Journal of Neurolinguistics 43 (2017) 95e106

Contents lists available at ScienceDirect

Journal of Neurolinguistics
journal homepage: www.elsevier.com/locate/jneuroling

Research paper

The emergence of recursion in human language: Mentalising

predicts recursive syntax task performance
Nathan Oesch*, Robin I.M. Dunbar
Social and Evolutionary Neuroscience Research Group (SENRG), Department of Experimental Psychology, University of Oxford, South
Parks Road, Oxford OX1 3UD, UK

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

Article history: Theory of mind, also known as mentalising, meta-representation, second-order inten-
Received 29 March 2016 tionality, or mindreading is the ability to attribute and reect on the mental states of
Accepted 22 September 2016 others. A number of investigators have noted that an important relationship exists be-
Available online 7 October 2016
tween child language development and childrens understanding of second-order inten-
tionality. However, although the ontogeny of theory of mind has been extensively studied
Keywords:
over the past few decades, only recently have we begun to understand more concerning
Mindreading
the limits of human mentalising ability in adults. For example, several studies have shown
Recursion
Mentalising
that the limits of mentalising ability for normal adults are consistently placed around fth-
Theory of mind order intentionality (i.e. I believe that you suppose that I imagine that you want me to
Evolution of language believe that...), forming a naturally recursive hierarchy which corresponds to increasingly
Intentionality embedded mindreading. Moreover, several psychologists have recently suggested the
Recursive syntax adult capacity for higher-order intentionality may have played a critical role in the evo-
lution of language, including especially the ability for recursive syntax comprehension and
production, according to a cognitive bootstrapping effect. Here, we used the Imposing
Memory Task (n 210 female and 204 male adults) to analyse the association and
interaction between higher-order intentionality capacity and performance on a recursive
syntax measure. Multiple regression analyses indicated that recursive syntax abilities are
lower than mindreading competences below fth-order, but then reverses at higher
values. In addition, a path analysis further suggested intentionality capacity as the likely
causal variable. Thus, these results seem to suggest that rst-order through fth-order
intentionality is necessary to assist the processing of simpler syntactic structures, but
beyond fth-order intentionality the cognitive scaffolding provided by recursive syntax
may be engaged to enable higher-order mentalising. In summary, this may explain in part
how and why many modern languages exhibit recursive syntax.
2016 Elsevier Ltd. All rights reserved.

1. Introduction

Theory of mind, also known as mentalising, metarepresentation, second-order intentionality, or mindreading is the ability
to attribute and reect on the mental states of others (Premack & Woodruff, 1978; Wellman, Cross, & Watson, 2001; Wimmer
& Perner, 1983). Previous studies have distinguished two types of theory of mind: implicit, an early-developing system that is
piecemeal and unconscious, and explicit, a later-developing system that is abstract and conscious (Apperly & Butterll, 2009;

* Corresponding author.
E-mail address: nathan.oesch@stx.oxon.org (N. Oesch).

http://dx.doi.org/10.1016/j.jneuroling.2016.09.008
0911-6044/ 2016 Elsevier Ltd. All rights reserved.
96 N. Oesch, R.I.M. Dunbar / Journal of Neurolinguistics 43 (2017) 95e106

Low, 2010). A number of investigators have noted that an important interaction exists between child language development
and children's understanding of explicit theory of mind. In general, language development and explicit theory of mind are
relatively strongly related, even when language measures are taken one or two years before children start mastering false
belief tasks around age 4, suggesting language acquisition may be more fundamental than theory of mind (Astington &
Jenkins, 1995; Dunn, Brown, Slomkowski, Tesla, & Youngblade, 1991; Farrar & Maag, 2002; Gale, de Villiers, de Villiers, &
Pyers, 1996; Rubio-Ferna
ndez & Geurts, 2013; Tomasello, Carpenter, Call, Behne, & Moll, 2005; Watson, Painter, &
Bornstein, 2002; de Villiers & de Villiers, 2000). On the other hand, more recent investigations using nonverbal
spontaneous-response tasks show that the ability for an implicit theory of mind is present in prelinguistic infants as young as
15-, 13-, 10- and even 7-months-old, suggesting theory of mind may precede, and hence be fundamental for the acquisition of
language (Baillargeon, Scott, & He, 2010; Kova
cs, Te

gla

s, & Endress, 2010; Luo, 2011; Onishi & Baillargeon, 2005; Surian, Caldi,
& Sperber, 2007). Unfortunately, comparable abilities of adults are much less well understood. The present study therefore
aims to investigate this gap by testing adult explicit, as opposed to implicit, theory of mind and it's relation to adult language
ability.
Due to such disparate ndings, many have argued that no solid basis currently exists for strong conclusions concerning the
direction of inuence between language acquisition and theory of mind, in the broadest sense (de Villiers, 2007; Wellman
et al., 2001). However, at least one inuential proposal suggests the conceptual developments of early theory of mind may
form an essential basis for helping to x at least word reference (de Villiers, 2007). After age 5, mastery of grammar has been
highlighted as a potential representational tool to enable theory of mind, although it remains debated whether the more
important factor is general syntactic development or more specic aspects of grammar, such as syntactic complementation
(Apperly, Samson, & Humphreys, 2009; de Villiers, 2007). In light of these and other recent developments, some have offered
the tentative conclusion that the causal interface between language acquisition and theory of mind is bidirectional (Miller,
2009; Milligan, Astington, & Dack, 2007; de Villiers, 2007).
Although the ontogeny of theory of mind has been extensively studied over the past few decades, only recently have we
begun to understand more concerning the limits of human mentalising ability in adults. For example, several studies have
shown that the limits of mentalising ability for normal adults are consistently placed around fth-order intentionality (i.e. I
believe that you suppose that I imagine that you want me to believe that ) (Kinderman, Dunbar, & Bentall, 1998; Stiller &
Dunbar, 2007). Moreover, despite full adult competency, other studies have shown that mentalising often fails to be reli-
ably deployed in every social context (Keysar, Lin, & Barr, 2003). Further, adult ability develops over a period of time between
age 5, when children rst acquire theory of mind or second-order intentionality, and early adolescence, when they nally
acquire fth-order adult competency (Henzi, Hawker-Bond, Stiller, & de Sousa Pereira, 2007). Beyond the fact that higher-
order intentionality abilities have not yet been acquired in children under the age of 5, research on adult theory of mind
and higher-order intentionality should be of interest to developmental psychologists for at least two additional reasons. First,
an account of the mature adult system is necessary to know when development is complete. Even after children pass
developmentally sensitive theory of mind tasks their abilities are often slower and less exible than those of adults
(Dumontheil, Apperly, & Blakemore, 2010; Keulers, Evers, Stiers, & Jolles, 2010). Second, an account of the mature adult
system provides critical information for understanding why researchers observe developmental relationships between
theory of mind and other abilities such as language and executive function (Apperly et al., 2009).
Indeed, while most attention has historically been drawn to the study of child language acquisition, more recent studies
have pointed to a signicant gap in our understanding of the relation between language and adult higher-order intentionality.
Moreover, several primatologists and evolutionary psychologists have recently argued the adult capacity for higher-order
intentionality may have been critical for the emergence of many uniquely human behaviors and cultural institutions
including communication, humor, religion, cooperation, story-telling, and even consciousness (Corballis, 2011; Dunbar, 2003,
2005, 2008; Dunbar, Launay, & Curry, 2016; Graziano, 2013; Krems, Dunbar, & Neuberg, 2016; Sperber, 2000; Tomasello,
2008). Indeed, both ancient and modern humans have long occupied a social ecological niche in which the ability to
monitor and manage social interactions, reason about other's motives and intentions, keep track of relationships, and decide
who can be trusted has been of critical importance (Byrne & Whiten, 1989; Dunbar, 2003; Humphrey, 1976). Perhaps most
critically, higher-order intentionality may have also played a critical role in the evolution of language, including especially the
ability for recursive syntax comprehension and production (Cheney & Seyfarth, 2007; Dunbar, 1998, 2009).
For instance, the most prominent evolutionary accounts of human communication argue that communication involves a
speaker's expression of an intention that the listener(s) recognise(s) as the speaker's intention to inform the listener(s) d and
that the listener(s) must understand these embedded intentions (Csibra, 2010; Sperber, 2000, pp. 117e137; Tomasello, 2008),
a capacity which non-human primates are unable to duplicate (Call & Tomasello, 2008). Further, observational studies have
also shown that within typical freely-forming natural conversations, fourth or fth-order mentalising is relatively common,
especially given that many conversations are not strictly dyadic: around half of conversations contain three or more persons
(Dunbar, Duncan, & Nettle, 1995) or refer to absent third-party individuals (Dunbar, Duncan, & Marriott, 1997; Krems et al.,
2016). In principle, intentional states therefore provide a natural platform for communication through mentalising capacity
(an individual's understanding of states of mind, typically exemplied by the use of words like believe, intend, suppose, think,
et cetera). Intentionality dened in this way forms a naturally reexive hierarchy which corresponds to increasingly
embedded mindreading (i.e. I suppose that you intend that I believe that you want me to understand that ). Representing
another's mental state is thus inherently recursive and orders of intentionality beyond second-order (formal theory of mind)
are even more recursive.
 N. Oesch, R.I.M. Dunbar / Journal of Neurolinguistics 43 (2017) 95e106 97

In accord with this argument, it is reasonable to assume that when natural selection favored higher-order intentionality, it
necessarily also favored recursive thinking, as it is logically necessary to represent the content of another's mental state
within the embedded structure of one's own mental state. According to this hypothesis, recursive thinking became the
necessary cognitive scaffolding upon which language and recursive syntax could later be bootstrapped (Cheney & Seyfarth,
2007; Dunbar, 1998, 2009; see Carey, 2004; Carey, 2009; Gentner, 2010; Gentner & Christie, 2010 for further discussion of
cognitive bootstrapping). If accurate, this may explain, at least in part, how and why many languages exhibit recursive syntax.
Of course, many modern languages display recursion, often apart from whether or not the sentence contains mental state
attribution. However, this does not belie the possibility that spoken language, and the relevant psycholinguistic software,
were originally adapted for processing social information, and only later co-opted for also processing instrumental infor-
mation about the environment.
At a more practical level, this further raises the question as to whether the kinds of tasks that have traditionally been used
to assess higher-order mentalising in adults, such as the Imposing Memory Task (IMT) (a robust task designed by Kinderman
et al., 1998 composed of written narratives describing different individuals' mind states), are simply a reection of the ability
to process embedded syntax, rather than intentionality per se. Part of the aim of this study, then, is to determine whether
higher-order mentalising is more than just unpacking syntactic recursion. Conventionally, the IMT includes intentionality
questions, as well as memory questions specically designed to assess basic comprehension of the factual detail in the stories.
Crucially, the memory questions also provide further insight into individual differences likely to be important in determining
the capacity to unpack both recursion and mentalising questions: if the essential facts of the story cannot be remembered,
then individuals will not be able to correctly answer recursion or mentalising questions. It is therefore important to determine
whether individual differences in intentionality or recursion performance isn't simply a consequence of individual differences
in factual memory, as well as determine whether either are constrained by simple cognitive processes like memory capacity.
If this hypothesis is correct, then three general predictions follow with respect to adult competences on intentionality,
recursive syntax and basic memory capacity on the IMT. First, if intentionality performance and recursive syntax performance
bootstrap one another, then they should be correlated. Second, if mentalising or recursion is more cognitively demanding
than simply remembering the facts of a story, then performance on simple fact-based socially-contextualised memory tasks
should exceed performance on either equivalent mentalising or recursive syntax problem-solving. Previous studies have
found that memory for factual details from a story are remembered better than mentalising details (Kinderman et al., 1998;
Stiller & Dunbar, 2007), but in these cases, memory questions have included a mixture of social and instrumental facts from a
story. As other studies have found that social facts are better remembered and conversationally transmitted than non-social
facts (Mesoudi, Whiten, & Dunbar, 2006; Redhead & Dunbar, 2013), we use only social facts (i.e. facts about the individuals in
the stories) in the present analyses so as to ensure a more appropriate comparison. Third, if either recursive syntax or
intentionality task problem-solving scaffolds the other, then one should predict or inuence performance on the other,
independently of working memory performance. Although several studies have demonstrated that intentionality is limited to
around fth-order (Kinderman et al., 1998; Stiller & Dunbar, 2007), involving signicant neurological resources and func-
tional connectivity (Klapwijk et al., 2013; Lewis, Rezaie, Browne, Roberts, & Dunbar, 2011; Powell, Lewis, Dunbar, Garca-
~ ana, & Roberts, 2010), to date no studies have yet determined how these two factors interact with one another. In addi-
Fin
tion, as previous research has shown that women tend to have higher performance than men in both language and men-
talising abilities (Hyde & Linn, 1988; Powell et al., 2010; Stiller & Dunbar, 2007), we also tested the prediction that females
perform better than men on both intentionality and syntactical recursion tasks.

2. Methods

2.1. Participants

A questionnaire was distributed to participants via the online survey and participant recruitment system Amazon Me-
chanical Turk (AMT), an Internet application that provides instant access to thousands of potential participants for
questionnaire-based psychology experiments for a small monetary payment. Several studies have demonstrated the validity
of online surveys and shown them to be as reliable as traditional paper-and-pencil laboratory questionnaires (Buchanan &
Smith, 1999; Epstein, Klinkenberg, Wiley, & McKinley, 2001; Gosling, Vazire, Srivastava, & John, 2004; Metzger, Kristof, &
Yoest, 2003; Preckel & Thiemann, 2003; Salgado & Moscoso, 2003; Smith & Leigh, 1997). The study was advertised as
being about memory and perspective-taking. All responses were anonymous, voluntary, and restricted to individuals over
18 years of age. All participants provided informed consent before participation and were paid $0.50 for completing the
20 min survey.
In total, 210 females and 204 male undergraduates, postgraduates and working professionals from a range of backgrounds
from India (59.18%), North America (35.51%), the Middle East (1.45%), Macedonia (1.21%), Australia (0.48%), the United
Kingdom (0.48%), and other industrialised nations with available Internet access completed the task. High school education
was completed by 98.8% of participants, with 94.2% having attained a diploma or some college experience, and 72.0% having
completed a bachelor's degree or higher. All participants were native English rst language speakers. To avoid developmental
effects and confounds due to declining social engagement in old age, participants were restricted to the age range 18e65
years (M 33.6, SD 11.1).
98 N. Oesch, R.I.M. Dunbar / Journal of Neurolinguistics 43 (2017) 95e106

2.2. Materials

The Imposing Memory Task (IMT) was designed by Kinderman et al. (1998) to assess mentalising ability in adults. Ad-
ditions and changes to the IMT were then undertaken by Stiller and Dunbar (2007) and Lewis et al. (2011). The IMT consists of
a series of stories involving social situations that require the listener to understand the perspective and intentions of the
actors. A series of true/false mentalising and memory questions test subjects' memory for the events described in the stories.
The mentalising questions require increasingly complex metarepresentational understanding of a given character's
perspective of a social situation, from rst-order intentionality to as high as ninth-order intentionality (Stiller & Dunbar,
2007). The IMT has been used as a measure of theory of mind in a number of studies of both normal and clinical adult
populations (Kinderman et al., 1998; Lewis et al., 2011; Paal & Bereczkei, 2007; Powell et al., 2010, 2014; Taylor & Kinderman,
2002; Nettle & Liddle, 2008; Sylwester, Lyons, Buchanan, Nettle, & Roberts, 2012). It has also been used on typically
developing children as a way of generating a more detailed developmental trajectory of children's increasing meta-
representational capacities, and enabling direct comparisons across preschoolers and preadolescents (Henzi et al., 2007;
Liddle & Nettle, 2006). Lastly, the IMT has shown uniformly consistent results across more than half a dozen previously
published studies among various different authors and research groups (Henzi et al., 2007; Kinderman et al., 1998; Lyons,
Caldwell, & Shultz, 2010; Nettle & Liddle, 2008; Paal & Bereczkei, 2007; Powell et al., 2014, 2010; Stiller & Dunbar, 2007);
meta-analyses have further found high correlations between many of these studies and consistent validity and reliability of
the task between various different research populations (Stylianou, 2007).
The design of the stories and questions used in the present study were based on Stiller and Dunbar (2007) and Kinderman
et al. (1998), except that all question series contained between one to nine levels of intentionality (included to ensure that all
participants reached the upper limit on their performance). Firstly, the stories were slightly modied to eliminate unclear or
ambiguous wording and address potential concerns brought up by one recent study (OGrady, Kliesch, Smith, & Scott-Phillips,
2015). In addition, since not all the original stories had the full range of questions, new questions were constructed (including
complete sets of recursive syntax, memory and sixth through ninth level intentionality questions) and the others rened to
eliminate ambiguities in the originals and improve overall readability. Secondly, as few empirical studies to date have spe-
cically targeted syntactic comprehension in children (Huttenlocher, Vasilyeva, Cymerman, & Levine, 2002; de Villiers &
Pyers, 2002) and even fewer have specically targeted recursive syntax comprehension in adults (but see Apperly et al.,
2009 on embedded complement clauses in adults as one recent exception), embedded syntactic sentences were devel-
oped for the current study, which varied in their levels of recursive syntax. For instance, the statement Betty is a roommate of
Brian, who walked into the bedroom of Frank, who received 10 from Brian, who likes lettuce and tomatoes for a salad, that Betty
and Brian helped to prepare, demonstrates an example of a sentence with ve embedded clauses. A validity and reliability
check of the modied materials on a small test sample (N 50) revealed the mean and variance on intentionality questions
(M 5.28) and memory questions (M 5.91) was similar to that previously reported (Kinderman et al., 1998; Lyons et al.,
2010; Stiller & Dunbar, 2007) and indicated that participants both remembered as well as fully understood the narratives
and test questions, despite the international sample. The full set of all three stories is provided as Supplementary material.
Altogether, these improvements to our narratives ensured that the questions could only be answered correctly if par-
ticipants had accurately understood the higher-order intentionality, recursive syntax and memory control questions. Firstly,
we rened each set of performance measures in order to ensure good construct validity for each of the three task questions.
More specically, while the higher-order intentionality and recursive syntax questions were very similar to one another, the
differences they did have were precise only to those aspects that involved higher-order intentionality. For example, one
fourth-order intentionality question stated Brian knew Frank thought that Brian believed that Betty wanted some in-
gredients for breakfast, while the corresponding fourth-order recursive syntax question stated Frank becomes confused by
Betty, who tells him not to worry about Brian, who has no preference for organic produce. Critically, the two sentences are
very similar in both syntax and semantics, yet the intentionality question includes mental-state attribution, while the
recursive syntax question contains no aspects of mentalising.
Secondly, we eliminated strategies unrelated to mentalising based on forms of simple substitutions, broken conceptual
chains or shortcuts based on parsing the sentence into simpler units, ensuring that participants would not succeed at levels
better than chance (O'Grady et al., 2015). More specically, because of the fact that at any point in the sentence, there could
potentially have been a negation or other qualier, this required participants to read and interpret the entire sentence, and
not just parse separate clauses. For example, in the fth-order mentalising question Frank believed Betty understood that
Brian was of the opinion that Betty thought that Frank believed all along that Brian was lazy about house duties, it could be
assumed that the sentence only requires parsing of the nal clause Brian was lazy about the house duties, a statement which
is in itself false according to the original narrative. However, the simple qualier that Frank believed Betty understood that
Brian was of the opinion that Betty thought that Frank didn't believe all along that Brian was lazy about house duties', renders
the entire sentence true. Accordingly, this means that in order to answer the question correctly, participants needed to parse
the entire sentence, which requires understanding four levels of intentionality. Similarly, if additional negations or qualiers
were also used at additional points in the construction (e.g. Frank believed Betty understood that Brian wasn't of the opinion
that Betty thought that Frank didn't believe all along that Brian was lazy about house duties'), then participants again needed
to change their evaluation of the truth value of the sentence. Indeed, several of the questions in the test battery included
statements with negations and other qualiers, such as in the sentence Peter certainly seems to believe that Fiona probably
 N. Oesch, R.I.M. Dunbar / Journal of Neurolinguistics 43 (2017) 95e106 99

thinks that Sophie likely knows that Peter actually thinks that he is unfortunately embarrassed by Gavin's resolute feeling that
Peter probably believes that he is not quite as qualied as Gavin is inclined to think he is with respect to the job.
Third, we also reduced ambiguities in the original narratives in order to ensure that participants correctly understood each
question. For example, in the fth-order recursive syntax question Betty is a roommate of Brian, who walked into the
bedroom of Frank, who received 10 from Brian, who likes lettuce and tomatoes for a salad, it could be assumed that the
sentence could be parsed as Betty is a roommate of Brian, and she walked into the bedroom of Frank, and she received 10
from Brian, and she likes lettuce and tomatoes for a salad, or alternatively as Betty is a roommate of Brian; Brian walked into
the bedroom of Frank; Frank received 10 from Brian; Brian likes lettuce and tomatoes for a salad. Notably, however, the
statements were not provided in isolation from the stories; indeed, the narratives were further rened and presented in order
to provide relevant context and ensure that all relevant information was available to distinguish among potential ambiguities.
Moreover, participants were informed at the outset to assume the most straightforward interpretation of each statement
possible, assuming normal rules of grammar and basic comprehension. In summary, our questions were designed to ensure
that participants had to parse and comprehend the entire narrative as well as the complete statement in order to perform at
levels above chance.

2.3. Procedure

Participants were rst asked to provide information about their age, sex, nationality, education, income, whether they
were native English speakers, and to give informed consent. Subjects were then asked to read a series of three short stories
(each approximately 200 words long) depicting a social situation. Participants were permitted to read over the narrative for as
long as they required, but once they began answering questions, were not permitted to again the view the story. After each
story had been read, participants answered a randomly ordered series of three different types of socially situated fact-based
memory questions: 1) levels 1e9 intentionality questions, 2) 1e9 embedded clause structure questions (containing
increasingly complex recursive syntax, but having no aspects of mindreading), and 3) 1e9 sequential clause structure
questions (containing consecutive single clauses with no syntactic embedding or mindreading). The sequential clause
structure questions were intended to target mainly working memory and served as a control for the other two series of
questions. Each question contained a statement and the participant was asked to decide whether the statement was in fact
true or false. The mean rate of time to complete the online survey was approximately 20 min, which agrees well with
completion times under live test conditions.
As previously noted, participants were instructed to assume the simplest and most straightforward interpretation of each
statement possible, and not to assume the questions were specically designed to be unnecessarily complicated or that they
were being deliberately deceived. All story questions were counter-balanced with an equivalent number of true or false
correct answers to avoid any systematic bias in guessing or answering unknown questions (see Supplementary material). In
addition, an instructional manipulation check (IMC) was implemented to detect satiscing and reduce statistical noise from
non-diligent participants (Oppenheimer, Meyvis, & Davidenko, 2009). Participants were excluded from analysis if they did
not complete the survey, failed the manipulation check or if they completed the survey in too short a time to have paid full
attention (<5 min). The manipulation check consisted of a question embedded in the survey in which participants were
requested to put a certain digit to conrm that they were paying attention. Participants who failed the manipulation check,
completed the survey in too short a time to have paid attention, or indicated they were non-native English speakers were
excluded (N 22). Participants were nally debriefed on the full aims upon completion of the study. The study was approved
by the local Research Ethics Committee.
To score performance on each task, we followed Stiller and Dunbar (2007) summing all correct answers and rescaling by
the number of stories (three, in this particular case) to vary between 0 and 9 (the number of levels being tested). While
questions of increasing length arguably become increasingly unnatural, we elected to include a full range of difculty in order
to ensure that subjects were taxed to their absolute limits. Moreover, because our study is the rst of its kind to investigate
this hypothesis, our main focus here was to demonstrate proof of concept, arguably at the expense of some degree of
ecological validity. Because all questions were structured in terms of units of information (essentially clauses, embedded or
otherwise), we interpreted the baseline cognitive load at any given level as equivalent across tasks, save the additional
cognitive load introduced by each particular cognitive task. In other words, we assume that fourth-order memory, syntactical
recursion and intentionality tasks share the same level of language comprehension processing, and that differences between
them therefore reect the added processing costs of the task itself.

2.4. Experimental design

The design of the stories and questions used in the present study were based on Stiller and Dunbar (2007) and Kinderman
et al. (1998), except that all question series contained between one to nine levels of intentionality (included to ensure that all
participants reached the upper limit on their performance). The stories were slightly modied to eliminate unclear or
ambiguous wording. In addition, since not all the original stories had the full range of questions, new questions were con-
structed (including complete sets of recursive syntax, memory and sixth through ninth level intentionality questions) and the
others rened to eliminate ambiguities in the originals and improve overall readability.
100 N. Oesch, R.I.M. Dunbar / Journal of Neurolinguistics 43 (2017) 95e106

The rst series of questions designed for the current study were composed of recursive syntax true/false statements.
Indeed, few empirical studies to date have specically targeted syntactic comprehension in children (Huttenlocher et al.,
2002; de Villiers & Pyers, 2002) and even fewer have specically targeted recursive syntax comprehension in adults (but
see Apperly et al., 2009 on embedded complement clauses in adults as one recent exception). Therefore, embedded syntactic
sentences were developed for the current study, which varied in their levels of recursive syntax. The level of complexity for
each embedded syntax question was calculated according to the number of embedded clauses for each sentence. For instance,
the statement: Betty is a roommate of Brian,1 who walked into the bedroom of Frank,2 who received 10 from Brian,3 who likes
lettuce and tomatoes for a salad,4 that Betty and Brian helped to prepare,5 has ve embedded clauses (subscripts count the
number of embedded clauses).
The next series of questions designed for the current study were composed of social memory true/false statements. The
social memory questions served as a control for the subject's comprehension of basic sentences and understanding of the
questioning procedure. In addition, they also provided a more general index of basic memory capacity. For these purposes, we
opted for the simplest possible measure of memory capacity (i.e. working memory for social facts) in order to differentiate as
clearly as possible between mentalising ability, complex recursion, and basic memory capacity (with recursive complexity
being held constant). As such, the level of complexity for each social memory question was calculated according to the
number of simple clauses for each sentence. For instance, the statement: The friends share a house,1 but Frank is lazy,2 so Brian
gives him 10,3 but this makes Frank confused,4 so Brian writes a list,5 has ve simple clauses (subscripts count the number of
simple clauses).
The last series of questions designed for the current study were composed of higher-order intentionality true/false
statements. Indeed, previous studies have found that for most normal healthy adults the highest level of mindreading
competency is around fth-order intentionality (Stiller & Dunbar, 2007). Beyond this level of intentionality, most adults fail
IMT questions (Kinderman et al., 1998; Stiller & Dunbar, 2007). The level of complexity for each higher-order intentionality
question was calculated according to the number of intentions or belief states for each sentence. For instance, the statement:
Simon believes1 that Martin thinks2 that Charlotte supposes3 that Jane knows4 that Simon thinks5 they had a lovely date at the
restaurant, has ve levels of intentionality (subscripts count the number of metarepresentations). Moreover, since not all of
the original stories had the full range of questions, new intentionality questions were constructed, including complete sets of
sixth through ninth level intentionality questions. Therefore, the most complex mentalising questions contained up to nine
orders of perspective-taking (including the participant's own perspective, dened as level one). The full set of all three stories,
and subsequent social memory, embedded syntax, and higher-order intentionality sentence questions, are provided as
Supplementary material.

2.5. Dependent and independent variables

The recursive clause structure questions required cognitive processing of linguistic structure and constituted the central
dependent variable. The higher-order intentionality questions required complex mentalising over a character's perspective in
a social situation and constituted the central independent variable. The social memory questions provided a more general
index of working memory capacity, served as a control for the basic understanding of the stories and constituted a second
central independent variable. Further, the social memory questions were also used both to reduce the demand characteristics
of the study (by obscuring the central hypothesis under consideration) and to provide some convergent evidence for the
effectiveness of the experimental manipulation. The study used a within-subjects design for all three dependent and inde-
pendent variables.

2.6. Statistical analysis

Regression analysis was used to derive correlations between age, ethnicity and education as control variables for all levels
of intentionality, recursive syntax and memory questions. Age, ethnicity and education were found to be non-signicantly
related to performance for all question types. In addition, probabilistic Bayesian network analysis was used to determine
the structural and causal relationships between the three core variables of memory, recursive syntax and intentionality.
Bayesian, or probabilistic causal networks, are directed acyclic graphs (DAGs) which calculate the conditional probabilities of
causality for the variables in an experiment (Pearl, 2009; Pearl, Glymour, & Jewell, 2016). Bayesian networks have the added
utility in that they can calculate these probabilities when directly manipulating such variables in an experiment is impossible
or unethical. Accordingly, it was therefore used here to identify the primary factors driving correlations in performance
among the three tasks.

3. Results

Overall, performance across all three narratives (indexed as the mean number of correct answers for each story) was
approximately normally distributed with a peak at level 5 for intentionality questions (M 5.26, SD 0.96) and recursive
syntax questions (M 5.64, SD 1.24) (Fig. 1). Performance on factual memory questions was slightly higher (M 5.93,
SD 1.13) with a peak that was noticeably more left-skewed (Fig. 1). The mean and variance in intentionality and memory
questions were similar to that previously reported (Kinderman et al., 1998; Lyons et al., 2010; Stiller & Dunbar, 2007).
 N. Oesch, R.I.M. Dunbar / Journal of Neurolinguistics 43 (2017) 95e106 101

Fig. 1. The level of intentionality (black bars), recursive syntax (white bars) and memory (stippled bars) at which participants achieved a particular average sum
of points for all three narratives over all nine levels of task measures. Individual values are rounded to the nearest whole number (N 414).

A one-way ANOVA was used to determine whether individuals performed at higher levels on one task compared to
another. The differences across the three variables are in fact highly signicant (one-way ANOVA F2,1239 37.46, p < 0.0001,
h2 0.06), with signicant pairwise differences between all three variables (Scheffe
tests: memory vs syntax, p < 0.001;
memory vs mentalising, p < 0.001; syntax vs mentalising, p 0.001). Notably, the effect size here indicates moderate dif-
ferences in performance for each task measure. As a result, the signicant differences suggest that although the tasks were
indeed very similar to one another, they were in fact measuring fundamentally different constructs. Lastly, in accord with
predictions, women had signicantly higher scores than men on both intentionality performance (M 5.49 vs. M 5.03,
respectively; independent samples t-tests, t412 4.996, p < 0.001, Cohen's d 0.492) and recursive syntax performance
(M 5.85 vs. M 5.43, respectively; t412 3.461, p < 0.001, Cohen's d 0.341), but not on memory performance (M 6.00 vs.
M 5.85, respectively; t412 1.381, p 0.168, Cohen's d 0.136) (Table 1).
Fig. 2 plots individual scores for intentionality regressed on recursive syntax scores, with the least squares regression line
as well as the line of equality. There is a signicant relationship between the two variables (F1,412 236.8, p < 0.0001), but the
slope is signicantly lower than the line of equality (i.e. a slope b 1) (t413 17.8, p 0.0001). The pitch of the regression
slope suggests that recursive syntax abilities are lower than intentionality competences below fth-order, but that the reverse
is the case at higher values indicating that intentionality asymptotes at a lower value than those for recursive syntax.
Finally, we used a probabilistic Bayesian network analysis from the pcalg package implemented in R version 3.0.1
(Colombo, Hauser, Kalisch, & Maechler, 2014) to determine the structural and causal relationships between the three core
variables, memory, recursive syntax and intentionality and identify the primary factors driving correlations in performance
among the three tasks. As shown in Fig. 3, there were robust positive interactions between memory and recursive syntax
(rpartial(414) 0.434, p < 0.001) and between intentionality and recursive syntax (rpartial(414) 0.416, p < 0.001), with a much
weaker interaction between memory and intentionality (rpartial(414) 0.255, p < 0.001).

4. Discussion

The results conrm previous ndings that there is an upper limit on performance on mindreading tasks at the equivalent
of fth-order intentionality (Kinderman et al., 1998; Stiller & Dunbar, 2007). Several other studies have also found a typical
fth-order limit on mentalising tasks (Lyons et al., 2010; Nettle & Liddle, 2008; Paal & Bereczkei, 2007) and there is evidence
to support the claim that mentalising tasks are signicantly more demanding of neurological resources and functional
connectivity than factual memory tasks (Klapwijk et al., 2013; Lewis et al., 2011; Powell et al., 2010). Inspection of Fig. 1
indicates that recursive syntax competences also exceed performance on intentionality, suggesting that they too may be

Table 1
Mean values of major independent and dependent variables.

Factual question condition

Mean Male Female

Intentionality 5.26 (SE 0.05) 5.03 (SE 0.07) 5.49 (SE 0.06)
Recursive syntax 5.64 (SE 0.06) 5.43 (SE 0.08) 5.85 (SE 0.09)
Memory 5.93 (SE 0.06) 5.85 (SE 0.08) 6.00 (SE 0.08)
102 N. Oesch, R.I.M. Dunbar / Journal of Neurolinguistics 43 (2017) 95e106

Fig. 2. Performance on intentionality plotted against performance on recursive syntax. The line of equality is indicated as a dashed line and least squares
regression line as a solid line. The regression equation is given for the regression line. The open symbols here represent the overlap of between 1 and 2 datapoints
while lled symbols represent between 3 and 12 datapoints.

Fig. 3. Bayesian network analysis of intentionality, recursive syntax and memory as dependent variables depicted as a graphical model.

cognitively less demanding (though more demanding than simple factual memory that does not involve recursive embed-
ding). In addition, inspection of the means in Table 1 reveals that females performed signicantly better than males on both
intentionality performance and recursive syntax performance. This is consistent with previous research showing that women
typically perform slightly better than men on both abilities (Hyde & Linn, 1988; Powell et al., 2010; Stiller & Dunbar, 2007).
Indeed, it is conceivable that sex differences in mentalising ability may partially account for sex differences in language ability.
The more important nding of this study, however, is that higher-order intentionality competence correlates with
recursive syntax performance. This result is consistent with the rst general prediction, as a correlational relationship
constitutes an initial prerequisite for a cognitive bootstrapping effect. The results are also consistent with the second pre-
diction. The ndings revealed a signicantly greater performance for memory of socially-situated sequential clause structure
questions than for either intentionality questions or embedded clause structure questions. This is in line with the suggestion
that as the more evolutionarily ancient primate cognitive trait (Rushworth, Mars, & Sallet, 2013), simple fact-based social
memory tasks should be cognitively less demanding than equivalent fact-based mentalising or embedded syntax problem-
solving (Kinderman et al., 1998; Stiller & Dunbar, 2007). This result is also consistent with previous studies which have shown
that social facts are better remembered and more easily transmitted conversationally than non-social facts (Mesoudi et al.,
2006; Redhead & Dunbar, 2013).
The third prediction was that there would be a scaffolding relationship between intentionality and recursive syntax.
Although this is implicit in the fact that the two are correlated, deciding which way the causal sequence runs requires a closer
examination of the ndings from the path analysis. Firstly, the mean value for recursive syntax performance was signicantly
greater than the mean value for intentionality performance (Fig. 1), suggesting that intentionality may be cognitively more
demanding than syntactical recursion per se. Secondly, the fact that the regression equation for intentionality regressed on
recursive syntax is signicantly shallower than the line of equality might lead us to suppose that recursive syntax scaffolds
intentionality, thereby perhaps allowing us to create the higher orders of intentionality via a propositional calculus
 N. Oesch, R.I.M. Dunbar / Journal of Neurolinguistics 43 (2017) 95e106 103

underpinned by grammatical recursion. This would be in accord with a linear interpretation of the results of the Bayesian
network analysis of Fig. 3: memory constrains recursive syntax, which in turn constrains intentionality.
However, an alternative interpretation of the path diagram in Fig. 3 is that memory and intentionality independently
determine recursive syntax, with both being independently determined according to a developmental process giving rise to
individual differences. This V-shaped interpretation would seem to t with the neuroanatomical evidence suggesting that
working memory is primarily associated with the prefrontal cortex (Cowan, 2008; Diamond, 2013; Malenka, Nestler &
Hyman, 2009) whereas mentalising competences and language are additionally associated with the temporoparietal junc-
tion (TPJ) and the temporal lobes, especially including the superior temporal sulcus (STS) (Decety & Lamm, 2007; Gallaher
et al., 2000; Hickok & Small, 2015; Samson, Apperly, Chiavarino, & Humphreys, 2004; Saxe & Kanwisher, 2003). The fact
that memory has a weak correlation with intentionality that is independent of its relationship with recursive syntax tends to
favour a V-shaped interpretation rather than a linear one. Accordingly, this tentatively suggests that higher-order inten-
tionality competence may predict recursive syntax performance independently of working memory capacity. If so, this
nding would support the claim that higher-order intentionality capacity provides the cognitive scaffolding with which
recursive syntax utilises for a cognitive bootstrapping effect, perhaps explaining in part how and why many modern lan-
guages exhibit recursive syntax.
That said, Fig. 2 can also be read as suggesting a complex bidirectional relationship: from rst-order through fth-order
intentionality, mentalising performance exceeds recursive syntax performance, but this reverses at higher orders of inten-
tionality. This could be interpreted as suggesting that, initially, intentionality provides the cognitive scaffolding which
recursive syntax utilises for a cognitive bootstrapping effect, but that recursive syntax may provide the scaffolding for yet
higher intentionality competences once a given level of language competence has been achieved. In effect, a positive feedback
system is achieved whereby a systematic linguistic schema facilitates a systematic conceptual schema. According to this
linguistic toolkit view, language provides new resources that augment human cognitive capacities, but do not replace pre-
linguistic abilities (Gentner, 2010; Gentner & Christie, 2010). This might explain how some individuals develop advanced
mentalising skills above fth-order intentionality. In other words, rst-order through fth-order intentionality may be
necessary to assist the processing of simpler syntactic structures, but beyond fth-order intentionality (once mentalising
ability has been taxed to its conventional uppermost limit) the cognitive scaffolding provided by recursive syntax may be
engaged to enable even higher-order mentalising in those individuals whose language skills allow them this capacity.
A potential objection to the results reported here is that the strong correlation seen between recursive syntax and
intentionality ability is merely a product of general intelligence; however, given the generally lower correlations typically
recorded between the various general intelligence subtests, such as vocabulary, comprehension, arithmetic, picture
completion, and digit span (Chabris, 2007), this interpretation seems unlikely. A second potential objection is that the cor-
relations between memory and intentionality, and memory and recursive syntax, introduce a confound into the relationship
between intentionality and recursive syntax performance. Certainly, as implied by Fig. 3, memory is a signicant predictor
variable, especially with respect to recursive syntax performance above fth-order intentionality. However, for many reasons,
this is hardly surprising; previous studies have shown working memory to be strongly related to performance on a wide range
of complex cognitive tasks, including reading comprehension, problem-solving, and general intelligence (Ackerman, Beier, &
Boyle, 2005; Conway, Kane, & Engle, 2003). More importantly, whatever effect differences in memory capacity might have,
the path analysis conrms that intentionality is independently correlated with syntax. A third potential objection is that both
the intentionality and recursive syntax task questions included linguistic embedding and were therefore correlated solely for
that reason. However, there are at least two reasons to think this unlikely. Firstly, as has been previously noted, the effect size
for all three question types indicated medium differences in performance for each measure. This suggests that although the
tasks were indeed very similar to one another they were in fact measuring fundamentally different constructs. Consequently,
it is unlikely that any differences in performance between the recursive syntax and intentionality tasks are due solely to the
syntactic demands of the mentalising questions, rather than to mindreading specically. The most likely interpretation of
these results is that they are sequentially inclusive, with each stage introducing an extra cognitive processing demand:
recursion depends on memory for the facts of the story, plus unpacking syntactical structure, and intentionality depends on
memory and recursion, plus unpacking mind states. Secondly, the different correlation strengths found between each of the
three factors in the Bayesian network analysis indicate signicantly different weightings driving performance for each of the
three variables. A perfect correlation between any of the three variables would indeed support the claim that any of the three
tasks are fundamentally measuring the same construct; however, this is not what we nd.
A fourth potential objection is that the Bayesian network analysis employed here cannot establish intentionality as the
primary causal variable. Although no single experiment can denitively establish causality, Bayesian network analysis is
nevertheless considered to be a valid and widely accepted statistical tool for calculating the conditional probabilities of
causality among several factors in an experiment (Pearl, 2009; Pearl et al., 2016). Moreover, our results agree well with
previous invasive lesioning studies, which suggest that impaired executive function has multiple effects on mature adult
theory of mind (Apperly et al., 2009), but that severely impaired grammar can leave theory of mind functionally intact
(Apperly et al., 2009; Varley & Siegal, 2000). For example, several studies of patients who have signicantly impaired
grammar on standard language tests, following left hemisphere lesions, nevertheless perform well on nonverbal theory of
mind tasks (Varley & Siegal, 2000; Varley, Siegal, & Want, 2001; Willems, Benn, Hagoort, Toni, & Varley, 2011). Nonetheless,
our results should be considered preliminary in lieu of further invasive experiments.
104 N. Oesch, R.I.M. Dunbar / Journal of Neurolinguistics 43 (2017) 95e106

A fth potential objection to our results is raised by O'Grady et al. (2015) who claim the current reported limit on human
mentalising capacity is too conservative; indeed, their results suggested a limit on human metarepresentation at seventh-
order intentionality. Their unusual results, it was further argued, were allegedly due to particular methodological aws in
previous research, including broken conceptual chains, simple substitutions and impossible choices among a few question
types. However, their conclusions are subject to question for at least ve critical reasons. (1) By their own estimation, the
proportion of awed intentionality questions in previous studies amounted to ~25% with broken conceptual chains, <1%
with simple substitutions, and <15% with impossible choices: however, even if these estimates are accurate, and might be
responsible for the overall level of variance across individuals, it should not substantively change the overall pattern of results.
In particular, if the broken conceptual chains argument is in fact true, in that a small proportion of awed test questions
actually test the conjunction of multiple cases of second-level mindreading rather than the intended levels of third, fourth
or fth-order intentionality, one might predict performance to be greater than the mean fth-order intentionality reported by
most studies (and lower, not higher, if corrected in the O'Grady et al. (2015) study); but, in fact, the results have been uni-
formly consistent across studies and various research groups (Kinderman et al., 1998; Lyons et al., 2010; Stiller & Dunbar,
2007). (2) The O'Grady et al. (2015) study used a comparatively small sample size (N 66), the vast majority of which
were female participants. Critically, females are consistently reported to have higher mentalising abilities than males in
earlier studies (Lyons et al., 2010; Stiller & Dunbar, 2007) as well as the present study. (3) According to their self-reported
condence ratings, condence actually went down signicantly at higher levels of intentionality, suggesting that partici-
pants may have in fact been successfully guessing; indeed, their particular methodological approach took few measures to
safeguard against random guessing (e.g. counter-balancing the number of true/false correct answers to account for potential
answering bias as was done in the current study). (4) Unlike most previous administrations of the IMT, where participants
were allowed a single exposure to each narrative and test question, the O'Grady et al. (2015) study permitted participants to
view the same test stimuli as often as they wished without limit in order to aid their recall and comprehension of the
narratives; predictably, participants did in fact choose to view the test stimuli more frequently at higher orders of inten-
tionality. Finally, (5) O'Grady et al. (2015) found no signicant differences in mentalising between their implicit and explicit
stimuli questions; in other words, they failed to replicate the results of several other previous studies (Kinderman et al., 1998;
Lyons et al., 2010; Stiller & Dunbar, 2007). In fact, the unusual ndings of the O'Grady et al. (2015) study remain alone against
the uniformly consistent results of many IMT studies across various different authors, research groups and cultures (Henzi
et al., 2007; Kinderman et al., 1998; Lyons et al., 2010; Nettle & Liddle, 2008; Paal & Bereczkei, 2007; Powell et al., 2014,
2010; Stiller & Dunbar, 2007); meta-analyses have further found high correlations between many of these studies and
consistent validity and reliability of the task between various different research populations (Stylianou, 2007).
A nal potential objection is that some task questions at the highest orders of recursive syntax and intentionality arguably
became increasingly unnatural, thereby sacricing some degree of ecological validity. However, given that grammar is a
discrete combinatorial system, regularly producing an unlimited number of completely distinct combinations consisting of an
innite range of properties, with most combinations never having been previously produced, there really is no statistically
typical type of sentence or construction (Harley, 2014). Indeed, as most spoken sentences are in fact novel constructions,
most of spoken language, including constructions adhering to the rules of good prescriptive grammar, consists of atypical
sentences (Pinker, 1998); accordingly, the questions presented here had good ecological validity.
In conclusion, this study supports the claim that adult higher-order intentionality capacity provides the cognitive scaf-
folding with which recursive syntax utilises for a cognitive bootstrapping effect, perhaps explaining in part how and why so
many modern languages exhibit recursive syntax. Additionally, our results further suggest a bidirectional relationship, with
higher-order intentionality capacities up to fth-order bootstrapping recursive syntax, and grammatical language capacities
bootstrapping intentionality so as to achieve unusually higher orders of mentalising. Understanding the association between
higher-order intentionality and recursive syntax comprehension brings us some way towards understanding the mechanisms
that have contributed to the communicative complexity of our species.

Acknowledgments

NO and RD are supported by a European Research Council Advanced Grant (Grant Number 295663) to RD.

Appendix A. Supplementary data

Supplementary data related to this article can be found at http://dx.doi.org/10.1016/j.jneuroling.2016.09.008.

References

Ackerman, P. L., Beier, M. E., & Boyle, M. O. (2005). Working memory and intelligence:The same or different constructs? Psychological Bulletin, 131, 30e60.
Apperly, I. A., & Butterll, S. A. (2009). Do humans have two systems to track beliefs and belief-like states? Psychological Review, 116, 953e970.
Apperly, I. A., Samson, D., & Humphreys, G. W. (2009). Studies of adults can inform accounts of theory of mind development. Developmental Psychology,
45(1), 190e201.
Astington, J. W., & Jenkins, J. M. (1995). A longitudinal study of the relation between language and theory-of-mind development. Developmental Psychology,
35, 1311e1320.
Baillargeon, R., Scott, R. M., & He, Z. (2010). False-belief understanding in infants. Trends in Cognitive Sciences, 14(3), 110e118.
 N. Oesch, R.I.M. Dunbar / Journal of Neurolinguistics 43 (2017) 95e106 105

Buchanan, T., & Smith, J. L. (1999). Using the Internet for psychological research: Personality testing on the world wide web. British Journal of Psychology, 90,
125e144.
Byrne, R. W., & Whiten, A. (1989). Machiavellian intelligence: Social expertise and the evolution of intellect in monkeys, apes, and humans. Oxford: Oxford
University Press.
Call, J., & Tomasello, M. (2008). Does the chimpanzee have a theory of mind? 30 years later. Trends in Cognitive Sciences, 12(5), 187e192.
Carey, S. (2004). Bootstrapping and the origin of concepts. Daedalus, 133(1), 59e68.
Carey, S. (2009). The origin of concepts. New York: Oxford University Press.
Chabris, C. F. (2007). Cognitive and neurobiological mechanisms for the law of general intelligence. In M. J. Roberts (Ed.), Integrating the mind: Domain
general versus domain specic processes in higher cognition (pp. 449e491). Hove, UK: Psychology Press.
Cheney, D. L., & Seyfarth, R. M. (2007). Baboon metaphysics: The evolution of a social mind. Chicago: University of Chicago Press.
Colombo, D., Hauser, A., Kalisch, M., & Maechler, M. (2014). pcalg: Methods for graphical models and causal inference. R Package Version, 2.0-3 http://pcalg.r-
forge.r-project.org.
Conway, A. R., Kane, M. J., & Engle, R. W. (2003). Working memory capacity and its relation to general intelligence. Trends in Cognitive Sciences, 7(12),
547e552.
Cowan, N. (2008). What are the differences between long-term, short-term, and working memory? Progress in Brain Research, 169, 323e338.
Corballis, M. C. (2011). The recursive mind: The origins of human language, thought, and civilization. Princeton, NJ: Princeton University Press.
Csibra, G. (2010). Recognizing communicative intentions in infancy. Mind & Language, 25(2), 141e168.
Decety, J., & Lamm, C. (2007). The role of the right temporoparietal junction in social interaction: How low-level computational processes contribute to
meta-cognition. The Neuroscientist, 13(6), 580e593.
Diamond, A. (2013). Executive functions. Annual Review of Psychology, 64, 135e168.
Dumontheil, I., Apperly, I. A., & Blakemore, S.-J. (2010). Online usage of theory of mind continues to develop in late adolescence. Developmental Science, 13,
331e338.
Dunbar, R. I. M. (1998). Theory of mind and the evolution of language. In J. R. Hurford, M. Studdert-Kennedy, & C. Knight (Eds.), Approaches to the evolution of
language (pp. 92e110). Cambridge: Cambridge University Press.
Dunbar, R. I. M. (2003). The social brain: Mind, language, and society in evolutionary perspective. Annual Review of Anthropology, 32, 163e181.
Dunbar, R. I. M. (2005). Why are good writers so rare? An evolutionary perspective on literature. Journal of Cultural and Evolutionary Psychology, 3(1), 7e21.
Dunbar, R. I. M. (2008). Mind the gap: Or why humans aren't just great apes. Proceedings of the British Academy, 154, 403e423.
Dunbar, R. I. M. (2009). Why only humans have language. In R. Botha, & C. Knight (Eds.), The prehistory of language (pp. 12e35). Oxford: Oxford University
Press.
Dunbar, R. I. M., Duncan, N. D. C., & Marriott, A. (1997). Human conversational behaviour. Human Nature, 8, 231e246.
Dunbar, R. I. M., Duncan, N. D. C., & Nettle, D. (1995). Size and structure of freely forming conversational groups. Human Nature, 6, 67e78.
Dunbar, R. I. M., Launay, J., & Curry, O. (2016). The complexity of jokes is limited by cognitive constraints on mentalizing. Human Nature, 27(2), 130e140.
Dunn, J., Brown, J., Slomkowski, C., Tesla, C., & Youngblade, L. (1991). Young children's understanding of other people's feelings and beliefs: Individual
differences and their antecedents. Child Development, 62, 1352e1366.
Epstein, J., Klinkenberg, W. D., Wiley, D., & McKinley, L. (2001). Insuring sample equivalence across Internet and paper-and-pencil assessments. Computers in
Human Behavior, 17(3), 339e346.
Farrar, M. J., & Maag, L. (2002). Early language development and the emergence of a theory of mind. First Language, 22, 197e213.
Gale, E., de Villiers, P., de Villiers, J., and Pyers, J. (1996). Language and theory of mind in oral deaf children. Paper presented at the 20th Boston University
Conference on Language Development, November, Somerville, MA.
Gallaher, H. L., Happe
, F., Brunswick, N., Fletcher, P. C., Frith, U., & Frith, C. D. (2000). Reading the mind in cartoons and stories: An fMRI study of theory of
mind in verbal and nonverbal tasks. Neuropsychologia, 38(1), 11e21.
Gentner, D. (2010). Bootstrapping the mind: Analogical processes and symbol systems. Cognitive Science, 34, 752e775.
Gentner, D., & Christie, S. (2010). Mutual bootstrapping between language and analogical processing. Language and Cognition, 2(2), 261e283.
Gosling, S. D., Vazire, S., Srivastava, S., & John, O. P. (2004). Should we trust web-based studies? A comparative analysis of six preconceptions about Internet
questionnaires. American Psychologist, 59(2), 93e104.
Graziano, M. S. (2013). Consciousness and the social brain. Oxford: Oxford University Press.
Harley, T. (2014). The psychology of language: From data to theory (4th ed.). New York: Psychology Press.
Henzi, S., Hawker-Bond, D., Stiller, J., & de Sousa Pereira, L. (2007). Look who's talking: Developmental trends in the size of conversational cliques. Evolution
and Human Behavior, 28(1), 66e74.
Hickok, G., & Small, S. L. (2015). Neurobiology of language. London: Academic Press.
Humphrey, N. K. (1976). The social function of intellect. In P. P. G. Bateson, & R. Hinde (Eds.), Growing points in ethology (pp. 303e317). Cambridge:
Cambridge University Press.
Huttenlocher, J., Vasilyeva, M., Cymerman, E., & Levine, S. (2002). Language input and child syntax. Cognitive Psychology, 45, 337e374.
Hyde, J. S., & Linn, M. C. (1988). Gender differences in verbal ability: A meta-analysis. Psychological Bulletin, 104, 53e69.
Keulers, E. H. H., Evers, E. A. T., Stiers, P., & Jolles, J. (2010). Age, sex and pubertal phase inuence mentalizing about emotions and actions in adolescents.
Developmental Neuropsychology, 35(5), 555e569.
Keysar, B., Lin, S., & Barr, D. J. (2003). Limits on theory of mind use in adults. Cognition, 89, 25e41.
Kinderman, P., Dunbar, R., & Bentall, R. P. (1998). Theory-of-mind decits and causal attributions. British Journal of Psychology, 89, 191e204.
Klapwijk, E. T., Goddings, A.-L., Heyes, S. B., Bird, G., Viner, R. M., & Blakemore, S.-J. (2013). Increased functional connectivity with puberty in the mentalising
network involved in social emotion processing. Hormones and Behavior, 64, 314e322.
Kova
cs, A.
M., Te
gla
s, E., & Endress, A. D. (2010). The social sense: Susceptibility to others' beliefs in human infants and adults. Science, 330(6012),
1830e1834.
Krems, J. A., Dunbar, R. I. M., & Neuberg, S. L. (2016). Something to talk about: Are conversation sizes constrained by mental modeling abilities? Evolution
and Human Behavior, 37(5), 1e25.
Lewis, P. A., Rezaie, R., Browne, R., Roberts, N., & Dunbar, R. I. M. (2011). Ventromedial prefrontal volume predicts understanding of others and social
network size. NeuroImage, 57, 1624e1629.
Liddle, B., & Nettle, D. (2006). Higher-order theory of mind and social competence in school-age children. Journal of Cultural and Evolutionary Psychology, 4,
231e246.
Low, J. (2010). Preschoolers' implicit and explicit false-belief understanding: Relations with complex syntactical mastery. Child Development, 81, 579e615.
Luo, Y. (2011). Do 10-month-old infants understand others' false beliefs? Cognition, 121, 289e298.
Lyons, M., Caldwell, T., & Shultz, S. (2010). Mind-reading and manipulation - Is machiavellianism related to theory of mind? Journal of Evolutionary Psy-
chology, 8(3), 261e274.
Malenka, R. C., Nestler, E. J., & Hyman, S. E. (2009). Higher cognitive function and behavioral control. In A. Sydor, & R. Y. Brown (Eds.), Molecular neuro-
pharmacology: A foundation for clinical neuroscience (2nd ed., pp. 313e321). New York: McGraw-Hill Medical.
Mesoudi, A., Whiten, A., & Dunbar, R. (2006). A bias for social information in human cultural transmission. British Journal of Psychology, 97, 405e423.
Metzger, M. M., Kristof, V. L., & Yoest, D. J. (2003). The world wide web and the laboratory: A comparison using face recognition. Cyberpsychology and
Behavior, 6(6), 613e621.
Miller, S. A. (2009). Children's understanding of second-order mental states. Psychological Bulletin, 135(5), 749e773.
106 N. Oesch, R.I.M. Dunbar / Journal of Neurolinguistics 43 (2017) 95e106

Milligan, K., Astington, J. W., & Dack, L. A. (2007). Language and theory of mind: A meta- analysis of the relation between language ability and false-belief
understanding. Child Development, 78(2), 622e646.
Nettle, D., & Liddle, B. (2008). Agreeableness is related to social-cognitive, but not social perceptual, theory of mind. European Journal of Personality, 22,
323e335.
O'Grady, C., Kliesch, C., Smith, K., & Scott-Phillips, T. C. (2015). The ease and extent of recursive mindreading, across implicit and explicit tasks. Evolution and
Human Behavior, 36(4), 313e322.
Onishi, K. H., & Baillargeon, R. (2005). Do 15-month old infants understand false beliefs? Science, 308(5719), 255e258.
Oppenheimer, D. M., Meyvis, T., & Davidenko, N. (2009). Instructional manipulation checks: Detecting satiscing to increase statistical power. Journal of
Experimental Social Psychology, 45, 867e872.
Paal, T., & Bereczkei, T. (2007). Adult theory of mind, cooperation, Machiavellianism: The effect of mindreading on social relations. Personality and Individual
Differences, 43(4), 541e551.
Pearl, J. (2009). Causality: Models, reasoning and inference (2nd ed.). Cambridge: Cambridge University Press.
Pearl, J., Glymour, M., & Jewell, N. P. (2016). Causal inference in statistics: A primer. London: Wiley.
Pinker, S. (1998). Words and rules. Lingua, 106, 219e242.
~ ana, M. (2014). Different association between intentionality competence and
Powell, J. L., Kemp, G. J., Dunbar, R. I. M., Roberts, N., Sluming, V., & Garca-Fin
prefrontal volume in left-and right-handers. Cortex, 54, 63e76.
Powell, J. L., Lewis, P. A., Dunbar, R. I. M., Garca-Fin ~ ana, M., & Roberts, N. (2010). Orbital prefrontal cortex volume correlates with social cognitive
competence. Neuropsychologia, 48(12), 3554e3562.
Preckel, F., & Thiemann, H. (2003). Online versus paper-pencil version of a high potential intelligence test. Swiss Journal of Psychology, 62(2), 131e138.
Premack, D., & Woodruff, G. (1978). Does the chimpanzee have a theory-of-mind? Behavioral and Brain Sciences, 1, 515e526.
Redhead, G., & Dunbar, R. I. M. (2013). The functions of language: An experimental study. Evolutionary Psychology, 11(4), 845e854.
Rubio-Ferna
ndez, P., & Geurts, B. (2013). How to pass the false-belief task before your fourth birthday. Psychological Science, 24(1), 27e33.
Rushworth, M. F. S., Mars, R. B., & Sallet, J. (2013). Are there specialised circuits for social cognition and are they unique to humans? Current Opinion in
Neurobiology, 23, 436e442.
Salgado, J. F., & Moscoso, S. (2003). Internet-based personality testing: Equivalence of measures and assesses' perceptions and reactions. International
Journal of Selection and Assessment, 11(2e3), 194e205.
Samson, D., Apperly, I. A., Chiavarino, C., & Humphreys, G. W. (2004). Left temporoparietal junction is necessary for representing someone else's belief.
Nature Neuroscience, 7(5), 499e500.
Saxe, R., & Kanwisher, N. (2003). People thinking about thinking people: The role of the temporo-parietal junction in theory of mind. NeuroImage, 19(4),
1835e1842.
Smith, M. A., & Leigh, B. (1997). Virtual subjects: Using the Internet as an alternative source of subjects and research environment. Behavior Research
Methods, Instruments and Computers, 29(4), 496e505.
Sperber, D. (2000). Metarepresentations: An interdisciplinary perspective. Oxford: Oxford University Press.
Sperber, D. (2000). Metarepresentations in an evolutionary perspective. In D. Sperber (Ed.), Metarepresentations: An interdisciplinary perspective (pp.
117e137). Oxford: Oxford University Press.
Stiller, J., & Dunbar, R. I. M. (2007). Perspective-taking and social network size in humans. Social Networks, 29, 93e104.
Stylianou, M. S. (2007). Does executive functioning training improve mentalising ability? University of Liverpool. Unpublished PhD thesis.
Surian, L., Caldi, S., & Sperber, D. (2007). Attribution of beliefs by 13-month-old infants. Psychological Science, 18, 580e586.
Sylwester, K., Lyons, M., Buchanan, C., Nettle, D., & Roberts, G. (2012). The role of theory of mind in assessing cooperative intentions. Personality and In-
dividual Differences, 52, 113e117.
Taylor, J. D., & Kinderman, P. (2002). An analogue study of attributional complexity, theory of mind decits and paranoia. British Journal of Psychology, 93,
137e140.
Tomasello, M. (2008). Origins of human communication. Cambridge, MA: MIT Press.
Tomasello, M., Carpenter, M., Call, J., Behne, T., & Moll, H. (2005). Understanding and sharing intentions: The origins of cultural cognition. Behavioral and
Brain Sciences, 28(5), 675e691.
Varley, R., & Siegal, M. (2000). Evidence for cognition without grammar from causal reasoning and theory of mind in an agrammatic aphasic patient.
Current Biology, 10(12), 723e726.
Varley, R., Siegal, M., & Want, S. C. (2001). Severe impairment in grammar does not preclude theory of mind. Neurocase, 7, 489e493.
de Villiers, J. (2007). The interface of language and theory of mind. Lingua, 117(11), 1858e1878.
de Villiers, J., & de Villiers, P. (2000). Linguistic determinism and the understanding of false beliefs. In P. Mitchell, & K. J. Riggs (Eds.), Children's reasoning and
the mind (pp. 191e228). Hove, England: Psychology Press.
de Villiers, J., & Pyers, J. (2002). Complements to cognition: A longitudinal study of the relationship between complex recursive syntax and false-belief-
understanding. Cognitive Development, 17, 1037e1060.
Watson, A., Painter, K., & Bornstein, M. (2002). Longitudinal relations between 2-year- olds' language and 4-year-olds theory of mind. Journal of Cognition
and Development, 2, 449e457.
Wellman, H. M., Cross, D., & Watson, J. (2001). Meta-analysis of theory of mind development: The truth about false belief. Child Development, 72, 655e684.
Willems, R. M., Benn, Y., Hagoort, P., Toni, I., & Varley, R. (2011). Communicating without a functioning language system: Implications for the role of
language in mentalizing. Neuropsychologia, 49, 3130e3135.
Wimmer, H., & Perner, J. (1983). Beliefs about beliefs: Representation and constraining function of wrong beliefs in young children's understanding of
deception. Cognition, 13, 103e128.