Edited by: Bert Timmermans, University Hospital Cologne, Germany
Reviewed by: Bert Timmermans, University Hospital Cologne, Germany; Martin J. Pickering, University of Edinburgh, UK; Ivana Konvalinka, Aarhus University, Denmark
*Correspondence: Ivan Toni, Donders Institute for Brain, Cognition and Behaviour, Radboud University Nijmegen, Kapitellweg 29, Nijmegen, Netherlands. e-mail:
This is an open-access article distributed under the terms of the
When established communication systems cannot be used, people rapidly create novel systems to modify the mental state of another agent according to their intentions. However, there are dramatic inter-individual differences in the implementation of this human competence for communicative innovation. Here we characterize psychological sources of inter-individual variability in the ability to build a shared communication system from scratch. We consider two potential sources of variability in communicative skills. Cognitive traits of two individuals could independently influence their joint ability to establish a communication system. Another possibility is that the overlap between those individual traits influences the communicative performance of a dyad. We assess these possibilities by quantifying the relationship between cognitive traits and behavior of communicating dyads. Cognitive traits were assessed with psychometric scores quantifying cooperative attitudes and fluid intelligence. Competence for implementing successful communicative innovations was assessed by using a non-verbal communicative task. Individual capacities influence communicative success when communicative innovations are generated. Dyadic similarities and individual traits modulate the type of communicative strategy chosen. The ability to establish novel communicative actions was influenced by a combination of the communicator's ability to understand intentions and the addressee's ability to recognize patterns. Communicative pairs with comparable systemizing abilities or behavioral inhibition were more likely to explore the search space of possible communicative strategies by systematically adding new communicative behaviors to those already available. No individual psychometric measure seemed predominantly responsible for communicative success. These findings support the notion that the human ability for fast communicative innovations represents a special type of complex collaborative activity.
Human communication relies heavily on complex skills acquired early in life (i.e., language), but we are also endowed with the ability to build new communicative systems from scratch when necessary. Dramatic examples of the latter ability are “home-sign” systems that can be developed by deaf children of hearing parents who have been deprived of access to conventional language (Goldin-Meadow,
Early descriptive studies of dialog (Clark and Carlson,
In this study, we quantified inter-individual variations in communicative skill by means of a controlled and validated experimental setting, the Tacit Communication Game (TCG) (Newman-Norlund et al.,
We considered a set of parameters that have been previously validated and used to characterize various cognitive and social abilities. These parameters were chosen on the basis of the following considerations. First, when solving a communicative problem, people need to identify not only what is ambiguous according to their viewpoint, but also what is ambiguous to their communicative partner. These might be different components of the problem. This aspect of interactive intelligence resembles abilities that have been proposed in the human emotional domain. For instance, empathy refers to the ability to identify other's feelings and emotions and to respond to these in an appropriate way. It could be that highly empathic individuals are better able to establish new communicative conventions. The empathizing quotient (EQ) is one way to measure empathy (Baron-Cohen and Wheelwright,
We tested 54 participants. They were right-handed male students (18–27 years), with normal or corrected to normal vision. This group of participants was selected from a larger pool of 285 subjects, contacted by means of e-mails and flyers, who completed two questionnaires, the EQ and SQ-R (Wheelwright et al.,
The participants were assigned into 27 communicative pairs, arranged in order to cover different combinations of EQ and SQ-R scores. Because type S contained the largest group and we wanted to obtain a good spread, we further divided type S into two sub-groups by introducing an extra boundary in the middle of type S. Pairs were then generated by randomly choosing participants from two different types out of the five categories. All participants gave informed consent according to the institutional guidelines of the local ethics committee (CMO region Arnhem-Nijmegen, The Netherlands). The participants received a financial payment or course credits for their participation.
We considered seven psychometric questionnaires, requiring forced-choice responses. Two questionnaires (EQ and SQ-R) were administered through a website, during subjects selection (see above), one to six months before performance of the TCG (Part I). Two questionnaires (Raven, NCS) were administered in the laboratory immediately after performance of the TCG (Part II). Three questionnaires (IRI, COSI, BIS/BAS) were administered at home, approximately 8 months after performance of the TCG (Part III). Part III of the experiment was conducted by forty participants (20 senders, 15 complete pairs) who returned the questionnaires.
Details on the construction of the EQ and SQ-R can be found in Baron-Cohen and Wheelwright (
The experiment was structured in three-parts. Part I was the web-based subject selection (see above). Part II took place at the Donders Institute for Brain, Cognition and Behaviour (Nijmegen, The Netherlands) and it consisted of a TCG training session, a TCG testing session, and a psychometric session, in this order. Part III involved completing three more questionnaires (IRI, COSI, BIS/BAS; see above). In the following sections we focus on the procedures of Part II. During the TCG training session (duration: 30 min), subjects were familiarized with the TCG. During this session, each communicative pair generated and learned a communicative rule for solving a set of TCG problems (see below). During the TCG testing session (duration: 40 min), each communicative pair solved both learned and new TCG problems. During the psychometric session (duration: 30 min), subjects were administered the Raven's test and the NCS in consecutive order. Below we elaborate on the procedures followed during the TCG training and testing sessions. In both sessions, the participants could not see or hear each other. Each participant used Logitech hand-held controllers to move an object shown on a computer monitor. The four face buttons of the controller were used for movements to the left, right, up, and down, two shoulder buttons were used to rotate the token clockwise and counter-clockwise, and another shoulder button was used as a start and end button. The TCG was programmed using Presentation version 10.1 and was run on a Windows XP personal computer.
The TCG training session was structured in three sub-sessions, sequentially presented. First, the participants were individually familiarized with the experimental setup (40 trials). Namely, each participant saw a blue triangle (the target) with a random rotation at a certain location on the game board. After the participant pressed the start button, the target disappeared from the game board, and a triangle that pointed upward appeared in the center of the game board (player's token). The participant had to position his token in the location and orientation of the target previously shown, by pushing the appropriate buttons on the hand-held controller. After the participant matched the target with his token, a new target was shown, in a pseudo-randomly chosen position and orientation on the game board.
Second, the participants were jointly introduced to the basic procedures of the TCG (10 trials). Each participant of a communicative pair was assigned the role of either sender or receiver, and he kept this role during the remainder of training and testing sessions. During this training sub-session, the participants were asked (by means of written instructions) to use their tokens to match the targets configuration shown on the game board (see Figure
Third, the participants were jointly introduced to the communicative aspects of the TCG (at least 25 trials). This training sub-session was identical to the second sub-session, apart from one important difference, namely only the sender could see the targets configuration. Each communicative pair was informed about this change with written instructions. This change meant that, to successfully complete a trial, the sender had to communicate to the receiver the location and in some cases the orientation of the receiver's token. Given the structure of the TCG, the sender could communicate this information to the receiver only by moving his own token around the game board. The sender was encouraged to think how to do so before pressing the start button.
If a communicative pair made a mistake during the last ten trials of this training sub-session, they had to complete ten extra trials until they had performed ten correct trials sequentially. This type of communicative problems was labeled as OLD, since by the end of this training sub-session each communicative pair was successful in solving these problems with a consistent communicative strategy.
To investigate the establishment of new shared communicative actions, we compared a situation in which communicative rules were already established (OLD problems) with a situation in which a communicative rule was yet to be established (NEW problems). During the testing session the pair played a version of the TCG consisting of such OLD and NEW trials. The old trials of this session were similar to the OLD trials of the third part of the training session. The similarity was based on the fact that the same communicative strategy could be applied. In contrast, the NEW trials entailed different problems. Namely, the sender had to indicate both location and rotation of the receiver's token with his own token, although the shape of the sender's token contained less rotation possibilities than the shape of the receiver's token (see Figure
At the start of the testing session, the players received a short written instruction with a summary of the most important game features experienced during the previous training session. These points were: only the sender can see the goal configuration; after pressing the start button you have 5 s to move; both location and rotation of the token need to be correct; try to be quick, but more importantly try to get as many trials correct as possible; press the end button after you have finished moving your token (for the receiver only).
The experimental session contained 84 trials; half were OLD trials, half were NEW trials. No more than either three OLD or three NEW trials were presented sequentially. For the OLD trials, the presentation of the shape combinations was intermixed. For the NEW trials, there were four shape combinations (Figure
Group differences between psychometric scores of senders and receivers were assessed with a One-Way ANOVA. The relations between the psychometric scores of each subject were investigated by means of bivariate correlation analyses.
The relation between psychometric scores of participants within a pair was quantified by means of a difference score (defined as the absolute value of the difference between sender and receiver scores), an indicator of the similarity of the two individuals that constitute a pair. Lower difference scores reflect larger similarities on that particular psychometric test.
For each pair, we considered two indices of TCG performance, i.e., mean accuracy across the testing session, and its rate of change. The mean accuracy of each pair was analyzed using repeated-measures ANOVA (threshold,
We tested for the influence of communicative strategy by using an ANOVA considering the effect of those strategies (categorized as COARSE and REFINED, see section “Results” for a full description) on success rate and frequency of occurrence. Each trial was replayed offline and categorized accordingly. We used repeated-measures ANOVAs to test whether COARSE or REFINED strategies influenced success rate and strategy occurrence.
To test whether the psychometric scores of the senders and the receivers influence performance and strategy choice during the NEW problems, linear regression analysis were performed. First, only the psychometric scores assessed for all participants (from part I and II) were included. Second, the psychometric scores of part III were included as well (15 complete pairs). The two measures of performance used as dependent variable are the mean accuracy scores and the learning rate of each pair. Strategy choice was defined by the occurrence of each strategy group. An estimate of change in performance (standardized beta value) was obtained by means of linear regression analyses with accuracy (moving average) as dependent variable and trial number (log transformed) as independent variable. The independent variables (psychometric scores) were entered into the linear regression model following a stepwise fashion, meaning that only those independent variables that explained a significant (and unique) part of the variance of the dependent variable were entered into the model. The significant models (
Overlap/differences in psychometric profiles of participants within each pair were quantified by creating “mismatch values” for each psychometric measure, defined as the absolute difference between the score of the sender and the receiver. Mismatch values of different psychometric scores were then entered in a linear regression model following a stepwise inclusion procedure.
One pair was excluded from analyses because of their poor performance on both OLD and NEW trials (79%, 0% correct, respectively), indicating an inability in establishing and maintaining a communicative system, especially on the NEW trials. The idiosyncratic behavior of this pair is described in Box
The receiver of one pair did not understand that he had to turn, but at the end of the experiment he did indicate that he knew he could turn. He did not turn his shape in any of the trials. Starting from the first NEW trial, the sender was using different strategies to indicate to the receiver that he needed to turn. The sender used 6 different strategies, but strategy IV was used the most (25 of the 42 times). If the sender had a rectangle he would rotate it, but he also moved along the whole row or column of the game board to indicate a pointing direction.
The receiver had the lowest Raven score of all participants. This corresponds with findings from other pairs, namely pairs composed of a receiver with low fluid intelligence are less effective at establishing a novel communicative system.
Mean accuracy scores showed a significant effect of problem,
During the NEW trials the pairs had to develop particular strategies to convey a message about location and rotation of the receiver's shape. These communicative strategies were divided into two main groups: COARSE, in which the desired rotation of the receiver's shape was indicated with little or no information, and REFINED, in which more elaborate movements indicated the rotation. The COARSE group consisted of three strategies: (1) the sender indicated the position of the receiver's shape only, ignoring its rotation; (2) the sender used the direction in which he moved away from the middle of the game board (sender's start position) to the receiver's target position as a marker for the desired orientation of the receiver's shape; (3) the sender used the direction in which he moved away from the receiver's target position to his own target position as a marker for the desired orientation of the receiver's shape. The REFINED group consisted of five strategies that explicitly indicate the movement and rotation of the token; (4) the sender moved to the receivers location, after which he moved one square in the pointing direction and back to indicate the desired rotation; (5) the sender first indicated the desired rotation of the receiver's token by moving in that direction (and back to the starting position) before moving to the receiver's location; (6) when the sender had a rectangle token, he indicated the desired rotation by rotating his rectangle the desired amount of rotations; (7) after moving to the receiver's desired position, the sender indicated rotation by moving his token along the whole row or column of the receiver's goal position; (8) the sender indicated rotation by imitating a rotation, namely moving his token along a square across the whole board (e.g., one square up, one to the right, one down and one to the left. We also considered two additional, independent categories; (9) other idiosyncratic strategies observed for a few trials only; (10) no definite strategy.
During the game, different pairs used different strategies, in different proportions, as illustrated in Figure
Some strategies had a higher success rate than others, and there were also differences in the number of times a strategy was used (Table
I | 0.20 | 17.0 |
II | 0.93 | 3.7 |
III | 0.91 | 3.2 |
IV | 0.53 | 41.0 |
V | 0.93 | 3.9 |
VI | 0.39 | 10.0 |
VII | 0.39 | 6.4 |
VIII | 0.76 | 1.9 |
IX | 0.67 | 8.0 |
X | 0.09 | 4.9 |
Figure
There was no significant relation for the sender. The pair's (dis)similarity did not influence overall performance or rate of change.
To investigate the influence of individual cognitive traits on usage of different communicative strategies, we considered the occurrence of COARSE and REFINED strategies. There was a negative relation between the Raven score of the senders and their use of COARSE strategies,
Finally, we investigated the influence of the overlap in psychometric profiles of participants within each pair on the usage of different communicative strategies. This analysis was based on the psychometric measurements of part III, completed by 15 pairs only. Participants mismatch on the SQ-R and BIS scores decreased the chances of using a REFINED strategy (SQ-R:
The aim of this study was to investigate the psychological traits leading to inter-subject variation in communicative skills. We operationalized communicative skill as the ability to build shared communicative innovations. We describe qualitative and quantitative indexes of communicative performance in pairs of participants engaged either in applying previously established communicative conventions, or in establishing new shared conventions. Three observations indicate that the experimental procedures were effective in capturing communicatively relevant variability in subjects' performance. First, when faced with new communicative problems, subjects' pairs progressed from communicative failure (early in the experiment) toward mutual understanding (late in the experiment). This improvement in communicative performance occurred despite the expansion of the set of problems faced by the participants, as NEW trials were progressively introduced, and previously established communicative conventions might have become ineffective. Second, there were large differences in the ability of the different pairs to establish shared communicative strategies. Some pairs quickly established a novel successful communicative strategy, while others had more difficulty in doing this. Third, pairs differed in their inclination to change communicative strategies during the course of the experiment, a sign of mutual adjustment during social interactions (Clark,
There are three main findings in this study. First, the ability of a pair to successfully establish novel communicative actions was influenced by a combination of the sender's need for cognition (NCS) and of the receiver's Raven's score. It is known that the learning strategies of individuals with high NCS are more flexible by virtue of being less biased by surface information (Cacioppo et al.,
Second, pairs with comparable systemizing abilities or behavioral inhibition were more likely to use refined communicative strategies. More precisely, pairs with high systemizing scores and particularly averse to negative feedback appear more likely to explore the search space of possible communicative strategies by systematically adding new communicative behaviors to the available conventions, i.e., safely building on pre-existing behaviors rather than violate pre-existing conceptual pacts (Brennan and Clark,
Third, measures of empathy and reward-related tendencies (BAS) were not able to account for significant portions of inter-subject variability in communicative performance. This negative result complement the finding of a previous study that, using a similar communicative challenge, reported a relation between empathy scores and audience design abilities (Newman-Norlund et al.,
It might be argued that the findings of this study are not relevant for understanding how humans try to modify the mental state of another agent according to their intentions. For instance, the same findings might have been obtained when the communicator were interacting with an artificial agent producing a pre-defined set of behaviors. In fact, collateral evidence clearly indicate that subjects engaged in this game consider the mental state of the other participant, as indicated by the presence of audience design effects (Newman-Norlund et al.,
We show that inter-individual variability in communicative skills is partially accounted for by a number of cognitive traits. Individual capacities influence communicative success, when communicative innovations are generated, while dyadic similarities as well as individual traits modulate the type of communicative strategy chosen. Given that no individual psychometric measure was predominantly responsible for communicative success, we infer that general-purpose cognitive abilities are unlikely to fully account for human communicative skills. Existing indexes of cognitive abilities fail to adequately capture elements of those skills. Accordingly, it appears relevant to develop novel and quantitative indexes of communicative skills, analogous to those recently introduced to quantify social skills in children and non-human primates (Herrmann et al.,
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.