Amy L. Weiss
Department of Speech Pathology and Audiology, University of Iowa, 120B SH C, Iowa City, IA USA 52242-1012
This study followed up a preliminary investigation (Weiss, 2002) that substantiated the presence of recasts in the child-directed language repertoires of parents of school age CWS and CWNS. Further analyses demonstrated that parents of CWS varied in the types and frequencies of their recast productions. As a group, children’s utterances following their parents’ recasts were not more significantly likely to contain disfluencies and as a group the children’s utterances were less lengthy and complex than the average utterances in their samples. For individual parent-child dyads, however, the use of recasts by their parents yielded attempts at more complex syntactic structures.
Weiss (2002) suggested that one type of child-directed language frequently studied in the repertoires of caregivers, that of recasts, had not been investigated in the conversations of CWS and their primary caregivers. Recasts are considered to be simple or complex and involve the caregiver’s use of an utterance produced by the child as a beginning point or “platform” for providing information about alternate ways to convey the same meaning, for inputting new lexical items, extending or correcting the child’s syntax, or merely imitating the child to acknowledge the acceptability of the utterance. For example, if a child were to say, “That’s a horse”, a caregiver could then say, “Yes, and that Stallion looks like it’S galloping at full speed.” In this case, the caregiver provided the child with another word for horse (stallion) and provided some information about what the horse was doing, perhaps using a term not yet part of the child’s lexicon. This would be an example of a complex recast. In simple recasts, only one clausal element is altered (i.e., “That’s a stallion”), and imitations by caregivers contain all of what the child produced, perhaps with an acknowledgement attached (i.e., “Yes, that’s a horse”). Her results demonstrated that the parents of school age CWS used recasts as frequently as did the parents of children who were not stutterers (CWNS), and that the parents of the CWS did not necessarily focus on stuttered utterances for use as the platform utterances. In fact, the parents were more likely to recast fluent utterances produced by their children.
In the normal and disordered language development literature, research results are equivocal with regard to whether recasts are an essential language-leaming tool whether they occur naturally or are purposely constructed by caregivers. As a result of her preliminary study, Weiss (2002) speculated that recast use with school age CWS might lead to a tendency on the part of the child to produce longer and more complex utterances. Given the converging data available, this would logically lead to a greater likelihood for production of disfluencies. Although recasting could be argued to be a useful strategy for modeling more sophisticated language, for CWS it might serve to make fluency maintenance more difficult.
To determine whether this is the case, the conversations of the same group of children used for the Weiss (2002) study were further analyzed to answer the following two basic questions. First, were the CWS more likely to produce utterances that were longer and more complex than their average sample utterances, following the caregivers’ use of a recast? Second, were the CWS more likely to produce disfluencies in their utterances following a caregiver’s recast? Further, did the complexity level of the recast produced by the caregiver yield differential results?
During the 20 minutes that the children conversed with one of their parents, they participated in 10 minutes of a self-selected, age-appropriate, structured activity (e. g., putting together a puzzle, building a structure from a kit), and 10 minutes of unstructured conversation (i.e., “Talk about anything you would like to talk about”). Two different conversation contexts were utilized with the hope that a representative sample of what typically transpired between child and parent would be collected. Following each sample’s collection on videotape, the transcripts of both child and parent were transcribed using CHAT conventions from the CHILDES program (MacWhinney, 1995). Each utterance produced by the children was coded re: the fluency of the utterance, specifically, “fluent”, “stuttered”, containing one or more example of within-word disfluency (Conture, 2001), or as a “no code”, indicating that the utterance was a routinized phrase. The “no code” utterances were left out of the analyses because of their potential biasing nature. Length and complexity measures were calculated for each sample using Mean Length of Utterance (MLU) (Brown, 1973), and a set of six, later-developing syntactic constructions (Weiss & Johnson, 1993). The parents’ utterances were also transcribed and analyzed for these features. In addition, the parents’ samples were analyzed for the quality and quantity of recasts employed in their conversation turns. A more complete summary of this information can be found in Weiss (2002).
In the present study, the children’s turns immediately following the parents’ turns that contained recasts were scrutinized. These turns were subjected to analyses of both length and complexity and compared with the average length and complexity measures calculated for the entire sample to determine whether these particular turns were significantly longer or more complex than what the children typically produced. Further, the presence of disfluencies in these utterances was tallied and compared with the type of recast most often preceding them.
The two conversation samples were analyzed separately because of their appreciable difference in size. Specifically, in the spontaneous conversation sample, there were a total of 267 turns produced by children following recasts made by parents (M = 20.54, SD = 9.86); in the structured conversation samples the total number of turns following recasts was 104 (M = 8.00, SD = 3.00), for a collapsed total across conversation context types of 371.
In the spontaneous conversation samples, the probability of turns following recasts to be greater than, less than, or equal to the average of utterances in the children’s samples was statistically different than would be expected by chance [X2(2) = 209.09, p < .001]. However, this represented three categories that were obviously quite different in terms of numbers of entries; there were 66 turns produced with MLUs at least 1.00 greater than the over all MLU for the sample, 176 that were produced with MLUS at least 1.00 less than the MLU for the entire sample and only 25 with MLUs within +/- 1.00 of the overall MLU. Thus, the two larger categories, greater than and less than, were compared and their difference was also shown to be greater than what would be expected by chance [X2(1) = 111.24, p < .001]. Although the numbers make it fairly obvious that the turns following recasts were more likely to be less than greater than the overall MLU for the sample, the Wilcoxon Signed Ranks Test (Siegel & Castellan, 1988) was applied to determine the statistical significance of the difference observed. This statistic (T* = 88, p < .01), demonstrated that there was a significantly greater likelihood of the children producing a turn following a recast that was at least 1.00 morpheme shorter than the overall sample’s MLU than one that was greater by the same or a greater margin.
In the structured conversation samples where the total number of turns following recasts was 104, 25 were greater than the average MLU for its respective sample, 64 were shorter, and 15 were within =/- 1.00 of the MLU for its sample. When comparing all three Categories of potential length changes, the probability of the turns following recasts falling into those three categories was significantly different [X7-(2) = 58.26, p < .001]. A comparison of the two larger categories of greater than and less than the average MLU, yielded the same significant results [X2(1) = 32.35, p < .001]. Again, the numbers of turns appear to reveal a trend but to substantiate it, the two larger categories were compared using the Wilcoxon Signed Ranks Test. The statistic (T* = 93.01, p < .01) also substantiated that in the structured conversation contexts, children were significantly more likely to produce utterances following recasts that were shorter than their overall sample MLUs by at least 1.00, than produce longer utterances. Thus, the results allow the conclusion that utterances following recasts were more likely to be shorter than the average utterance when the entire sample was considered.
When the MLUs for all 26 of the samples (across contexts) were averaged, the group demonstrated an MLU of 4.89, with a SD of 2.27. In absolute values, average MLU deviations following complex recasts were 4.17 (with 4.42 for utterances greater than average and 3.92 for those less than average), 3.38 following simple recasts (with 2.24 for utterances greater than average and 4.27 for those less than average), and 3.40 for utterances following imitation (with 3.45 for those greater than average and 3.37 for those less than average).
In terms of complexity, six different constructions, shown to be later-developing in children learning English, were tallied when they appeared in the samples: coordinate conjunctions (n = 16), complements (n = 2), noun + post modification (n = 4), infinitives (n = 8), subordinate adverbial clauses (n = 3), and modals (n = 8). These numbers in parentheses represent the totals across both sample types. Given that there were so few of them, it did not appear that they affected the likelihood for longer or shorter turns following recasts. However, when utterances containing within-word disfluencies that contained one or more of the six constructions and MLUS greater then the average for each child’s respective sample were delineated, only 32 or 9% of the turns following recasts met these criteria. For the spontaneous conversation samples, the average MLU deviation for these utterances was 5.39, with a range of 1.00 to 16.00. In the structured conversation samples, the MLU deviation ranged from 1.00 to 3.00, with an average of 1.50 morphemes.
The second question addressed whether the three recast types the parents produced differentially affected either the likelihood of disfluent utterance productions or the length and complexity of the turn produced by the children. A description of the distribution of the three recast types across the two different samples is provided in Table 1. As can be observed from the data in Table 1, the children were more likely to follow complex recasts with turns than they were when presented with simple recasts or imitations on the part of the parent, regardless of conversation context.
Table 1. Distribution of the Three Recast Types “Follow-Up” Utterances Across Spontaneous and Structured Conversation Samples
The data were also analyzed in terms of whether the children’s utterances that followed the three different recast types were produced fluently, stuttered, or not coded because of the automaticity of the response. These data are presented in Table 2 below.
Table 2. Distribution of the “Follow-Up” Utterances By Recast Type and Fluency Code.
The data in Table 2 yield the following information. Of the 371 utterances produced following a recast, the 13 CWS who participated in this study produced only 54 or 14.5% that were considered stuttered. Interestingly, almost two-thirds of these followed complex recasts produced by their parents (11 = 35, or 64.8%). Although complex recasts constituted the largest recast type produced by the parents (11 = 218), the children’s stuttered follow-up utterances made up only 16% of this group.
These data analyses provided a counterpoint to the investigation published by Weiss (2002). Instead of studying the parents’ child-directed language, this study proposed to examine the outcome of parents’ use of different recast types in terms of their potential co-occurrence with changed length, complexity, and the exacerbation of disfluent behaviors. What was observed was that the total rate of disfluencies contained in utterances following recasts was quite low and even the complex recasts, most often followed by the disfluent utterances, represented less than one-fifth of the stuttered rejoinders.
One striking finding was that parents’ recasts were most often followed by children’s utterances that were shorter and less complex than their typical utterance productions, when the sample’s MLU was considered. One explanation could be that the CWS, or any child, might view a recast as an attempt on the part of the parent for correction. What follows this correction might logically be more carefully prepared and in doing so produced with syntax and semantic features that are already well-learned by the child. Longer utterances may contain syntactic constructions and perhaps lexical items that the child is still in the process of learning. In a shorter, less complex utterance, the child may be taking less of a risk for producing within-word disfluencies. At least on average, this is what appears to be happening and the finding fits with those of other investigators who have demonstrated that there is a relationship between greater utterance length and complexity, with the production of disfluencies in 1anguage-learning children (Gaines et al., 1991; Logan & Conture, 1995; Weiss & Zebrowski, 1992). It is important to heed the eterogeneity of variance that is apparent in this population of CWS. That is, children who are diagnosed as CWS not only demonstrate a wide range of performance profiles but the language environments in which they live may differ greatly. Although the general pattern of shorter and less complex follow-up utterances to a recast should not suggest that every child lacks risk taking in terms of production. It also does not necessarily follow that the child has not derived the benefit of exposure to more sophisticated syntactic structures and new lexical items that recasts can provide.
In the corpus of data analyzed for this study, one of the youngest subjects produced eight follow-up utterances with MLUS greater than his average sample containing at least one disfluency. Of these eight, five also contained one or more of the six later-developing syntactic constructions tallied (e.g., “You know what I can do?”). In this example, the child produced a part-word repetition on the word “can”, as he produced a sentence containing both a complement and a modal. One of the older children in the study, however, produced no examples of stuttered follow-up utterances with MLUs that were greater than the average of his entire conversation sample.
In conclusion, recasts produced by the parents of CWS, whether they are complex, simple or imitative, are consistently followed up by their children. Their presence as a parent’s conversational turn does not appear to exacerbate the child’s propensity to stutter. Quite the-opposite appears to occur. For the majority of children in this study, a parent’s recast was followed by utterances that were fluent, possibly aided by the fact that there was a tendency for these utterances to be shorter and less complex than the child’s average utterance in the sample.
Brown, R. (1973). A first language. Cambridge, MA: Harvard University Press.
Conture, E. (2001). Stuttering: Its nature, diagnosis, and treatment (2”d ed. ). Boston: Allyn & Bacon.
Gaines, N., Runyan, C., & Meyers, S. (1991). A comparison of young stutterers’ ï¬ uent versus stuttered utterances on measures of length and complexity. Journal of Speech and Hearing Research, 34, 37-42.
Johnson, W., Darley, F., & Spriestersbach, D. (1963). Diagnostic methods in speech pathology. New York: Harper & Row.
Logan, K., & Conture, E. (1995). Length, grammatical complexity and rate differences in stuttered and fluent conversational utterances of children who stutter. Journal of Fluency Disorders, 20, 35-61.
MacWhinney, B. (1995). The CHILDES project: Tools for analyzing talk (2"" ed. ). Hillsdale, NJ: Lawrence Erlbaum Associates.
Proctor-Williams, K., Fey, M., & Loeb, D. (2001). Parental recasts and production of copulas and articles by children with specific language impairment. American Journal of Speech-Language Pathology, 10, 155-168.
Snow, C., & Ferguson, C. (Eds.). (1977). Talking to children: Language input and acquisition. Cambridge: Cambridge University Press.
Weiss, A. (2002). Recasts in parents’ language to their school-age children who stutter: a preliminary study. Journal of Fluency Disorders, 27, 243-266.
Weiss, A., & Johnson, C. (1993). The relationship between narrative and syntactic competencies in school-aged, hearing-impaired children. Applied Psycholinguistics, 14, 35-59.
Weiss, A., & Zebrowski, P. (1992). Disfluencies in the conversations of young children who stutter: Some answers about questions. Journal of Speech and Hearing Research, 35, 1230-1238.