In Search of a Grammatical Marker of Language Impairment in Children This discussion of language assessment is based on the findings from a current program of research carried out in my lab in collaboration with Ken Wexler of the Massachusetts Institute of Technology and a group of talented student research assistants and staff personnel, supported by funding from the National ... Article
Free
Article  |   May 01, 1998
In Search of a Grammatical Marker of Language Impairment in Children
Author Notes
Article Information
Models of Assessment
Article   |   May 01, 1998
In Search of a Grammatical Marker of Language Impairment in Children
SIG 1 Perspectives on Language Learning and Education, May 1998, Vol. 5, 3-7. doi:10.1044/lle5.1.3
SIG 1 Perspectives on Language Learning and Education, May 1998, Vol. 5, 3-7. doi:10.1044/lle5.1.3
This discussion of language assessment is based on the findings from a current program of research carried out in my lab in collaboration with Ken Wexler of the Massachusetts Institute of Technology and a group of talented student research assistants and staff personnel, supported by funding from the National Institute of Deafness and Communicative Disorders. In this area of investigation, we are exploring the grammatical abilities of young children with Specific Language Impairment (SLI) and the ways in which certain grammatical competencies differentiate youngsters with SLI from unaffected age peers and younger control children. One outcome of this line of investigation is the identification of a possible clinical marker for young children with language impairment.
Clinical Markers
What is a clinical marker and how is it different from conventional standardized test assessment?
Fundamental to the difference is the way we think of variation across children in their language abilities. Standardized language tests assume, for a given age, an underlying normative distribution of children on a general language dimension. This is based on the well-known bell-shaped curve (see Figure 1 on page 4). In this distribution, individuals are scattered along a range of performance levels such that a few people score very high values (along the right-hand side of the scale), a few people score very low values (along the left-hand side), and most (about 66%) score in the middle. Furthermore, the bellshaped curve has well known distributional properties, such that 16% of the people will score one standard deviation or more below the mean and 2% will score two or more standard deviations below the mean. For example, if we were to examine the size of children’s vocabularies, say all 6-year-olds, we could expect that a few children would have a large number of words in their vocabularies, a few would have very few words, and most children would have a moderate number of words in their vocabularies.
There are important limitations to the assumption of normative distribution. A big one is that there is no intrinsic criterion for where to draw the line between “normal” and “affected,” shown on the left in Figure 1. Instead, a certain arbitrariness is inevitable. How can we know who does not have enough words? Who is “impaired” in vocabulary development? Without an obvious way to separate “normal” and “affected,” we may falsely identify as affected those children who are not affected, or we may falsely identify children as unaffected who really have an underlying language deficit. Second, there is no obvious way to interpret the actual test score in terms of particular linguistic content. Because the tests are constructed according to a general description of language (e.g., “expressive” vs. “receptive” modes, or “vocabulary” vs. “grammatical morphemes”), usually there are no clear indications of how to translate a child’s score into particular linguistic competencies that may or may not be affected. In other words, one cannot go directly from a test score to a set of known or unknown linguistic competencies. One cannot know what to teach by knowing a child’s standardized score. Also, it is not possible to interpret a child’s performance relative to the expected adult model of language. How can you tell if a child with a standard score of 85 is mostly in line with an adult grammar or mostly far away? Often we encounter clinical use of “language age” scores, which are the mean scores for a particular age group. So, when a 5-year-old child’s score is the same as the average for children of 3 years of age, the child is said to have a “language age” of 3 years. But what does that mean? Do we know if the 5-year-old missed the same items as the 3-year-olds? If the 5-year-old had scored several items better, would that make his performance that of a 4-year-old? The point is that, although the notion of normative distribution is a vitally important one for identification of young children with language impairments, it does have serious interpretive limitations.
Figure 1

Bell Curve.

Bell Curve.
Figure 1

Bell Curve.

×
In Search of a Grammatical Marker
Distributional properties. Now let us consider the way grammar works. Can we have a normative distribution of individuals with varying understanding of the fundamental rules of grammar? For example, consider the rule for plurals, such that regular nouns that refer to more than one must be marked with -s to denote plurality. Under this rule, a phrase like 1. *these ball is ungrammatical (here I will follow the convention that utterances marked with an * are ungrammatical). It is not something about which speakers show only partial knowledge or can choose to ignore; it is obligatory. In like fashion, speakers must insert a form of “be” in a sentence where a copula is required (e.g., 2. She is walking vs. 3. *She walking). Likewise, the third person singular -s morpheme is required in contexts such as 4. She walks outside; 5. *She walk outside is ungrammatical in Standard English. Other examples are: 6. Yesterday she walked outside; 7. *Yesterday she walk outside; 8. Does she like to walk? 9. *She like to walk? Grammatical properties such as these are understood by people who know English. In order to work as a conventional language, the community of users know these properties and follow them. Grammar users are not distributed as in a Bell curve; instead they are all bunched at the top end of the distribution, because they know these grammatical principles.
Now let us consider children. By the time they go to kindergarten, children can be expected to know the properties of grammar illustrated above (i.e., rules for plurals, the use of “be” forms, regular third person singular -s, regular past tense -ed, auxiliary “do”). But children do not show this knowledge at the outset of their language in their first simple sentences. We know children differ from adults. Over time they come to be like adults in their use of these grammatical properties. In order to identify children with language impairments, we must find those children who are not following the expected timeline for learning the adult grammar.
Clinical Characteristics of a Grammatical Marker
If we now turn our attention to the ways in which young children’s grammar approximates adult grammar, we no longer need to think in terms of the Bell curve but can instead think of a clinical marker in the domain of grammar that would have the following properties:
  1. By a certain age, grammatical markers would show little variation across unaffected children. That is, children would show the adult grammar or a close approximation of the adult grammar. They would know that certain grammatical morphemes are obligatory. They would cluster at the upper ends of the distribution.

  2. Affected children would perform below the unaffected children. They would cluster at the lower ends of the distribution.

  3. Because of this bimodal distribution of children, grammatical markers would have high levels of sensitivity and specificity. Sensitivity is the rate of identifying true cases of affectedness. Specificity is the rate of identifying true cases of unaffectedness. Just as we would want to have cancer testing with methods of high specificity and selectivity, we would prefer clinical methods with high accuracy for identifying true cases of language impairment that do not falsely identify unaffected children as having a language impairment.

  4. The content of assessment would be meaningful for interpretation of a child’s language deficits. It would be possible to see which grammatical knowledge was affected, which could in turn be used to plan intervention, to know what language competencies to teach.

  5. The child’s performance would be interpretable in terms of the adult grammar. It would be possible to see which gaps persisted as a child moved toward full grammatical competence.

  6. Grammatical markers would persist over time. Long-standing grammatical differences would increase the likelihood that children would be identified, because they could be detected at different age levels.

Figure 2

Distribution of individual children’s performance on a composite TNS marking score: SLI and age controls.

Distribution of individual children’s
performance on a composite TNS marking score:
SLI and age controls.
Figure 2

Distribution of individual children’s performance on a composite TNS marking score: SLI and age controls.

×
Extended Optional Infinitives
Our work has targeted this question: Can we identify children with SLI because they are very slow in acquiring a part of the grammar that unaffected children are also slow in acquiring? Ken Wexler’s earlier work identified tense-marking as a part of the grammar that is relatively slow to emerge in unaffected young children. This period is known as an Optional Infinitive stage. This stage is interesting because new models of the adult grammar, proposed by Noam Chomsky (1995), focus on how this part of the grammatical system is crucial for the formulation of grammatical sentences, across many languages. So there is a direct link between models of the adult grammar and children’s grammar, which shows us which grammatical morphemes are likely to be affected and which ones are likely to be unaffected. One consequence is that we do not expect a pervasive grammatical limitation, but instead a constrained deficit. The verbal morphemes of interest here are those illustrated above, in examples 2–9 (i.e., past tense, 3rd person singular -s, “be,” and “do”). For this set of morphemes, our prediction is that, as shown above, the children can be expected to sometimes use them and sometimes omit them. We expect younger, unaffected children to do this, but to move out of this immature grammar faster than the children with SLI, who seem to get stuck with an immature grammar for an extended time. Therefore, we call it an Extended Optional Infinitive (EOI) stage (Rice, Haney, & Wexler, in press; Rice, Wexler, & Cleave, 1995).
Findings
An important initial finding is that children with SLI have much lower levels of performance on tensemarking morphemes than do control children. Figure 2 shows the performance of two groups of 5-year-old children (in the year before kindergarten), one group of 37 children identified as expressive/receptive SLI, and the other, a control group of 45 children of the same age (cf. Rice & Wexler, 1996). The measure is the percentage correct for the obligatory use of the tensemarking morphemes (indexed here by a composite measure, summed across the different morphemes). Clearly, the SLI group falls below the 60% accuracy level, whereas the control children are at 80% or better. Only one child in each of the two groups moved across the border area. If the cut-off is set for 80%, 97% of the true cases are identified (sensitivity) and 98% of the true non-cases are identified (specificity). These are very encouraging findings, suggesting that it may be possible to identify young children with language impairments on the basis of certain grammatical morphemes.
We just reported the first longitudinal study of young children with SLI and control children, for the age range of 3–8 years, in this area of the grammar. Longitudinal evidence shows that the children with SLI remain behind their control groups from preschool through second grade. Figure 3 shows this pattern (cf. Rice, Wexler, & Hershberger, in press). In this figure, the line on the lower left—beginning at 3 years and going to 8 years—shows the growth in tensemarking for young, unaffected children (N = 20) at the same mean length of utterance as the SLI group (N = 21) at the beginning of the study. We can see that they reach adult-like levels somewhere between 4 and 4½ years. The consistency of performance levels across age levels is shown in the top line on the right, which is the unaffected age peers (N = 23) of the SLI group. These children maintain adult-like performance once they get to that level. On the other hand, the SLI group, the lower right curve, is lower than the control children throughout the time of the study, and they continue to be lower at age 8 years. Although the unaffected children have been at adult levels for years, the children with SLI continue to fall behind even in elementary school. This shows that they do not “outgrow” their problem in this domain. Formal analyses show that the pattern of growth is similar for the SLI group and the younger, unaffected children, but the SLI group does not “catch up” to their peers. Detailed analyses show that growth in this linguistic domain is not predicted by a child’s mother’s education (although this is known from other studies to predict vocabulary growth), a child’s initial receptive vocabulary, or performance on nonverbal intelligence tests. In other words, a child’s grammatical growth did not depend on initial vocabulary scores or a nonverbal intelligence score, or the education levels of his or her mother. A weak predictor (1% of the variance) was a child’s initial mean length of utterance. Mostly, however, grammatical growth was not related to these predictors.
Figure 3

Composite tense.

Composite tense.
Figure 3

Composite tense.

×
It is important to know that not all grammatical morphemes show this pattern. For example, the acquisition of regular plural -s (as in “cats”) is relatively unaffected, even though it is phonetically very similar to the third person -s verbal morpheme. Figure 4 (see page 7) shows the same age period for performance on plurals, where it is clear that this morpheme is quite stable during this time, and very near adult-like levels of performance, even for the SLI group. Thus, we can see that the problem is not a general one of being unable to acquire grammatical rules, nor is it likely that the children cannot perceive the difference in morphemes. We also know, based on grammaticality judgment evidence, that the children’s limitations are not restricted to production alone, but are also evident in comprehension of the affected morphemes.
Figure 4

Regular -s plurals.

Regular -s plurals.
Figure 4

Regular -s plurals.

×
Implications for Assessment
What these findings point toward are possible advances in clinical tools. If subsequent studies confirm that certain grammatical competencies can differentiate affected from unaffected children, we could then develop new tests to quickly identify children at high risk for language impairment, including children who currently are undetected. Such tests, for example, would be especially useful for the preschool period, as a way of identifying young children in need of early language intervention, even if no other developmental impairments are evident, and before the child experiences academic frustrations. Although this would be a welcome advance, it would supplement, not replace, traditional evaluations. One major reason that conventional assessment methods will continue to be essential for full evaluation is because clinical markers are likely to be detectable in some, but not all, of the dimensions of children’s language performance.
Of great scientific import, the identification of grammatical markers enhances our understanding of the nature of language impairment and possible causal factors. If some children maintain a particular form of an immature grammar for an extended time, then we know that certain parts of the grammar can be weak. If this weakness is evident across the morphemes thought to share an underlying function, we can see that a particular grammatical function, not an individual morpheme, is implicated. If the affected morphemes do not share similar surface features, then we can see that it is probably not primarily accounted for by problems of surface structure production or processing. Our investigations lead us to these conclusions.
The big question is: What causes these grammatical impairments? Is it the same agent that is responsible for the faster-maturing grammars of unaffected children? Our hypothesis is that there is an inherited contribution to grammatical acquisition, and children who show a grammatical marker may well be children who have inherited a faulty mechanism for grammatical growth. For the sample of children in the SLI group shown in Figures 2 and 3, we know the likelihood is greater that someone in the affected children’s family had a speech and language impairment (22%) than in the control families (7%). The question of possible inherited speech and language impairments is under investigation in several scientific labs around the world. The means of identification of affected individuals is an absolutely crucial factor in this search. The contribution of the work on a grammatical marker, summarized briefly here, is to point toward the special properties of grammar, how this changes our views of the way individuals can be expected to vary, how grammatical measures clearly point toward affected children, and what this might mean. It is likely that during the clinical careers of today’s practitioners, we will arrive at new means of assessment, new interpretations of why some children fail to acquire grammar in the expected timeline, and how to identify children as early as possible.
References
Chomsky, N. (1995). The minimalist program. Cambridge, MA: MIT Press.
Chomsky, N. (1995). The minimalist program. Cambridge, MA: MIT Press.×
Rice, M. L. Haney, K. R. & Wexler, K. (in press). Family histories of children with SLI who show extended optional infinitives. Journal of Speech, Language, and Hearing Research..
Rice, M. L. Haney, K. R. & Wexler, K. (in press). Family histories of children with SLI who show extended optional infinitives. Journal of Speech, Language, and Hearing Research.. ×
Rice, M. L. & Wexler, K. (1996). Toward tense as a clinical marker of specific language impairment in English-speaking children. Journal of Speech and Hearing Research, 39, 1239–1257. [Article] [PubMed]
Rice, M. L. & Wexler, K. (1996). Toward tense as a clinical marker of specific language impairment in English-speaking children. Journal of Speech and Hearing Research, 39, 1239–1257. [Article] [PubMed]×
Rice, M. L. Wexler, K. & Cleave, P. L. (1995). Specific language impairment as a period of extended optional infinitive. Journal of Speech and Hearing Research, 38, 850–863. [Article] [PubMed]
Rice, M. L. Wexler, K. & Cleave, P. L. (1995). Specific language impairment as a period of extended optional infinitive. Journal of Speech and Hearing Research, 38, 850–863. [Article] [PubMed]×
Rice, M. L. Wexler, K. & Hershberger, S. (in press). Tense over time: The longitudinal course of tense acquisition in children with specific language impairment. Journal of Speech, Language, and Hearing Research..
Rice, M. L. Wexler, K. & Hershberger, S. (in press). Tense over time: The longitudinal course of tense acquisition in children with specific language impairment. Journal of Speech, Language, and Hearing Research.. ×
Figure 1

Bell Curve.

Bell Curve.
Figure 1

Bell Curve.

×
Figure 2

Distribution of individual children’s performance on a composite TNS marking score: SLI and age controls.

Distribution of individual children’s
performance on a composite TNS marking score:
SLI and age controls.
Figure 2

Distribution of individual children’s performance on a composite TNS marking score: SLI and age controls.

×
Figure 3

Composite tense.

Composite tense.
Figure 3

Composite tense.

×
Figure 4

Regular -s plurals.

Regular -s plurals.
Figure 4

Regular -s plurals.

×
We've Changed Our Publication Model...
The 19 individual SIG Perspectives publications have been relaunched as the new, all-in-one Perspectives of the ASHA Special Interest Groups.