Now Thal's work had some limitations, she was using a non-representative sample. They stopped following the kids at 36 months, so you're wondering where are we with the acquisition of the adult model? She really had no robust measure of language at three to four years and beyond.
The UK study by Philip Dale and group, looking at twin pairs, again noted that the poor prediction of outcome from two year measures is a disappointing result for those interested in identifying children at risk for continuing language difficulty. So there's a holy grail out there that says, "got to get in here at two and at an early point and see if we can predict who might not be on a trajectory" and these are large studies with cautionary tales around this.
Okay, so we're back here. Now we're back to the staging post. Does starting late matter? Are there late talker effects on later language acquisition? Are there environmental effects and what does LLE predict? This is the way that we have started to look at this. This particular piece of the NIH work that we're doing focuses upon our work with singletons in a study called the RASCAL Study. This study was designed as a cross-sectional population sample of non-Aboriginal singleton children who were born in 1995 and '96.
We sampled from midwives' notification papers, 4007 families, we sampled with replacement in order to get this sample together. That comprised the original cross-sectional population. That work was productive in looking at [point prevalence] estimates of certain conditions and maternal factors in those years. At that point we thought it prudent to develop out of the cross-sectional study a longitudinal capability and didn't have the money to handle all 4007, so we randomly sampled, from the 4007, 2800 families and invited them to participate in a longitudinal study from age one to age eight. That's what was on offer and 2200 of them agreed to participate.
For the work we're presenting today 1880 caregivers completed a two-year questionnaire. On looking at those, some of the families that were in this sample had to be excluded, our interests were in the English-speaking component of them and so this particular work features 1700 children. We excluded children with known neurodevelopmental disorders and children with exposure to languages other than English. They came out, so we've got English-speaking singletons with no known syndromal or neurodevelopmental disorders.
In terms of the sample that we have, the social benchmarks are all within four per cent of the population benchmarks. That broadly means that the participants at two years of age were little more likely to be families that were earning more than - remember these days - $25,000 per year, a little more likely to be from married households with a higher level of maternal education. These percentages again are within four percentage points of population benchmark.
Aside from that there were no significant differences for the mother's place of birth, the number of children or adults in the household, father absence or the receipt of government benefits. It's a pretty good approximation to the population at the time. We asked the parents to rate the child's ability using the ages and stages questionnaire in terms of the child's ability to point to pictures, use phrases, carry out simple directions, name simple objects, point to body parts on request and use personal pronouns like me and I. The choices that the parent had was "not yet", "sometimes' and "yes"; Fred can do this.
The ASQ measure of language, then we had to ask is it sensitive to language growth and boy, is it ever! It's surprising how effective a parent-rated measure of early language emergence is. These are the Rausch scale results from the scale score. It effectively discriminates early language emergence. What you're wanting in this particular response curve is a good ability to get the low end of the ability spectrum. As kids get more and more of the items, they hit the ceiling which is exactly what you would expect.
In this sample we had a good representation of children who were on that early language emergence spectrum. It's quite effective at doing that. That permitted us to establish a cut-off to define late language emergence at the first standard deviation. We were able to say alright, we'll declare the child to have late language emergence on this measure and that produced in the RASCAL sample a prevalence of 13 per cent of language delay at age two. In our sample that's about 238 children which equates to about 3250 two-year-olds in each Western Australian birth cohort who would be defined as having late language emergence.
Fortunately we were also able to gather a more direct measure of vocabulary on these children and in the sample we observed that the children with late language emergence had a mean vocabulary of 62.5 words with a standard deviation of 52. So we had some pretty quiet kids in that group from the standpoint of word use, whereas the children without LLE had a mean vocabulary nearly three times that large. So the ASQ is discriminating quite well kids with smaller vocabularies and the absence of two-word combinations from the others.
Any questions about that? It's an important feature of the work. We've got a definition, an early indicator, that works. Okay. Now we're going to start the process of what predicts LLE status? So our interest here is, is it possible to predict LLE status? I have about half an hour, is that right? Forty minutes? Is that allowed? Okay, good.
We asked questions about gender, age, race, gestational age, foetal growth, birth weight and time. There are several components to this model. The birth characteristics that predicted late language status were gender, gestational age and foetal growth. The odds of late language emergence were higher for males, about two and a half times higher, and for infants with foetal growth restriction and those who were premature.
From the standpoint of maternal characteristics, we had a raft of maternal characteristics that looked at the mother's age, education, her country of birth, employment, parenting style, mental health, cigarette use in pregnancy. None of those predicted late language emergence. This was a show-stopper in the literature. Up until then there'd been a heavy narration around environmental inputs predict late language and this brought things to a stop in the publication. We published quite successfully in it but this was a real heads up.
In terms of family characteristics, we looked at family structure, size, income and function, the local SEIFA indicators and a family history of Late Talking, family size and a family history of late talking were predictive. Children were more likely to have late language emergence if they came from larger families. They were more likely to have late language emergence if there was a family history of late talking.
Finally we looked at the characteristics of the child at aged two in terms of their temperament. We had other ASQ measures of motor function, adaptive personal-social functions, and the child behaviour checklist for mental health and day care status. These children were slower on gross motor, fine motor, adaptive function and personal-social functioning. So the ASQ measures basically said slower here. Where are the clocks?
So the story so far is that we can identify late language emergence at the age of two years from a parent report measure. We can go, does the child apparently have late language now - yes we can do that, we can say we've got a pretty good phenotype that captures it. It's the earliest point at which Late Talking can reliably be identified. The early predictors of late language at two are almost entirely internal to the child, gender, foetal growth and prematurity. Family history is a significant predictor and children with LLE at age two are also likely to show concurrently poor motor development and adaptive/social development.
While LLE can be identified at age two, the data here do not lead though to useful recommendations that would allow an efficient prediction of which child might have LLE from another set of prior observations. You're better to measure it at two rather than predict from birth or from circumstances in the child's environment. The findings are notable for the absence of environmental drivers of language emergence. It's not a matter of priming the pump. There's not strong evidence for environmental priming.
Having at least identified those two-year-olds with LLE, what does LLE status at age two predict about later language development? Now that story, and I'll check with my compare here, requires about another five or six minutes. Is it okay if we go there? Oh we've got thumbs up from someone, so this is the point where if you're going to flee, now is your chance. Are you with me in the story so far? Okay.
So having defined LLE at age two, what does it predict about later language development? How are we going to measure later language development? How are we going to do that? That becomes a really critical methodological issue, because your choice as a measurer is to have an omnibus measure that, as it were, aggregates or agglomerates a whole bunch of information about language into a score, or measure the development in the component systems.
Everything that we have about language indicates you better be measuring the individual component systems because language isn't one thing. It's a set of component systems and we've got pretty good evidence that they become asynchronised, that something's happened that the synchrony isn't there. So the threat of asynchrony in the component systems of language requires that we measure phonemics, syntax, morphosyntax and semantics in order to go, "what's happening here' as the child moves from age two to age seven.
If you've got an omnibus measure of language it will conceal asynchrony and result in differing mixes of results, depending on which component parts are summed to make the composite. That's a big deal, because you'll get, "well I didn't find a difference". Well, "what was your composite measure?" Oh well, your composite measure is mostly semantics. That system's not cock-a-hoop here. Does this make sense? Okay.
So I'm going to skip a bit of this other than to say of the 228 with LLE we had to go and locate them because the study had finished. We located 169 and recruited 128 for intensive examination at age seven. By the way there was no significant difference in the proportion of children with LLE in the age two and the age seven. That's important. We kept that together. We located 765 with normal language and randomly sampled 109 of them in the sample.
All participants with both late language emergence and NLE (ie Normal Language Emergence) at age two were given a general omnibus language measure, a speech examination, a specific examination of components, nonverbal intellectual function and a hearing test. At total of 17 measures of speech and language were gathered in this study and I won't detail that, we can go there if we have to.
Now what we're going to look at here is, I should say, poor performance at age seven. I've left a word off this slide - poor performance at age seven - for groups differentiated by age two - language emergence. We could find no significant difference between those kids with late language and normal language at age seven, in their nonverbal IQ or in their language semantics. They were fine, completely comparable. In terms of looking at poor performance at age seven on an omnibus language score, the test of language development, spoken language picked up significant differences. All of the syntax, third person singular and past tense use, picked up a significant difference and the speech articulation test picked up a significant difference in the LLE but only insofar as that some of these words were used for grammatical marking. So there's a confound there.
The other thing that I wanted to show you is here. Only one per cent of kids with normal language emergence had any difficulty distinguishing the third person singular. They arrive at the adult model and 99 per cent of them are doing it fine, they're good. There's a 10-fold increase in that for children with late language emergence. Remember though, that these kids with normal language emergence at two are also being picked up at the rate of 11 per cent of them having late language at seven.
So the gateway to late language (at age 7) isn't, for all children, late language emergence (at age 2). Those kids have normal language emergence at two, but some of them are arriving into categories with late language at seven. So the hope that late language emergence at two is going to be the universal gateway by which kids get to late language is not a hope that one should be hanging on to. That's part of what the study is finding.
I'll conclude now and then we'll have a bit of a talk. The risk for what does LLE status predict at two, the risk for persistent language development differs depending on the component language system measured. Omnibus measures of language at seven will reduce or inflate estimates of persistence depending on the component language measured. Semantic and phonemic measures are going to fail to discriminate late language and language delay at seven, but syntax and grammatical tense marking particularly, is extremely vulnerable as a marker of both late language emergence and persistent language impairment.
We noted no gender differences in language impairment at seven. That's a huge surprise, because at two, there were major gender differences. What caused the boys to catch up? Where are the clocks there? Late language impairment at 24 months is not the first diagnostic symptom of language impairment for all children. To model the predictive validity of our measure of LLE benchmarked against the most sensitive measure of language delay at seven is out next step. That's what the next piece of our published work will be about and we'll contrast that predictability with the other component in omnibus measures.
If I have to leave a parting suspense story around this, it's that the predication from age two late language status to later language status is extremely poor. You wouldn't want to hang a hat on it. The general picture emerging here is that the best time to measure specific language impairment may be at age four and earlier prediction of specific language impairment before the age of four is - we're predicting on the basis of what we've got here - will not be efficient or reliable.
There are plenty of things that we can be doing with children in the epoch from two to four that are of great benefit, but the specific intervention around language with the notion that somehow what that's going to do is cause a catch-up, would need to be looked at with caution. And that's the story as it's emerged. Thank you.-- end --
IMPORTANT INFORMATION - PLEASE READ
The transcript is provided for information purposes only and is provided on the basis that all persons accessing the transcript undertake responsibility for assessing the relevance and accuracy of its content. Before using the material contained in the transcript, the permission of the relevant presenter should be obtained.
The Commonwealth of Australia, represented by the Australian Institute of Family Studies (AIFS), is not responsible for, and makes no representations in relation to, the accuracy of this transcript. AIFS does not accept any liability to any person for the content (or the use of such content) included in the transcript. The transcript may include or summarise views, standards or recommendations of third parties. The inclusion of such material is not an endorsement by AIFS of that material; nor does it indicate a commitment by AIFS to any particular course of action.