Facilitating referential communication among children: The listener as teacher

Facilitating referential communication among children: The listener as teacher

JOURNAL OF EXPERIMENTAL Facilitating CHILD PSYCHOLOGY 29, 357-370 (1980) Referential Communication among Children: The Listener as Teacher CHAR...

865KB Sizes 0 Downloads 42 Views

JOURNAL

OF

EXPERIMENTAL

Facilitating

CHILD

PSYCHOLOGY

29, 357-370 (1980)

Referential Communication among Children: The Listener as Teacher CHARLOTTE J. PATTERSON AND CHRISTOPHER M. MASSAD University

of Virginia

This study was designed to test the hypothesis that systematic variations in listener behavior can have an important influence on both speaker behavior and communicative success among children. In particular, we investigated the idea that systematic variations in listener behavior might have not only within-trial effects on the adequacy of speakers’ messages and the accuracy of communication among children, but also cumulative effects on speakers’ initial messages across trials. Effects of stimulus complexity were also examined. Pairs of 7- and 9-year-old children participated in a referential communication game, with the younger child serving as speaker and the older one as listener. Half of the listeners were given a plan for effective listening which emphasized the importance of asking questions if the speakers’ messages were ambiguous. Replicating earlier findings, the plan manipulation was successful in encouraging listeners to ask questions when necessary. The major result was that listener questions not only had the expected trial-by-trial effect on message adequacy and communicative accuracy, but also showed a cumulative effect on speaker performance. When exposed to systematic listener feedback, speakers improved their initial messages over trials. Stimulus complexity was not a major determinant of performances. These findings suggest that provision of systematic listener feedback may be an effective method for teaching speaker skills to young children.

Early studies of children’s referential communication revealed considerable development of skill in the speaker role over the childhood years, and emphasized the speaker’s responsibility for communication failures among young children (Glucksberg, Krauss, & Higgins, 1975). Recent studies (Cosgrove & Patterson, 1977; Ironsmith & Whitehurst, 1978; Patterson, Massad, & Cosgrove, 1978) have also documented the development of an important listener skill: the ability to respond effectively This research was supported by NSF Grant BNS77-15277 to the first author. We wish to thank Debbie Brlggs, Cathy LaBrasca, Deirdre Lavieri, Carolyn O’Brien, Lam-in Talley, and Betsy Woelfel for their assistance, and Sister Odile McKenna, principal of St. Bridget’s School in Richmond, Virginia, for her cooperation. Requests for reprints should be sent to Charlotte J. Patterson, Department of Psychology, Gilmer Hall, University of Virginia, Charlottesville, VA 22901. 357 0022~0965/80/020357-14$02.00/0 Copyright 0 1980 by Academic Fess, Inc. All rights of rrpmduction in any fom reserved.

358

PATTERSON

AND

MASSAD

to messages which are initially ambiguous. For example, Cosgrove and Patterson asked preschool, kindergarten, second-grade, and fourth-grade children to play the listener’s role in a referential communication game. An adult confederate took the role of the speaker and produced standardized messages which varied in their informational adequacy. Consistent with the earlier findings (cf. Glucksberg et al., 1973, even the preschoolers were quite successful in using the speaker’s informative messages. When the message was ambiguous, however, only the fourth graders were likely to spontaneously request clarification. Since the speaker complied with such requests, fourth graders received more information and made more correct referent choices than younger children. Cosgrove and Patterson (1977) also attempted to facilitate listener performances. Following the pretest described above, half of the children at each age were given a plan for effective listening which emphasized the importance of asking questions if the initial message was unclear. Although the plan had no effect on the behavior of the preschoolers, children in each of the three older age groups who received the plan performed significantly better on a post-test than those in a control condition. A central question raised by these data is whether or not such systematic alterations in the listener’s behavior can be used to facilitate communicative success among young speaker-listener pairs. Of course, in the Cosgrove and Patterson (1977) study, the speaker was an adult confederate who had been instructed to provide more information in response to listener questions. Whether or not systematic changes in the listener’s behavior can affect communicative success among children will depend on the manner in which young speakers respond to listener questions. Although findings on this point are mixed and often difficult to interpret, the literature generally seems to indicate that young speakers are most likely to reformulate ambiguous messages when the listener’s questions are explicit in indicating what information is needed (Alvy, 1968; Flavell, Botkin, Fry, Wright, & Jarvis, 1968; Glucksberg & Krauss, 1967; Karabenick & Miller, 1977; Peterson, Danner, & Flavell, 1972). Since these are exactly the kinds of questions that children in the Cosgrove and Patterson (1977) study were most likely to ask, the suggestion is that systematic variations in listener behavior might have a substantial impact on the success of communication among young speaker-listener pairs. While some of the data (e.g., Karabenick & Miller, 1977) suggest that listener’s questions may influence the success of referential communication among children, only a few attempts to facilitate referential communication among children using systematic manipulation of listener behavior have been made (cf. Asher & Wigfield, Note 1; Whitehurst & Sonnenschein, Note 2). One aim of the current study was to demonstrate such facilitation. Accordingly, we chose third-grade children, who should profit maximally from the training (Cosgrove & Patterson, 1977), as lis-

REFERENTIAL

COMMUNICATION

359

teners. First graders were chosen as speakers, since earlier work (Cosgrove & Patterson, Note 3) has shown that children of this age produce generally inadequate messages in the present setting. There are at least two ways in which listener questions might facilitate referential communication among children. Questions might serve on a trial-by-trial basis to indicate the listener’s informational needs, but have no lasting impact on a speaker’s performance. That is, listener feedback might prompt the speaker to provide additional information, but only in response to individual questions. We will call this the “trial-by-trial effect” hypothesis. On the other hand, questions might have a cumulative effect on the speaker’s performance, teaching the speaker what kinds of information should be included in his or her messages on subsequent trials. We will call this the “cumulative effect” hypothesis. To investigate these two hypotheses, we broke speakers’ messages down into “initial” messages (everything the speaker said before the listener asked a question) and “final” messages (everything the speaker said before the listener made a choice among referents, including any answers to question). If the trial-by-trial hypothesis is correct, final messages should show improvement with listener feedback, but initial messages should remain unchanged. If, on the other hand, the cumulative effect hypothesis is correct, then initial messages should show improvement over trials as a result of listener feedback. We also examined the effects of stimulus complexity on children’s performances as communicators. In much of the referential communication literature, the possible effects of stimulus factors have remained unclear (Patterson & Kister, Note 4). Some studies have employed relatively simple, nameable stimulus items (e.g., Peterson et al., 1972), others have used hard-to-label nonsense shapes (Glucksberg & Krauss, 1967), and others have used quite complex, heterogeneous stimulus arrays (e.g., Karabenick & Miller, 1977). Although effects on speakers have been examined (e.g., Ford & Olson, 1975; Whitehurst & Sonnenschein, 1978), there has been little systematic study of the effects of stimulus complexity on the performances of young listeners. In the present study, we examined the effects of three levels of stimulus complexity on the communicative performance of both speaker and listener. METHODS Design

The study employed a 3 x 2 between-subjects design, which varied stimulus complexity (low, medium, and high) and the nature of instructions to the listener (plan vs no plan). Dependent measures assessed the impact of these manipulations on the behavior of both the speaker and the listener during a referential communication task.

360

PATTERSON

AND

MASSAD

Subjects and Experimenters The subjects were 41 first and 41 third graders from a predominantly middle-class parochial school. There were 17 boys and 24 girls at each grade level. The children were tested in same-sex pairs, with a first grader serving as speaker and a third grader as listener. The average age of the speakers was 7 years 0 months; for the listeners, it was 9 years 1 month. Three adult females alternated in the role of Experimenter l(El), and three alternated in the role of Experimenter 2(E2). Each of the experimenters tested approximately equal proportions of subjects in each condition. Materials There were three different sets of stimulus materials for the communication game, one for each level of the complexity factor. Each set consisted of 18 pairs of identical cards; 16 pairs were used for the test trials and 2 were used for the practice trials (see below). In the low-complexity condition, each card contained 4 potential referents. In the mediumcomplexity condition, each card contained 8 potential referents, and in the high-complexity condition each contained 16. The potential referents always varied systematically in terms of binary attributes (e.g., striped-plain, tall-short) so that information about two (low complexity condition), three (medium complexity), or four (high complexity) crucial attributes would allow definitive identification of the target referent. For example, one item in the low-complexity condition contained a picture of a blue house with a chimney, a red house with a chimney, a blue house without a chimney, and a red house without a chimney. In addition, in the medium-complexity condition, half of the houses had a front door, and half did not. In the high-complexity condition, an additional factor was whether or not the house had windows. One additional pair of cards, termed the “introductory item,” contained four pictures of common animals (a bird, an elephant, a dog, and a rabbit) and was used during the introduction to the game for all of the children. Reward certificates (“Good Player Awards”) were identical to those described elsewhere (Cosgrove & Patterson, 1977). Procedure Each session began as El seated the listener at one end of a small table. A divider was placed across the middle of the table so that persons seated at either end could see only each other’s faces. El explained the referential communication game, telling the listener that he or she would be trying to pick out the target referent on each trial and that the object was to choose correctly as often as possible. After the child appeared to understand, El administered a practice trial using the introductory item.

REFERENTIAL

COMMUNICATION

361

El produced a fully informative message, and all subjects chose the correct picture. Each child was then administered a standard interview, designed to assess his or her understanding of the rules of the communication game. All of the children indicated that they understood the rules; in particular, all agreed that it would be permissible to ask questions during the game. At this point, half of the listeners (plan condition) were given a plan for effective listening, much as described by Cosgrove and Patterson (1977). The plan emphasized that “whenever you are not sure which answer is the right one, then you can ask questions to help yourself figure it out.” Each child was then given two practice trials. El served as speaker, and on each trial she produced a message that was lacking in information about two of the target referent’s crucial attributes. In the low-complexity condition, these messages simply specified one or more characteristics common to all four potential referents (e.g., “it’s a house”). Children in the no-plan (control) condition did not receive instructions about the plan, but did participate in these two practice trials. E2 then entered the room with the child who was to play the speaker’s role. E2, who was blind to the experimental condition to which the listener had been randomly assigned, conducted the remainder of the procedure; while El collected the data. E2 introduced the speaker and listener, and explained the communication game to the speaker. She informed the children that both speaker and listener would be awarded one point for each correct choice by the listener, and that if they won enough points, each would receive a Good Player Award like one she displayed. All children indicated their desire to do so. After the speaker indicated his or her understanding of the rules, E2 initiated the test trials. She placed the first pair of stimulus items in front of both speaker and listener, and showed the speaker which picture was the target referent. Target referents were initially chosen at random for each pair of stimulus items, but were the same for all subjects. After the speaker produced a message and the listener made a choice among the referents, E2 indicated whether or not the listener’s choice was correct. Each pair of children participated in a total of 16 trials. After the last trial, each child was given a Good Player Award, thanked for participating, and returned to the classroom. Dependent Measures The main dependent measures were the number of questions asked by each listener, the number of correct referent choices made by each listener, the number of crucial attributes of the target referent mentioned in the speaker’s initial messages, and the number of crucial attributes mentioned in the speaker’s final messages. The initial message was defined as everything the speaker said before the listener made a verbal request for

362

PATTERSON

AND

MASSAD

more information. The final message was defined as everything the speaker said before the listener marked his or her choice among referents. If the listener asked no questions, the initial and final messages were, of course, identical. We also coded the contents of listener questions, using four categories. A question was coded as “specific” if it isolated a relevant dimension or attribute and included a relevant value within the question (e.g., “is it green or yellow?“). A “general” question was a request for further information which specified the dimension, but not the values (e.g., “what color is it?“). Questions were coded as “repeats” if they asked for information already given in the speaker’s message. Assignment to this category took precedence over the first two, so that if a question asked for specific information which had already been given by the speaker, it was coded as a “repeat” question. Any questions which did not fit in the above categories were coded as “other.” All questions were coded by two independent raters using transcripts of each experimental session. The coders were in exact agreement on the classification of %% of all questions; the codings of the first rater were used for all analyses. Preliminary analyses revealed no effects for sex of subject, and so this factor was dropped from the main analyses. RESULTS

A multivariate analysis of variance on the raw scores for each of the four main dependent measures revealed a significant main effect for plan instructions, F(4, 32) = 16.22, p < .OOl, a main effect for stimulus complexity, F(8, 64) = 10.47, p < .OOl, and a plan x complexity interaction, F(9, 64) = 2.39, p < .05. The means are presented in Table 1. Effects of the Plan on the Listener Univariate analysis showed that, as expected, listeners in the plan condition asked more questions than those in the no-plan condition, F( 1, 35) = 27.46, p < .OOl. This result replicates our earlier findings (Cosgrove & Patterson, 1977; Patterson et al., 1978) by demonstrating the impact of the plan on listener behavior. It is of interest in the present context mainly as a manipulation check-i.e., insofar as it indicates that the plan did indeed have the intended effect on listener behavior. Examination of the types of questions asked by listeners indicated that, as one might expect from the findings reported above, most questions were both useful and task related. Of the 406 questions asked, 85% were specific requests for more information, 5% were general requests for more information, 8% were requests to have previously stated information repeated, and 2% were coded as falling into the “other” category. A multivariate analysis of variance demonstrated that the significant effect of plan instructions on the number of questions asked by listeners was due

Low (n =6) Moderate (n=8) High (n=S)

Low (n=5) Moderate (?I=7) High (n=7)

Plan

No plan

0

0.4

5.4

1.4

2.4

16.3

0

0

17.6

0

0

G

4.5

s

TABLE

1

Listener

0.3

0

0

1.4

2.1

0.3

R

0.1

0

0

0.6

0

0

0

2.3

5.4

0

20.6

19.8

4.8

Total

NUMBER AND

.50

.45

.49

.64

58

.77

Proportion

33.5

27.1

15.9

56.3

44.5

29.3

Raw score

.52

.57

.49

.88

.95

.91

Proportion

questions. All values are means.

31.9

21.7

15.8

41.3

27.8

24.7

Raw score

Number of crucial attributes in final messages

Speaker

OF CRUCIAL ATTRIBUTES STIMULUS COMPLEXITY

Number of crucial attributes in initial messages

0 = “other”

9.7

9.7

9.6

11.8

14.6

14.7

Number of correct choices

OF CORRECT REFERENT CHOICES, AND AS A FUNCTION OF INSTRUCTIONAL CONDITION

Number of questions asked

ASKED,NUMBER MESSAGES

Note. S = Specific questions, G = general questions, R = requests to repeat information,

Stimulus complexity

Instructional condition

OF QUESTIONS IN INITIAL AND FINAL

NUMBER

364

PATTERSON

AND MASSAD

TABLE 2 NUMBEROFTRIALS ON WHICH LISTENERS DID ORDID NOTASKFORADDITIONAL INFORMATION AS A FUNCTION OF STIMULUS COMPLEXITY AND NUMBER OF CRUCIAL BITS OF INFORMATION OMI~ED IN THE SPEAKER'S INITIAL MESSAGE (PLAN CONDITION ONLY)

Stimulus complexity Low Moderate High

Listener response Question’ No question Question No question Question No question

Number of crucial bits of information omitted in speaker’s initial message I

2

3

4

22 11 29 9 21 28

3 2 40 1 33 0

14 0 17 0

3 0

u Question = Listener made one or more requests for more information. No question = Listener did not request more information.

to an increased number of specific requests, F(l, 3.5) = 18.93, p < .OOl, and an increase in requests for repetitions, F( 1, 35) = 7.10, p < .05. Effects of Message Ambiguity

on the Behavior

of Listeners

We also investigated the extent to which the degree of ambiguity in the speaker’s initial description of the referent affected the likelihood that the listener would ask questions. Specifically, we examined the probability that the listener requested additional information on each trial as a function of the number of bits of crucial information omitted in the speaker’s initial message for that trial.’ As can be seen in Table 2, listeners were more likely to ask questions when the message was more ambiguous. A log-linear regression analysis on these data showed that a model consisting of only a linear component provided a good fit with the observed frequencies, x2 = .lS, p = .93, indicating that the probability of listener questions increased with the degree of ambiguity in speaker’s initial messages. Effects of the Plan on the Speaker Univariate tests showed that speakers in the plan condition named more of the crucial attributes of target referents in both their initial, F( 1, 35) = 6.43,~ < .05, and final messages, F(1, 35) = 38.85,~ < .OOl. The latter finding reflects the responsiveness of speakers to listener questions. That is, listeners made 368 requests for additional information, and speakers provided the relevant information 92% of the time. There can be r Since there were so few listener questions in the no-plan condition, these analyses were conducted using only the data from the plan condition.

REFERENTIAL

COMMUNICATION

365

little doubt that listener questions resulting from introduction of the plan had a positive, trial-by-trial effect on the adequacy of speaker’s messages. More important in the present context, however, is the finding that providing listeners with the plan had a beneficial impact on speakers’ initial messages. The fact that speakers in the plan condition mentioned more crucial attributes in their initial (as well as final) messages suggested that listener questions might have had more than simply a trial-by-trial effect. To examine the possibility of a cumulative effect, we divided the 16 trials into 4 consecutive trial blocks, and examined the number of crucial attributes mentioned in the speakers’ initial messages during the 4 blocks. Means are presented in Table 3. A 2 (plan or no plan) x 3 (high, moderate, or low stimulus complexity) analysis of variance with 4 (trial blocks) repeated measures yielded significant main effects for plan, F(1, 35) = 7.01, p < .Ol, and trial blocks, F(3, 105) = 7.24, p < .OOl. The plan x trial blocks interaction was also significant, F(3, 105) = 3.41, p < .05, indicating that the effect of the plan varied across trial blocks. A multivariate trend analysis revealed the presence of a significant linear trend, F( 1,39) = 37.83,~ <: .OOl , indicating that the plan x trial blocks interaction was due to an increasing effect of the plan manipulation over trial blocks. A multivariate analysis of variance on the speakers’ initial messages over trial blocks also yielded the expected main effect for plan, F(4, 32) = 4.08, p < .Ol. The subsequent univariate analyses, however, showed that the effect was not present in the first,F(l, 35) = 1.81, n.s., or second trial blocks,F(l, 35) = 1.84, n.s., but appeared in the third trial block, F(1, 35) = 10.00, p < .005, and was maintained through the fourth, F(l, 35) = 9.63, p < .005. These results provide considerable support for the cumulative effect hypothesis, indicating that, over trials, listener feedback served to teach speakers to produce more adequate initial messages. Effects

of Stimulus

Complexity

Univariate tests showed that stimulus complexity had a significant effect on the number of questions asked, F(2,35) = 6.43,~ < .005, but not on the number of correct choices by listeners. Complexity also had a significant effect on the number of crucial attributes mentioned by speakers in both their initial, F(2,35) = 9.71,~ < .OOl, and final messages, F(2, 35) = 18.64, p < .OOl. None of the univariate tests for the stimulus complexity x instructional condition interaction approached significance. It seemed likely that some of the main effects for stimulus complexity might be artifact&. There were 32 crucial attributes to be identified in the low-complexity condition, 48 in the moderate-complexity condition, and 64 in the high-complexity condition. As a result, speakers in the highcomplexity condition were faced with more attributes to describe than those in the moderate and low conditions. From the listener’s point of

Low Moderate High

Low Moderate High

Plan

No Plan

Note. All values are means.

Stimulus complexity

Instructional condition

5.3 6.4 9.4 4.2 5.3 7.9

Raw score

.53 .44 .49

.67 .53 .59

Proportion

First trial block

4.4 5.6 7.4

6.2 5.9 9.4

Raw score

4.2 4.4 7.7

6.7 7.8 11.1

.77 .49 .59 .55 .46 .46

Raw score

4.7 8.9 12.1 4.0 6.4 9.6

.53 .37 .48

Raw score

.50 .54 .60

.83 .74 .75

Proportion

Fourth trial block

.83 .65 .70

Proportion

Third trial block

Proportion

Second trial block

Number of crucial attributes named by speakers

TABLE 3 NUMBEROF CRUCIAL ATTRIBUTES NAMED BY SPEAKERS IN INITIAL MESSAGES ON EACH OF THE FOUR TRIAL BLOCKS AS A FUNCTION OF INSTRUCTIONAL CONDITION AND STIMULUS COMPLEXITY

F z %

z9

s

s 3 E

REFERENTIAL

COMMUNICATION

367

view, the same level of certainty regarding the identity of the target referent would require information about different numbers of crucial attributes in different conditions. Thus, raw scores might provide biased estimates of children’s abilities across stimulus complexity conditions. To control for these factors, corrected scores were calculated for all of the main dependent measures except the number of correct referent choices.2 Corrected scores were computed by dividing the raw scores by the product of the number of crucial attributes per trial (2, 3, or 4, depending on complexity condition) and the number of trials. Whereas the analyses reported above are based on the absolute amounts of information conveyed, use of the corrected scores allows examination of the amounts of information conveyed relative to the total amount needed to identify a target referent. In other words, these corrected scores allow us to ask: What proportion of the total amount of necessary information was provided by the speaker or requested by the listener? A multivariate analysis of variance on the corrected scores showed that the main effect for plan remained essentially unchanged, F(4, 32) = 14.98, p < .OOl. This analysis also yielded a significant main effect for complexity, F(8, 64) = 2.15, p < .05, but no instructional condition x stimulus complexity interaction, F < 1. Univariate analyses revealed that stimulus complexity had a significant effect only on the number of questions asked, F(2, 35) = 4.15, p < .05; listeners in the high-complexity condition asked more questions than those in the low-complexity condition, but fewer than those in the moderate-complexity condition. Stimulus complexity had no significant effect on the quality of speakers’ initial or final messages or on the number of correct referent choices made by listeners. These analyses demonstrate that most of the significant effects of stimulus complexity on raw scores were artifactual. When proportional measures were examined, a significant effect of stimulus complexity appeared for only one of the four main dependent measures. Overall, we conclude that stimulus complexity was not a major determinant of the success of children’s communicative performances in this situation. DISCUSSION

The most important result of the present study was that systematic alterations of the behavior of young listeners had a cumulative impact on the adequacy of young speakers’ messages over trials. Listeners who received the plan were able to transform naive first graders into more effective speakers over the course of only 16 trials. The success of these children in teaching young speakers to produce more informative initial * It was not necessary to correct the scores for number of correct referent choices since there was, of course, only one correct choice per trial, regardless of the level of complexity.

368

PATTERSON

AND

MASSAD

messages stands in marked contrast to the results of most efforts to train children to be better speakers (cf. Glucksberg et al., 1973, and suggests systematic manipulation of listener behavior as a particularly promising method for teaching effective speaker skills, From a practical standpoint, the most significant application of the present findings may be to the growing practice of peer tutoring. While peer teaching has often been shown to have beneficial effects on tutors, evidence of a positive impact on tutees’ mastery of curricular material is sparse (Feldman, Devin-Sheehan, & Allen, 1976). Considering the findings on children as speakers (Glucksberg et al., 1975), it seems reasonable to assume that tutors may be relatively ineffective in communicating the relevant materials. If so, it might be possible to improve the efficacy of peer-tutoring procedures by training tutees in effective listening strategies, using methods analogous to those employed here. Further empirical work will be needed to evaluate this intriguing possibility. In accord with earlier findings (Glucksberg et al., 1973, our data show that young speakers’ initial messages were often inadequate to allow identification of the target referent. In contrast to data reported by some investigators (e.g., Glucksberg & Krauss, 1967; Karabenick & Miller, 1977), however, our results show that when listeners requested more information, speakers almost always provided it. This difference in findings may be attributable to the fact that listeners in the present study most often asked questions that were both clear and specific. This, in turn, might be due to the fact that the crucial attributes of referent items were both highly visible and quite salient, or to the fact that the stimuli encountered by any one child were relatively homogeneous across trials. Of course, one cannot be certain exactly why young speakers have sometimes appeared unwilling to reformulate unsatisfactory initial messages. However, the current findings demonstrate that, at least under some conditions, young speakers do offer additional information in response to listener feedback, and that this clarification leads to more effective communication. A somewhat surprising result of this study was that stimulus complexity had no consistent effects on the proportion of necessary information provided by speakers or requested by listeners. Of course, we examined performance across only a relatively limited range of complexity; the most complex items required the speaker to transmit information about four attributes of the target referent. The success of children in the plan condition clearly indicated their ability to handle this level of complexity. Research employing a greater range of stimulus complexity would be more likely to demonstrate significant effects on performance. In summary, the present findings indicate the significance of listener behavior in influencing the success of referential communication among children. Systematic listener questions not only had immediate, trial-by-

REFERENTIAL

COMMUNICATION

369

trial effects on speaker performance and communicative accuracy, but also had a cumulative effect on the quality of speakers’ initial messages. Thus, the results suggest that provision of systematic listener feedback can be an effective method for teaching speaker skills and improving communicative accuracy among children. REFERENCES Alvy, K. T. Relation of age to children’s egocentric and cooperative communication. Journal of Genetic Psychology, 1968, 112, 275-286. Cosgrove, J. M. & Patterson, C. J. Plans and the development of listener skills, Developmental Psychology, 1977, 13, 557-564. Feldman, R. S., Devin-Sheehan, L., & Allen, V. L. Children tutoring children: A critical review of the research. In V. L. Allen (Ed.), Children as teachers: Theory and research on tutoring. New York: Academic Press, 1976. Flavell, J. H., Botkin, P. T., Fry, C. L., Wright, J. C., & Jarvis, P. E. The development of role-taking and communication skills in children. New York: Wiley, 1968. Ford, W. & Olson, D. The elaboration of the noun phrase in children’s description of objects. Journal of Experimental Child Psychology, 1975, 19, 371-382. Glucksberg, S. & Krauss, R. M. What do people say after they have learned how to talk? Studies of the development of referential communication. Merrill-Palmer Quarterly, 1967, 13, 309-316. Glucksberg, S., Krauss, R. M., & Higgins, E. T. The development of referential communication skills. In F. D. Horowitz (Ed.), Review of child development research (Vol. 4). Chicago: Univ. of Chicago Press, 1975. Ironsmith, M. & Whitehurst, G. The development of listener abilities in communication: How children deal with ambiguous information. Child Development, 1978,49,348-352. Karabenick, J. D. & Miller, S. A. The effects of age, sex, and listener feedback on grade school children’s referential communication. Child Development, 1977, 48, 678-683. Krauss, R. M. & Glucksberg, S. The development of communication: Competence as a function of age. Child Development, 1969, 40, 255-266. Patterson, C. J., Massad, C. M., & Cosgrove, J. M. Children’s referential communication: Components of plans for effective listening. Developmental Psychology, 1978, 14, 401-406. Peterson, D. L., Danner, F. W. & Flavell, J. H. Developmental changes in children’s response to three indications of communicative failure. Child Development, 1972, 43, 1463-1468. Whitehurst, G. J. & Sonnenschein, S. The development of communication: Attribute variation leads to contrast failure. Journal of Experimental Child Psychology, 1978, 25, 490-504.

REFERENCE NOTES 1. Asher, S. R. & Wigfield, A. Training referential communication skills. Paper given at the National Institute of Education Conference on Children’s Oral Communication Skills, Madison, Wisconsin, 1978. 2. Whitehurst, G. J. & Sonnenschein, S. The development of informative messages in referential communication: Knowing when versus knowing how. Paper given at the National Institute of Education Conference on Children’s Oral Communication Skills, Madison, Wisconsin, 1978.

370

PATTERSON

AND MASSAD

3. Cosgrove, J. M., & Patterson, C. J. Adequacy of young speakers’ encoding in response to listener feedback. Unpublished manuscript, University of Virginia, 1978. 4. Patterson, C. J. & Kister, M. C., The development of listener skills for referential communication. Paper given at the National Institute of Education Conference on Children’s Oral Communication Skills, Madison, Wisconsin, 1978. RECEIVED: March 8, 1979; REVISED: May 17, 1979.