Abstract
This work provides initial evidence of reciprocity in conversation. We tested whether conversations with contribution imbalances brought on by task demands contained attempts to redress the created imbalance. Pairs of participants identified public art via phone communication. One member of the pair, the director, gave instructions using a map while the other, the follower, walked around a small town finding public art pieces. Later, trained raters coded the participants’ transcribed conversational turns as either on-task or off-task. As observed in similar studies, directors spoke more in on-task portions of the dialogue. We newly found that in off-task communication, followers spoke more than their directors and used a greater number of words per turn than their directors. We interpret the pattern as reflecting behaviors leading toward balance in contributions across the conversation as a whole, a process we refer to as reciprocity in conversation.
1 Introduction
Contributions to a conversation can be short, such as replying uh huh or yeah. They can also be long, as when explaining the intricacies of a complex board game or describing the nuance of an unpopular belief. Conversationalists may not always notice when they contribute more than each other, but do they contribute in a systematic way? We propose that interlocutors naturally balance the amount contributed over the course of a conversation, a process that can be observed under certain dialogic conditions. One such condition is when people are put into situations and roles that lead to contribution imbalances. If there is an opportunity to make up for that imbalance, such as through a stretch of dialogue unconstrained by the activity driving the imbalance, interlocutors may adjust their conversation to compensate. We refer to this compensation as reciprocity in conversation.
2 Communication accommodation theory
Reciprocal behaviors in conversation can be understood within the framework of communication accommodation theory, which lays out several strategies people use to communicate with different addressees based on interpersonal orientations (e.g., social group memberships, relationship history, personal values, potential interpersonal conflict) and conversational goals (e.g., building rapport, distancing oneself; Gallois et al., 2005). Individuals choose these communication strategies based on the notions they have of both themselves and their interlocutors. In some scenarios, people choose strategies in order to complete an objective. Convergence is a strategy in which interlocutors match each other’s communicative behaviors throughout the course of conversation, whereas divergence is a strategy in which interlocutors emphasize differences between their communicative behaviors (Gallois et al., 2005). There are two forms of divergence in the research literature (divergence and speech complementarity). See Table 1 for a summary of terminology used in the current report.
Speaking strategy definitions.
2.1 Convergence
Each communicative strategy is associated with different purposes. People use convergent communication strategies when they want to be viewed positively by their interlocutors (Gallois et al., 2005). And indeed, interlocutors are viewed more positively when convergence occurs: Greater convergence by individuals placed into a less powerful role in a conversation was associated with greater ratings of rapport and attractiveness (Muir et al., 2016). As another example, mutual likability scores increased as convergence of interlocutor articulation rate increased (Schweitzer & Lewandowski, 2013). Similarly, convergence in speech rate and response latency was associated with higher ratings of attractiveness (Putman & Street, 1984). Although convergence was associated with increases in positive perceptions of interlocutors, in some situations interlocutors may want to socially distinguish themselves from one another. In these cases, divergence may be preferred.
2.2 Divergence
If communicative interactions are deemed negative, either through social group membership, initial orientations, or perceived conflict, divergence may be used. For example, Bourhis and Giles (1977) had an English individual speaking in Received Pronunciation question Welsh individuals’ motivations for learning the Welsh language, seemingly challenging their ethnic identity (calling Welsh a dying language with a dismal future). After this comment by the Received Pronunciation English speaker, Welsh participants accentuated their accents, and even introduced Welsh English words into their responses. In this instance, Welsh speakers chose to diverge from their interlocutor, distinguishing themselves from the Received Pronunciation English speaker in a linguistically unique way. Speakers may choose to diverge from their interlocutor for social reasons or based on conversational roles and contextual constraints.
2.3 Speech complementarity
If conversation contexts lead to imbalanced interlocutor roles, individuals may implement speech complementarity (Giles, 1980). Speech complementarity refers to instances in which individuals linguistically diverge (e.g., differences in speech rate, length of turn), but do so to work towards convergence in other ways (Giles, 1980). If a situation arises in which speech complementarity is to be used, interlocutors accept differences in communicative styles due to conversational roles (Gallois et al., 2005). In these situations, conversational contributions differ not as a way to create social distance (as in divergence) but as a strategy to draw socially closer through cooperation. For example, a professor explaining a concept to a student has more information to share with the student than the student has to share with the professor. When the professor says more, this accepted divergent behavior is perceived as accommodating; neither the student nor the professor takes this divergent behavior negatively.
Researchers have previously explored the differentiation of contribution behaviors due to conversational roles. Differentiation in social status led to a more dominant speaker contributing more words compared with their subordinates, while interlocutors sharing social status often converged on the number of words spoken (Niederhoffer & Pennebaker, 2002), reinforcing the idea that different social roles have immediate influences on contribution behaviors. Similarly, participants who were put into a directing role during a collaborative task spoke for a longer duration compared with their following partner, even after participants exchanged roles several times throughout the course of completing the task (Pardo et al., 2013). The authors argued that participants who were placed in the directing role dominated the conversation due to their original role assignment.
However, common everyday conversations rarely take place in situations calling for such well-defined conversational roles. Instead, colleagues may weave in and out of particular topics, including topics in which one is more of an expert than the other. Two close friends may stumble into a conversation about a particular country and find out that one of them had lived there as a child. The natural rhythms of conversational shifts may bring forth situations where convergence is preferred, while also bringing forth situations where speech complementarity is preferred, all within the same conversation. The question we aim to address here is how similar people, when put in a situation calling for speech complementarity, will respond to the contribution imbalances when the roles of the complementarity situation are removed. We argue that people make up for task-demanded contribution imbalances when roles are less consequential. In essence, interlocutors will reciprocate.
2.4 Social group membership, convergence, and divergence
For convergence, divergence, and speech complementarity, interlocutors’ initial orientations influence their selection of communication strategies. Perhaps the most important contributor to strategy selection is social group membership. Social group characteristics play key roles in the likelihood that interlocutors will converge or diverge (Giles et al., 1987, 1991; Fitch & Hopper, 1983). For example, phonetic convergence occurred between roommates enrolled at the same college (Pardo et al., 2012), but not between speech communities that share a turbulent history (Bourhis & Giles, 1977). Alternatively, when differences in group membership are expected within a conversation, individuals may use speech complementarity. For example, doctors talk for longer periods of time compared with patients (Street, 1991). Here, because there are expectations associated with the knowledge and skill of a doctor compared with a patient, it is expected and generally accepted that the doctor will talk more during this interaction.
While the influences of convergence and divergence have been explored, the direct influence of speech complementarity has received limited attention. There is some evidence that conversational imbalances have negative influences on task performance: Fox Tree and Clark (2013) found that the more a directing individual (director) spoke while guiding a single individual or group of individuals (matcher) through a referential communication task, the worse the group performed at the task. The authors argued that feedback from matchers helped shape information from the directors so that it was more understandable by all. But increasing contribution balance in a situation calling for speech complementarity may also have played a beneficial role in other ways. For instance, increases in matcher contributions, which enhanced contribution balance, may have worked to improve engagement with the task, yielding greater success. People may have felt more comfortable in more balanced conversations and therefore paid more attention. The participants in the study were students at the same university with presumably similar a priori knowledge of the task. In this scenario, convergent communication behaviors would be the expected norm going into a conversation. The introduction of asymmetrical task roles creates a situation that encourages speech complementarity in the form of differences in how much each interlocutor contributes. Adjusting the contribution balance back toward symmetry could accommodate for the imbalance from speech complementarity, increasing conversational enjoyment and comfort during the conversation.
We predict that interlocutors will respond to speech complementarity by rebalancing the conversation. This rebalancing is most likely to happen when interlocutors are given the opportunity to adjust the imbalance, such as by being afforded time for both on-task and off-task communication. Ultimately, when reciprocity takes place, interactions should be more positively viewed by the interlocutors involved.
An alternative hypothesis is that people do not adjust for imbalance in a conversation. In one study, researchers observed that participants failed to alter contribution behaviors after initial roles had been assigned (Pardo et al., 2013). In an asymmetrical task that switched participant roles several times, individuals who were assigned the directing role at the beginning tended to talk more across the entire conversation, despite the fact that the roles were later switched. At first glance, these results may seem to contradict our hypothesis that interlocutors will naturally balance. But because both parts of the study included well-defined roles, the task never gave participants the opportunity to step outside of those roles and interact as peers. It could be that people are more likely to reciprocate when they step outside of speech complementarity roles altogether.
We tested for the presence of conversational reciprocity by analyzing on-task and off-task dialogue in the Artwalk corpus (Liu et al., 2016). This unique corpus provided the appropriate conversational contexts because participants belonged to the same overall social group but were required to take on roles that gave them asymmetrical information. In addition, unlike in many laboratory tasks, participants also were afforded the opportunity to engage in off-task dialogue.
3 Artwalk corpus
The Artwalk corpus (Liu et al., 2016) is a collection of transcripts in which dyads completed a structured task while also having time to engage in small talk. The corpus allowed us to test to what extent interlocutors’ contribution differences change as role salience changes over the course of a conversation. Research assistants living in Santa Cruz transcribed Artwalk sessions at the University of California Santa Cruz.
In the Artwalk task, participants were either assigned the role of director or follower. The director was seated in a participant booth in a lab. The follower walked around downtown Santa Cruz. Ten public art installations located in the downtown area were identified as target locations for the director to guide the follower to. The goal of the task was to take a series of photographs of the art targets. The director was shown images of the art targets with maps of where they were located. Followers did not see the photographs of the targets, nor were they given access to maps. To complete the task, directors guided followers to the particular art installations by finding and communicating a path and describing the artwork. Once the follower confirmed that they had arrived at the described art installation, they took a photo of the art installation and continued on toward the next art installation.
This task is similar to tasks done in laboratory settings, such as the tangram task, where directors instruct matchers about the order in which to place a series of abstract shapes (Clark & Wilkes-Gibbs, 1986), and the map task, where givers instruct followers to identify a route that is on the giver’s map but not on the follower’s map (Anderson et al., 1991). In laboratory versions of these tasks, participants are focused on getting the task completed. They do not usually talk about activities outside of the realm of the task, as close investigation of participants’ contributions to the conversations makes clear (Tolins et al., 2018). In the Artwalk task, however, there were many periods of time where participants were walking from one art installation to the next. These periods could be used to talk about off-task topics and were frequently used in this way (Liu et al., 2016). For example, participants often talked about school-related information, topics of common interest, or the weather:
(1) D: so what college are you at
um I’m actually this year living over like by the um Redwood Grove Apartments
oh cool
I was I was Merrill last year but uh yeah I dunno it’s chillin’ though over at the Grove
what year are you
I’m a sophomore this year
oh cool cool cool
yeah how ’bout you
I’m a freshman I’m at Stevenson
oh ok right on cool
(2) F: um well I’ve been leaning towards doing med school and since I haven’t done like any of the science stuff for that I’m going to probably to end up doing catch up program
yeah I think that’s really upsetting I feel like you have to know in high school if you want to be a doctor otherwise you get stuck and behind
well I almost feel that way about everything I mean I’m personally a transfer student and it took me like three years just because it was so hard to get the classes once I knew which classes I needed to transfer as a psych major so
yeah yeah I think it’s just it’s such a shame how you like yeah how you have to play catchup unless you like really know early on
(3) F: oh you you you listen to that good band
which band you mean Boston
yes
I love Boston
no way me too!
what do you think about The Who
uh yeah they’re decent
(4) F: no it’s cool I was just wondering you having a nice day
um I’m having an okay day could have gotten some better sleep how ’bout you
oh yeah kind of the same but it’s really sunny
hmm
which is nice
(5) D: oh yeah I know it’s been really warm I like it
I definitely definitely dig it though we’re supposed supposed to get rain on Sunday
oh yeah I heard that too so I’m gonna try to go to the beach on Saturday
what’s that
so I’m going to try to go the beach on Saturday before
oh yeah definitely to like soak up as much sun before the weather starts to foul up
It is because of the special nature of the Artwalk corpus, where people had time to talk about off-task topics, that we are able to assess reciprocity.
4 Current research
We propose that contribution imbalances created by speech complementarity are frequently corrected through reciprocity. In the Artwalk corpus (Liu et al., 2016), directors and followers have access to different types of information that they must relay to each other, but otherwise they are on the same footing. This creates good conditions for speech complementarity, in which the participants assume roles that are distinct in their expected contribution behaviors. In similar laboratory tasks with speech complementarity, interlocutors do not usually have the opportunity to naturally rebalance toward symmetrical contributions. In the Artwalk task, however, interlocutors were able to talk about off-task topics because the activity included substantial time during which the followers physically walked to destinations. We hypothesized that the Artwalk participants would rebalance their contributions during off-task conversation back toward symmetrical contributions overall. We expected directors to contribute more words and take longer turns in on-task conversation. We hypothesized that followers would say more words and take longer turns in off-task conversation to adjust for this contribution imbalance.
Because we were working with data from a previously run experiment, we were not able to directly measure our hypothesized outcome for reciprocity, conversation enjoyment. We did, however, have two similar measures available from a post-experiment questionnaire: task-enjoyment ratings and comfort with interlocutor ratings. We tested how contribution balance affected these measures, predicting that more contribution imbalance remaining at the end of the conversation would be associated with lower reported ratings of task enjoyment and comfort with partners. We caution, however, that these measures do not directly assess conversation enjoyment.
5 Method
5.1 Materials
We analyzed 69 transcripts: 48 from the existing Artwalk corpus (Liu et al., 2016) plus an additional 21 transcripts that were not released because participants failed to complete the task (e.g., they failed to identify at least 8 of the 10 targets, or they experienced technical errors during their completion of the task).
5.2 Participants
We assessed language use for 138 participants (69 dyads). They were recruited from the Psychology Department participant pool at the University of California Santa Cruz, as well as via word of mouth and on-campus ads. Participants were paired by signing up for the same time slot and were assigned to be either the director (who came to the lab) or the follower (who went to downtown Santa Cruz), receiving either course credit or a $10 gift card for their participation. Of the 69 dyads, 21 dyads consisted of females only, 12 dyads were male only, 15 dyads were mixed genders, and 21 were unknown gender. The 21 dyads in which gender information is not known were transcripts that were not released with the Artwalk corpus. All participants were native English speakers. Participants expressed moderate familiarity with Santa Cruz (M = 4.89, SD = 1.31 on a 1–7 scale, with 1 being not at all familiar and 7 being very familiar), very high comfort completing the task (M = 6.05, SD = 1.06, with 1 being very uncomfortable and 7 being very comfortable), and very high comfort with the other participant (M = 6.36, SD = 1.09, with 1 being very uncomfortable and 7 being very comfortable). Together, these facts suggest that Artwalk participants felt like part of the same social group.
5.3 Procedure
Two trained coders evaluated each transcribed turn as being either on-task or off-task. On-task was defined as conversation that directly related to the completion of the task (e.g., the director giving the follower directions to the next art installation, the follower clarifying these directions, the follower clarifying descriptions of the art installations, etc.), whereas off-task was defined as conversation that did not pertain to the completion of the task (e.g., discussion of college majors, upcoming social plans, etc.). Out of 69 transcripts, 10 were coded by one of the two independent coders as entirely on-task and were removed from the analyses, leaving a total of 59 transcripts.
Inter-rater reliability for the two independently trained coders was κ = 0.61 across the 59 transcripts. While this κ is moderately high, it has limited interpretability because of how task-relatedness was coded: To understand when interlocutors were contributing to on-task or off-task conversation we determined that the turns must be coded sequentially, within the context in which they occurred. By using sequential coding for on-task or off-task conversation, each coding of a turn was likely influenced by the coding of the preceding turn, making each coding partially dependent on the content in the previous turns. For example, a simple response turn like “yeah” would be coded as on-task if it were preceded by “are you on Cathcart street?” and as off-task if preceded by “you said you’re taking 15 units.” While raters were instructed to code individual turns, the nature of the conversations resulted in coding that usually occurred in multi-turn blocks of on-task and off-task speech. Because of this phenomenon, κ may be artificially inflated, but it also may be artificially deflated by the relative sparsity of off-task communication (e.g., Viera & Garrett, 2005). For these two reasons, we also performed a nominal reliability analysis of on-task and off-task ratings. We grouped consecutive turns of off-task communication into blocks as they occurred within the conversation, and then determined how many blocks of off-task exchanges each coder identified. Coder 1 reported 357 blocks of off-task communication, and Coder 2 reported 777 blocks of off-task communication. Importantly, 98.3% of Coder 1’s off-task blocks (350 blocks) completely overlapped with Coder 2’s blocks, suggesting that Coder 1 and 2 strongly agree on Coder 1’s ratings.
To arrive at a completely coded set of transcripts, a third independent coder resolved all turn-level disagreements between Coder 1 and Coder 2. Once disagreements were reconciled, we compared Coder 3’s ratings of the disagreements with the ratings of Coder 1 and Coder 2. Coder 3 agreed with Coder 2 on 87.02% of the disagreements (2548 of 2928 turn disagreements). The final coded data set (across all 59 included dyads) consisted of 24,331 turns with 18,753 of these coded as on-task and 5,578 as off-task.
It is worth considering the difference between the off-task conversation coding of Coder 1 and Coder 2. While we did observe a 98.3% overlap of Coder 1’s blocks with Coder 2’s blocks, Coder 2 identified 420 more off-task blocks than Coder 1 did (more than twice as many). After Coder 3’s corrections, off-task contributions accounted for 22.93% of total turns contributed across all dyads. Coder 1 may have simply expected less off-task conversation to occur, leading them to be less likely to label a turn as off-task.
6 Results
To calculate the average turn length, we divided the total number of words spoken by each interlocutor by the total number of turns that interlocutor produced. For example, if a director spoke 1,000 words over the course of 100 turns during on-task conversation, their average turn length would be recorded as 10 words per turn for on-task conversation. A turn refers to the complete articulation of a contributed utterance made by an interlocutor, with an utterance consisting of only backchannels considered a complete turn. A word refers to the complete verbalization of a word during a turn. We removed all partial words from the transcripts before conducting analyses. We used repeated-measures ANOVAs and paired-sample t-tests to test hypotheses. In our first analysis, we tested whether directors’ and followers’ word counts changed over conversation type. In our second analysis, we tested whether directors’ and followers’ turn lengths changed over conversation type. We then tested the relationship between contribution balance and task enjoyment, as well between contribution balance and comfort with their partners. Finally, we ran post hoc analyses to compare the included 59 dyads with the excluded 10 dyads.
6.1 Number of words contributed
To test whether contribution balance changed over on-task and off-task conversation, we conducted a role (director vs. follower) by conversation type (on-task vs. off-task) repeated-measures ANOVA on the number of words contributed during conversation (see Figure 1). We found a significant role by conversation type interaction, F(1, 58) = 127.56, p < 0.001, ηp2 = 0.687. Pairwise comparisons of simple main effects revealed that directors (M = 1,977.85, SD = 655.85) spoke a greater number of words than their followers (M = 1,252.81, SD = 559.23) during the on-task conversation, t(58) = 9.53, p < 0.001, 95% CI for the difference, [572.71, 877.36], d =1.19, whereas followers (M = 441.37, SD = 436.58) spoke a greater number of words than their directors (M = 335.59, SD = 304.43) during the off-task conversation, t(58) = 3.04, p = 0.004, 95% CI for the difference, [36.20, 175.36], d = 0.28.
Number of words contributed by each interlocutor. Error bars represent 95% confidence intervals.
Due to large differences in the number of words that dyads used, we also conducted follow-up comparisons using the percentage of conversation that was attributable to one of the interlocutors. We divided the number of words contributed by each interlocutor in each conversation type by the total number of words contributed by both interlocutors in that conversation type. For example, the number of words contributed by the director in task-related conversation was divided by the total number of words contributed by both the director and the follower during task-related conversation. Directors (M = 61.84%, SD = 8.55%) accounted for a greater percentage of the conversation than their followers (M = 38.16%, SD = 8.55%) during on-task conversation, t(58) = 10.63, p < 0.001, 95% CI for the difference [19.22%, 28.14%], d = 2.77. Further, followers (M = 55.33%, SD = 12.83%) accounted for a greater percentage of the conversation than their directors (M = 44.67%, SD = 12.83%) in off-task conversation, t(58) = 3.19, p = 0.002, 95% CI for the difference [3.97%, 17.34%], d = 0.83.
6.2 Average turn length
To compare the directors’ and followers’ average turn length during both on-task and off-task conversation, we conducted a role (director vs. follower) by conversation type (on-task vs. off-task) repeated-measures ANOVA on the average number of words per turn contributed during each type of conversation (see Figure 2). We found a significant role by conversation type interaction, F(1, 58) = 135.90, p < 0.001, ηp2 = 0.701. Pairwise comparisons of simple main effects revealed that the average length of directors’ turns (M = 12.90, SD = 3.89) during on-task conversation was higher than the average length of their followers’ turns (M = 7.80, SD = 1.85), t(58) = 8.95, p < 0.001, 95% CI for the difference, [3.96, 6.24], d = 1.67, whereas the average length of followers’ turns during off-task conversation was higher (M = 7.83, SD = 3.05) than the average length of their directors’ turns (M = 6.87, SD = 2.63), t(58) = 2.18, p = 0.033, 95% CI for the difference, [.08, 1.84], d = 0.34.
Average words per turn by each interlocutor. Error bars represent 95% confidence intervals.
6.3 Association between contribution balance, task enjoyment, and partner comfort
To understand how contribution balance was associated with participants’ enjoyment of the task and comfort with their interlocutor, we correlated a ratio calculation evaluating the contribution behaviors of the dyad with the dyads’ average task enjoyment rating (1 being very unenjoyable and 7 being very enjoyable), and the dyads’ reported average comfort with the other participant (1 being very uncomfortable and 7 being very comfortable). Because we were working with data from a previously-run experiment, direct measures related to our hypothesized reason for reciprocity (ratings of conversation enjoyment) were not available. We used task enjoyment and comfort with interlocutor ratings from the post-experiment questionnaire as approximations of conversation enjoyment.
For these analyses, we calculated a contribution balance score:
In this contribution balance score, D is the number of words spoken by the director and F is the number of words spoken by the follower. For example, if the director spoke a total of 1,000 words, while the follower spoke 500 words, this would yield a contribution balance score of 0.67. A score of 1 would be perfectly balanced, and a score of 0 would mean that only one person spoke.
6.3.1 Contribution balance and task enjoyment
We evaluated the relationship between contribution balance scores for the entire conversation and average task enjoyment. Average task enjoyment was high (M = 5.56, SD = 0.95). Overall contribution balance scores were related to average task enjoyment reported by the dyad, r(57) = 0.26, p = 0.047 (see Figure 3), such that the more balanced the conversations were, the higher the average task enjoyment.
Relationship between contribution balance score and dyad average task enjoyment.
We repeated this analysis using the same contribution balance score calculation outlined above, but for only on-task or off-task contribution behaviors. There was no significant relationship observed between balance scores in on-task conversation and task enjoyment, r(57) = 0.19, p = 0.155, nor was there a significant relationship observed between balance scores in off-task conversation and task enjoyment, r(57) = 0.19, p = 0.143. This shows that the overall effect observed was not exclusive to one type of conversation.
6.3.2 Contribution balance and comfort with partner
Next, we evaluated the relationship between contribution balance scores and average comfort with partner. Dyads’ average reported comfort with one another was very high (M = 6.43, SD = 0.84). Overall contribution balance scores were not significantly related to dyads’ average reported comfort, r(57) = –0.009, p = 0.948.
We repeated this analysis to evaluate the relationship between partner comfort and contribution balance in on-task and off-task portions of the conversation. There was no relationship observed between dyads’ average comfort and contribution balance scores during on-task conversation, r(57) = –0.005, p = 0.968, nor was there a relationship observed between dyads’ average comfort and contribution balance scores during off-task conversation, r(57) = 0.193, p = 0.144.
6.4 Comparison of included and excluded dyads
It is worth further consideration that in 10 of 69 transcripts (15%) off-task conversation was not identified by Coder 1. The low prevalence of off-task conversation may relate to the salience of participating in an empirical study, leading participants to limit their deviation from on-task conversation. As mentioned, off-task conversation accounted for only 22.93% of total contributed turns among the 59 included dyads.
We ran post hoc analyses to determine whether there were systematic differences between the included and excluded dyads. Because the excluded 10 transcripts were not included in our planned analyses, we did not have coding differences reconciled by our third coder. Therefore, for the purposes of these post hoc analyses, we assumed that all turns taken by the interlocutors in the excluded dyads were on-task. We calculated contribution balance scores for both included and excluded transcripts using the same methodology described above, but only on-task conversation was included in the comparison. Because these analyses were post hoc, as well as because they compare 10 dyads with 59 dyads, these analyses and their outcomes should be interpreted as purely exploratory. Included dyads (M = 0.79, SD = 0.14) did not differ from excluded dyads (M = 0.77, SD = 0.14) on their contribution balance scores, t(67) = 0.506, p = 0.615. Included dyads (M = 5.56, SD = 0.95) also did not differ from excluded dyads (M = 5.15, SD = 1.51) in average task enjoyment, t(67) = 1.15, p = 0.256. Finally, included dyads (M = 6.10, SD = 0.74) did not differ from excluded dyads (M = 5.75, SD = 1.01) in average dyad comfort, t(67) = 1.32, p = 0.19.
7 Discussion
Sometimes interlocutors diverge from one another due to differences in knowledge, status, or expertise. Other times they diverge because they assume situational roles to support cooperative interaction. In the research presented here, directors and followers diverged in their contribution amounts, highlighting the influence that knowledge and role differences have on resulting speech behaviors. We hypothesized that contribution imbalances created through speech complementarity would result in a drive towards rebalancing, which in our corpus would manifest as increased contributions from the interlocutor in the follower role during off-task conversation.
Within the context of the Artwalk corpus (Liu et al., 2016), directors were instructed to guide followers to specific locations. We hypothesized that directors would contribute a greater number of words during on-task conversation compared with their followers and that followers would contribute a greater number of words during off-task conversation compared with their directors. Specifically, we took more words contributed by the director in on-task conversation to be evidence that speech complementarity had taken place, while more words contributed by the follower in off-task conversation to be evidence of reciprocity. We also hypothesized that the average length of turns spoken by directors would be greater in on-task conversation compared with the average length of turns spoken by their followers, and that the average length of turns spoken by followers in off-task conversation would be greater compared with the average length of turns spoken by their directors. Here, we took longer average turns contributed by the director in on-task conversation as further evidence of speech complementarity, and longer average turns by the follower in off-task conversation as further evidence for reciprocity. These hypotheses were supported.
In contrast to other researchers’ observations that participants failed to reconverge toward more similar contribution levels when repeatedly switching roles (Pardo et al., 2013), we found that participants did reduce imbalance. One way to reconcile the seemingly disparate findings is by noting that the opposite of imbalanced roles may not be to switch roles, but rather to remove roles entirely and allow interlocutors to interact as peers.
The effects of role removal may be observable in another existing corpus that is similar to the Artwalk corpus, the Walking Around corpus (Brennan et al., 2013). Both corpora involve referential communication outside of the laboratory. In both tasks, one participant was designated as a director and the other was designated as a follower. In both, the director guided the follower to a specific physical location where they were instructed to take a photo. The Walking Around corpus was described as having the navigational dialogue transcribed, and the transcript examples in the Walking Around paper were of on-task dialogue (Brennan et al., 2013). The transcript examples in the Artwalk paper included several examples of off-task dialogue, including discussion of events taking place concurrently with the task (such as interactions with passersby) and discussion of Santa Cruz attractions (Liu et al., 2016). If there is off-task dialogue in the Walking Around corpus, we predict reciprocity findings would be similar to those observed here.
While the association between conversational roles and contribution behavior has been observed previously (e.g., Street, 1991; Pardo et al., 2013), the communicative response to conversational imbalances created through speech complementarity have not been documented. We show that participants move towards more similar contribution levels, a process we call conversational reciprocity. The small observed effect of reciprocity suggests that interlocutors may find importance in demonstrating their equal social footing with each other outside of role-driven speech complementarity. While participants were able to complement each other’s communicative roles during the task, they showed an olive branch of reciprocity during off-task communication.
An important characteristic of contribution balance brought forth by the use of reciprocity in speech complementarity situations is that more balanced conversations are predicted to be more enjoyable. We hypothesized that dyads would report higher levels of task enjoyment and comfort with their partner when conversations were more balanced. In the analyses presented here, this was examined by evaluating the relationship between contribution balance scores (with higher contribution balance scores indicating more balanced conversations) with both dyad task enjoyment and dyad comfort with partner ratings. While the hypothesized relationship between contribution balance and task enjoyment was supported, the hypothesized relationship between contribution balance and comfort with interlocutor was not. In addition, neither the task enjoyment nor comfort with partner measures showed an effect when the data was split into on-task and off-task portions. The presence of a relationship between balance and task enjoyment across conversation type but not within conversation type shows that the effect was not driven by one kind of conversation.
While interlocutors seem to be adjusting toward balance, the relative difference between contributions was small, indicating that reciprocity in conversation may simply be an allowance to contribute more during a conversational subset, rather than completely making up for contribution imbalances. It is also important to note the differences in setting between the director and follower when interpreting our results. Specifically, the director was in a participant booth on campus with access to directions and images of targets, allowing them to contribute more during on-task conversation, while the follower was in a more lively environment downtown, possibly facilitating their increased off-task conversation contributions (e.g., describing the visual information they were observing). Future researchers can test reciprocity under conditions that match participants’ environments. Additionally, researchers can also test the circumstances in which reciprocity occurs, how much imbalance is generally corrected for, and how this correction behavior may vary across contexts.
Understanding the dynamics of role acceptance in speech complementarity situations should prove useful for developing a clearer picture of when and why reciprocity is used. Is reciprocity strictly used because the contribution imbalances created by speech complementarity drive the interlocutors to remind each other of their equal social footing? The work presented here allows for this to be the case, but it is also possible that reciprocity is used by interlocutors to signal non-acceptance of a conversational role that they feel is being forced upon them, which requires them to be in an undesired position. There are also questions about the implications of reciprocity. For example, if contribution imbalances are left uncorrected in successive conversations, what are the effects of this imbalance on the personal relationship between the interlocutors, extending beyond the immediate conversational encounter? What factors lead to heightened or dampened reciprocity? We are only at the beginning of understanding the phenomenon of conversational reciprocity.
Footnotes
Acknowledgements
We thank our research assistants Blaire Hobbs, Emily Truong, and Lauren Zartner for their help on this project. We thank Dr. Kris Liu for discussions about this topic and Dr. Travis Seymour for comments on an earlier version of this manuscript.
Funding
This research was supported by the Professor Bruce Bridgeman, Ph.D. and Diane Bridgeman, Ph.D. Graduate Award in Cognitive Psychology.
