Evolution and Human Behavior 36 (2015) 313–322

Contents lists available at ScienceDirect

Evolution and Human Behavior

journal homepage: www.ehbonline.org

Original Article

The ease and extent of recursive mindreading, across implicit and

explicit tasks
Cathleen O’Grady a,1, Christian Kliesch b,1, Kenny Smith a, Thomas C. Scott-Phillips c,⁎,2
a
School of Philosophy, Psychology, and Language Sciences, University of Edinburgh
b
Department of Psychology, Lancaster University, Lancaster LA1 4YF
c
Evolutionary Anthropology Research Group, Durham University

a r t i c l e i n f o a b s t r a c t

Article history: Recursive mindreading is the ability to embed mental representations inside other mental representations e.g. to
Initial receipt 6 May 2014 hold beliefs about beliefs about beliefs. An advanced ability to entertain recursively embedded mental states is con-
21 December 2014 sistent with evolutionary perspectives that emphasise the importance of sociality and social cognition in human
Final revision received 22 January 2015
evolution: high levels of recursive mindreading are argued to be involved in several distinctive human behaviours
and institutions, such as communication, religion, and story-telling. However, despite a wealth of research on
Keywords:
Mindreading
first-level mindreading under the term Theory of Mind, the human ability for recursive mindreading is relatively
Recursive mindreading understudied, and existing research on the topic has significant methodological flaws. Here we show experimentally
Mentalizing that human recursive mindreading abilities are far more advanced than has previously been shown. Specifically, we
Theory of mind show that humans are able to mindread to at least seven levels of embedding, both explicitly, through linguistic de-
Metarepresentation scription, and implicitly, through observing social interactions. However, our data suggest that mindreading may be
Intentionality easier when stimuli are presented implicitly rather than explicitly. We argue that advanced mindreading abilities are
Social cognition to be expected in an extremely social species such as our own, where the ability to reason about others' mental states
is an essential, ubiquitous and adaptive component of everyday life.
© 2015 Elsevier Inc. All rights reserved.

“Humans have the ability to represent representations… This meta- and commonly held view is that high-level recursive mindreading (i.e.
representational ability is as distinctive of humans, and as important beyond first or second level) is cognitively demanding, and perhaps be-
in understanding their behaviour, as is echolocation for bats” yond normal human abilities (e.g. Clark, 1996; Gómez, 1994). Yet theo-
(Sperber, 1997, p.69). retical explanations of many important human behaviours and
institutions, such as communication, religion, story-telling, and culture
itself either argue or assume that humans can and do process high levels
of recursive mindreading routinely and without difficulty (Dunbar,
2003, 2008; Sperber, 2000a; Tomasello, 2008). Furthermore, our natu-
1. Introduction
ral ecology is a social one, in which both collaboration and competition
are everyday activities (Byrne & Whiten, 1989; Dunbar, 2003; Hum-
Mindreading is the ability to mentally represent others' mental rep-
phrey, 1976). In such an environment, the ability to monitor and
resentations. It is also known as mental metarepresentation, or theory
manage one's social environment, by reasoning about the motives and
of mind. Recursive mindreading is the ability to embed further levels
intentions of others, keeping track of others' relationships, deciding
of mental representation inside existing mental representations (e.g. I
who to trust, and so on, is of critical importance. From this perspective,
think0 that you believe1 that he thinks2 that she wants3… and so on;
we should expect humans to be able to process mental
subscripts count the number of metarepresentations 3). An intuitive
(meta)representations with relative ease, at least when those represen-
tations are encountered within this social ecology.
⁎ Corresponding author. Department of Anthropology, Dawson Building, South Road, There is a large literature on various aspects of first-level
Durham, CH1 3LE.
mindreading (e.g. the ability to reason about the mental state of
E-mail address: [email protected] (T.C. Scott-Phillips).
1
Joint first authors. another: I think0 that you believe1 some proposition). This includes,
2
@tscottphillips (twitter). most prominently, its development in children (Baillargeon, Scott, &
3
There is inconsistency in the mindreading literature regarding how to count the levels. He, 2010; Wellman, Cross, & Watson, 2003), its role in some social cog-
Some studies include the focal individual's perspective; others exclude it. This is the
nitive disorders (Baron-Cohen, 1995; Chevallier, Kohls, Troiani, Brodkin,
difference between counting the number of representations (I think1 that Mary thinks2…)
or only the number of metarepresentations (I think0 that Mary thinks1…). Most adult
& Schultz, 2012), and its presence or absence in non-human primates
studies use the former practice, whereas the developmental literature uses the latter. (Call & Tomasello, 2008; Premack & Woodruff, 1978). In contrast,
We follow this latter practice in this paper. there is far less research dealing specifically with recursive

http://dx.doi.org/10.1016/j.evolhumbehav.2015.01.004
1090-5138/© 2015 Elsevier Inc. All rights reserved.
314 C. O’Grady et al. / Evolution and Human Behavior 36 (2015) 313–322

mindreading, despite its theoretical importance for human social life. Roberts, 2010), in studies designed to investigate the relationship
The handful of studies in adults that do exist report a prominent drop between mindreading and various cognitive disorders (e.g. Frith
in performance after four levels of recursive mindreading (Kinderman, & Corcoran, 1996; Kerr, Dunbar, & Bentall, 2003), and in studies de-
Dunbar, & Bentall, 1998; Lyons, Caldwell, & Shultz, 2010; Stiller & signed to investigate the relationship between mindreading ability
Dunbar, 2007). There is also a small literature on second- and third- and other aspects of social psychology, in both adults and children
level tasks in children's development, which finds that the ability to per- (e.g. Henzi et al., 2007; Liddle & Nettle, 2006; Nettle & Liddle, 2008;
form these higher-level tasks emerges later in development than Paal & Bereczkei, 2007; Sylwester, Lyons, Buchanan, Nettle, & Rob-
competence in first-level tasks (see Miller, 2009 for a review). erts, 2012). The IMT involves stories which are read aloud to partic-
However, previous research on high-order recursive mindreading ipants, followed by a series of true-or-false or forced-choice
may have significantly underestimated the extent of human recursive mentalising questions, designed to test participants' understand-
mindreading abilities, for at least two reasons. First, the stimuli used ing of the levels of recursive mindreading involved. Control ques-
in previous studies have a number of shortcomings serious enough to tions are designed to test participants' ability to remember details
raise doubts about their validity. We detail these issues in the next sec- of the stories that are unrelated to mental states, but which contain
tion. Second, previous studies tested recursive mindreading ability only a matched number of elements to be remembered. We analysed the
explicitly, by presenting stimuli either as text to be read, or narration to stimuli used in the IMT 5 and identified five main problems that,
be heard, and by testing understanding with direct questions, and not collectively, are significant enough to cast doubt on the conclusions
implicitly, by presenting stimuli as social events to be observed, and drawn in these studies regarding the extent of human recursive
testing understanding by measuring reactions to those events. It may mindreading ability. We describe these issues in the following par-
be the case, especially given the ecological perspective outlined above, agraphs. A full breakdown of which of these criticisms apply to
that human mindreading abilities are fully expressed only when they which questions in the IMT is provided as supplementary informa-
are employed within social contexts i.e. when encountered implicitly tion (available on the journal's Website at www.ehbonline.org).
(as opposed to being encountered as explicit, disembodied descriptions
of those same contexts). This possibility is supported by findings in the
developmental literature which shows that children pass implicit first- 2.1. Broken conceptual chains
level mindreading tasks (false belief tasks) far earlier than they do
equivalent explicit tasks: around the first birthday vs. around the fourth In some cases, mental questions are constructed in a way that
birthday (see Baillargeon et al., 2010 for a review of implicit false-belief allowed them to be processed in ‘chunks’, rather than as a single
tasks). Precisely what causes this dramatic difference is an unresolved metarepresentational unit. For example, the following sentence is
issue in developmental psychology, but whatever the explanation, intended to test fourth level mindreading: ‘Simon imagined1 that
these results show that the mode of presentation can make a dramatic Betty wanted2 to marry Edward but that Edward really wanted3 to
difference to performance, at least in children. Based on this finding, marry Susan, whom Jim would like4 to have married’ (here and else-
we might expect that adult performance on recursive mindreading where in this paragraph, we have omitted the participant's own mental
tasks could also be facilitated by implicit presentation. state, which, if we had included it, would have had the subscript 0 (i.e.
In sum, recursive mindreading plays an important role in explana- ‘The participant believes0 that Simon imagined1…’)). However, this sen-
tions of many major human behaviours, yet there are reasons to think tence does not contain one continuous chain of mental representations.
that we may not currently know or appreciate the full extent of this abil- Rather, it consists of three statements, joined by logical relationships:
ity in adult humans. In this paper, we present a new study of recursive (i) Simon imagined1 that Betty wanted2 to marry Edward; (ii) Simon
mindreading, which has two major advances on previous research. imagined1 that Edward really wanted2 to marry Susan; and (iii) Simon
First, we use new stimuli designed to avoid the various methodological imagined1 that Jim would like1 to have married Susan. Consequently,
issues we have identified in previous studies (detailed below). Second, constructions of this sort do not test 4th level mindreading; they test
we use a 2 × 2 design of implicitly and explicitly presented stories, the conjunction of multiple cases of 2nd level mindreading. 13 of 50
crossed with implicitly and explicitly presented questions. As such, the mental questions in the IMT are constructed like this.
key novelty here is the use of implicit stimuli, which have not previously
been used in the study of recursive mindreading in adults. We expected
that, at least in conditions featuring implicit presentation, participants 2.2. Simple substitution
would succeed at recursive mindreading tasks at levels higher than
those reported in previous studies. Correspondingly, our design in- Some stimuli are constructed in such a way that the entire sentence
cludes questions of up to seven levels of mental metarepresentation, did not need to be processed in order to be answered correctly. An exam-
three levels higher than the typical level of successful performance in ple is the forced choice between ‘The girl whose car Simon works on
previous tasks (e.g. Kinderman et al., 1998; Lyons et al., 2010; Stiller & practices dance with the person who is a loan officer in Edward's bank’
Dunbar, 2007). and ‘The girl whose car Simon works on practices dance with the person
who is a computer consultant in Edward's bank’ (from Rutherford,
2004). The only difference here is between ‘loan officer’ and ‘computer
2. Problems with previous research
consultant’. In many cases, one of these alternatives simply did not ap-
pear in the story at all. As such, the question can be answered by simply
Previous studies of recursive mindreading ability used versions of
spotting the unfamiliar item: the full sentences, and the complex propo-
the Imposing Memory Task (IMT) 4 (e.g. Kinderman et al., 1998; Lyons
sitions they convey, do not need to be understood. This occurs in 6 of 50
et al., 2010; Stiller & Dunbar, 2007). The IMT has also been widely
control questions, and 1 of 50 mental questions.
used as a measure of mindreading ability in studies designed to identify
brain regions involved in mindreading (e.g. Lewis, Rezaie, Brown,
Roberts, & Dunbar, 2011; Powell, Lewis, Dunbar, García-Fiñana, &
5
With one exception, none of the currently published studies that we are aware of pro-
vide a complete list of the specific questions used. The exception (Liddle & Nettle, 2006)
used a version modified for children. We therefore analysed the complete set of IMT ques-
tions sent to us by R. Dunbar. This set of questions is an updated version of the materials
4
Although based on the IMT, some of the later studies do not use the name IMT. Here, used in the earliest IMT studies, and forms the basis for the materials used in later studies.
we use IMT to refer to all studies based on the same general idea, and set of questions used. As such, the items we analysed are representative of the stimuli used in this literature.
 C. O’Grady et al. / Evolution and Human Behavior 36 (2015) 313–322 315

2.3. Impossible choices in the IMT are more recursive than the control questions, both
syntactically and conceptually.
Some questions in the IMT cannot be answered based on the infor-
mation included in the story, or based on reasonable inference from 2.6. Summary
the story. For example, in one story we are told that: Henry gave Sam
faulty information; that Henry is a prankster, and Sam suspects him of Although there have been consistent findings across different stud-
playing a trick; and that their colleague Pete does not think Henry was ies that use the IMT, pointing perhaps to its internal consistency,
trying to trick Sam. Crucially, however, Henry's actual motivations when taken as a whole the problems discussed above cast some doubt
(rather than just Sam's suspicions about them) are never mentioned, on its validity as a measure of recursive mindreading ability. Partly in
and cannot be reasonably inferred. Nevertheless, one true/false ques- light of this, and partly because we wished to use implicit as well as ex-
tion was the statement ‘Henry wanted to play a trick’. 7 of 50 mental plicit tasks (as detailed above, the IMT uses explicit questions only), we
questions and 2 of 50 control questions were impossible to answer in developed new stimuli for our study.
this way.
3. The current study
2.4. Syntactic complexity
Our new study was designed to address the following questions:
The mentalising questions in the IMT are more syntactically recur- (i) Does the step-change in performance after four levels of
sive than the corresponding control questions. To measure this, we metarepresentation reported in previous studies still occur with differ-
counted the number of embedded clauses in each of the questions ent stimuli, which avoid the issues identified above?; (ii) If not, does
(e.g. subordinate clauses such as “Susan wants to marry Edward” in performance decline gradually, or not at all?; and (iii) Is any effect spe-
the sentence “Jim thinks that Susan wants to marry Edward”). We cific to implicit mindreading tasks, or does it generalize to non-
found that the average number of embedded clauses was significantly mindreading tasks and/or explicit tasks, which have less ecological va-
higher in the mentalising questions than the control questions (Median lidity than implicit tasks?
level of embedding for control questions: 0; median for mental ques-
tions: 2; Mann–Whitney U Test, p b .001). This difference in syntactic 4. Methods
complexity is not an issue for studies that use control questions only
as a way to test participants' comprehension of the stories. However it 4.1. Participants
is an issue if these controls are used as an experimental contrast with
mental questions, as some IMT studies have done. For instance, neuro- We recruited 66 participants (41 F, 25 M; average age 22y 10 m). All
imaging studies use the control questions as a baseline task (e.g. Lewis participants were recruited from the student population of the Univer-
et al., 2011; Powell et al., 2010). Consequently, it is possible that any dif- sity of Edinburgh, through the University Careers Service Student and
ferences in the brain regions associated with the two conditions may be Graduate Employment online database, and paid £7 for their participa-
due to the additional syntactic demands of the mental questions, rather tion. Participants were screened to ensure that they (1) were native En-
than to mindreading specifically. glish speakers, and (2) did not know any of the actors in the stories.

2.5. Inappropriate control questions 4.2. Materials

Finally, the control questions used in the IMT are arguably not ap- We wrote four original stories, each of which had a plot involving
propriate controls in the first place, since they do not involve conceptual seven levels of recursively-embedded mental representation, and
embedding. Recursion is the repetition of a given feature, with each rep- seven levels of a non-mental recursive concept, such as possession.
etition embedded inside a previous instance of that feature (Karlsson, The stories were written in two different formats: as scripts, to be
2009). In a linguistic context, this would involve the embedding of a acted; and as a narrative, to be read by a single story-teller. All details
phrase within a similar phrase, for example: of plot, character, and so on were identical in both formats; only the
manner of presentation differed. The scripts were then performed by
[NP The book [PP on [NP the desk [PP in [NP the corner [PP of [NP the
actors (implicit story presentation) and filmed; the narrative was read
room]]]]]]]
out by a single story-teller and filmed (explicit story presentation).
This sentence involves a noun phrase (NP) ‘the room’ embedded We also created, for each of the levels of mental and non-mental recur-
within a prepositional phrase (PP) ‘of the room’ embedded within a sion, two scenes to follow the main story. One of these scenes was con-
larger NP ‘the corner of the room’, embedded within a larger PP, and sistent with the relevant mental/non-mental aspect of the story, the
so on. The syntax is recursive, and the concept itself is also recursive: other not. Again, these additional scenes were filmed both as scripts
each location is contained within another location. However, although performed by actors (implicit test question presentation), and as narra-
syntactic recursion is often used to express conceptually recursive con- tives read by a single story-teller (explicit test question presentation).
cepts, such as possession, recursive locations, or mental This gave a total of 14 test questions for each story (7 levels of recursion,
metarepresentations (e.g. “Jake thinks that I believe that Mary feels each in mental and control conditions).
sad”), it is possible to express conceptual recursion without heavy syn- These stimuli were designed to avoid each of the problems identified
tactic recursion, through parataxis. For instance, the syntactically and in the previous section. First, we did not use any broken conceptual
conceptually recursive “Portia's dog Fido's ball” can also be expressed chains. Second, we took a number of steps to ensure that simple substi-
as “Portia has a dog. That dog is called Fido. Fido has a ball”, which is con- tution was not possible, despite minimal differences between correct
ceptually but not syntactically recursive. As such, an appropriate control and incorrect answers in our explicit stimuli (see below). Third, the
for recursive mindreading, which is conceptually recursive, would be to only differences between correct and incorrect stimuli involved differ-
use control questions that are also conceptually recursive, while con- ences in mental state attribution (e.g. ‘wants’ vs. ‘doesn’t want’), rather
trolling for syntactic recursion across both types of question. However, than other aspects of the scenario (e.g. ‘loan officer’ vs. ‘computer con-
control questions in the IMT do not contain the same level of syntactic sultant’), thus ensuring that there was no confound of syntactic com-
recursion as mental questions (see ‘Syntactic complexity’, above). Men- plexity. Fourth, we ensured that there were no impossible choices.
tal questions also do not use recursive concepts in a controlled manner Fifth, we used control questions that had the same levels of conceptual
(see ‘Broken conceptual chains’, above). Instead, the mental questions and syntactic recursion as our experimental questions.
316 C. O’Grady et al. / Evolution and Human Behavior 36 (2015) 313–322

Collectively, these aspects of our stimuli ensured that the two op- 4.5. Data analysis
tions could only be differentiated if participants had correctly under-
stood the embedded levels of recursive mindreading involved. In Data for a single story from three participants could not be used due
particular, non-mindreading strategies based on forms of simple substi- to computer error. We analysed three dependent variables: success (i.e.
tutions should not succeed at better than chance levels, and in any case identifying the correct video) on test; the confidence ratings that partic-
are not possible for implicit questions. More specifically, while the cor- ipants attached to those responses; and the number of times that partic-
rect and incorrect choices in the explicit stimuli were very similar to one ipants viewed each question video before answering (which may index
another, the differences they did have were precise and only those that whether participants found certain types of questions more difficult
involved recursive mindreading. For example, one 4th-level mental than others, and had re-watch them in order to answer). For the statis-
question involved the choice between ‘Stephen knows that Elaine tical analysis, multilevel models with random effects were employed:
knows that Bernard feels she doesn’t know him well enough to date’ responses were fitted to a binomial distribution for the binary success
and ‘Stephen doesn’t know that Elaine knows that Bernard feels she DV, and a Poisson distribution for the confidence rating and number of
doesn’t know him well enough to date’. The only difference here is in additional video views DVs. 6 Analyses were conducted in the R pro-
the first intentional verb (‘Stephen knows…’ vs. ‘Stephen doesn’t gramming environment (version 3.1.1, R Core Team, 2014) using the
know…’). This means that in order to answer the question correctly, lme4 package (version 1.1–7; Bates, Maechler, Bolker, & Walker,
participants must parse everything that follows this verb — which in it- 2014). Lme4 was used at default settings, except for the usage of the
self involves interpreting three levels of mental metarepresentation — `bobyqa'-optimizer and increasing the maximum iterations to 100,000.
and then determine whether this is something that Stephen knows or The theoretically-motivated factors level of question complexity (1–7,
not (a fourth level mental metarepresentation). Similarly, if correct indicating required level of metarepresentation), condition (mental or
and incorrect stimuli were also different at other points in the construc- control), story presentation (implicit or explicit) and question presenta-
tion (e.g. ‘Stephen knows that Elaine knows that Bernard feels she tion (implicit or explicit) were included as fixed effects for all DVs. Ad-
doesn’t know him well enough to date’ and ‘Stephen doesn’t know ditionally, we included number of additional story views (i.e. beyond
that Elaine doesn’t know that Bernard feels she doesn’t know him well the first, obligatory viewing of the story) as a predictor, in order to con-
enough to date’), then participants need only parse what follows trol for the effects this might have on performance. Following Barr, Levy,
those differences (i.e. they need only know if Elaine does or doesn’t Scheepers, and Tily (2013) we used a maximum random effects model,
know that Bernard feels she doesn’t know him well enough to date), and consequently included subject (64 levels) and story (4 levels) as
and nothing before that. In short, our stimuli were designed to ensure random effects, with by-subject and by-story random intercepts and
that participants had to parse and comprehend the whole scenario in random slopes for level, question type, story presentation and question
order to perform at above chance levels. presentation, which represented the most complex converging model.
One of our stories is provided in the Appendix A, in both narrative For further analyses, we reduced the random effects structure further,
and script forms, with a complete set of questions for each. The full set if the models did not converge. The validity of the mixed effects analyses
of all four stories is provided as supplementary information (available were assessed by computing likelihood ratio tests comparing models
on the journal's Website at www.ehbonline.org), and the videos them- containing effects with null models that contained the intercept and
selves are available at http://hdl.handle.net/10283/609. the random effect structure only (c.f. Mundry, 2011). Due to the nature
of multilevel analysis, F-ratios will not be reported.

4.3. Design
5. Results
Each participant viewed and was tested on all four stories, with
5.1. Success rates
manner of story presentation and test question presentation fully
crossed within subjects (i.e. participants saw one story with implicit
The model predicting success did significantly better than the null
presentation of the story and implicit testing, one with explicit story
model (χ2(15) = 28.244, p = .02), and showed slight underdispersion
presentation and implicit testing, and so on), in a fully counter-
(dispersion parameter of 0.771). Overall success rates for all questions
balanced design, such that each story appeared in each position (1st
were well above the level expected under chance performance (inter-
viewed, 2nd viewed, etc.) an equal number of times, and each story ap-
cept in log-odds space of 2.356, corresponding to odds of greater than
peared before and after each other story an equal number of times.
10:1 of answering correctly, i.e. answering around 90% of all questions
correctly). Moreover, there was very little evidence of any effect of
level or condition on success (level: β = − 0.037, SE = 0.055,
4.4. Procedure
p = .494; condition: β = 0.092, SE = 0.329, p = .779; see Fig. 1), nor
of any interaction between these two factors (β = − 0.073, SE =
Participants were first presented with the story video. They were
0.066, p = .266).
able to watch this as many times as they liked before proceeding to
There were two significant interactions between predictors of accura-
the questions, after which they could not watch the story again (this
cy (Fig. 2). First, there was an interaction between condition and story
mimics, in a different modality, the procedure used in several previous
presentation (β = 0.727, SE = 0.282, p = .01). Post-hoc tests using
IMT tasks, in which participants read stories on paper). The fourteen
dummy-coded data and taking control-explicit story as the baseline
test questions for each story (7 mental, 7 control) were presented in
showed that only the mental-explicit combination (i.e. the combination
random order. For each question, two video frames were shown on
investigated in the IMT) performed worse than baseline (mental-explicit:
the screen simultaneously, presenting the two forced choice options
β = −0.366, SE = 0.173, p = .034; control-implicit: β = −0.526, SE =
for that question, with left/right presentation randomized. Participants
0.391, p = .179; mental-implicit: β = −0.067, SE = 0.398, p = .866).
were able to watch both videos as many times as they liked (again,
Second, there was an interaction between story presentation and
this mimics procedures from previous research), until they wished to
question presentation (β = 0.885, SE = 0.304, p b .001). Post-hoc
identify, by mouse click, the video which they thought was consistent
with the story. After each selection, participants were asked to rate 6
As confidence ratings were generally towards the upper end of the 1 to 10 scale, con-
their confidence, on a scale of 1 (lowest) to 10 (highest), that they had fidence measures were subtracted from 10, in order to avoid predictions outside the pos-
chosen the correct answer. After making a selection, participants were sible range. Note that this reverses the expected sign of the coefficient estimates in the
not able to return to that question. models for confidence.
 C. O’Grady et al. / Evolution and Human Behavior 36 (2015) 313–322 317

Mean proportion of correct responses 1.0

0.8

Q u e s tio n Ty p e

Control

Mental

All

0.6

1 2 3 4 5 6 7
Level of recursion

Fig. 1. Mean proportion correct (by participant) at each level of mindreading. Error bars give 95% CIs. These data show that performance on both mental and control tasks does not decrease
as the level of embedding increases.

tests revealed no difference between the two modes of question presen- 5.2. Judgements of confidence
tation for explicitly-presented stories (β = 0.089, SE = 0.603, p = .883),
but for implicitly-presented stories performance was significantly For the confidence data the fitted model did significantly better than
worse if the questions were presented explicitly (β = 0.767, SE = the null model (χ2(15) = 275.87, p b .001), and had a dispersion pa-
0.364, p = .035). In other words, an implicit story followed by an explic- rameter of 1.356 suggesting overdispersion within the acceptable
it question was the most difficult combination of story presentation and range. Further analysis of confidence ratings revealed several effects of
question presentations. All other main effects and interactions were not our manipulations. There was a small but significant effect of level
significant (p N .204). (β = 0.106, SE = 0.022, p b .001; see Fig. 3): participants' confidence
Finally, additional views of the story videos lead to a significant dropped as the levels increased even though, as noted above, their actu-
improvement in success (significant effect of number of additional al level of accuracy remained high. There were also a number of signifi-
story views: β = 0.457, SE = 0.138, p b .001). While the inclusion of cant two- and three-way interactions involving level. Confidence
number of additional story views as a factor does improve model fit ratings for mental but not control questions decrease with level (main
significantly (χ2(1) = 11.312, p b .001), a model lacking number of model reveals a significant level × condition interaction, β = 0.058,
additional story views as a factor produced qualitatively similar results SE = 0.012, p b .001; post-hoc tests using multilevel models on subsets
to those outlined above. of the data show a significant effect of level for mental questions, β =
In summary, participants were able to successfully process recursive 0.070, SE = 0.0546, p = .009, but not for control questions, β =
mental concepts even at high levels of recursion, and this was no more 0.045, SE = 0.030, p = .134). This interaction is further modulated by
difficult than other, non-mental recursive concepts. both story presentation and question presentation (level × condition

Condition x Story Presentation Story Presentation x Question Presentation

1

Explicit Story Explicit Question

Implicit Story Implicit Question
* ***
Mean proportion correct
0.9

0.9
0.8

0.8

Control Mental Explicit Story Implicit Story

Fig. 2. The condition × story presentation (left) and story presentation × question presentation (right) interactions for accuracy. Bars give means of the by-participant mean success rates;
error bars indicate 95% CIs. These results show that for control questions, performance was lower when the stories were presented implicitly rather than explicitly (left-hand side), and
that performance was reduced when stories were presented implicitly and questions explicitly, in comparison to all other possible combinations of story presentation and question pre-
sentation (right-hand side). These were the only significant interactions for accuracy.
318 C. O’Grady et al. / Evolution and Human Behavior 36 (2015) 313–322

10
Q u e s tio n Ty p e

Control

Mental

9 All
Mean confidence

1 2 3 4 5 6 7
Level of recursion

Fig. 3. Mean confidence (by participant). Error bars give 95% CIs. These data show that although actual performance (i.e. accuracy) does not decrease as the level of embedding increases
(see Fig. 1), our participants' confidence in their answers does decrease.

× story presentation interaction: β = 0.076, SE = 0.024, p = .001; level question videos as levels increased. There were also significant
× condition × question presentation interaction: β = − 0.136, SE = effects of condition (β = − 0.272, SE = 0.075, p b .001: mental
0.024, p b .001). Post-hoc tests on control questions reveal no two- questions required fewer additional plays) and question presentation
way interactions between level and story presentation ((β = −0.016, (β = −0.473, SE = 0.086, p b .001: implicitly-presented questions re-
SE = 0.017, p = .325) or level and question presentation (β = 0.020, quired fewer additional plays), and various two-way interactions in-
SE = 0.016, p = .206): however, these two-way interactions are signif- volving level, condition, and question presentation. Ultimately these
icant for mental questions (level × story presentation: β = 0.082, SE = are best explained by considering the significant three-way interaction
0.019, p b .001; level × question presentation: β = −0.116, SE = 0.018, between level, question presentation and condition (β = −0.171, SE =
p b .001): confidence on mental questions decreases faster with increas- 0.054, p = .001; see Fig. 4). Post-hoc tests on subsets of the data show
ing level when the story is presented implicitly, or when the questions that the number of times the question videos were viewed increased
are presented explicitly. Finally, there was also a three-way interaction with level for explicit questions (there were strong effects of level in ex-
between condition, story presentation and question presentation (β = plicit–control and explicit–mental conditions: explicit–control, β =
0.358, SE = 0.098, p b .001). Post-hoc tests using dummy-coded data 0.207, SE = 0.030, p b .001; explicit–mental, β = 0.278, SE = 0.038,
and taking explicit story–explicit question as the baseline showed that p b .001) and for implicit–control questions (β = 0.131, SE = 0.057,
for control questions, participants had significantly lower confidence p = 0.022), but not for implicit–mental questions (β = 0.056, SE =
for the implicit story–explicit question combination (β = 0.464, SE = 0.100, p = .577), which showed no significant increase in additional
0.164, p = .005), with implicit story–implicit question (β = 0.104, question viewings with increasing complexity.
SE = 0.270, p = .700) and explicit story–implicit question (β = There was also a significant interaction between story presentation
0.281, SE = 0.201, p = .125) not significantly different from the explicit and question presentation (β = − 0.404, SE = 0.121, p b .001).
story–explicit question intercept (β = 0.438, SE = 0.146, p b .001); in Follow-up analyses using dummy coded data with explicit story–explic-
contrast, for mental questions, participants had higher confidence for it questions as the baseline revealed that, compared to this baseline,
the implicit story–implicit question combination than for explicit combinations of implicit story–implicit questions required significantly
story–explicit question (β = −0.537, SE = 0.179, p = .003; other com- fewer views before answering (β = −0.539, SE = 0.129, p b .001), as
parisons n.s., p N .137). did explicit stories–implicit question combinations (β = − 0.397,
Finally, while there was a tendency for participants' confidence SE = 0.122, p = .001). Implicit story–explicit question combinations re-
rating to increase with number of additional story views, this effect quired more views than baseline before answering, albeit not signifi-
was not significant (β = − 0.086, SE = 0.049, p = 0.079). cantly so (β = 0.170, SE = 0.113, p = .131).
Furthermore, including this factor did not significantly improve Finally, while there was a tendency for participants' to require fewer
model fit (χ2(1) = 2.253, p = .133) over an equivalent model views of the question videos if they had watched the stories more often,
lacking this factor, and the simpler model yielded qualitatively this effect was not significant (β = − 0.042, SE = 0.075, p = 0.574).
similar results to those discussed above. Furthermore, including this factor did not significantly improve model
fit (χ2(1) = 0.306, p = .580) over an equivalent model lacking this fac-
tor, and the simpler model yielded qualitatively similar results to those
5.3. Number of additional video views
discussed above.

The third dependent variable was the number of times each question
video was viewed. The fitted model did significantly better than the null
model (χ2(15) = 172.52, p b .001) and had a dispersion parameter of 6. Discussion
1.095, suggesting low overdispersion within the acceptable range.
This model revealed a small but significant effect of level (β = 0.169, Our study produced several results worthy of note, of which two are
SE = 0.016, p b .001): participants' required additional views of the of particular importance.
 C. O’Grady et al. / Evolution and Human Behavior 36 (2015) 313–322 319

1.2 Q u e s tio n P re s e n ta tio n −Ty p e

Number of additional question video plays
Explicit−Control

1.0 Explicit−Mental

Implicit−Control

0.8 Implicit−Mental

0.6

0.4

0.2

0.0

1 2 3 4 5 6 7
Level of recursion

Fig. 4. Number of additional video views for each combination of condition and question presentation. Error bars give 95% CIs. These results show that for all control questions (i.e. both
implicit and explicit), and for mental questions that were presented explicitly, participants chose to watch the videos increasingly often as level increased. This was, however, not true for
mental questions presented implicitly, for which there was no corresponding increase with level. Given that this reduction in the number of views did not lead to any reduction in actual
performance (see Figs. 1 and 2), this is tentative evidence that the implicit mental questions were processed more easily than all other types of question. In the Discussion we suggest that
this may be because recursive mindreading is actually easy, when presented within is natural social context.

First, performance on mindreading tasks was high throughout (see experiments have now shown that in implicit contexts, we track the be-
Fig. 1). The design of our stimuli ensured that this level of performance liefs of others automatically, as part of our intuitive monitoring of the
is unlikely to be due to guesswork or any other strategy that did not in- world around us, and that like our perceptual experiences, these
volve recursive mindreading (see Methods). These findings run counter representations of others' mental states fade quickly if we do not focus
to the intuition that high level recursive mindreading tasks are cogni- on them (e.g. Kovács, Téglás, & Endress, 2010; Samson, Apperly,
tively demanding, and counter to results obtained in previous research Braithwaite, Andrews, & Bodley Scott, 2010; van der Wel, Sebanz, &
that suggest that performance on mental questions decrease markedly Knoblich, 2014). Our results tentatively suggest that the same may be
after level 5. One way to understand this result is by analogy with our true of recursive mindreading. In particular, we found that although
perceptual skills: a formal description of what is involved in, say, vision, participants in general increased their number of views of the question
is complex, but this does not mean that seeing is a cognitively demand- videos as level increased, suggesting an increase in the level of difficulty,
ing activity, beyond the ken of typical human abilities. Our results sug- this was not true of implicit mental questions.
gest that the same may be true of recursive mindreading, even at high It is also instructive to compare our results with the developmental
levels of recursion. Interestingly, the intuition that high levels of recur- literature. First, note that the classic Sally-Anne false belief task uses
sive mindreading are particularly cognitively demanding extends even an implicit story (acted out using dolls, albeit with explicit commentary
to the individuals involved: while actual performance remained high attached), followed by an explicit question. Our results show that this
across all levels, confidence levels declined as level of embedding in- implicit–explicit combination is the most challenging combination for
creased, for mental questions but not control questions (see Fig. 3). adults in recursive mindreading tasks (see Fig. 2). The robustness of
Second, we found that participants viewed the videos more often as this finding, and the exact reasons for it, are topics for future research,
the level of complexity increased — except for implicitly presented but it does raise the possibility that the classic false belief task involves
mental questions i.e. except in those contexts that are most ecologically the most cognitively demanding combination of story and question pre-
valid (see Fig. 4). This tentatively suggests that recursive mindreading is sentation possible. More generally, our results suggest two possible
especially easy when employed within its natural environment, and lines of future research on the development of mindreading abilities:
that it is otherwise no more or less easy than recursive tasks in general. (i) the use of explicit–explicit and/or explicit–implicit methods, in
As we emphasised in the Introduction, humans' natural ecology is social. order to make comparisons with the existing implicit–explicit and im-
Correspondingly, prominent accounts of the evolution of human cogni- plicit–implicit approaches; and (ii) the investigation of higher-level, re-
tion emphasise the importance of specifically social cognition, including cursive mindreading abilities in children, using implicit–implicit
mental state attribution (Sterelny, 2003; Tomasello, 2014; Tooby & methods. Implicit–Implicit methods have dramatically re-shaped our
Cosmides, 1992). Our results make sense from this ecological perspec- understanding of the development of simple mindreading abilities,
tive: recursive mindreading is an essential, ubiquitous, and adaptive but this advance has not yet been extended to the development of re-
component of everyday life, and as such, we should expect that we are cursive mindreading abilities.
good at it. A natural extension of our study, which would increase the These results differ somewhat from previous research on adult
ecological validity, would be to limit the participants to single views of recursive mindreading, which found a prominent drop in perfor-
both the story and the question videos (rather than the mance after four levels of metarepresentation. We suggest two pos-
multiple views that, following previous research, we allowed for in sible reasons for this. The ecological validity of our implicit tasks
the current design). cannot fully explain this, since we also find high levels of perfor-
These results are consistent with the picture emerging from the mance on explicit tasks. A more likely explanation is the various
literature on adult first-level mindreading, which shows that methodological problems we have identified with the IMT, which
mindreading may be less like thinking, and more like perception i.e. previous studies used as a measure of mindreading ability (see
something that we do unconsciously, as part of the background cogni- Problems with previous research, above). These methodological
tion that manages much of our daily lives (Apperly, 2011). Several problems raise the possibility that previous results may not
320 C. O’Grady et al. / Evolution and Human Behavior 36 (2015) 313–322

accurately reflect human mindreading abilities, and their Mental questions for narrated version
relationship to other aspects of social psychology, as accurately
as possible. Further investigation into the exact reasons for
1. A. Elaine likes Bernard
the differences between our study and previous research may
B. Elaine likes Stephen
be warranted. 2. A. Megan knows that Lauren wants to ask Stephen out
The broader implications of our findings are several. In particular, B. Megan doesn’t know that Lauren wants to ask Stephen out
our results should reduce concerns that some theoretical explana- 3. A. Elaine doesn’t know that Bernard feels that she doesn’t know him well
tions of many important human behaviours and institutions are im- enough to date
B. Elaine knows that Bernard feels that she doesn’t know him well enough to date
plausible precisely because they invoke high level recursive
4. A. Stephen knows that Elaine knows that Bernard feels she doesn’t know him
mindreading abilities. For example, the most prominent theoretical well enough to date
accounts of human communication argue that it involves the expres- B. Stephen doesn’t know that Elaine knows that Bernard feels she doesn’t
sion, on the part of the speaker, of an intention that the audience rec- know him well enough to date
5. A. Megan knows that Stephen doesn’t know that Elaine knows that Bernard
ognises that the speaker has an intention to inform the audience —
feels that she doesn’t know him well enough to date
and that the audience must recognise these embedded intentions B. Megan knows that Stephen knows that Elaine knows that Bernard feels
(Csibra, 2010; Grice, 1969; Sperber, 2000b; Tomasello, 2008). Sever- that she doesn’t know him well enough to date
al researchers have argued that, while theoretically cogent, this anal- 6. A. Chris knows that Megan knows that Stephen knows that Elaine knows that
ysis is empirically implausible, precisely because it depends upon Bernard feels that she doesn’t know him well enough to date
B. Chris doesn’t know that Megan knows that Stephen knows that Elaine
high levels of recursive mindreading, which are assumed to be cogni-
knows that Bernard feels that she doesn’t know him well enough to date
tively demanding (e.g. Breheny, 2006; Clark, 1996; Glüer & Pagin, 7. A. Megan wants Lauren to know that she, Megan, knows that Stephen knows
2003; Gómez, 1994). Our results suggest that these concerns are that Elaine knows that Bernard feels she doesn’t know him well enough to
likely unfounded: at least in the contexts we explored here, recursive date, so that Lauren asks Stephen out
B. Megan doesn’t want Lauren to know that she, Megan, knows that Stephen
mindreading poses no particular challenges for adult humans, even
knows that Elaine knows that Bernard feels she doesn’t know him well
at high levels of embedding. The same point applies to numerous enough to date, so that Lauren doesn’t ask Stephen out
other activities that have been argued to depend upon recursive
mindreading, such as language, story-telling, culture, and even con-
sciousness (Corballis, 2011, Dunbar, 2003, 2005, 2008; Graziano,
Control questions for narrated version
2013; Sperber, 2000a).
1. A. Lauren is Megan's sister
B. Lauren is Megan's friend
Supplementary materials 2. A. Stephen has Biology with Megan's sister Lauren
B. Stephen was in the school play with Megan's friend Chris
Supplementary data to this article can be found online at http://dx. 3. A. Chris used to be best friends with Stephen, who has Biology with Megan's sister
B. Chris used to be best friends with Bernard, who acted in the school play
doi.org/10.1016/j.evolhumbehav.2015.01.004.
with Megan's sister
4. A. Megan is best friends with Chris, who used to be best friends with Stephen,
Acknowledgements who has Biology with Megan's sister Lauren
B. Megan's sister Lauren used to go out with Chris, who is best friends with
Elaine, who has Biology with Stephen
COG gratefully acknowledges financial support from the Skye Foun- 5. A. Bernard acted in the play with Megan, who is the best friend of Chris, who
dation. TSP gratefully acknowledges financial support from the used to be best friends with Stephen, who takes Biology with Lauren who
Leverhulme Trust and the ESRC. We would like to thank Tamsin Saxton is Megan's sister
for valuable discussion of the methods, Robin Dunbar for sending us the B. Bernard acted in the play with Lauren, who used to go out with Chris, who
used to be best friends with Stephen, who lives next door to Elaine who is
IMT used in previous studies, and Mateo Obregón and Roger Mundry for
Lauren's cousin
statistical advice. 6. A. Elaine lives next door to Bernard, who acted in the play with Megan, who is
the best friend of Chris, who used to be best friends with Stephen, who
takes Biology with Lauren, who is Megan's sister
Appendix A. Example script & narrative (story: ‘Bernard’) B. Elaine lives next door to Bernard, who is the best friend of Chris, who used
to be best friends with Stephen, who was in the play with Lauren's sister
Narrated version Megan, who has Biology with Elaine
7. A. Megan's cousin is Elaine, who lives next door to Bernard, who acted in the
play with Megan, who is the best friend of Chris, who used to be best
One evening, Megan finds out that her sister Lauren wants to go out friends with Stephen, who has Biology with Megan's sister Lauren
with a boy in her biology class, Stephen. Megan tells Lauren that Ste- B. Megan's cousin is Elaine, who has Biology with Bernard, who acted in the
phen used to be best friends with a boy called Chris, who is now play with Stephen, who used to go out with Megan, who is the best friend
Megan's best friend. Lauren tells Megan that she saw Stephen smiling of Chris, who has Biology with Megan's sister Lauren

and flirting with their cousin, Elaine, and so she thinks Stephen might
want to go out with Elaine. Because Lauren thinks Stephen likes some-
one else, she is too nervous to ask him out. Scripted version
Megan talks to Elaine at school and finds out that Elaine actually
wants to go out with Bernard, whom Megan knows from the school
play. Megan learns that Elaine and Bernard are next-door neighbours,
and that Bernard thinks that Elaine doesn’t know him well enough to Scene 1
date. Elaine tells Megan that Stephen knows how Elaine feels about Megan: Lauren, your phone just buzzed.
Bernard and how Bernard feels about Elaine. Lauren snatches up her phone
Megan later talks to her friend Chris about the situation, Megan: Ooooh, is that a booooy?
realising that if Lauren knew about Elaine's situation, and knew Lauren: Maybe…OK, OK, it’s Stephen. You know, the one in my
that Stephen knows about it too, Lauren would realise that Stephen biology class.
doesn’t want to go out with Elaine, and might work up the courage Megan: Oh, I didn’t know you had a thing for him. Did you know he
to ask him out. Megan plans to tell Lauren about everything used to be best friends with Chris?
that evening. Lauren: Which Chris?
 C. O’Grady et al. / Evolution and Human Behavior 36 (2015) 313–322 321

B. Elaine: I worked up the guts to ask Bernard out today. He said no. He
Megan: The Chris who’s my best friend. Yeah, they were friends in
reckons I don’t know him well enough
primary school. 4. A. Megan: Hey, Stephen, I thought you should know something.
Lauren: That’s weird. Anyway…I was actually thinking about working Stephen: Yeah?
up the guts to ask him out, but I’m nervous. Megan: It’s about Elaine – you know she has a thing for Bernard?
Megan: Haha, you’d better not let our Dad find out…Anyway, why are Stephen: Yeah, and he doesn’t want to go out with her because he thinks
she doesn’t know him enough? Poor girl, she was crushed when
you nervous? I’m sure he’ll say yes. she found out
Lauren: Well, I saw him chatting to Elaine at lunchtime the other day, B. Megan: Oh, I have some hot gossip – did you hear about Elaine?
and he was all smiley and flirty, so I think he’s planning to ask Stephen: No – what about her?
her out. Megan: She’s really into Bernard, but Bernard doesn’t like her back
because he thinks Elaine doesn’t know him well enough.
Stephen: Seriously? That’s a weird reason not to go out with someone.
Why didn’t Elaine tell me about this?
Scene 2 5. A. Megan: So Elaine told me today that she has a thing for Bernard, but
Megan: Hey Elaine! Bernard doesn’t like her back because he thinks Elaine doesn’t
know him well enough.
Elaine: Oh, hey cousin, could you hold this for a second?
Chris: Elaine has a thing for Bernard? But I saw her flirting with Stephen!
Megan: So, I saw you flirting with Stephen the other day…he’s cute! Megan: Mm-hmm, and Stephen has no idea, either that Elaine likes
Stephen walks past them, they pause awkwardly and giggle Bernard, or that Bernard doesn’t like her back.
Elaine: Haha, yup, he is, but there’s nothing going on. Chris: Yikes, that’s not cool – she needs to tell him.
B. Megan: So Elaine told me today that she has a thing for Bernard, but
Megan: Why not? Bernard doesn’t like her back because he thinks Elaine doesn’t
Elaine: I kind of have a thing for Bernard… know him well enough.
Megan: Bernard? The one who was in the school play with me? Chris: Elaine has a thing for Bernard? But I saw her flirting with Stephen!
Elaine: Yeah, that’s the one. Megan: Oh, yeah, he knows, she told him.
Chris: Knows which bit?
Megan: Oh, I didn’t know you guys knew each other.
Megan: Knows that Elaine likes Bernard and that Bernard told Elaine
Elaine: I live next door to him. Anyway, he doesn’t like me back. that he won’t go out with her.
Megan: Why on earth not? 6. A. Chris: Yeah, so Megan heard some juicy gossip today…Stephen too?
Elaine: Well, I dunno if it’s that he doesn’t like me, it’s more just that he Haha, I bet it’s the same gossip – they both know about this,
thinks I don’t know him well enough to really want to date him. anyway… So you know Elaine? She has a mega thing for
Bernard… uh-huh… and Bernard told her – yeah, told Elaine –
He says if I knew him better I wouldn’t want to… really weird. that he doesn’t want to go out with her because Elaine doesn’t
Megan: Ugh, he’s just being melodramatic. I’m sure he’ll come round. know him well enough. I know, it’s weird, right? So, wait, you
Does Stephen know? He looked pretty into you when you spoke to Megan earlier and asked her about Elaine and Stephen,
were chatting. and Megan said she didn’t know anything about it? That’s
weird, she definitely knows.
Elaine: Yup, I told him about it ages ago, we were just laughing about
B. Chris: Yeah, so what’s this hot gossip? Elaine and Bernard? Why not? Be-
Mr Murray's new haircut. cause she doesn’t know him well enough? That’s weird…does Ste-
Megan: Ah, OK. Anyway, I’d better run – hope Bernard realises what phen know? Ok, that’s good, it would be really awful of Elaine if he
he’s missing! didn’t. And Megan knows too? What? Has she heard that Stephen
knows about all of it? That’s weird, Megan didn’t tell me anything.
7. A. Megan: So, I’m thinking that Lauren needs to know what I heard. Right?
Scene 3 Because if she knows what I know right now, about Elaine’s
crush, and Bernard’s rejection, and that Stephen knows the
Megan: …so it turns out that Elaine and Stephen aren’t into each
whole thing…she’ll work up the guts to ask him out! So I’m
other at all. Elaine is actually really into Bernard. going to tell her tonight.
Chris: The acting one? B. Megan: Well, if you think about, if Lauren knew what I heard today –
Megan: Yeah. Bernard’s super melodramatic, too – he doesn’t want to and if she knew that Stephen knew all about it too, about
date Elaine because he reckons Elaine doesn’t know him well Elaine’s crush and Bernard’s weird reason for rejection and
everything – she’d ask Stephen out. But I don’t want her to do
enough or something. that, so I’m not going to tell her.
Chris: That’s weird. And Stephen knows that Elaine’s not into him?
Megan: Mm, Elaine told him about the whole Bernard thing ages ago.
Chris: So Lauren… Control questions for scripted version
Megan: Right! So, Lauren doesn’t want to ask Stephen out because
1. A. Megan: So I was chatting to my friend, Lauren…
she thinks he’s into Elaine – but if she knew that Stephen
B. Megan: So I was chatting to my sister, Lauren…
knows that Elaine likes Bernard, and that Stephen knows 2. A. Megan: You know Stephen? The one who has Biology with my sister
that Elaine’s not into him, she might work up the guts to ask Lauren?
Stephen out. B. Megan: You know Stephen? The one who was in the school play with my
Chris: I guess…so are you going to tell her? friend Chris?
3. A. Megan: Did you know that Chris used to be best friends with Stephen?
Megan: Yeah, I’m going to tell her the whole thing tonight. Elaine: Which Stephen?
Megan: The one who has Biology with my sister Lauren.
B. Megan: Did you know that Chris used to be best friends with Bernard?
Mental questions for scripted version Elaine: Which Bernard?
Megan: The one who was in the school play with my sister Lauren.
1. A. Elaine: Ahhh, there goes Bernard, he’s so cute…can’t remember the last 4. A. Megan: You know Stephen, who has Biology with my sister? He used to
time I liked someone so much. be best friends with my best friend, Chris.
B. Elaine: Ahhh, there goes Stephen, he’s so cute…can’t remember the last B. Megan: You know my sister Lauren used to go out with Chris, who’s best
time I liked someone so much. friends with Elaine? She has Biology with Stephen.
2. A. Chris: I heard your sister wants to ask Stephen out! 5. A. Elaine: No, the other Megan, the one whose best friend Chris used to be
Megan: What? She hasn’t told me that! best friends with this guy Stephen who has Biology with Lauren,
B. Chris: Lauren told me today that she wants to ask Stephen out. Megan’s sister…yeah, she knows Bernard from the school play.
Megan: Oh, yeah, she told me about it last night B. Elaine: No, the other Lauren, the one who used to go out with that Chris
3. A. Elaine: I worked up the guts to ask Bernard out today. He said no, but I guy who used to be best friends with Stephen…yeah, Stephen who
have no idea why he won’t go out with me lives next door to me. Yup, that’s the Lauren who’s my cousin.
(continued on next page) (continued on next page)
322 C. O’Grady et al. / Evolution and Human Behavior 36 (2015) 313–322

6. A. Elaine: You know Bernard, who was in the play with Megan? No, the Megan Henzi, S. P., de Sousa Pereira, L. F., Hawker-Bond, D., Stiller, J., Dunbar, R. I. M., & Barrett, L.
whose best friend is that Chris guy, the one who used to be best (2007). Look who's talking: Developmental trends in the size of conversational
friends with Stephen…nuh-uh, the other Stephen, the one who has cliques. Evolution and Human Behavior, 28(1), 66–74.
Biology with Megan’s sister Lauren. Yeah, that Megan. Remember Humphrey, N. K. (1976). The social function of intellect. In P. P. G. Bateson, & R. Hinde (Eds.),
she co-starred with Bernard? That’s my next-door neighbour. Growing Points in Ethology (pp. 303–317). Cambridge: Cambridge University Press.
B. Elaine: You know Bernard, Chris’s best friend? No, the Chris who used to Karlsson, F. (2009). Syntactic recursion and iteration. In H. van der Hulst (Ed.), Recursion and
be best friends with Stephen, that guy who starred in the play Human Language: Studies in Generative Grammar (pp. 43–67). Berlin: Mouton de Gruyter.
with Lauren’s sister Megan, who’s in Biology with us. Yeah, that Kerr, N., Dunbar, R. I., & Bentall, R. P. (2003). Theory of mind deficits in bipolar affective
disorder. Journal of Affective Disorders, 73(3), 253–259.
Bernard is my next-door neighbour.
Kinderman, P., Dunbar, R., & Bentall, R. P. (1998). Theory‐of‐mind deficits and causal at-
7. A. Megan: My cousin, Elaine, the one who lives next door to Bernard? Argh,
tributions. British Journal of Psychology, 89(2), 191–204.
you know Bernard. He was in the play with me and my best
Kovács, Á. M., Téglás, E., & Endress, A. D. (2010). The social sense: Susceptibility to others’
friend Chris…Chris who used to be best friends with Stephen, the beliefs in human infants and adults. Science, 330(6012), 1830–1834.
guy in Biology with my sister Lauren. Lewis, P. A., Rezaie, R., Brown, R., Roberts, N., & Dunbar, R. I. M. (2011). Ventromedial pre-
B. Megan: You know my cousin Elaine, who has Biology with Bernard? frontal volume predicts understanding of others and social network size. NeuroImage,
Argh, you know Bernard…he acted in the play with my ex, 57(4), 1624–1629.
Stephen, and my best friend Chris…Chris who has lives next door Liddle, B., & Nettle, D. (2006). Higher-order theory of mind and social competence in
to me and my sister Lauren. school-age children. Journal of Cultural and Evolutionary Psychology, 4, 231–246.
Lyons, M., Caldwell, T., & Shultz, S. (2010). Mind-reading and manipulation - Is Machia-
vellianism related to theory of mind? Journal of Evolutionary Psychology, 8(3),
261–274.
References Miller, S. A. (2009). Children’s understanding of second-order mental states. Psychological
Bulletin, 135(5), 749–773.
Apperly, I. A. (2011). Mindreaders: The cognitive basis of ‘Theory of Mind’. New York, NY: Mundry, R. (2011). Issues in information theory-based statistical inference – A commen-
Psychology Press. tary from a frequentist’s perspective. Behavioral Ecology and Sociobiology, 65(1), 57–68.
Baillargeon, R., Scott, R. M., & He, Z. (2010). False-belief understanding in infants. Trends in Nettle, D., & Liddle, B. (2008). Agreeableness is related to social-cognitive, but not social-
Cognitive Sciences, 14(3), 110–118. perceptual, theory of mind. European Journal of Personality, 22, 323–335.
Baron-Cohen, S. (1995). Mindblindness: An essay on autism and theory of mind. Cambridge, Paal, T., & Bereczkei, T. (2007). Adult theory of mind, cooperation, Machiavellianism: The
MA: MIT press. effect of mindreading on social relations. Personality and Individual Differences, 43,
Barr, D. J., Levy, R., Scheepers, C., & Tily, H. J. (2013). Random effects structure for 541–551.
confirmatory hypothesis testing: Keep it maximal. Journal of Memory and Language, Powell, J. L., Lewis, P. A., Dunbar, R. I. M., García-Fiñana, M., & Roberts, N. (2010). Orbital
68(3), 255–278. prefrontal cortex volume correlates with social cognitive competence.
Bates, D., Maechler, M., Bolker, B., & Walker, S. (2014). lme4: Linear mixed-effects models Neuropsychologia, 48(12), 3554–3562.
using Eigen and S4. R package version 1.1–7. URL http://CRAN.R-project.org/ Premack, D., & Woodruff, G. (1978). Does the chimpanzee have a theory of mind?
package=lme4 Behavioral and Brain Sciences, 1(4), 515–526.
Breheny, R. (2006). Communication and folk psychology. Mind & Language, 21(1), R Core Team (2014). R: A language and environment for statistical computing. Vienna,
74–107. Austria: R Foundation for Statistical Computing (http://www.R-project.org).
Byrne, R. W., & Whiten, A. (1989). Machiavellian intelligence: Social expertise and the Rutherford, M. D. (2004). The effect of social role on theory of mind reasoning. British
evolution of intellect in monkeys, apes, and humans. Oxford: Oxford University Press. Journal of Psychology, 95(1), 91–103.
Call, J., & Tomasello, M. (2008). Does the chimpanzee have a theory of mind? 30 years Samson, D., Apperly, I. A., Braithwaite, J. J., Andrews, B. J., & Bodley Scott, S. E. (2010). Seeing
later. Trends in Cognitive Sciences, 12(5), 187–192. it their way: Evidence for rapid and involuntary computation of what other people see.
Chevallier, C., Kohls, G., Troiani, V., Brodkin, E. S., & Schultz, R. T. (2012). The social Journal of Experimental Psychology: Human Perception and Performance, 36(5), 1255.
motivation theory of autism. Trends in Cognitive Sciences, 16(4), 231–239. Sperber, D. (1997). Intuitive and reflective beliefs. Mind & Language, 12(1), 67–83.
Clark, H. H. (1996). Using language. Cambridge: Cambridge University Press. Sperber, D. (Ed.). (2000). Metarepresentations: An interdisciplinary perspective. Oxford:
Corballis, M. C. (2011). The recursive mind: The origins of human language, thought, and Oxford University Press.
civilization. Princeton, NJ: Princeton University Press. Sperber, D. (2000b). Metarepresentations in an evolutionary perspective. In D. Sperber
Csibra, G. (2010). Recognizing communicative intentions in infancy. Mind & Language, (Ed.), Metarepresentations: An Interdisciplinary Perspective (pp. 117–137). Oxford: Ox-
25(2), 141–168. ford University Press.
Dunbar, R. I. M. (2003). The social brain: Mind, language, and society in evolutionary Sterelny, K. (2003). Thought in a hostile world: The evolution of human cognition. Oxford:
perspective. Annual Review of Anthropology, 32, 163–181. Wiley-Blackwell.
Dunbar, R. I. M. (2005). Why are good writers so rare? An evolutionary perspective on Stiller, J., & Dunbar, R. I. M. (2007). Perspective-taking and memory capacity predict social
literature. Journal of Cultural and Evolutionary Psychology, 3(1), 7–21. network size. Social Networks, 29(1), 93–104.
Dunbar, R. I. M. (2008). Mind the gap: Or why humans aren’t just great apes. Proceedings Sylwester, K., Lyons, M., Buchanan, C., Nettle, D., & Roberts, G. (2012). The role of theory of mind
of the British Academy, 154, 403–423. in assessing cooperative intentions. Personality and Individual Differences, 52, 113–117.
Frith, C. D., & Corcoran, R. (1996). Exploring ‘theory of mind’ in people with schizophre- Tomasello, M. (2008). Origins of human communication. Cambridge, MA: MIT Press.
nia. Psychological Medicine, 26(03), 521–530. Tomasello, M. (2014). A natural history of human thinking. Cambridge, MA: Harvard Uni-
Glüer, K., & Pagin, P. (2003). Meaning theory and autistic speakers. Mind & Language, versity Press.
18(1), 23–51. Tooby, J., & Cosmides, L. (1992). The psychological foundations of culture. In J. H. Barkow,
Gómez, J. C. (1994). Mutual awareness in primate communication: A Gricean approach. In L. Cosmides, & J. Tooby (Eds.), The Adapted Mind: Evolutionary Psychology and the Gen-
S. T. Parker, R. W. Mitchell, & M. L. Boccia (Eds.), Self-Awareness in Animals and eration of Culture (pp. 19–136). Oxford: Oxford University Press.
Humans (pp. 61–80). Cambridge: Cambridge University Press. van der Wel, R. P. R. D., Sebanz, N., & Knoblich, G. (2014). Do people automatically track
Graziano, M. S. (2013). Consciousness and the social brain. Oxford University Press. others’ beliefs? Evidence from a continuous measure. Cognition, 130(1), 128–133.
Grice, H. P. (1969). Utterer’s meaning and intention. The Philosophical Review, 78(2), Wellman, H. M., Cross, D., & Watson, J. (2003). Meta-analysis of theory-of-mind develop-
147–177. ment: The truth about false belief. Child Development, 72(3), 655–684.