Language Impairment in Children who Speak Nonmainstream Dialects Gaining and applying a deeper understanding of dialect differences in terms of rate, context, and function to better distinguish African American English and Southern White English speaking children with specific language impairment from their typically developing peers. Presentation Video
Presentation Video  |   November 01, 2014
Language Impairment in Children who Speak Nonmainstream Dialects
Author Affiliations & Notes
  • Janna Oetting
    Louisiana State University
  • Presented at the 24th Annual Research Symposium at the ASHA Convention (November 2014).
    Presented at the 24th Annual Research Symposium at the ASHA Convention (November 2014). ×
  • The Research Symposium is hosted by the American Speech-Language-Hearing Association, and is supported in part by grant R13DC003383 from the National Institute on Deafness and Other Communication Disorders (NIDCD) of the National Institutes of Health (NIH).
    The Research Symposium is hosted by the American Speech-Language-Hearing Association, and is supported in part by grant R13DC003383 from the National Institute on Deafness and Other Communication Disorders (NIDCD) of the National Institutes of Health (NIH).×
Article Information
Cultural & Linguistic Diversity / Research Issues, Methods & Evidence-Based Practice / Language Disorders / Specific Language Impairment / Clinical Practice Research / Diagnostics, Screening and Assessment Research
Presentation Video   |   November 01, 2014
Language Impairment in Children who Speak Nonmainstream Dialects
CREd Library, November 2014, doi:10.1044/cred-pvd-c14001
CREd Library, November 2014, doi:10.1044/cred-pvd-c14001

The following is a transcript of the presentation video, edited for clarity.

I'm going to be speaking about language impairment in children who speak non-mainstream dialects of English. I'd like to recognize my collaborators. We always have a full, fun lab and two of my collaborators are Dr. Michael Hegarty and Dr. Janet McDonald who are at LSU and then the other people listed are a number of PhD students who I've been honored to work with and I'm not listing the many MA students and BA students who do a lot of our transcription and our coding for us. I want to disclose some funding sources. No, non-financial interest, that does sound so bad. I'm very interested in this topic, even though I don't make a lot of money on it, but those are my disclosures.
The Diagnostic Conundrum
I'm going to start with Harry Seymour's semantical article where he talked about the diagnostic conundrum and you're going to see parallels in my work and in Joanne Paradis' work, but he described the diagnostic conundrum as how do you find children with language impairment when they speak a non-mainstream dialect because linguistic structures of non-mainstream dialects that are totally appropriate for the dialect can appear identical to symptoms of childhood language impairment. Some very common ones would be the zero copula BE, the zero auxiliary BE, the zero verbal -s, zero past tense, and zero auxiliary DO.
If you trained in the 80s or maybe 90s, you know of this as the dialect versus disorder debate. When I've gone around the country and I've talked to different people, if they know anything about dialects, they know about the dialect versus disorder debate, and they know not to make the mistake of calling a child who speaks a non-mainstream dialect a language-impaired child. In fact, yesterday at a round table I had a number of new clinicians, and they were very proud to say that as teachers refer children to them and they're working in a school where everyone is a non-mainstream speaker, they just send them immediately back because they're non-mainstream speakers and the teachers are misdiagnosing the dialect difference as an impairment.
Okay, there's Harry's work. The reason I've moved to it is he had a solution for this diagnostic conundrum, and his solution was to recognize that dialects have two types of structures in them. They have the contrastive structures: those that differ across dialects and those would be the ones that I just showed you. And then there are these other structures that are non-contrastive that all dialects share. Like the use of articles, the use of conjunctions, demonstratives, locatives, negatives, etcetera. He said the way to solve that diagnostic conundrum is to just work with those patterns that are occurring in all dialects because you've evened the playing field. They're all using them, and that's where you want to find the language impairment. With the idea being that if the child has a language impairment, it's going to show up everywhere.
Contrastive Structures Have Dialect Specific and Dialect Universal Features
I guess we'll do some activity here. Raise your hand if you've heard about the diagnostic evaluation of language of variation test or the DELV? Oh that's fabulous. This is a test that was created to be dialect neutral, to identify language impairment in children. It's a test set and there's a screener that you use the first section to decide if you have a child who speaks a non-mainstream dialect, and then the second part to decide if the child is at risk for impairment. If they're at risk, then you would give the second, the norm-reference test. I want to disclose that we own ten copies of this test, and so we like this test very much. If you look into the manual, and manuals are not our favorite, right? Manuals are manuals, but it reports in its normative data the sensitivity and specificity are very good using one point, minus one standard deviation below the mean.
My point is not to criticize this test or to even criticize this approach to use the non-contrastive structures, however, somebody who trained with me or as someone who loves Larry Leonard's work, the study of cross-linguistic studies of childhood language impairment -- it killed me to think that I was going to move to Louisiana and throw out those parts of language that are theoretically interesting, that play such a large role in studies of childhood language impairment. So when you move to those non-contrastives you remove yourself from the table of all that good discussion that's happening in the field. I decided very early on in the early 90s that I was going to work with the contrastive structures or I was certainly never going to throw anything out, that I was going to study them all. So that's what I'd like to share with you today. I'm going to share with you some stuff that is published and then I'm hoping to end with some new findings from a project that we've been working on.
Okay so when we started working with the dialect, the contrastive structures, we read as much literature as we could, and we found out that there were all these contrastive ways in which dialects, these non-mainstream dialects, differed from mainstream American English. One of the things we learned very early was that if you studied one of these contrastive structures, it wasn't 100% different across dialects. That, in fact, these contrastive structures, there are things about them in the way children use them that are very dialect specific, so specific to their Louisiana dialect, and then they would use those structures in ways that were very dialect universal, in ways that were typical of speakers of general American English or mainstream American English.
Here's an example with relative clauses in African American English and in Southern White English. You can zero mark the relative clause, I mean it's a subject. So you can say: I paid the student, made the stimuli. Where in Standard English we don't say that. You can also use the what marker. I ain't got a sister what I fight much. Where in mainstream and general American English we don't typically say that. So those are dialect specific.
But now look at the dialect universal. In African American English and in Southern White English, the two Louisiana dialects that I'm going to talk about the most today, when a child speaking in those dialects uses a relative clause marker, and the subject that the relative clause is referring to is human, you're going to get who. Or let's say you're never going to get "who" unless the subject is human. If you hear the word where it will always refer to something that is a locative and non-human, and if you look at where you get your zero marking of that relative clause, you're much more likely to get it when it's a non-subject and a non-human. So, this is the bottle you gave me, which in general American English you can also say.
Okay, so you look at these dialect specific ways of this contrastive structure, and these dialect universal ways, and you look at how often it happens. Well look, 100% of the time African American English speaking children and Southern White English speaking children are doing the dialect universal features, just like general American English. Some 20% of the time or less there's zero marking that subject relative clause when you don't do that in general American English, and less than 5% of the time they're using the what instead of a that or a who.
So what you can see with this contrastive structure is that perhaps we don't need to perseverate so much on how it's different across languages, but more on what's universal across dialects, even in these structures that are considered contrastive. That lead us to say, okay we definitely want to keep moving this way. We're going to do this. Instead of using Harry Seymour's diagnostic conundrum metaphor, we decided to use the cross linguistic study of SLI metaphor, which says if I were going to study SLI French speaking children or Dutch or German, this is the question I would ask: how do same dialect speaking children with and without SLI differ from each other? That's exactly what you would say if you were studying French. How do French speaking children with and without SLI differ?
This is opposed to the diagnostic conundrum, which asks a very different question. It asks, does this child have a dialect difference or a disorder? So it forces you to ask a question that you have to figure out which box to put the child in versus the cross-linguistic approach says no, this child can be both. They're all going to be non-mainstream dialect speakers and now I want to figure out who is impaired in that group. So there's no either/or.
Typical Dialect Variation
So to do this, anyone who is doing cross-linguistic work, the first thing you have to know is your language or your dialect. You have to study typical dialect variation before you can begin to think about well, what would an impairment in those dialects look like?
The two dialects that we studied are African American English and Southern White English, and our children are predominantly girls. The first set of studies that I'm going to show you, the children lived 30 minutes to an hour away from the Baton Rouge area. Baton Rouge is a university town, it's large. When I started, Baton Rouge had about a quarter of a million, and now it has about three-fourths of a million people in it, so it is not rural by any means, but you very quickly get to rural areas. So those studies that we have published a lot on are from children half an hour to an hour away, and then in our new work children live an hour to an hour and a half away, and you also cross the Mississippi. If you know anything about dialect, things like rivers are nice and natural boundaries for language. So you get differences depending on what side of the river you live on.
So I'm going to tell you a little bit about typical dialect variation. One of the first things we learned that wasn't well documented in the literature was that these dialects share a similar inventory of non-mainstream grammatical forms. We've found in the literature 35 to 36 different non-mainstream forms that are supposed to occur in African American English or Southern White English. Now nobody had compiled these for us, we just would search the literature, and if somebody talked about one we would throw it on the list. We're always, whenever we hear a child say something, we'll go see if we can find it in the literature. It sounds dialectical, can we find it? But anyway, we found these and our first study had 93 children, as I mentioned, and the purple are the structures that only our AAE speaking children produced, and the red are the structures that only our Southern White English speakers produced. All of the ones in black, both groups produced. So there's a lot of overlap and in our newest set of data, which is 252 children, and all these children or most of these children are in kindergarten. Our new work is all kindergarten based, we're replicating this finding.
Then you could even look at what are the frequent structures of these dialects. So we did, we looked at the ten features that are occurring the most to see how the dialects differed on these. What we found is that the two dialects are sharing the same structures that are high frequency. So in this table, the structures that are red are shared or are on both lists, and the structures that are black are unique to the lists. Okay, so even though they're in both, these are the ones that are the most high, occurred most frequently, and then the percentages in the parentheses are the percent of children in that group that produced them. So you see very high percentages for these structures. So even though these dialects are sharing structures, they are not the same dialect.
They differ in many ways, and the three ways that we have come to think about these differences are in rate of use, so how often do speakers produce these non-mainstream forms, the context in which speakers choose to use these forms, and then the function that the structure plays for that speaker in their language production. And I'd like to say that if you move out of the US literature, and you move to the British English literature, one of the things you see in the dialect world worldwide is this: they recognize that this is how American English and British English differ as well. So these are universal properties of dialect variation that extend beyond the United States.
I'm going to show you a little bit of data about each of these so that you can get a sense of how these dialects differ. First we're going to look at rate. How often do children produce what would be considered a non-mainstream form of these structures? And here you see two different, two sets of numbers. This is a dialect density measure where you have a language sample, and you count up all the non-mainstream structures and you divide it by the number of utterances. That's a very crude measure. One of the reasons why we do it that way, rather than percent correct, is that there are some non-mainstream structures that don't have obligatory contexts. So it was a very early 2002 or 2001 way that we said, well this will at least get us a sense of rate. And here you see the mean for the AAE speaking children is .29, the mean for the SWE speaking children is .12. You look at the standard deviations, but now look at the range. Both dialect groups have nice wide ranges -- a little bit of overlap, but not a whole lot of overlap.
Now I want to point out that that was one measure we used, but then in previous studies and as well by other research labs, it doesn't matter what dialect density measure you want to pick, you're going to find this variation within a dialect group, individuals who use these non-mainstream structures very little and individuals who use them frequently, and anytime you have a black/white, two dialects together you're going to see this difference, so it's a very robust finding. You can use language sample data, you can use blind listener judgements where you ask people to just listen to a minute and they rate the density, or you can use the DELV screener, which has that.
The second way dialects differ is the context in which speakers choose to produce a non-mainstream form. This is kind of the textbooks from sociolinguistics that talk about the BE system for African American English. The BE has been studied very frequently. I think Brandi Newkirk in her dissertation found 20 studies that were done on the structure of auxiliary and copula BE in African American English, and in all those works, especially in the sociolinguistics works, they talked about how zero marking of BE is not random, it's highly systematic, and there are constraints placed on where you zero mark and when you don't. In fact, they often times refer to this as these are the linguistic constraints of zero BE, and so there are three constraints; person, tense, and number, constructability and grammatical function. This will play a role in whether you're going to hear a zero marked form or an overtly marked form.
So I'm happy, you're much more likely to have an overt marking, you'll have lower rates of overt marking for he's happy, lowest rates of marking for you're happy. Past tense, you're much more likely to have overt marking than present tense. Uncontractible, you're much more likely to have marking than contractible. Now notice, for those of you who have studied typical development of general American English. In mainstream American English, the contractible usually comes in before the uncontactable. So this is a place where the dialect is a little different than typical mainstream American English development.
Then finally, the copula versus the auxiliary. You're more likely to get overt marking if it's functioning as a copula as compared to an auxiliary. And in a recent paper by Joe Roy that we published, we looked at our children's language samples and we found that both of these dialects children are constrained by these variables, but not in the same way.
So, the red numbers are statistically different from the black numbers. What you see for African American English is that all three variables, or all three contexts affects rate of marking for the speaker and the direction of the constraint hierarchy is that is is overtly marked at higher rates than are, am, was, and were are marked -- that is versus are is the wrong way. Are and is should be flipped. Are is marked less than is, is and are are marked less than am, was, and were. In SWE, you only get two of the constraints, so you get that person number, but it's only influencing are. All of the other structures are at high rates.
For contractibility, AAE speakers are affected by contractibility, Southern White English speakers are not. And then finally, for grammatical function, that affects both groups, and you see those are the numbers, the percentages. So this is a percent of overt marking or proportion of overt marking. Now, we can move into counting all of the contexts, just like we do in general American English studies, all of the BE contexts, and then what percent of them are overtly marked. And so a sociolinguist would look at this and say the constraint hierarchies of these two dialects are different and both number that are effected in the magnitude of them.
Then finally, one of my favorite things to talk about is that dialects may look like they share the same feature, but it's functioning differently for the dialect. So here's a typically developing six year old who is telling a story within her conversation, and it went like this. "My mama said she was about to go to Bible study, and on the way back her car had stopped. Then she had called the house because somebody let her use the phone. That she had called the house and I said, 'Hello, who's this?' Then my mama said, 'It's your mama. Let me talk to your daddy.' Then she had told my daddy to come with us and bring a big rope so they could pull the car home. So we got a new car."
Yesterday I stood by a poster and I just asked those who came if they know that structure. Anyone who works with African American English speaking children will shake their head and say "oh yeah, I've heard that structure." No one who came up to my poster though knew what it was. It was always described as "yeah, they're using it wrong. It's a wrong past tense form." I'm like, actually it's not a wrong past tense form, it's a form that African American English has that Standard English does not. In standard American English, we use the had plus the verb to be the past perfect, that it happened before something else in the sentence. But in African American English, it can be used for the past perfect, but it can also be used for the simple past. So it has an expanded version.
We didn't come up with this. John Rickford and his student in 1996 noted this, he had, they had samples from California, teenagers, and they had narratives and they found 52 cases of this had plus verb and they looked at it and they concluded that 96% of the time it was being used to refer to the simple past or the absolute past relative instead of the past perfect. One hundred percent of the time it was when the speaker was telling a personal narrative, and 94% of the time it happened in the complicating action clause of the narrative. So it's used to add drama to the story. Now, that was in California, we're in Louisiana on the Mississippi, where you don't collect data on Fridays because your kids will be fishing. And here's what we found. The first row is these had plus verb structures. Our Southern White English speaking children who go to the same schools as our African American English speaking children didn't produce this structure, but our African American English speaking children did. Nine percent of their past tense expressions were with this had structure, and then we had individual differences and so subject number 52, who gave you that nice narrative of getting a new car, she had 28% of her past tense expressions were this preterite had. And so what you see by looking at her data relative to the AAE is that she adjusted, you know as she had more of these than she had less of the others.
Okay, we also then said well, when are children using these? And 90% of the time it was when the children were producing a narrative and unlike Rickford, who had interview data where they specifically prompted for stories, we had conversations with play samples. So our narratives are inner-conversational. So they were just coming up through our prompts, we'd love to have the prompt, I bet you've been in a car wreck before. Okay, different prompts like that. I bet your brother is really mean to you. And so then you get these personal narratives, and 90% of ours were in narratives, and 84% of them in the six year olds in kindergarten were in the complicating action clause. Now I'm confident that none of the kids, that in this first study, had ever been to California. This is a robust structure in African American English.
I've been told that in New York, you hear it across many different dialects. Languages and dialects evolve. I, living in the south in a rural area, our dialects would be slower to change. If you read the sociolinguistics literature, it would say you get the most creative innovation in urban settings. Some authors will highlight women as being the movers and the shakers, and that in rural areas you're much slower and you're more likely to keep some of those structures that are consider archaic structures, like the relative what that kind of has disappeared in urban areas, but we still have it in rural area. So I wouldn't be surprised at all if this had structure doesn't cut across other dialects and doesn't evolve because it really draws you in. It has a nice pragmatic function, and although I don't have the slide up here, but certainly in the paper we published, what we found was children who'd use this were not the heaviest dialect users. What distinguished them is that they had better narrative skills. So if you see this marker, it's a marker of linguistic ability rather than linguistic weakness.
So to sum up this part of my talk, I'd like to say that one of the things, if you're going to work in dialects or dialect variation, you need to appreciate that our dialects share a number of non-mainstream and mainstream structures. But they do differ in rate of use, context of use, and function of use. And anytime you are thinking about writing a lecture in dialects or you're thinking about or you're reading something about dialects, if you read a sentence that doesn't tell you about rate of use, context of use, or function of use, you have a bad reference because these dialects are much more interesting than what we've been writing for the last 20 years.
When you hear, "oh in this dialect you can, you have final constant cluster reduction," which you do in both of these dialects. What you want to ask -- what we would ask as a consumer is -- "Okay, so that happened. So that's interesting. So how often does it happen? On what words does it happen? What contexts and what function does that play for the speaker?" And if we start asking that, we're going to know a lot more about these dialects. And when you then talk to a child who speaks a non-mainstream dialect, you immediately have so much more respect for what they're saying, then, if you think of their dialect as linguistically complex instead of "I know I'm supposed to think it's equal to others, but I just can't stand how it sounds." Children know that difference and so if you can embrace the academic world and what we know about these dialects, you'll probably find yourself listening and being really interested in when children are using these structures and when they're not.
Linguistic Profile of SLI within AAE and SWE
So now we're going to move to the study of SLI within these two dialects, recognizing everything I've just told you about dialect variation. To do this type of work, you would need four groups. You would need a typical, where you would need a dialect group with typical children and children with SLI in African American English, and then you would need another group of typically developing SWE speaking children and SWE with SLI. And if you wanted to add mainstream, you would have two more groups and so that's what we, that's what we do in our lab. It's always a four-group study.
To start, we went back to our inventory of non-mainstream structures. Remember these are supposed to be those taboo structures that vary across languages. So we're not supposed to be able to find an impairment in these, and we said, well let's look at how our four groups are using these and here's what we found. This was in the study in 2001. If you counted up the frequency of those 35 structures and divided each by the number of utterances produced, and you threw them all into a discriminate function, 97% of the time you could tell who was an AAE speaker and who was a SWE speaker. So using those 35, that's great classification. If you did the exact same thing and asked the computer to try and figure out who is impaired and who is typical, you're at 90%, if you throw in all of those.
That wasn't supposed to happen. They were all, "you shouldn't have been able to figure out who is typical and who is impaired if these were problematic structures." Then we asked the computer, well now what would be the smallest number of structures you could use to differentiate the groups? And so the reduced model was 94% accurate and it identified these four structures as showing different, being able to differentiate the dialects, and then with the reduced model for impairment it was 82% accurate, not as accurate, and it identified these four and so what we saw from this is that the lists were not identical. If you had a diagnostic conundrum and these structures were affected by impairment in the same way it was affected by dialect, you should've gotten the same structures. You also shouldn't have been able to do that, but we were able to.
And then finally, the last part of that paper, we said, well maybe you can get better sensitivity and specificity or diagnostic accuracy if you run it just on the African American English speaking children and say okay, how do you tell who is impaired and who is typical? So when you do it on just the AAE speaking children, your model, your reduced model is 82% accurate and that's your sensitivity and specificity, not great and Southern White English, it's a little better, 91% accuracy with sensitivity and specificity, close to 90 and those structures pop out. If you then take the model and switch it, so you take the AAE model and you try to identify impairment in the SLI group, it falls apart and if you take the SWE model and throw it over and try to identify who is impaired and who is not in the AAE, it falls apart. So, as any cross-linguistic researcher says, if you're going to study SLI in French, you would have French controls. You wouldn't use a Dutch model to find it in French. So exactly here, you want to stay within your dialect when you're identifying who is impaired and who is not.
Then I just want to point out three other studies because this study was based on count up the frequency, divide by the number of utterances, very crude. Now, we're going to go to more traditional SLI researchers where you calculate percent of use in obligatory context and what you see here are three studies that have been published. The SLI group in one column and the typically developing in the other column and you'll see that they're all red because the SLI percentages are lower than the typically developing. So you keep dialect speaking controls for your typical group and you get a difference. I also want to highlight that the typically developing kids, because they're AAE speakers or SWE speakers, they are not speaking Standard English. Those are not what a six year old speaking general American English would produce, but we're not comparing these kids to general American English speaking children, we're comparing them to each other.
Finally, I'm going to talk about this fourth study that we published in 2013. This was on verbal -S. If you know anything about the non-mainstream literature, you would know that the verbal -S, first of all the reason why it's called verbal -S in the non-mainstream world is because you can have this -S with verbs when the subject isn't a third person singular. So you can have "they walks", "so I says to him" blah, blah, blah, and so people are more comfortable just calling it a verbal -S marker than an actual marker of number. When Leslie Cleveland looked at verbal -S, she found that she could not differentiate SLI in typical African American English. So 14 is not any different than 21, statistically, but in the Southern White English literature you could, or in our data you could, and so I wanted to throw this out here.
I don't want to make this seem like this research is easy and that you can just go out and collect these structures and you're going to find a difference. You definitely need to be careful and go slow. This is one that is not panning out to be useful. If you look at verbal -S and you look at the context and you split verbal -S up and you look at regular marking, irregular marking, and then irregular marking with negation, in particular the word doesn't, which is always produced as don't. He don't want to do it. You find that even in Southern White English, you can identify a difference in the regular and the irregular, but the group effect washes out with don't. Actually Mabel identified that years before we did this, even in the Kansas City population. That they throw out there doesn't, don't context because that's a dialect variant, definitely in Kansas City, definitely in Louisiana, perhaps in your area too.
And there are other things that we do that I'm not highlighting. There are certain words that we don't put in the analyses. So to give you an example in standard or general American English, when we're looking at past tense we never use the verbs put, hit, or cut. Why do they not get included? Because their present and their past are identical. We would never count those are zero marked forms because general American English doesn't mark those, and so one of the things we know in these dialects is that's what's happening with don't, you're always going to get don't and you're going to throw it out. We throw out the verb used when it's with used to, when we're looking at past tense, not so much because it's always zero marked, but because we can't reliably transcribe it. Used to, used-to, used to, used ta. It's just too hard. We can't tell when it's marked and when it's not and so until somebody does an acoustic analysis of that verb, we throw that verb away.
So if you're going to do this type of work, pay attention to your sample and pay attention to what you're putting in there and make sure you don't have specific lexical effects that are impacting your results.
AAE-speaking Children Reared in Poverty
Then I want to highlight real quickly before I get to our new study, Sonia Pruett was a PhD student with me and she said, well yeah that's nice work, but maybe your typically non-mainstream speakers are, you know, precocious? And I said, well, their test scores aren't precocious. But she says well what about children reared in poverty? AAE speaking children who are reared in poverty, maybe they have something that parallels SLI. And I said well okay, why do you think that? And she said well, children reared in poverty, and children with SLI or children with poverty earn low scores in standardized tests, and they also have high dialect densities. So they would be a really hard group to figure out who is impaired, and whose language is depressed because of poverty but they don't have a clinical impairment. And so I said well do that for your dissertation. And so she did.
She posed the question, do children AAE speaking children reared in poverty, do they show an SLI grammar profile? So the SLI grammar profile she looked for would be low rates of marking of past tense, absence of errors of commission because SLI children don't make errors of commission, neither do typical children -- high rates of errors of commission. Absence of high rates of over regularizations. We have a task where you ask children to past tense irregular verbs like fly, so it'd be a deverbal, it's a verb that came from the verb category today I fly a plane, yesterday I flew and then you can also create new verbs from nouns. A denominal or a verb that came from a noun -- one example would be the word fly as in the fly that buzzes around. And so if that word is fly as buzzes around, you can create a situation where you can turn that into a verb and the way we did it is we had a bunch of verbs that were from verbs, and then verbs that we created from nouns and for the fly one we would say, can I put flies on your arm and every kindergartener loves flies on their arm. So they would hold their arm out, and we'd put all these paper flies. And then we'd say, so tell me what we did? "You flied by arm." Versus when you do the fly, you get he flew the plane. So one of the things we know in typical general American English development is children very young in preschool can do this differential marking. They will regularize the denominal verbs so we can cheese a hamburger, we can meat hamburger. So you meated it. Versus Cookie Monster can meet Big Bird, he met Big Bird. Okay so that's just another example. Language impaired children, children with SLI can actually do this test too, even though they don't mark past tense at high rates. So I said just throw that in there too because that's what professors, mentors do.
These were her participants. She had 45 typically developing AAE speakers, she had a low SES group and a middle SES group. So they're all kindergarteners, they're all African American, they all speak African American English. Her low SES group, her mother's did not finish high school and then children went to schools where 90% of the children at the school received free lunch, and also those schools had low test scores relative to the state of Louisiana, which we're pretty low. So that was even lower. Then her mid-SES group, mothers had to have finished high school and have some college, and they went to schools where less than 10% of the children there received free lunch. So differences, but all were typically developing, so you see a maternal ed difference, you see a PPVT difference. I think her low SES group they also couldn't have a PPVT above 90. I think she only ended up having to throw out two kids who had PPVTs higher, but the sense was that if you were going to be reared in poverty, poverty was going to affect your language, that you would have a depressed vocabulary score and her dialect density measurement was also higher for her low-income group.
And so she went, and she did a past test probe and a couple other past tense things, and this is here what she found. If you look at the low SES group with language samples, low and mid, and then you look at some experimental probes she did for past tense, low and mid. One of the things you're going to see is that the low and mid look very similar. In fact she had no effect for SES in this study, she did have a main effect for verb type, and that all of her children overtly marked irregular past tense at lower rates than regular, but that she attributed to dialect. She did not have an SES effect.
The other thing that I wanted to point out is Harry Seymour had already published a paper with AAE speaking children with SLI where his SLI children in that study were only overtly marking 50% of the time and so when you look at her data, her low SES, her children reared in poverty were marking at 85% regular past, where the SLI children were at 50. The low SES group also had very high rates of over regs, which SLI children do not have, and so when she compared her data on children reared in poverty to the literature on SLI, the profiles don't look the same.
New Study: Tense and Agreement by Dialect and Clinical Status
The new study is a five-year project funded by NIH. It's a four-group study, all kindergarteners. Our goal was to have 240 children in the study in five years. We did a number of tests on these kids, we did a non-word wraps, size judgement. If it's in the literature, we tried to do it, but the linguistic component included a 200 utterance language sample. We created dialect appropriate video probes and so we selected verbs. Often times, especially for past tense and third person, they end in vowels or glides, they don't end in a consonant because we don't want that consonant cluster issue of their dialects. We created a dialect appropriate sentence imitation probe, and we gave the test of early grammar impairment, the TEGI, which the literature would tell us and even the test developers of the TEGI have a caveat in the manual that says this is for children who speak general American English, this is not for non-mainstream speakers. But given our interest in these structures we thought well we're going to give it to the children anyway, and I know you would love to see all of that data.
These are our participants. So I can tell you something about the sample. We've only had, we've only really classified our kids in the last couple weeks. So we do our study a little differently than when you work with general American English. General American English, you often times know what group they're going to go in, but because we didn't trust our tests, because we don't have good test data with minority children, we kind of knew. But we were going to wait until the end, and so I can tell you that in these schools during the five years we've been there, 600 kindergarteners were enrolled in the schools. We had a 78% rate of consent return, and that is because people in Louisiana absolutely love LSU. In fact, one of our schools has framed posters of people that went to their school that then went on to LSU. Unfortunately, they don't have graduation hats and robes. What they have on are football uniforms, and those are players that went onto LSU and then probably went on to the NFL. So football in Louisiana, sports in general, people always want to participate in things related to LSU. That might be the only benefit of football, we get high rates of consent. I'm kidding. So there's 600 children enrolled. We had consent forms on 78%, that's like 460 some. We were able to do at least one or two days of testing on 400 of them, but they all don't fit our criteria, and so now we're trying to figure out who should be in our study.
What I'm presenting here is the most conservative groupings. This is 106 of the children. If you look you'll see we have 53 that are SLI. If you were to divide 53 by 600, what do you get? You get like 8% prevalence and if you divide this number by the number of consents we had or 400, you get a prevalence of about 13%. So we think just post hoc reflecting, we don't think we're over identifying language impairment, but these children that we are putting in the SLI category have multiple tests. So their PPVT is low, their DELV norm referenced, which is dialect appropriate, is low, that has a mean of 10 and a standard deviation of three. And their TOLD scores, we don't have TOLD scores on everybody. We picked the TOLD up late. We were hoping to just use the DELV, but we were getting so many sevens on the DELV, which is right at the one standard deviation, that the prevalence is too high if you put them all in the SLI group and so we thought, well let's just add another test and we're going to look for convergence. Then the typically developing all have very nice scores.
Okay, now the plan would be that this is how the data would look, and you would have no dialect effects, no group effects, and I want to just be honest to say that what is now in purple are SLI children have statistically lower maternal ed than our control, and we have a race effect with the PPVT and that in the typically developing children, the SWE children's PPVT is higher. So I think it might be getting to test bias, it might be getting into disparities. If you work in this area, these are issues you have to deal with. I can also tell you the study I'm going to show, even if I were to show you the data for all 252 children that collected it, the group effect is robust. It's always there. Okay, so what we, what I'm not showing by going to the 106, is some of those discrepancies that we have in the typical group between the black and the white children. So we're trying to see how to deal with those either statistically or from group selection.
I'm going to show you the sentence imitation probe. The children get 36 sentences. The pool of sentences are 60, and they strategically get different versions. So they each get 36, 12 of the sentences have tense in the sentence, 12 have tense and negation and 12 have tense negation and a complementizer, and so I have three of our sentences. It's on a laptop, it's presented through the laptop, and we've controlled number of words, the ones with the three functional categories range from seven to nine, so we can look at sentence length.
The first thing we did was score this sentence recall or sentence imitation task like Sean does, where you give a sentence a two if it's recalled exactly the same, a one if there are less than four errors? Two errors? Four errors? Three errors? Okay we did exactly what you told us to. And then a zero if you had four or more. That's where I got the four. Four or more. And we also accepted dialect appropriate is for are and was for were, but no other dialect issues. Look at the percent for the impaired versus the typical. Very robust finding, and we get that even if we include the 252 children. With no interaction with dialect and no dialect mean effect here.
Then we asked are the errors related to tense. So we coded the errors and these were actually the ungrammatical. So if a child recalled the sentence, and recalled it grammatically for general American English, so maybe the question was "Ernie wondered if" and the child said "Ernie was wondering if". That was scored low, but in this analysis was considered grammatical and we didn't include it, but of those that were ungrammatical, then we asked, is it with one of those target grammatical structures, is it with something else, or is it both? And then we did an analysis, if it's with one of those functional grammars, is it tense negation or complementizing.
And so one of the things you see for proportion of error -- so they each equal 100 -- is that the language impaired children have much higher proportions of error, which would be usually an omission of tense, negation or complementizer. The typically developing do not, and they are omitting words that are not those. And then for both, the SLI children have higher both as well, but those both include functional and non-functional. So you get a tense component there, and then when you separate them for all groups and this includes only those sentences that had all three. It wouldn't be fair to put the sentences that only had tense, you would have an inflated score, but of those sentences that had all three for all of the children, if you're going to have an ungrammatical sentence where you omit one of these, it's going to be in tense rather than negation or with the complementizer.
Conclusion and Discussion
So my conclusions are that I think we can say that multiple studies that I've shown you reveal grammar deficits to be a component of language impairment in children who speak non-mainstream dialects of English. The grammar deficit is best characterized as lower rates of overt marking for their age, and compared to same dialect speaking peers, not mainstream American English, but same dialect speaking peers. And then finally, the grammar deficit is most salient when you create materials and scoring systems that are dialect appropriate, when you're really understanding the dialect, understanding rate, context, and function when you go to create your experiments.

Audience Question

What are your thoughts in terms of separating or discussing dialects like AAE which can occur among such different populations in an urban setting, and comparing SLI to typically developing peers that might be influenced in different ways because of the multiple ethnicities who are speaking AAE, but are also influenced by those other things going on.

I think I would still stay with that cross-linguistic approach to ask what are other children, their peers who seem to be doing well. How are they using language? And then comparing that as your base relative to children that you think are weak. I mean, one place to start -- the DELV is not perfect. The test developers know that it's not perfect. But it is definitely a start and if you look at sensitivity and specificity, children who are typical do well on the test. If it errs, it errs on missing a kid. So if you find a child on the DELV who fails the DELV, then you've used a dialect appropriate test, so you've shown that you understand non-mainstream, and the child still coming up with an impairment. That's one of the reasons why we like to use it because we can say there's an impairment in spite of the fact that we're using a test that's dialect appropriate. So that's one thing I would do.
And then I also would not worry about what do we call these dialects? We like to call what our children speak Louisiana dialects. People are proud of where they're from, they're proud of their community, and so if you ask families or communities, or you just observe in your urban area, you may be able to use a different label. So like my student is doing some work in South Carolina in an area where Gullah/Geechee or Gullah is the heritage language. So she's referring to it as I want to study children who have Gullah/Geechee heritage because she doesn't know what their dialect is. She doesn't know if it's Gullah, Gullah influenced AAE, AAE, Gullah influenced standard. So she's kind of dealing with some of the same issues you might be dealing with as multiple dialects. I don't know what to call it, but you may not have to worry about what you call it. Kids don't care what you call it, parents don't really care what you call it. What you want to do, you know, I want to study the language you're using, or let's talk about your use of language.

Audience Question

To go back to one of your opening remarks and your anecdote about people here at the conference who assured you that when children were referred to them with dialects, they sent them back with no help. Are you endorsing that?

I'm not endorsing that at all. I see that often. Bruce Tomblin's epidemiology study found that less than 30% of the kindergarteners who could qualify for services had teacher or parent concern, right? We have less than 30. So our children are completely under-identified, and people don't know how to find an impairment there. What will happen is you start to have a conversation about impairment within a dialect, and everyone is more comfortable talking about dialect versus difference and so the conversation then moves there. It's hard to get people to talk about no, no, no they're all dialect speakers, now how do we figure out who is impaired versus who is typical? And it's just not a paradigm that we have in our profession and we need to have that.
The "intervention question" is what most people want to know. Okay now that you've found who is impaired, what do you do? If you work on BE or past tense, people are going to think you're trying to change the child's dialect.
What I say is go read this book by Vershawn Young, Other People's Children: Code Switching, Code Meshing, and Language Literacy Achievement. I do not know him, I make no money off of him, but it's my favorite book right now. There are about four authors on it and what he says is don't teach code switching. Teach code meshing. Code meshing is what individuals who are bilingual do and this is what a number of really articulate people do, is they bring all of their language to the task. We have nice clips in there about celebrities doing it, Obama doing it, senators doing it, on Twitter we do it. Code mesh. Use the language that is most effective. There's a really nice chapter in there that a teacher has written about fourth graders and eighth graders. The two things that I love in that book is these are the two things she does. If let's say she's doing writing samples or the child is talking and she hears maybe a AAE structure or an SWE structure, she will then say, "Are you code meshing right now? Because if you're code meshing, let's keep it. Just like that. But if you're not code meshing, let's talk about it. Just so I know you know the difference."
The idea is that you're never telling a child not to use some of the language that they have. Very similar to, we like to teach babies sign and oral language. We don't say stop one to the other, same with bilingual. We don't say stop using one language for the other. So it's in that same principle of let that child bring everything to the task. Then the other thing that she does is she turns students into scientists and they do a lot of activities where they go out and they try to find a piece of language that maybe girls say, but boys don't. Or that boys do and girls don't. And then a parent does or an older person does and they don't. And then what might be codes, you know something in African American English versus something at school. So turning them into scientists, not saying any of it is wrong or right, but getting them to be meta about it. So that's where I kind of send people if they have intervention questions. I didn't mean to plug that, but that's usually the first question that people say is I can't work on those structures because people will think I'm trying to change the dialect. So I always say you can work to the target, which is the typical developing AAE speaking child or the typically developing SWE. That typically developing child is using past tense markers, you know 80% to 90% of the time, maybe just 70%, but if you have an SLI child who is only using it 20 or 30, that child should be able to get up to 70.