14
Joint Attention and Conversation in Mother-Infant-Sibling Triads Author(s): Michelle E. Barton and Michael Tomasello Source: Child Development, Vol. 62, No. 3 (Jun., 1991), pp. 517-529 Published by: Blackwell Publishing on behalf of the Society for Research in Child Development Stable URL: http://www.jstor.org/stable/1131127 Accessed: 24/04/2009 15:42 Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/action/showPublisher?publisherCode=black. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We work with the scholarly community to preserve their work and the materials they rely upon, and to build a common research platform that promotes the discovery and use of these resources. For more information about JSTOR, please contact [email protected]. Blackwell Publishing and Society for Research in Child Development are collaborating with JSTOR to digitize, preserve and extend access to Child Development. http://www.jstor.org

Tomasello Joint Attention and Conversation

Embed Size (px)

Citation preview

Page 1: Tomasello Joint Attention and Conversation

Joint Attention and Conversation in Mother-Infant-Sibling TriadsAuthor(s): Michelle E. Barton and Michael TomaselloSource: Child Development, Vol. 62, No. 3 (Jun., 1991), pp. 517-529Published by: Blackwell Publishing on behalf of the Society for Research in ChildDevelopmentStable URL: http://www.jstor.org/stable/1131127Accessed: 24/04/2009 15:42

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available athttp://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unlessyou have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and youmay use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained athttp://www.jstor.org/action/showPublisher?publisherCode=black.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printedpage of such transmission.

JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We work with thescholarly community to preserve their work and the materials they rely upon, and to build a common research platform thatpromotes the discovery and use of these resources. For more information about JSTOR, please contact [email protected].

Blackwell Publishing and Society for Research in Child Development are collaborating with JSTOR to digitize,preserve and extend access to Child Development.

http://www.jstor.org

Page 2: Tomasello Joint Attention and Conversation

Joint Attention and Conversation in Mother-Infant-Sibling Triads

Michelle E. Barton and Michael Tomasello

Emory University

BARTON, MICHELLE E., and TOMASELLO, MICHAEL. Joint Attention and Conversation in Mother- Infant-Sibling Triads. CHILD DEVELOPMENT, 1991, 62, 517-529. The current study investigated the general nature of joint attentional and conversational interaction in mother-infant-sibling triads. 9 19-month-old infants and 9 24-month-old infants were videotaped during 20 min of free play with their mothers and preschool-aged siblings around a common activity. Analyses revealed that even 19-month-old infants were capable of participating in triadic interactions and conversa- tions, and that the proportional frequency of both these measures increased with age. Triadic conversations were nearly 3 times longer and elicited nearly twice as many infant turns per conversation as dyadic conversations. Infants were more likely to join into an ongoing conversa- tional topic than to initiate one themselves, and they were more likely to take a turn in those conversations if they were in a joint attentional state with the speaker. Infants were just as likely to respond to a comment or request directed to another person as they were to one directed to themselves, indicating reliable comprehension of language not addressed to them. These results suggest that the mother-infant-sibling interactive context differs in important ways from the mother-infant dyadic context and that it is a richer language learning environment than pre- viously supposed.

Most research examining the effects of social interaction on language development has focused on Western, middle-class mother-infant dyads. Converging lines of ev- idence from this literature have led to the belief that frequent dyadic interactions with a responsive, nondirective adult are condu- cive to early language learning (e.g., Della Corte, Benedict, & Klein, 1983; Nelson, 1973; Olson, Bayles, & Bates, 1986; Rice, 1989; Tomasello & Farrar, 1986). Dyadic in- teraction with adults is clearly not necessary for language acquisition, however, as many children in non-Western cultures acquire language with few, if any, such interactions. Many of these children are immersed from birth almost exclusively in multispeaker contexts, especially those involving multiple children (Ochs, 1982; Schieffelin, 1979; Schieffelin & Ochs, 1983). This pattern is typical of many later-born children in West- ern culture as well, and it is becoming more prevalent for firstborn children, who spend more and more of their time with peers in day-care or nursery school settings. The multichild context is clearly an important one in the language acquisition of most of the world's children.

The few existing studies of multichild contexts have focused almost exclusively on how the presence of another child affects adult-child dyadic interactions. Several studies have reported that the presence of another child reduces both the overall quan- tity and the overall quality of adult-child linguistic and nonlinguistic interactions. Adults in multichild contexts directly ad- dress each child with fewer utterances (Jones & Adamson, 1987; Woollett, 1986), and they become more directive in their in- teractive styles (Schaffer & Liddell, 1984; Tomasello, Mannle, & Kruger, 1986)-both well-known negative correlates of early lan- guage growth (e.g., Tomasello & Farrar, 1986; Wells, 1981). These "negative" influ- ences on the adult in multichild contexts have thus been posited as a possible expla- nation for the slower rate of vocabulary growth found in twins (e.g., Tomasello et al., 1986) and, to a lesser extent, in later-born children (Jones & Adamson, 1987; Nelson, 1973; Nelson, Baker, Denninger, Bonvillian, & Kaplan, 1985). The clear implication is that the multichild context is a less than opti- mum language learning environment be- cause the language learning children have

The authors would like to thank the mothers and children who made this study possible. Thanks also to Ann Kruger and Amy Ledeberg for helpful comments on a previous version of the manuscript. Portions of this research were presented at the Conference on Human Development, Richmond, VA, March 1990. Requests for reprints may be addressed to either author at the Department of Psychology, Emory University, Atlanta, GA 30322.

[Child Development, 1991, 62, 517-529. ? 1991 by the Society for Research in Child Development, Inc. All rights reserved. 0009-3920/91/6203-0010$01.00]

Page 3: Tomasello Joint Attention and Conversation

518 Child Development

to "share access" to caregivers with their co-twins, siblings, or peers.

Recently, however, it has been sug- gested that there may be positive effects of participating in multichild (especially sib- ling) interactive contexts. The posited bene- fits in all cases involve pragmatic skills rather than the more strictly linguistic skills, such as vocabulary size, focused on by previ- ous investigators. For example, Woollett (1986) argued that while homes with older siblings may lessen adult sensitivity to indi- vidual children, they also provide each child with a more stimulating linguistic environ- ment with a variety of salient language mod- els and communicative styles (see also Schaffer, 1989). Similarly, Mannle and To- masello (1987) argued that communicating with siblings forces young children to adapt their linguistic skills for speakers who are less able (and perhaps less willing) than pri- mary caregivers to take their perspective or assent to their wishes. Also, the multi- speaker context may allow the language learning child to "overhear" language among other people, which may provide lin- guistic models helpful in such areas as the acquisition of deictic terms (e.g., see Oshima-Takane, 1988) and third-party refer- ence (see Forrester, 1988).

One very obvious benefit of multi- speaker contexts for pragmatic skills is the opportunity they provide for participation in multispeaker conversations. The home set- ting with siblings is an obvious first step in this direction, but, surprisingly, almost no research has been conducted on parent con- versations with multiple siblings. The only study to focus on such conversations was by Dunn and Shatz (1989), who investigated toddlers' abilities to join into ongoing mother-sibling conversations. They re- corded the behavior of 2-3-year-olds in a home context with both mother and a preschool-age sibling present, focusing on the younger siblings' utterances that were "intrusions" into the conversations of their mothers and older siblings. They found that the younger siblings were quite capable of understanding conversations not directly in- volving them, as evidenced by their increas- ing ability to join in. Moreover, the propor- tion of toddler statements contributing new information to the conversation was actually higher when they were intruding than when they were responding to speech addressed directly to them. This presumably indicates that the joining in process encouraged the children to use their most sophisticated lin-

guistic skills in order to be accepted into the ongoing conversation.

The current study attempted to look more deeply into the dynamics of mother- infant-sibling triadic interactions. We fo- cused on triads containing either a 19- or 24-month-old infant, along with their preschool-age sibling. Our first aim was to document the nonlinguistic joint attentional interactions that took place in these triads. Joint attention has been shown to provide important nonlinguistic scaffolding for the linguistic interactions of 1-2-year-old chil- dren in dyadic contexts (see Tomasello, 1988, for a review), and so it was expected that it might be important in triadic contexts as well. Our second aim was to establish some basic quantitative parameters of the conversations that took place in mother- infant-sibling triadic contexts, that is, basic quantitative information on the relative fre- quency of mother-infant, mother-sibling, infant-sibling, and mother-infant-sibling conversations, and the average length of each of these. We hypothesized that conver- sations involving all three participants would occur at both ages but would be more frequent and longer for the 24-month-olds. Our final aim was to determine more spe- cifically the ways in which infants under 2 years of age, whose linguistic and conversa- tional skills are considerably more modest than those of the children studied by Dunn and Shatz, participate in conversations in the triadic context. We sought to determine how often they initiated, joined into, and contin- ued conversations in this context, as well as some of the linguistic and nonlinguistic fac- tors that might facilitate their participation.

Method

Subjects Subjects were recruited from a psychol-

ogy department subjects file composed of parents who responded voluntarily to a let- ter (mailed to parents of newborns) soliciting cooperation for studies in child develop- ment. Mothers were contacted by phone and asked about their infants' language progress; only infants who were regularly producing language were invited to participate. Two groups ofmothet-infant-sibling triads served as subjects. Thk first group of subjects in- cluded nine infants (four males and five fe- males) between the ages of 19 and 20 months (M = 19.2 months, SD = .4 months; mean MLU 1.28 words), their preschool- aged siblings between the ages of 3 and 5 years (M = 4.4 years, SD = .5 years), and

Page 4: Tomasello Joint Attention and Conversation

Barton and Tomasello 519

their mothers. The second group of subjects included nine infants (five males and four females) between the ages of 23 and 25 months (M = 24.2 months, SD = .4 months; mean MLU 1.70 words), their preschool- aged siblings between the ages of 3 and 5 years (M = 4.3 years, SD = .5 years), and their mothers. (Two additional triads, one in each group, were dropped from the study when they produced almost no language in the observation session.)

Observational Procedure Each mother-infant-sibling triad was

videotaped for 20 min of free play in the psy- chology department's child observation playroom. The play situation was centered on tactile exploration activities using a bin (86 x 71 x 10 cm) of uncooked rice and age-appropriate sand toys such as cups, scoops, trucks, and sieves. A cameraperson and the experimenter were also present in the room but remained as unobtrusive as possible.

Mothers were instructed to play freely with their children as they typically would at home, except that on four occasions (at 5-min intervals) the experimenter gave the mother or the sibling a new toy to introduce to one of the two others. This procedure was meant to simulate four types of naturally oc- curring dyadic exchanges: mother-infant, sibling-infant, sibling-mother, and mother- sibling. This ensured that, for each infant, both the mother and the sibling directly ini- tiated interaction with the infant at least once, and that there were at least two oppor- tunities for the infant to overhear conversa- tions between the mother and sibling.

Coding Procedure Each videotape was coded in a number

of ways, falling into three general categories: joint attention, all conversations, and infant- involved conversations.

Joint attention.-First, the videotapes were coded for joint attentional episodes (JAEs) using a procedure based on that of Tomasello and Todd (1983). JAEs were de- fined as social interactions between two or three participants who shared the same at- tentional focus (usually an object) at the same time for at least 3 sec (brief diversions to a nonsocial activity within a shared focus were ignored). To qualify as a social interac- tion, at least one participant had to acknowl- edge during the joint focus the participation of the other(s), either by establishing eye contact or by making an appropriate verbal or nonverbal response.

The first author and a trained research assistant coded the videotapes continuously for JAE type: mother-infant, sibling-infant, mother-sibling, and triadic. (Sibling-infant interactions, however, never occurred and hence were not analyzed, nor were portions of the tapes where no interaction occurred.) For purposes of reliability analyses, 16 5-min segments were randomly chosen from the videotapes, with the restriction that no two segments came from the same triad; these were coded independently by the two coders. Agreement was calculated on a second-by-second basis, where agreement meant that both coders identified the same type of JAE for a given second (Bakeman & Gottman, 1986). Percent agreement was .85 and Cohen's kappa was .77, both well within acceptable ranges.

For each JAE type, three quantitative measures were computed for each triad: (1) proportional frequency, computed as the number of a given type as a proportion of total JAEs; (2) proportional time, computed as the time spent in a given type as a propor- tion of total time in JAEs; and (3) average time, computed as the number of seconds in a given type divided by the number of JAEs of that type.

All conversations.-Two pairs of re- search assistants made written transcriptions of all language occurring during each ses- sion, along with notes on nonverbal ex- changes such as nods, gestures, and vocal- izations. These were checked and corrected where necessary by the first author. All con- versational measures were coded from these transcripts by the first author.

The second coding was aimed at as- sessing the overall conversational behavior of the triads. Conversations were defined as a series of exchanges between two to three speakers (at least one turn each) that shared the same topic. Topic was defined as what was being talked about-this could be as specific as "the funnel" or as general as "the eating game"-and coders wrote down a brief verbal description of the topic of each conversation. Conversation type was determined by the participants and thus in- cluded the same four possibilities as JAE type: mother-infant, sibling-infant, mother- sibling, and triadic. (Sibling-infant conversa- tions were so rare, occurring less than 2% of the time, that they were not analyzed, nor was language that was not a part of any con- versation.)

Page 5: Tomasello Joint Attention and Conversation

520 Child Development

For purposes of reliability, 16 5-min segments were randomly chosen from the transcripts, with the restriction that no two segments came from the same triad. These segments were coded independently by a trained research assistant. Agreement with the original coder was calculated on an utterance-by-utterance basis, where both coders had to agree on both type and topic for a given utterance for that observation to be considered as an agreement. Judgments of topic were considered in agreement if both coders mentioned the same object or same general activity (e.g., "making rice piles" was considered the same as "making rice castles"). Percent agreement was .86. (This coding procedure was not amenable to Cohen's kappa, because the topic code did not consist of a finite set of categories that could be specified prior to coding.)

For each conversation type, three quan- titative measures were computed for each triad: (1) proportional frequency, computed as the number of conversations of a given type as a proportion of all conversations; (2) average length, computed as the average number of turns per conversation for a given type (a turn was defined as one speaker's utterance[s] bounded by another speaker's turn or at least 5 sec of silence; a turn could be any vocalization, unless otherwise noted); and (3) average number of partici- pant turns, computed as the average number of each person's turns per conversation for a given type.

Infant-involved conversations.-The third and final set of measures was aimed at assessing the conditions under which infants participated in dyadic and triadic conversa- tions with their mothers and siblings. We be- gan by identifying infant utterances and de- termining how they were used in terms of topic maintenance. There were four types of infant turn. (1) If the infant produced an ut- terance that was the first turn on a topic, and that turn was preceded by more than a 5-sec silence, the turn was coded as Initiate Topic. (2) If the infant produced an utterance that was the first turn on a topic, and that turn was not preceded by more than a 5-sec si- lence (i.e., the others were talking about something else when the infant took her turn), the turn was coded as Change Topic. (3) If the infant made a contribution to a topic already established by the mother or sibling, her first turn was coded as Join Topic. (4) All subsequent infant turns on an established topic were coded as Continue Topic. This variable was called turn type.

For each infant turn in the first three of these turn type categories (i.e., all infant first turns), it was determined whether or not that turn was successful at continuing or starting a conversation, that is, whether or not the subsequent speaker took a turn on the in- fant's topic. Thus the ratio of success in each of these categories was the number of infant turns of a given turn type that were followed by an on-topic response by mother or sib- ling, divided by the total number of infant turns of that same type.

Both the turn type and success classifi- cations required purely mechanical tallying from the transcripts that had already been coded for general conversational topic, and thus further measures of reliability beyond the original coding into topics were consid- ered unnecessary.

Singled out for further analysis were those infant-involved dyadic and triadic con- versations where the infant joined into an ongoing conversational topic. Of particular interest were differences between the utter- ances immediately preceding an infant turn (either the Join Topic turn or the subsequent Continue Topic turns) and other utterances that were not followed by an infant turn. With this in mind, each mother and sibling utterance within these conversations was coded as to: (1) addressee (infant-directed and other-directed, based on who was being addressed); (2) speech act (question, com- ment, or request); and (3) joint attention (whether or not the infant was in a joint at- tentional state with the speaker). Combining these codes yielded a total of 12 different utterance types to which the infant might or might not respond: infant-directed ques- tions, comments, or requests inside JAEs; infant-directed questions, comments, or re- quests outside JAEs; other-directed ques- tions, comments, or requests inside JAEs; other-directed questions, comments, or re- quests outside JAEs. For each infant, the probability of responding to each utterance type was calculated as the number of times the infant verbally responded on topic to a particular utterance type divided by the total number of that type (i.e., number of re- sponses as a function of number of opportu- nities to respond).

Reliability for these three coding schemes was determined as follows. Ad- dressee was coded by the transcribers as they transcribed from videotape. Four ran- domly chosen videotapes (20%) were then independently transcribed and coded for ad-

Page 6: Tomasello Joint Attention and Conversation

Barton and Tomasello 521

dressee by the first author. Agreement was calculated on an utterance-by-utterance ba- sis, where both coders had to agree on both who was speaking and who was being ad- dressed. Percent agreement was .96. Speech act was coded by the first author. A trained research assistant independently coded a randomly chosen 20% of the target conversa- tions. Percent agreement was .96. The joint attention determination required only a matching of the time at which an utterance occurred with the times in the JAE coding, and thus reliability analyses were deemed unnecessary.

Results

Results are presented in three sections: joint attention, all conversations, and infant- involved conversations. In all cases in which the dependent variable was a proportion, analyses were done both on the original pro- portions and on the values resulting from an arcsine transformation. Statistical values for the transformed data are reported in the text, whereas the tables and figure present the original, untransformed means and standard deviations. In all cases the analyses on the transformed and untransformed data yielded the same results.

Joint Attention During the 20 min of taped interaction,

the triads engaged in an average of 36 JAEs when infants were 19 months of age and an average of 35 JAEs when infants were 24 months of age. The proportional frequencies of the JAE types were different at the two

ages, however, as may be seen in Table 1. Because proportions for each subject summed to unity across types, an overall 2 x 3 ANOVA could not be performed on these data. Since our primary interest was in developmental changes, our analytic strat- egy was to perform a t test on the JAE type showing the least developmental change (mother-infant), and to perform a 2 (age) x 2 (JAE type) mixed ANOVA on the other two types (mother-sibling and mother-sibling- infant). The t test found no significant change in the mean proportion of mother- infant JAEs as a function of child age (ap- proximately 34% at both ages).

The 2 (age) x 2 (JAE type) mixed ANOVA performed on the remaining data found a main effect of JAE type. The mean proportion of mother-sibling JAEs (.37) was greater than that of mother-infant-sibling JAEs (.29), F(1,16) = 5.48, p < .05. The sig- nificant interaction, F(1,16) = 10.40, p < .01, however, suggests that the main effect of JAE type is due mainly to differences at the younger age only. Tukey's (a) test for uncon- founded means (Linton & Gallo, 1975) con- firmed this finding: only at 19 months was the mean proportion of mother-sibling JAEs greater than that of mother-infant-sibling JAEs (p < .05). The Tukey test also revealed a significant increase in the mean proportion of mother-infant-sibling JAEs with age (p < .05); the mother-sibling decrease was not a significant one. (The pattern of results was similar when the average time of JAEs was analyzed; however, these comparisons did

TABLE 1

MEAN PROPORTIONS (and Standard Deviations) OF JOINT ATTENTION AND CONVERSATION TYPES AS A FUNCTION

OF INFANT AGE

INFANT AGE

INTERACTION TYPES 19 Months 24 Months

Joint attention: Mother-infant .................... .35 (.10) .33 (.13) Mother-sibling.................... .42 (.08)a .32 (.11) Mother-infant-sibling........... .24 (.08) .35 (.09)b

Conversation: Mother-infant .................... .23 (.12) .28 (.07) Mother-sibling.................... .62 (.12)a .43 (.15)e Mother-infant-sibling........... .15 (.09) .27 (.16)b

NOTE.-Joint attention proportions are based on a range of 20 to 51 JAEs per triad. Conversation proportions are based on a range of 16 to 38 conversations per triad.

"a Mother-sibling greater than mother-infant-sibling, p < .05. b 24 months greater than 19 months, p < .05. c 19 months greater than 24 months, p < .05.

Page 7: Tomasello Joint Attention and Conversation

522 Child Development

not reach significance. This indicates that triads spent roughly equal average lengths of time in JAEs of all types at both ages.)

We were also interested in how triadic JAEs started. In particular, we were inter- ested in whether the majority of triadic JAEs began by the infant joining into an ongoing mother-sibling dyadic interaction, by the sibling joining into an ongoing mother- infant dyadic interaction, or by all three members participating equally from the be- ginning. To assess these possibilities, the proportional distribution of mother-infant, mother-sibling, and "null" JAEs (i.e., peri- ods of no interaction) immediately preced- ing each triadic JAE was calculated. Because 23% of the triadic JAEs began at the onset of a 5-min recording session when a new toy was introduced, the preceding joint atten- tional state was not comparable in these cases (one participant had left the room to get the new toy), and hence these triadic JAEs were excluded from analysis. The 2 (age) x 3 (JAE type) mixed ANOVA per- formed on the remaining data found a main effect of the JAE type immediately preced- ing each triadic JAE, F(2,32) = 4.03, p < .05. A Tukey post hoc analysis found that the mean proportion of null episodes preceding the triadic JAEs (.36) was significantly greater than that of mother-infant JAEs (.16), p < .05. The mean proportion of mother- sibling JAEs preceding the triadic JAEs (.24) was not significantly different from those of null or mother-infant episodes. It is thus clear that the infant was an active participant in the initiation of the vast majority of triadic JAEs.

All Conversations Proportional frequency.--During the

20 min of taped interaction, the triads en- gaged in an average of 27.3 conversations of one type or another when infants were 19 months of age and an average of 26.7 conver- sations when infants were 24 months of age. The proportional frequencies of the differ- ent conversation types, however, differ as a function of child age, as can be seen in Table 1. Once again, because proportions across type for each subject summed to unity, an overall 2 x 3 ANOVA could not be per- formed on these data. Thus our analytic strategy was to perform a t test on the con- versation type showing the least develop- mental change (mother-infant) and to per- form a 2 (age) x 2 (conversation type) mixed ANOVA on the other two types (mother- sibling and mother-infant-sibling). The t test found no change in the mean proportion of

mother-infant conversations as a function of child age (approximately 25% at both ages).

The 2 (age) x 2 (conversation type) mixed ANOVA found a main effect of con- versation type such that the mean proportion of mother-sibling conversations (.53) was greater than that of mother-infant-sibling conversations (.21), F(1,16) = 28.75, p < .01. The significant age x type interaction, F(1,16) = 6.67, p < .05, however, suggests that the main effect of conversation type is due to differences at the younger age only. Tukey's (a) test for unconfounded means confirmed this finding (p < .05): only at 19 months was the mean proportion of mother- sibling conversations greater than that of mother-infant-sibling conversations. In ad- dition, the decrease in mother-sibling con- versations with age and the increase in mother-infant-sibling conversations with age both were significant changes (p < .05).

Average length.-Table 2 presents the average length of each conversation type (in turns). The data on which these values are based were subjected to a 2 (age) x 3 (type) mixed ANOVA. While there was no main ef- fect of age, there was a significant main ef- fect of conversation type, F(2,32) = 28.53, p < .01. Means for each type were: mother- infant, 3.79; mother-sibling, 5.07; and mother-infant-sibling, 13.72. A Tukey post hoc analysis found that the average length of mother-infant-sibling conversations was sig- nificantly greater than both mother-infant and mother-sibling conversations (p < .01), while the latter two types did not differ sig- nificantly. The interaction was not sig- nificant.

Turns.-Also presented in Table 2 is the average number of turns taken by each participant in each conversation type. In the case of dyadic conversations, the turns of each participant should be highly correlated with the total length of conversations of that same type; this is because turns were de- fined as alternating, and thus no one partici- pant could take more than 50% of the turns unless they both began and ended conversa- tions. In the case of triadic conversations, however, there are no constraints on the number of turns any one participant could take (outside of the stipulation that each in- dividual take at least one turn). In order to evaluate individuals' contributions to triadic conversations, we therefore analyzed the av- erage number of each participant's turns per conversation type using age x conversation type mixed ANOVAs, one for each partici-

Page 8: Tomasello Joint Attention and Conversation

Barton and Tomasello 523

TABLE 2

MEAN NUMBER OF PARTICIPANT TURNS (and Standard Deviations) AS A FUNCTION OF CONVERSATION TYPE

CONVERSATION TYPES

Mother-Infant- PARTICIPANT TURNS Mother-Infant Mother-Sibling Sibling

Mother ....................... ............................. 2.11 (.79) 2.52 (.79) 6.70 (3.38)a Infant ........................... .................... 1.85 (.74) ... ... 3.01 (2.00)b Sibling .................................. ........... .. .. ... 2.54 (.83) 4.00 (2.06)b

Total average length (in turns)........... 3.79 (1.13) 5.07 (1.56) 13.71 (6.66)a

NOTE.-The number of conversations on which the means are based range from 2 to 12 for mother-infant, 7 to 29 for mother-sibling, and 1 to 17 for mother-infant-sibling.

"a Mother-infant-sibling greater than mother-infant and mother-sibling, p < .01. b Mother-infant-sibling greater than mother-infant or mother-sibling, p < .05.

pant. (These were 2 x 2 ANOVAs for infants and siblings since they only participated in two conversation types; because mothers participated in all three conversation types, their data were analyzed in a 2 x 3 ANOVA.) There was no main effect of age on the aver- age number of participant turns in any of the analyses. In all three analyses, however, there was a main effect of conversation type, such that each participant took more turns in triadic than in dyadic conversations: infant, F(1,16) = 4.95, p < .05; sibling, F(1,16) = 8.29, p < .05; mother, F(2,32) = 23.79, p < .01. (Tukey post hoc analysis revealed that the mothers took more turns in triadic con- versations than in mother-infant or mother- sibling conversations, the latter two not be- ing significantly different.) None of the interactions was significant.

Further analysis of triadic conver- sations.-The findings that triadic conversa- tions were nearly three times as long as dyadic conversations, with each participant taking many more turns, were so striking that two further possibilities were consid- ered. First, it is possible that the mother- infant-sibling conversations were not truly triadic because one participant (especially the infant) may not have participated equally with the others. The first thing to note in this regard is that the infants took nearly as many turns per triadic conversation (3.01) as did the siblings (4.00) (see Table 2). In addition, however, we examined the triadic conversa- tions in more detail in order to analyze each participant's relative contribution. Three subtypes of triadic conversation emerged. The first type was characterized by full re- ciprocation in which each participant both addressed and was addressed by each of the other participants; these comprised 28% of the triadic conversations. The second type

was characterized by partial reciprocation in which each participant either addressed or was addressed by each other participant; these comprised 50% of the triadic conversa- tions. In roughly half of these partially recip- rocating conversations, the infant both ad- dressed and was addressed by each of the other participants. The last type was charac- terized by no reciprocation between two of the participants (i.e., they never addressed each other); these comprised the remaining 23% of the triadic conversations. In almost all cases the "missing link" in these conver- sations was between the infant and sibling. Thus, in terms of the infant's participation overall, in 78% of the triadic conversations (all of the fully and partially reciprocating ones) the infant either addressed or was ad- dressed by each of the other participants; in 53% of the triadic conversations (all of the fully reciprocating and half of the partially reciprocating ones) the infant both ad- dressed and was addressed by each of the other participants.

The second possibility is that the in- fants' dyadic conversations with their moth- ers were shorter than triadic conversations, with fewer infant turns, because they were interrupted by siblings attempting to begin a new topic. There is some a priori validity to this hypothesis, as on average 48% of the mother-infant conversations ended because they were interrupted by siblings (of the re- maining mother-infant conversations, 24% ended naturally with a pause of silence and 30% ended when mother or infant decided to change topic). However, comparison of in- terrupted and uninterrupted mother-infant conversations showed them to be of nearly identical length, 3.5 turns for interrupted (SD = 2.7 turns) and 3.4 turns for uninter- rupted (SD = 2.2 turns). Thus, when the av-

Page 9: Tomasello Joint Attention and Conversation

524 Child Development

erage length and number of infant turns of the uninterrupted mother-infant conversa- tions are compared to triadic conversations, the pattern of results did not change: (i) the average length of the triadic conversations (13.7 turns) remained greater than that of the mother-infant dyadic conversations across both ages (3.4 turns), F(1,16) = 34.6, p < .001; and (ii) the average number of infant turns in the triadic conversations (3.0 turns) remained greater than that in the mother- infant dyadic conversations at both ages (1.5 turns), F(1,16) = 7.9, p < .02. Thus, the shorter mother-infant conversations, with fewer infant turns, were not due to the sib- lings directly depressing the lengths of the dyadic exchanges by interrupting.

Infant-involved Conversations The following three analyses concern

only those dyadic and triadic conversations in which the infant participated (approxi- mately 38% of all conversations at 19 months and 55% of all conversations at 24 months). These were of special concern because of our interest in the kinds of turns infants took and the factors influencing infant turns.

Turn type.-Infants took an average of 21.9 turns in infant-involved conversations at 19 months of age; they took an average of 37.6 turns at 24 months of age. This analysis defined turns as the use of English words by infants because judgments of topicality were necessary. A 2 (age) x 3 (turn type) mixed ANOVA found no main effect of age on the mean proportion of infant turns in infant- involved conversations. There was a sig- nificant main effect of turn type, however, F(2,32) = 13.71, p < .01. Mean values for turn types are presented in Table 3. A Tukey test found that the mean proportion of Join Topic was significantly greater than that of both Change Topic and Initiate Topic, and that the latter two were also significantly dif-

ferent (p < .05). The interaction was not sig- nificant.

Success.-An overall 2 (age) x 3 (turn type) mixed ANOVA was not performed on the success data because of low frequencies of Initiate Topic: four of the younger infants and two of the older infants did not initiate any new topics. The mean success rates for those infants who did attempt to initiate con- versations were .77 for the five younger in- fants and .69 for the seven older infants. Ini- tiations were thus omitted from further analysis, and a 2 (age) x 2 (turn type) mixed ANOVA was performed on the success ratios of the other two turn types (Join Topic, Change Topic). This analysis found no main effect of age. There was a significant main effect of turn type: the success rate for join- ing a conversation with an already estab- lished topic was greater than that for chang- ing the topic, F(1,16) = 8.32, p < .05. Means are presented in Table 3. The interaction was not significant. (Some subjects' propor- tions were based on small denominators be- cause they displayed only a few instances of a particular type. When these subjects were eliminated from analysis, the results re- mained the same.)

Joining conversations.-Focusing on the dyadic and triadic conversations in which the infant joined into a conversational topic, a final analysis was conducted to de- termine differences in the characteristics of mother and sibling utterances followed by an infant turn and mother and sibling utter- ances not followed by an infant turn. Using the probability of infant response as the de- pendent measure, the full design for this analysis would be a 2 (age) x 2 (joint atten- tion) x 2 (addressee) x 3 (speech act) mixed ANOVA, with repeated measures on the last three factors. Because of small frequencies in some of the cells, however, the overall

TABLE 3

MEAN PROPORTIONS AND SUCCESS RATES (and Standard Deviations) OF INFANT FIRST TURNS AS A FUNCTION OF TURN TYPE

TURN TYPE

INFANT FIRST TURNS Join Topic Change Topic Initiate Topic Mean proportiona ........... .32 (.16) .19 (.09) .11 (.11) % Successfulb ................ 92 (.15) .70 (.36)

"a Mean proportion of Join Topic, Change Topic, and Initiate Topic all significantly different from one another, p < .05. (Proportions are based on a range of 5 to 72 infant turns.)

b Success rate of Join Topic greater than Change Topic, p < .05. (Successful Join Topic proportion is based on a range of 3 to 17 Join Topic attempts; successful Change Topic propor- tion is based on a range of 1 to 14 Change Topic attempts.)

Page 10: Tomasello Joint Attention and Conversation

Barton and Tomasello 525

0.6-

0 - 0.5

: 0.4-

0.3

0.2

0 0.1,

E 0.o question comment request

speech act

U infant-directed 0 other-directed

FIG. 1.-Mean probability of infant response as a function of addressee and speech act

ANOVA could not be run. Hence, the analy- sis was broken down into two parts, one as- sessing the effects of the nonlinguistic factor of joint attention and the other assessing the effects of the linguistic factors of addressee and speech act, in each case in combination with the effects of age.

The 2 (age) x 2 (joint attention) mixed ANOVA found no main effect of age on the probability of infant response. There was a significant main effect of joint attention: the mean probability of infants responding was greater when engaged in a JAE with the speaker (.23) than when not so engaged (.13), F(1,16) = 10.37, p < .01. The interaction was not significant.

The 2 (age) x 2 (addressee) x 3 (speech act) mixed ANOVA found a main effect of age: the mean probability of infant response was greater at 24 months (.22) than at 19 months (.14), F(1,16) = 4.77, p < .05. There was also a significant main effect of ad- dressee: the mean probability of infants' re- sponding was greater when they were di- rectly addressed (.24) than when they were not addressed (.13), F(1,16) = 18.50, p < .01. The significant speech act x addressee in- teraction, F(2,32) = 5.38, p < .01, however, indicates that only when a question was

asked did addressee have a significant effect on the probability of infants' responding; Tukey's (a) test for unconfounded means confirmed this finding (p < .05).' In the cases of comments and requests, the infants were just as likely to respond when they were not addressed directly as when they were. This interaction is shown in Figure 1. No other interactions were significant.

Discussion

The current study attempted to docu- ment the dynamics of linguistic and nonlin- guistic interaction in mother-infant-sibling triads. Three sets of findings are of interest.

First, it was found that children as young as 19 months of age were able to en- gage in triadic joint attentional episodes (JAEs) and triadic conversations. This is the youngest age at which such interactions have been systematically documented, with the exception of Bruner's (1977) observa- tions of a single 13-month-old infant learn- ing to give objects to both his mother and an experimenter. Further, the proportion of triadic JAEs and conversations increased significantly from 19 to 24 months of age, indicating that important developments in these skills are still taking place during the

'Because some cells were still based on small frequencies (range 1-108), the analysis was repeated, deleting subjects who had any cell with fewer than five opportunities to respond. Results remained the same. Thus, the statistical values and data reported are for the entire sample of 18.

Page 11: Tomasello Joint Attention and Conversation

526 Child Development

second half of the second year of life. It is also important that mother-infant dyadic in- teraction in the current study did not change over time, and that sibling-infant JAEs and conversations were virtually nonexistent at both ages. Thus, all increases in the infants' interactions and conversations with their mothers or siblings in the triadic context during this developmental period were due almost exclusively to their increased in- volvement in triadic interactions and con- versations.

The second set of findings was quite un- expected. Mother-infant-sibling conversa- tions were almost three times longer than either mother-infant or mother-sibling con- versations, with all three participants taking more turns than they did in dyadic conversa- tions. This was unexpected because previ- ous research has reported that infants pro- duce more utterances overall in dyadic contexts than in triadic contexts (Jones & Ad- amson, 1987; Woollett, 1986). More utter- ances does not mean longer conversations, however, and thus the increased length of triadic conversations presumably reflects a change in the dynamics of conversation when three as opposed to two people are involved.

Our analyses of infant turns established that the increased length of triadic conversa- tions was not due to some dynamic that ex- cluded the infant. It was not the case that many triadic conversations reflected the mother's carrying on two essentially sepa- rate conversations with her two children. The vast majority of triadic conversations had some form of reciprocation between all participants, including the infant. Nor were the longer triadic conversations due to some depressive effect of the triadic context on dy- adic conversations. Interrupted mother- infant conversations were not any shorter than uninterrupted ones, and, moreover, the dyadic conversations in the current study are of roughly comparable length (or slightly longer), with the same number of infant turns, as the mother-infant dyadic conversa- tions of children of this age in purely dyadic contexts. For example, in a study of 10 18- 24-month-old later-born children, Tomasello and Mannle (1985) found an average conver- sation length of 2.6 turns for mother-infant dyads (3.8 in the current study), with an av- erage of 1.3 infant turns (1.8 in the current study). Our conclusion is thus that the longer triadic conversations, with a greater number of infant turns, are not due to some artifact of our triadic coding scheme, nor to a disrup-

tive effect of the triadic context on mother- infant dyadic interactions; mother-infant- sibling triadic conversations are longer than mother-infant dyadic conversations.

The reasons for this greater length can- not be directly determined by data from the current study. However, we may adduce at least some indirect support for the specula- tion that in a triadic conversation less pres- sure is put on each of the children. That is, during mother-infant dyadic conversations, roughly half of the conversational weight rests on the child; in these contexts the child has to take appropriate turns at all of the ap- propriate junctures if the conversation is to be maintained. In triadic conversations, on the other hand, when a child does not under- stand or have anything to say she may simply stay silent. The other two participants may continue the conversation, and the silent member might then rejoin later when she finds a comprehensible and interesting place to make a contribution. This interpre- tation is supported by the fact that mothers in the current study took about one-half of the turns in both dyadic and triadic conver- sations; each child took roughly one-half of the turns in their dyadic conversations with mothers, but only one-quarter of the conver- sational turns in triadic conversations. It seems plausible that such dynamics might account for longer conversational interac- tions in certain multichild contexts.

One alternative hypothesis is that the triadic conversations in our study were longer because they provided the infants with their only opportunity to converse with older siblings, since dyadic conversations and interactions between the children were almost nonexistent. This might have made the triadic conversations more motivating for the infants and thus more likely to sustain their interest and attention. The main evi- dence against this hypothesis is that triadic conversations were longer, with more child turns, for siblings as well as infants. It is un- likely that interacting with the infants in this setting was especially motivating for the older siblings, however, as previous re- search for a similar context (i.e., the presen- tation of new toys) has shown that preschool- ers prefer playing alone with the toys than conversing with their infant siblings (Toma- sello & Mannle, 1985).

It would thus seem most plausible to at- tribute the greater length of triadic conversa- tions in the current study to differences in the basic dynamics of dyadic and triadic in-

Page 12: Tomasello Joint Attention and Conversation

Barton and Tomasello 527

teractions. It must be kept in mind, however, that these dynamics only obtain when the participants have reached some critical level of linguistic skill, as evidenced by the fact that Tomasello et al. (1986) found that 18- 24-month-old twins (average age of 21 months) did not engage in triadic conversa- tions with their mothers beyond a few very short episodes. The linguistic skills of these twins were somewhat less than those of even the 19-month-olds in the current study, which would seem to indicate that our sub- jects might have just recently attained the requisite level of linguistic skill necessary for lengthy triadic conversations.

The third and final set of findings con- cerns the conditions that facilitate infant par- ticipation within the triadic context. Because joining a conversation (dyadic or triadic) is in some sense a more difficult task than ini- tiating a topic in that it requires the infant to understand another speaker's topic, it was striking to find that nearly one-third of the infants' first turns joined into an ongoing conversation, and an additional one-third of their turns continued a conversation. In con- trast, the presumably easier task of simply saying whatever is on one's mind (initiating a topic or interrupting to change the topic) comprised only 11% and 19% of the infants' first turns, respectively. This result suggests that the ongoing conversation offers impor- tant scaffolding for the infant by indicating a topic the speaker is interested in and there- fore likely to continue. The success of join- ing a conversation indicated that this was precisely the case: 92% of the time infants joined a conversation, that conversation con- tinued, while changing the topic was suc- cessful for infants only 70% of the time. Dunn and Shatz (1989) also found that in- fants were more successful when their intru- sions were topic-relevant (joined) than when they were topic-nonrelevant (changed topic). These results thus confirm that chil- dren nearing their second birthdays are quite good at joining into an ongoing conver- sational topic, which presumably requires comprehension of that topic.

Joint attention was clearly an important factor facilitating infant participation in tri- adic conversations. Infants of both ages were more likely to take a turn when they shared a joint attentional focus with the speaker, thus indicating that nonlinguistic information about the focus of interaction somehow facil- itated their verbal contributions. Although in some ways this is an unsurprising finding, it nevertheless goes a long way in explaining

why infants are able to join into only some types of ongoing interactions among other participants. Thus, it is presumably the case that 19-24-month-old infants are not able to join into many adult-adult conversations- even those that contain simple language- because these conversations are not about perceptually available topics that can be specified nonverbally. In the current study with siblings and mothers and a new set of toys, on the other hand, they joined in readily. It may thus be that the concreteness of topics in mother-sibling interactions makes joint attentional interactions quite natural for the infant, which in turn makes possible her participation in triadic conver- sations. One implication of this, however, is that the 19-24-month-old infant's participa- tion in triadic conversations may still be somewhat fragile, and thus it may not mani- fest itself in less propitious circumstances than those provided for them in the current study.

Nevertheless, just how in tune infants were in the current context is indicated by the fact that they made relevant verbal con- tributions to comments and requests equally often whether or not they were directly ad- dressed. (The fact that infants responded differentially to questions depending on whether or not they were addressed shows that this was not due to their inability to tell who was being addressed.) This result re- flects in part the simple fact that when moth- ers make comments or requests for action, they often tend not to have any particular expectation that the child take a subsequent verbal turn-at least not as much as with questions. But, in combination with the findings that infants joined into ongoing con- versations quite successfully and that joint attention facilitated infant conversational participation overall, it also implies that chil- dren younger than 2 years of age can process and respond to a substantial amount of lan- guage not directly addressed to them if that language is used in a comprehensible situa- tional context.

The findings of the current study clearly demonstrate that triadic interactions have their own dynamics that cannot be reduced to principles of dyadic interaction. Having three participants changes the demands placed on each participant and, in some cir- cumstances at least, leads to more extended conversational interactions. Individual par- ticipants in these interactions are provided with numerous opportunities and some prac- tice at joining into ongoing conversational

Page 13: Tomasello Joint Attention and Conversation

528 Child Development

interactions between other persons, which is simply not possible in dyadic interactions of any type. In addition, it may be that the mother-infant-sibling triadic context in par- ticular, as one instance of a triadic multichild context, has some unique features that en- courage infant participation in multispeaker conversations. In particular, the mother- infant-sibling context may facilitate infant participation more than triadic contexts with two adults because the conversational topics of adults often do not concern things that lend themselves to a nonlinguistic joint at- tentional focus among all participants. On the other hand, mother-infant-sibling inter- action may also encourage infant participa- tion more than triadic contexts with an adult and same-aged peer (including a twin) in cases where neither child is able to carry much of the conversational load.

The unique features of mother-infant- sibling interaction may or may not have im- portant effects on infant communicative competence; we collected no data on this question. One suggestive finding, however, is reported by Vandell and Wilson (1987) who found that infant experiences with sib- lings predicted their subsequent turn-taking exchanges with an unfamiliar peer. It might be, therefore, that mother-infant-sibling in- teraction of the type observed here fosters the development of communication skills that serve as a bridge for young children's interactions with peers in which many of the same dynamics obtain, both in terms of hav- ing child interactants and in terms of multi- speaker contexts. In "any case, the mother- infant-sibling context is clearly a rich and important language-learning environment for many of the children in most of the cul- tures of the world. It is thus clearly impor- tant that we go beyond the mother-child dyad to investigate this and all of the other contexts of interaction in which children ac- quire their linguistic skills.

References

Bakeman, R., & Gottman, J. M. (1986). Observing interaction: An introduction to sequential analysis. Cambridge: Cambridge University Press.

Bruner, J. (1977). Early social interaction and lan- guage acquisition. In H. Shaffer (Ed.), Studies in mother-infant interaction (pp. 271-290). London: Academic Press.

Della Corte, M., Benedict, H., & Klein, D. (1983). The relationship of pragmatic dimensions of mothers' speech to the referential-expressive

distinction. Journal of Child Language, 10, 35-43.

Dunn, J., & Shatz, M. (1989). Becoming a conver- sationalist despite (or because of) having an older sibling. Child Development, 60, 399- 410.

Forrester, M. A. (1988). Young children's polyadic conversation monitoring skills. First Lan- guage, 7, 145-158.

Jones, C. P., & Adamson, L. B. (1987). Language use in mother-child-sibling interactions. Child Development, 58, 356-366.

Linton, M., & Gallo, P. S., Jr. (1975). The practical statistician: Simplified handbook of statis- tics. Belmont, CA: Wadsworth.

Mannle, S., & Tomasello, M. (1987). Fathers, sib- lings, and the bridge hypothesis. In K. E. Nel- son & A. van Kleeck (Eds.), Children's lan- guage (Vol. 6, pp. 23-42). Hillsdale, NJ: Erlbaum.

Nelson, K. (1973). Structure and strategy in learn- ing to talk. Monographs of the Society for Re- search in Child Development, 38(1-2, Serial No. 149).

Nelson, K. E., Baker, N. D., Denninger, M., Bon- villian, J. D., & Kaplan, B. J. (1985). Cookie versus Do-it-again: Imitative-referential and personal-social-syntactic-initiating language style in young children. Linguistics, 23, 433-454.

Ochs, E. (1982). Talking to children in western Samoa. Language in Society, 11, 77-104.

Olson, S. L., Bayles, K., & Bates, J. E. (1986). Mother-child interaction and children's speech progress: A longitudinal study of the first two years. Merrill-Palmer Quarterly, 32, 1-20.

Oshima-Takane, Y. (1988). Children learn from speech not addressed to them: The case of personal pronouns. Journal of Child Lan- guage, 15, 95-108.

Rice, M. (1989). Children's language acquisition. American Psychologist, 44, 149-156.

Schaffer, H. R. (1989). Language development in context. In S. von Tetchner, L. Siegel, & L. Smith (Eds.), The social and cognitive aspects of normal and atypical language develop- ment (pp. 1-22). New York: Springer-Verlag.

Schaffer, H. R., & Liddell, C. (1984). Adult-child interaction under dyadic and polyadic condi- tions. British Journal of Developmental Psy- chology, 2, 33-42.

Schieffelin, B. B. (1979). Getting it together: An ethnographic approach to the study of the de- velopment of communicative competence. In E. Ochs & B. B. Schieffelin (Eds.), Develop- mental pragmatics (pp. 73-110). New York: Academic Press.

Schieffelin, B. B., & Ochs, E. (1983). A cultural

Page 14: Tomasello Joint Attention and Conversation

Barton and Tomasello 529

perspective on the transition from prelinguis- tic to linguistic communication. In R. M. Golinkoff (Ed.), The transition from prelin- guistic to linguistic communication (pp. 115-132). Hillsdale, NJ: Erlbaum.

Tomasello, M. (1988). The role of joint attentional

processes in early language development. Language Sciences, 10, 69-88.

Tomasello, M., & Farrar, M. J. (1986). Joint atten- tion and early language. Child Development, 57, 1454-1463.

Tomasello, M., & Mannle, S. (1985). Pragmatics of

sibling speech to one-year-olds. Child Devel-

opment, 56, 911-917. Tomasello, M., Mannle, S., & Kruger, A. C. (1986).

Linguistic environment of 1- to 2-year-old

twins. Developmental Psychology, 22, 169- 176.

Tomasello, M., & Todd, J. (1983). Joint attention and lexical acquisition style. First Language, 4, 97-212.

Vandell, D., & Wilson, K. (1987). Infant's interac- tion with mother, sibling, and peer. Child De-

velopment, 58, 176-186. Wells, G. (1981). Learning through interaction:

The study of language development. Cam-

bridge: Cambridge University Press. Woollett, A. (1986). The influence of older sib-

lings on the language environment of young children. British Journal of Developmental Psychology, 4, 235-245.