Effects of group entitativity on young English-speaking children’s interpretation of inclusive We

Jared Vasil; Camryn Capoot; Michael Tomasello

doi:10.1371/journal.pone.0306556

Abstract

The pronoun we can be used to refer to various collections of people depending on various pragmatic factors. This article reports the results of two online experiments that investigated children’s interpretation of inclusive we, in which the child-listener is part of the intended referent of we. 128 2- and 4-year-olds collaborated with three partners in a coloring task. Before they played together, one partner informed participants that, e.g., “we can color!” Participants had their own markers and had to choose to how many partners to distribute (virtual) markers. In the first experiment, the partners appeared more like an aggregation of individuals than a collaborative group. The second experiment flipped this so that the partners appeared more like a collaborative group. Contrary to expectations, there was relatively little evidence for development in children’s interpretation of we. Additionally, participants did not sharply distinguish their interpretations of we from those of we both or we all. Rather, participants were more likely to choose group interpretations when contextual cues indicated that their partners were a collaborative group than an aggregation of individuals. Interestingly, this interpretational distinction was sharpest for the pragmatically ambiguous we, compared to the relatively unambiguous we both and we all. These results are informative about the kinds of cues that shape young children’s interpretation of pragmatically ambiguous pronominal reference.

Citation: Vasil J, Capoot C, Tomasello M (2024) Effects of group entitativity on young English-speaking children’s interpretation of inclusive We. PLoS ONE 19(7): e0306556. https://doi.org/10.1371/journal.pone.0306556

Editor: Barbara T. Rumain, Touro University, UNITED STATES

Received: August 12, 2023; Accepted: June 19, 2024; Published: July 9, 2024

Copyright: © 2024 Vasil et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: All data and code for the studies are freely available at: https://osf.io/5cs9p/.

Funding: MT: Jacobs Foundation Klaus J Jacobs Research Prize (https://jacobsfoundation.org/klaus-j-jacobs-research-prize/) The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests: The authors have declared that no competing interests exist.

Introduction

Human psychology is characteristically “groupminded.” Adults uphold commitments, enforce norms, and share fairly. Groupmindedness relies on the ability to conceive of social groups and to leverage the group’s perspective to guide behavior. Groupminded thinking has an ontogenetic trajectory. While children younger than 3 years of age jointly engage in coordinated social play and reengage disengaged partners [1, 2], only after they turn 3 do children begin to uphold commitments [3], enforce norms [4], and share fairly [5]. That is, children younger than 3 years can only easily leverage a dyadic perspective on collaboration. In contrast, children older than 3 can, also, easily leverage a group perspective on collaboration, too (for distinctions between dyads and groups, see [6]). Three years of age is thus a watershed moment in the ontogeny of mature human psychology [7]. At this age, the kinds of easily conceptualizable social entities changes qualitatively–after 3 years, children can easily conceive of dyads or groups.

Does the groupminded shift at 3 years of age influence language development? To answer this question, research has focused on linguistic reference. Successful reference requires that listeners form inferences about speakers’ intended referents, given linguistic form. How listeners solve this problem depends, in part, on the entities that they can conceptualize [8, 9]. For example, if the groupminded shift expands the set of conceptualizable social entities (thus, conceptualizable referents), then children older than 3 years may form three-person “group interpretations” of plural person reference more easily than children younger than 3. Instead, those younger children may favor 2-person “dyadic interpretations.”

Indeed, research on English-speaking children’s comprehension of we has supported this hypothesis. Use of the word we only partially specifies speaker referential intentions. While use of we indicates that speakers intend to refer to themselves and to other(s), it is ambiguous whether the latter referential intention includes only one individual, or two or more individuals. Note that this is a kind of “pragmatic” ambiguity in referential intentions (see also [10]), and not semantic ambiguity (i.e., as “ambiguity” is understood to pertain to truth conditions). Nonetheless, resolution of this ambiguity is psychologically important for listeners because the former referential intention warrants a dyadic interpretation, whereas the latter warrants a group interpretation.

[11] investigated how 2- and 4-year-olds resolve this pragmatic ambiguity in use of we. Those authors found that 4-year-olds systematically formed dyadic or group interpretations of we, depending on the spatial position of a speaker relative to two other individuals. In contrast, 2-year-olds favored dyadic interpretations. Importantly, 2- and 4-year-olds displayed an adult-like understanding of the first-person semantics of we by reliably including the speaker as one of the individuals referred to by use of we (number was controlled, which precluded investigating children’s understanding of the plural semantics of we). Consequently, [11] interpreted this developmental shift in referential interpretation as evidence that a groupminded shift at 3 years expands the set of conceptualizable social entities. Corroboratory results for referential production are reported by [12].

However, there was a limitation of the study of [11]. Participants in that study were excluded from (i.e., were third parties to) the intended referent of we. Rather, in that study, speakers referred only to themselves and other (i.e., non-participant) individuals. This limitation is important because children may have more experience interpreting we when they are included in the intended referent of we. That is, children may have less experience with the “exclusive we” used in [11] than they do with the “inclusive we.” Indeed, the fact that young children hear we infrequently relative to most other pronouns arguably makes this limitation more pressing. For example, [12] found that we and other first-person plural personal, possessive, and reflexive pronouns constitute about 6% of mothers’ spontaneous pronouns and occur in only about 3% of mothers’ spontaneous utterances. The present studies remedied this limitation by examining children’s interpretations of inclusive we, in which children were part of the intended referent of we.

Some languages formally mark the “clusivity” of first-person plural pronouns [13]. However, English does not. For instance, the Mandarin Chinese first-person plural form wǒmen indicates a listener-inclusive intended referent, whereas the listener-exclusive form is zánmen. In contrast, English requires the use of multiword constructions to mark intended clusivity, e.g., we but not you, you and I, us two, etc.; or else, listeners rely on contextual cues (e.g., joint commitment) to appropriately disambiguate intended clusivity. The present experiments leveraged the latter option.

To do this, Study 1 implemented a novel, virtual method for investigating children’s interpretation of ambiguous reference during collaboration. Next, Study 2 built on Study 1 by increasing the entitativity of participants’ partners [14], that is, their appearance as a group working together towards a shared goal. In both studies, participants collaborated with three partners. One of the partners used we, we both, or we all to refer to the collaborators. Participants had to infer the speaker’s intended referent. First, developmental change in was predicted in the we condition, only. Specifically, 2-year-olds were predicted interpret we like we both (i.e., dyadic interpretations), whereas 4-year-olds were predicted to interpret we like we all (i.e., group interpretations). Second, it was predicted that increasing the group entitativity of participants’ partners would increase participants’ tendency to form group interpretations of we, only, and not we both or we all. This prediction was made because we is, as noted above, formally ambiguous with respect to speaker referential intentions. Thus, it was predicted that altering contextual cues to increase the group entitativity of participants’ partners would pull for participants to assign group interpretations to we. However, because we all and we both are not formally ambiguous in this way, no such effect was expected for those forms. Analyses presented in Study 1 and Study 2 investigated the first prediction. Analyses presented in the Comparison of Study 1 and Study 2 investigated the second prediction. Both studies also investigated effects of sex on interpretation of we. Previous studies of pronoun development suggest that males may form group interpretations of we more often than females [11]. Complete code and data used in the analyses reported in this article are freely available at OSF (https://osf.io/5cs9p/?view_only=e3c045f73d2b4ef08897b4f4718a9cd0). The experimenter scripts used in the procedures of the present studies are available at the link, too.

Study 1

Methods

Participants.

There were 71 participants. The final sample included 64 participants, 32 2-year-olds (median = 2.54 years, range = 2.25–2.74 years, 18 males) and 32 4-year-olds (median = 4.60 years, range = 4.29–4.76 years, 13 males). Seven participants had all three trials excluded (N = 3 2-year-olds, N = 4 4-year-olds; exclusion criteria below). Caregivers predominantly indicated that households made more than $100,000 per year (N = 42) or between $60,000-$100,000 per year (N = 17). Caregivers predominantly identified participants as White (N = 49) or Biracial-Asian/White (N = 7). Caregivers received a $10 Amazon gift card. Participants received a certificate. Study design and procedure were approved by the Duke University Campus Institutional Review Board (protocol 2021–0604). Participants were sampled from October 8, 2021 to March 16, 2022. Written informed consent was obtained from participants’ parents or legal guardian prior to participation in the procedure. Participants were minors and were typically developing (per caregiver statement).

Design.

A three-level independent variable was manipulated within-subjects. The levels of this variable were labelled we, we both, and we all. The we condition was presented first. The order of presentation of the we both and we all conditions was counterbalanced between participants. This was intended to preempt carryover effects (e.g., from we all to we). There were seven warmup trials and three test trials. The order of presentation of the warmup trials was counterbalanced between participants (individuals first or groups first; see below). The procedure was remotely moderated [15] via Zoom [16]. Specifically, the experimenter (E) interacted with participants while the latter’s “interactions” with their puppet partners were, in fact, with pre-recorded videos surreptitiously controlled by E and contingent on participant behavior.

Materials.

Three hand puppets (a lion, a giraffe, and a monkey), several 8 x 11-inch sheets of paper, a set of monochromatic blocks, a set of coloring markers, and a small table. A male native speaker of English voiced lion and two females voiced giraffe and monkey, respectively. No effect of puppet voice on participants’ responses was found, so analyses collapse across this variable.

Procedure.

E greeted caregivers and participants on Zoom before initiating a Zoom Setup Phase. During the Zoom Setup Phase, E showed a PowerPoint to caregivers that illustrated how their screen should appear (i.e., full, self-view hidden, appropriate volume). Caregivers were asked to remain silent during the procedure, except to refocus participants’ attention onscreen if it wavered.

The Zoom Setup Phase was followed by a Warmup Phase. In the Warmup Phase, E told participants that they will “meet some new friends.” Then, E played a video in which each of three puppets was presented individually, in sequence, onscreen. Each puppet greeted participants, in turn, by saying “Hi, I’m [e.g.] Lion!” followed by E pausing the video to say, “Hi Lion!” After greeting each partner, participants viewed a screen in which all three partners were aligned behind a table, roughly equidistant from one another. Then, E said “Hey [participant’s name], I think that Lion, Giraffe, and Monkey want to play with you, [participant’s name]!” E then stated that they have blocks that “you” (i.e., the puppets) can play with. Several images of colorful wooden blocks then appeared onscreen below the frame. Next, E asked participants whether they have their blocks. (Prior to participation, E asked caregivers to bring a set of blocks.) Participants showed or indicated the presence of their blocks. Next, E said “I think [e.g.] Lion wants to play with you! [Participant’s name], who do you think I should give blocks to?” Participants could respond freely. During the warmup, E always ‘gave’ the blocks to the puppet(s) who E suggested wanted to play with participants, regardless of how participants responded. Specifically, E said “Hm… I think I will give the blocks to [e.g.] Lion” before ‘giving’ the blocks to Lion by dragging the image of the blocks immediately in front of the Lion puppet. E then asked participants if they were ready “to start playing with” the partner(s) who possessed blocks. Once participants agreed, E played the appropriate video of the puppet(s) playing with the blocks. This video lasted approximately 10s and depicted the puppet(s) stacking a real set of blocks that resembled those depicted in the pictured blocks. Puppet(s) to whom E did not distribute blocks remained onscreen, standing still while the other puppet(s) built with the blocks. Participants were free to play with their own blocks during this time. Once the video completed, E indicated that another puppet(s) wanted to play with the participant, e.g., Giraffe. The above was repeated for each combination of puppets. That is, participants played blocks with each combination of puppets. This resulted in seven warmup trials. Counterbalanced between participants was order of warmup presentation, individuals first or groups first. The individuals first order began with participants playing with each of the three puppets, in turn, before playing with each puppet dyad, in turn, (e.g., Lion and Giraffe, etc.) and then all three puppets at once. The groups first order proceeded in reverse order. E concluded the Warmup Phase by displaying a screen in which the blocks were absent. E noted the absence of the blocks and suggested that participants color with their partners, next. In sum, the Warmup Phase introduced participants to their puppet partners and familiarized them with the task employed in the Test Phase. Specifically, the warmup familiarized participants with the structure of the task (e.g., that E could move the images onscreen to ‘give’ the toy) and demonstrated that every puppet, and combination of puppets, could receive the toy.

The Test Phase followed the Warmup Phase. E began by saying that E had to briefly leave but that, while E was gone, participants could talk to the puppets because “I think they want to color with you!” E then left the camera’s view while surreptitiously controlling the videos. There were three test trials. Each test trial featured a different speaker puppet, e.g., on Test Trial 1 the speaker puppet was Lion, on Test Trial 2 Giraffe, etc. (Fig 1). The speaker puppet was the puppet with whom participants ‘spoke’ during the trial. Specifically, on Test Trial 1, e.g., Giraffe said “Do you like to color too? [E pauses video, participant responds] Great! I love to color. Coloring is so much fun. Do you want to color? Does coloring sound fun to you?” [E pauses video, participant responds]. If participants did not respond to a question asked by the speaker puppet, E repeated the question to participants by saying that, e.g., “Giraffe asked if coloring still sounds fun to you. Does coloring still sound fun to you?” In the we condition, the speaker puppet said “Great! We can color! I think we can color! We can color together!” (The bolded terms were replaced with we all or we both in the we all condition and we both condition, respectively.) Next, the speaker puppet remarked on how there were no markers present by asserting “Oh no, wait! There are no markers! There’s no coloring without markers! I really wish there were markers, but there are no markers. If there were markers, then we could color together!” While the speaker puppet said this, E reappeared onscreen. Once the speaker puppet finished speaking, E pointed out that “I have some markers!” to color with. Then, three images of markers appeared onscreen (Fig 1), identically to how the images of blocks appeared in the Warmup Phase. The speaker puppet reacted happily to the appearance of the markers. The speaker puppet then asked participants whether they had their markers, too: “Oh, hey, wait, did you bring your markers, too?” [E pauses video, participant responds] “Great! You have your markers, too. Now, we can color! Now, we can color together!” With the markers displayed onscreen, E then asked participants “who does [e.g.] Giraffe want to color with? Who does Giraffe want to give markers to?” Participants responded freely. E clicked and dragged the image(s) of the markers to the puppet(s) who participants indicated. Once participants finished responding, the speaker puppet said “Thanks! Now, time to color!” Then, E chose the video that depicted the appropriate puppet(s) drawing with markers (i.e., those indicated by participants). E echoed that it was “Time to color!” and played the appropriate video. The video lasted approximately 10s and depicted the appropriate puppet(s) drawing on a sheet of paper placed in front of each puppet (Fig 1). Puppet(s) to whom participants did not distribute markers remained onscreen, standing still while the other puppet(s) colored. Participants colored freely. After 10s, E said “Okay, time’s up, I need my markers back!” just before the speaker puppet said “Okay, here’s your markers back” at which point the puppet(s) in the video placed their markers down (i.e., so that the speaker puppet’s speech and act of putting the markers down appeared contingent on E’s request). The second and third trials of the Test Phase were identical to the above, save for the use of we all or we both instead of we. The second and third trials were introduced by a new speaker puppet commenting to the participant that “Coloring still seems like so much fun! Do you still want to color?”

Download:

Fig 1. Timeline of a test phase trial.

https://doi.org/10.1371/journal.pone.0306556.g001

Coding.

To be included in the final sample, participants’ caregivers had to indicate that the participant’s primary language was English. Trial-level data was excluded due to sibling or caregiver interference, ambiguous/unclear participant response, or if participants did not include the speaker puppet in their response (in Test Phase trials). This latter follows from the fact that referents of uses of we, we all, or we both must include speakers. Participants could indicate their responses linguistically or nonlinguistically, e.g., via pointing. When possible, nonlinguistic responses were clarified by caregivers. Otherwise, nonlinguistic responses were excluded when ambiguous. In the final analysis, 153 trials were included (N = 64 participants) and 39 trials were excluded (N = 26 of the 64 participants). Of the 153 included trials, 38 participants contributed three trials (114 trials), 13 contributed two trials (26 trials), and 13 contributed one trial (13 trials). All data was initially coded by a research assistant who was blind to the hypotheses of the study. Reliability coding was performed on a random 25% of participants’ responses (N = 48 trials) by the first author. Responses were coded according to whether participants chose to give markers to 1, 2, or 3 puppets, and whether one of those puppets was the speaker puppet. Interrater reliability for the number of puppets to whom participants chose to distribute markers was excellent, κ = 0.96, and agreement was excellent, 96% of trials.

Data analysis plan.

Hierarchical Bayesian models were fitted to the data [17] via the Stan [18] front-end interface brms [19]. The dependent variable was dichotomous (“dyadic interpretation” or “group interpretation”). The dyadic interpretation corresponded to participants having selected only the speaker puppet to receive the markers. In contrast, the group interpretation corresponded to participants having selected the speaker puppet and at least one of the other two puppets to receive the markers. Thus, there were two group interpretations and one dyadic interpretation that participants could choose.

Main analysis. Two models were fitted and analyzed in the main analysis. A control variables model included first-order predictors of order of presentation of conditions (we all second, we both second), test trial (first, second, third), speaker puppet (giraffe, monkey, lion), and participant sex (male, female). A main model included the key predictor of condition (we, we all, we both), participant age group (2 years or 4 years), and their interaction. Both models included a random intercept of participant and random slope of condition. The control variables model formula was:

The main model formula was:

Variance components correlations were not modeled because they equaled zero due to counterbalancing. Control model predictors were sum coded. Main model predictors were treatment coded. The baseline groups in the main model were the we condition (Condition) and 2-year-olds (Age). Priors were β_intercept∼ t(10,0,1),β_predictors∼ N(0,0.75),σ∼ t(10,0,0.4). Posterior parameter estimates were characterized by their 95% highest density (HDI) and proportion of density over positively signed parameter values. For interpretational ease, qualitative assertions about the posterior evidence for positively signed values are provided. “Weak” means that between 10% and 90%, “moderate” between 5% and 10% or 90% and 95%, and “strong” that less than 5% or more than 95% of posterior mass covers values greater than 0. Stability of posterior estimates was ensured via sensitivity analysis with flatter and peakier fixed effects priors. All R-hats equaled 1.00 and effective sample sizes were adequate (following [17]).

Priors were chosen based on general domain knowledge and an understanding of the response distribution defined by the link function so as to be “weakly informative” about the most likely outcomes. This strategy is considered best practice in Bayesian modeling [e.g., 17, 20–22, provides an introduction] and has been used previously [11, 23].

Follow-up analyses. Simpler models were fitted to portions of the dataset investigated in the main analysis. All models reported included the same random effects structure as included in the model reported in the main analysis, or a simplified version thereof, as appropriate. Fixed effects structures are detailed before reporting the fitted model. A first follow-up analysis investigated associations between interpretation and age within conditions; a second investigated the effect of condition on interpretation within ages; and a third investigated interpretations against chance.

Results

Data.

The distribution of participant interpretations is depicted by condition in Fig 2. Two- and 4-year-olds generally favored dyadic over group interpretations. That is, Study 1 participants tended to distribute markers to the speaker puppet, only, and not to the other two puppets. Interestingly, 2-year-olds in the we all condition made group interpretations relatively more often than any other condition x age pair, with 9 of 20 2-year-olds choosing group interpretations of we all. S1 Fig displays the dependent variable as a trichotomous outcome (i.e., participants’ distribution of markers to 1, 2, or 3 puppets).

Download:

Fig 2. Trial-level data by condition.

Trial-level distribution of participants’ interpretations, by condition, in Study 1 and Study 2 (top and bottom facets, respectively). Study 1 facet displays data for Test Trials 1, 2, 3. Study 2 facet displays data for Test Trials 1 and 2 (because of a reliable association between the DV and the order of the we both and we all conditions; see Study 2).

https://doi.org/10.1371/journal.pone.0306556.g002

Main analysis.

The posterior control variables model was assessed, first. Most predictors were unreliably related to interpretations (see S1 Table for posterior parameter estimates). The only exception was a reliable posterior association that suggested that participants chose group interpretations less often when giraffe, rather than Monkey, was the speaker puppet, -0.48, [-1.17,0.19], Pr(β > 0.00|D) = .08. There was weak evidence that females chose group interpretations more often than males (Fig 3), 0.11, [-0.64,0.89], Pr(β > 0.00|D) = .61. S2 Fig reports participant-level data, by sex.

Download:

Fig 3. Trial-level data by sex.

Trial-level distribution of participants’ interpretations, by participant sex, in Study 1 and Study 2 (top and bottom facets, respectively). Study 1 facet displays data for Test Trials 1, 2, 3. Study 2 facet displays data for Test Trials 1 and 2 (because of a reliable association between the DV and the order the we both and we all conditions; see Study 2). S2 Fig depicts participant-level data.

https://doi.org/10.1371/journal.pone.0306556.g003

The posterior main model was assessed next. Developmental change was predicted in interpretation of we. Specifically, 4-year-olds were predicted to choose group interpretations more often than 2-year-olds. It was also predicted that 2-year-olds would interpret we like we both, whereas 4-year-olds would interpret we like we all.

Based on a model fitted to the data reported in Fig 2, mixed support for these predictions was found. There was strong evidence that 2-year-olds in the we condition chose group interpretations below chance levels, -1.70, [-2.89,-0.63], Pr(β > 0.00|D) = .00, and 4-year-olds did not reliably diverge from this pattern, 0.16, [-0.93,1.28], Pr(β > 0.00|D) = .62. That is, 2- and 4-year-olds favored dyadic interpretations of we. The lack of developmental change represented in this latter finding discords with our predictions. Nonetheless, 2-year-olds made group interpretations of we both at similarly depressed rates, -0.03, [-1.06,0.97], Pr(β > 0.00|D) = .48. This finding accorded with predictions. Taken with the first finding, 2-year-olds favored dyadic interpretations at similar rates in the we and we both conditions. Moreover, there was some evidence that 2-year-olds chose group interpretations of we all more often than of we, 0.69, [-0.32,1.68], Pr(β > 0.00|D) = .91. This finding accorded with predictions, though was not particularly robust. There was weak evidence that 4-year-olds interpreted we all and we both, relative to we, differently than 2-year-olds, we all: -0.01, [-1.15,1.15], Pr(β > 0.00|D) = .49; we both: 0.02, [-1.11,1.12], Pr(β > 0.00|D) = .51. Altogether, 2- and 4-year-olds, in situations like that of the present study, form largely similar interpretations of we, we both, and we all. Specifically, young children tend to interpret these forms referring to dyads, not groups. Interestingly, 2-year-olds’ interpretations distinguished between we and we all more than those of 4-year-olds.

Follow-up analyses.

Three sets of follow-up analyses were conducted. For readers’ ease, only the proportion of posterior mass greater than 0 is reported. Summary statistics of marginal posteriors are reported in S2 Table. A first set of follow-up analyses investigated developmental change by condition. One model was fitted to the data in each condition. Each model included a fixed effect of age group, only. The strongest evidence for developmental change was found in the we condition. 81% of posterior samples indicated that 4-year-olds chose group interpretations more often than 2-year-olds, Pr(β > 0.00|D) = .81. This pattern accorded with the predictions of the current study but was less robust than anticipated. As expected, there was little evidence for reliable associations with age in the other conditions, we both condition: Pr (β > 0.00|D) = 55; we all condition, Pr(β > 0.00|D) = .33. Overall, this first set of follow-up analyses broadly aligns with the predictions of this study. The strongest evidence of developmental change was in participants’ interpretation of we. Specifically, there was some evidence that suggested 4-year-olds chose group interpretations of we more often than 2-year-olds.

A second set of follow-up analyses investigated the effect of condition within age group. One model was fitted to the data in each age. Both models included a fixed effect of condition, only. There was weak evidence that participants chose group interpretations at different rates in the we and we both conditions, 2-year-olds: Pr(β_condition>0.00|D) = .55, 4-year-olds: Pr(β_condition>0.00|D) = .38. That is, the distribution of 2-year-olds’ interpretations of we and we both were similar (as predicted), but so were 4-year-olds’ (unpredicted). This pattern accords with the findings of the main analysis, above. There was weak evidence that 4-year-olds chose group interpretations at different rates in the we and we all conditions, Pr(β_condition>0.00|D) = .73. Thus, as predicted, 4-year-olds’ interpretations of we resembled their interpretations of we all. In contrast, there was moderate evidence that 2-year-olds chose group interpretations at a higher rate in the we all condition than in the we condition, Pr(β_condition>0.00|D) = .92. Though less robust than expected, this finding accorded with predictions– 2-year-olds chose group interpretations of we at lower rates than of we all. Overall, this second set of follow-up analyses suggested mixed evidence for the predictions of this study. While 2-year-olds’ interpretations of we resembled their interpretations of we both more than we all (as predicted), 4-year-olds’ interpretations of we were similar across conditions (not predicted).

A third set of follow-up analyses compared the distribution of participants’ responses against chance. Three models were fitted to each condition * age group pair (i.e., there were six models in total). Each model only modeled the intercept (baseline) rate of group interpretations. Conceptually, “chance” responses in the referring situation captured in the procedure can be defined in two ways. First, the chance rate of choosing group interpretations was defined as 0.50. This definition stipulated that 3- and 4-person groups were not perceived by participants as distinct interpretations. Under this definition of chance, 2- and 4-year-olds mostly picked group interpretations less often than chance, 2-year-olds: Pr(β_we > 0.00|D) = .00; Pr (β_{we both} >0.00|D) = .01; 4-year-olds: Pr(β_we>0.00|D) = .02; Pr(β_{we both}>0.00|D) = .01; Pr (β_{we all}>0.00|D) = .08. Unexpectedly, the sole exception to this pattern was 2-year-olds in the we all condition, who did not reliably choose dyadic over group interpretations, Pr(β_{we all}>.00|D) = .34. Next, the chance rate of choosing group interpretations was defined as 0.66. This analysis assumed that 3-person and 4-person groups were perceived by participants as distinct interpretations. Under this definition of chance, 2- and 4-year-olds picked group interpretations below chance levels, 2-year-olds: Pr(β_we>0.69|D) = .00; Pr(β_{we both}>0.69|D) = .00; Pr(β_{we all}>0.69|D) = .02; 4-year-olds: Pr(β_we>0.69|D) = .00; Pr(β_{we both}>0.69|D) = .00; Pr(β_{we all}>0.69|D) = .00. Overall, this third set of follow-up analyses suggested that participants generally favored dyadic interpretations in Study 1. Interestingly, however, 2-year-olds did not reliably favor dyadic or group interpretations of we all (while though 4-year-olds favored dyadic interpretations).

In sum, Study 1 participants typically interpreted we, we both, and we all dyadically. The follow-up analyses provided mixed support for the study predictions. Notably, there was some evidence that 4-year-olds made group interpretations of we, and not we both or we all, more often than did 2-year-olds. This key finding accorded with the predictions of the current study.

Study 1 Discussion

Study 1 investigated children’s interpretation of we. To do this, participants’ behavior was investigated in a context in which collaboration relied upon participants interpreting uses of we, we both, or we all (this was thus an “inclusive” we). Based on a hypothesized groupminded shift at 3 years [7], 2-year-olds were predicted to prefer dyadic interpretations of we. In contrast, 4-year-olds were predicted to prefer group interpretations of we. That is, 2-year-olds’ interpretations were predicted to resemble their interpretations of we both and 4-year-olds’ those of we all. However, participants mostly favored dyadic interpretations (one exception was 2-year-olds’ interpretation of we all, where group interpretations were comparatively frequent). There was little evidence for associations between other predictors and participant interpretations.

What explanation might account for the lack of evidence for the present set of hypotheses? Perhaps the context of the procedure used in Study 1 pulled too strongly for dyadic interpretations, resulting in a floor effect. If this were the case, a context that pulls more for group interpretations may increase the size of observed behavioral differences between conditions.

Study 2

Study 2 implemented a procedure that was essentially identical to that of Study 1. Changes targeted the group entitativity of participants’ puppet partners. These were enacted by modifying contextual cues to increase the partners’ appearance as a “task group” [14]. Task groups are constituted by partners with shared goals and plans. The puppets’ appearance as a task group in the Study 2 contrasted with their appearance as a “loose association” [14] in Study 1, e.g., individuals in close spatial proximity. Five- and 6-year-olds rate task groups as more like “real groups” compared to loose associations [14] and so, e.g., more often expect task group partners to help one another. Study 2 predictions were identical to those of Study 1.