Young Minds and Computer Coding

Institute of Education, University of London
September 2012
Computer programmers and the “bilingual advantage”: Enhanced
executive control in non-linguistic interference tasks.
Hannah Wright
A dissertation submitted in partial fulfilment of requirements for the degree of
MSc in Child Development
This dissertation may be made available to the general public for borrowing,
photocopying or consultation without the prior consent of the author.

2COMPUTER PROGRAMMING & THE BILINGUAL ADVANTAGE
Acknowledgements
I am enormously grateful to all of the participants – programmers and non-programmers
– who gave up their time to make this study happen. Thank you.
I am also indebted to those who helped with participant recruitment – Emma
Mulqueeny, Lawrence Job and Josh Pickett at Young Rewired State; Adam Cohen-
Rose; Tara Bloom; Tracey Herald; Nikki Sims; Adrian Wright and Catherine Erskine –
and to the developers and computer scientists who kindly shared experiences and ideas:
Paul Battley, Duncan Gough, Alex Maccaw, Dr John Murnane and Prof. Richard Noss.
Thank you to my supervisor, Prof. Andy Tolmie, for very helpful feedback throughout.
Sincere thanks to my family and friends, and particularly to Paul Sims – without your
support, this would have been much, much harder.

Abstract
Bilingualism is associated with life-long cognitive advantages. It is well established
that bilinguals perform better at non-verbal tasks requiring enhanced executive control.
Bilinguals typically record faster response times than their monolingual peers; this is
thought to result the development of greater efficiencies in conflict monitoring network,
which develop in response to the additional demands of managing two competing
language systems. The present study investigates whether this “bilingual advantage” is
also associated with the frequent use of computer programming languages. The
performance of 10 professional computer programmers (aged 22–25) and 10 adolescent
computer programmers (aged 14–17) is compared to age-matched and IQ-matched
controls in two executive control tasks. In the Attention Networks Test, as predicted,
programmers recorded faster global reaction times than their monolingual peers; the
difference was significant. In the Stroop colour-word task, programmers recorded
slower reaction times; however, these results were not significant. Overall, the results
suggest that extensive computer programming experience may, like bilingualism, be
associated with enhanced executive control. Whatever the direction of this relationship,
it could have important implications for education; these are discussed, along with areas
for future research.

Table of Contents
Abstract
3

Introduction
5

Overview
5

Part
I:
Executive
control
and
bilingualism
8

Section
II:
Computer
programming
and
cognition
15

Section
III:
The
present
study
23

Method
24

Participants
24

Design
26

Materials.
26

Procedure
30

Results
31

WASI
32

Attentional
Networks
Test
33

Stroop
Colour-‐Word
Task
37

Discussion
43

References
48

Introduction
Overview
Plasticity – the ability to change as a result of experience – is a defining and
enduring feature of the brain across the lifespan (Huttenlocher, 2002). Like any other
intensively practiced skill – for example, music (Pascual-Leone, 2001), taxi-driving
(Woollett & Maguire, 2011), or meditation (Brefczynski-Lewis, Lutz, Schaefer,
Levinson, & Davidson, 2007) – the use of language affects the structure and function of
the brain. For example, babies are born with the ability to distinguish between the
sounds of all languages, but by the end of their first year of life, any redundant
connections between neurons (synapses) have been pruned, so that their phonetic
perception is strongly biased towards the sounds that make up the languages to which
they have been exposed (Kuhl, 2004). Some researchers have even suggested that the
use of language is “what makes us smart” (Spelke, 2003). It is certainly difficult to
imagine a skill that is practiced more frequently and acquired more universally.
Bilinguals are defined as people who require and use two (or more) languages in
their every day lives (Grosjean, 1992). The term applies to individuals exposed to two
languages from birth, who undergo what Meisel (1989) described a bilingual first
language acquisition, and also to those who are immersed in additional languages later
in life, who must overcome interference from their more entrenched first language
through some degree of neuronal reorganisation (Hernandez, Li, & MacWhinney,
2005). Globally, bilingualism is the norm, not the exception: Crystal (1997) estimates
that some two thirds of the world’s population growing up in bilingual or multilingual
environments.
In stark contrast to early suspicions that bilingual children were at risk of
retardation or at best, “mentally confused” (Bialystok, 2005), recent research links

bilingualism to cognitive reserve and suggests it may offer protection against dementia
in old age. Cognitive reserve describes a kind of resilience which appears to mediate
the relationship between brain pathology and the clinical expression of that pathology; it
is thought that this resilience derives from more efficient use of brain networks and/or
the ability to deploy differential brain networks to a given task (Stern, 2002). Using a
sample drawn from a memory clinic, Bialystok, Craik and Freedman (2007) found that
bilingual patients presented symptoms of dementia three to four years later than
monolingual patients. A second study, featuring a different sample, found that the first
clinical appointments of bilinguals occurred 4.3 years later and the estimated age of
onset (based on self or relative reports) was 5.1 years later than monolinguals.
Although these results should be viewed with some caution due to social and cultural
differences between participants, the difference is dramatic and has understandably
caused much excitement about the potential impact of bilingualism on the efficiency
and resilience of the brain.
The suggestion that bilingualism could enhance general cognitive performance
is not new. Following the first reports of positive cognitive effects in Peal and
Lambert’s (1962) landmark study, bilingualism has been linked to numerous cognitive
benefits including improved metalinguistic awareness (Ben-Zeev, 1977; Ianco-Worrall,
1972), creativity (Kessler & Quinn, 1987; Ricciardelli, 1992) and problem-solving
(Kessler & Quinn, 1980). In recent years, an influential area of research has explored
the relationship between bilingualism and executive control, that is, the ability to
selectively attend to stimuli and to inhibit inappropriate responses in order to achieve
desired goals. A large number of studies (reviewed below) have found that bilinguals
outperform their monolingual peers in tests involving cognitive conflict. Most of this
research has focused on individuals who were bilingual from birth or early childhood,
but a recent study found similar advantages in late-proficiency Chinese-English

bilinguals who acquired their second language between the ages of 12 and 19 years old
(Tao, Marzecova, Taft, Asanowicz, & Wodniecka, 2011). It is widely hypothesized that
the bilingual advantage in these tasks arises because the bilingual brain places
additional demands on a domain-general aspect of executive control in order to
selectively attend to two competing language systems (e.g. Bialystok, Craik, Green &
Gollan, 2009; Costa, Hernandez, Costa-Faidella, & Sebastian-Galles, 2009; Hinchley &
Klein, 2011).
Like bilingualism, computer programming has often been argued to convey
cognitive advantages; however, the results to date are equivocal (e.g. Pea & Kurland,
1984; Palumbo, 1990; Liao & Bright, 1991). Research into the cognitive consequences
of computer programming has largely focussed on the problem-solving domain
(Ormerod, 1990); in contrast, the present paper explores computer programmers’
cognition from the perspective of language acquisition, an approach recommended by
Murnane (1993; 2006). Like bilinguals, expert computer programmers successfully
manage two or more separate lexicons, grammars and divergent concepts, avoiding
inadvertent transfer between the two. Numerous studies of novice programmers
indicate that they struggle to do achieve this division; transfer from natural language
creates bugs (e.g. Soloway and Spohrer, 1989; Witschital, 1995). The present study
therefore considers whether the “bilingual advantage” in executive control is found in
computer programmers.
Section I explores the evidence linking bilingualism to enhanced executive
control. Empirical support for two reported advantages in conflict tasks is considered:
first, faster global reaction times; second, reduced interference in incongruent
conditions. Two theories relating these advantages are considered: the bilingual
inhibitory control hypothesis and conflict monitoring theory. Finally, a recent study
indicating that the bilingual advantage is also found in late proficiency bilinguals is

presented. Section II argues that the possibility that computer programming languages
could convey similar cognitive benefits to bilingualism merits empirical investigation.
It first considers the similarities and differences between natural languages and
computer programming languages. It then presents evidence from the study of novice
programmers, which indicates that successfully preventing transfer between natural
language and programming languages is crucial to the development of programming
expertise. Section III introduces the present study, which investigates the performance
of adolescent and young adult computer programmers in two executive control tasks
associated with the bilingual advantage.
Part I: Executive control and bilingualism
When communicating, bilinguals must successfully manage two conflicting
languages; one must be accessed whilst the other is suppressed, in order to avoid
involuntary language switching. The cognitive demands of this task are thought to be
the origin of the bilingual advantage in executive control.
A series of studies have demonstrated that bilinguals outperform their peers on
tests of non-linguistic interference. Bilingual children, middle aged adults and older
adults consistently record faster global reaction times in the Simon task (Bialystok,
Martin and Viswanathan, 2005; Martin-Rhee and Bialystok, 2008), the spatial
Stroop/Simon arrows task (Bialystok, 2006; 2008), and flanker arrows tasks such as the
Attention Networks Task (Costa, Hernandez and Sebastian-Galles, 2008; Carlson and
Meltzoff, 2008; Emmorey et al., 2009). These computer-based neuropsychological
tasks all require participants to respond to a series of stimuli as quickly and accurately
as possible; the interference comes from conditions where the stimuli and the response
required are incongruent, typically resulting in slower reaction times. (See Box 1.)

9COMPUTER PROGRAMMING & THE BILINGUAL ADVANTAGEBox 1. Non-linguistic interference tasks
Key features of three non-linguistic interference tasks commonly used in bilingual research – the Simon task,
the spatial Stroop task (sometimes referred to as a Simon arrows task) and the flanker task – are described
here. It should be noted however different studies interpret these tasks slightly differently; this causes
difficulties in directly comparing studies.
Congruent trial: Incongruent trial:
Stimuli:
Response:
A. Simon Task. The participant’s task is to press one of two buttons depending on the colour of the square, as
quickly and accurately as possible. The relationship between the position of the arrow on the screen and the
location of the correct response determines congruency.
Stimuli:
Response:
B. Spatial Stroop. The participant’s task is to press one of two buttons depending on the direction that the
arrow is pointing, as quickly and accurately as possible. The relationship between the direction of the arrow
and the location of the correct response determines congruency.
Stimuli:
Response:
C. Flanker Task. In this task, the correct response is determined by the direction in which the central arrow is
pointing. Congruency is determined by the direction of the flanker arrows.

The bilingual advantage is less pronounced in young adults: studies have
reported that they recorded faster global RTs in spatial Stroop (Bialystok & DePape,
2009) and flanker tasks (Costa, Hernandez, & Sebastian–Galles, 2008; Tao, Marzecova,
Taft, Asanowicz, & Wodniecka, 2011); others have found that participants in this age
group only perform faster than their peers in tasks involving frequent switching between
congruent and incongruent trials, and not in low-switch versions of the same task (e.g.
Bialystok, 2006; Costa et al., 2009). A study using the Simon task with young adults
found no difference between the global RTs of bilingual and monolingual participants
(Bialystok, 2006). Overall, however, bilinguals tend to record faster global RTs in non-
linguistic interference tests; in a recent and impressively detailed empirical review,
Hilchey & Klein (2011) concluded that this finding was robust. The advantage,
however, does not appear to extend to bimodal bilinguals; Emmorey et al. (2008) found
no difference in the performance of speech-sign bilinguals and monolinguals on a
flanker task.
Some studies have also reported that the bilinguals were not only faster overall
but also less affected by the conflict conditions of these non-linguistic interference tests;
the difference in their reaction times in congruent versus incongruent conditions was
significantly smaller than for their monolingual peers. For example, Bialystok, Craik,
and Luk (2008) found that the Simon effect (the increase in reaction times in
incongruent conditions) was significantly smaller for older bilinguals than for age- and
IQ-matched controls. The authors observed similar differences in performance on the
Stroop colour-word naming task; bilinguals were less affected by the Stroop effect than
their monolingual peers (Bialystok, Craik and Luk, 2008). For many years, it was this
apparent advantage (rather than the global RT advantage) that attracted the greater
interest from theorists; this resulted in the bilingual inhibitory control advantage
(BICA) hypothesis. Based on Green’s (1998) inhibitory control theory, the BICA

hypothesis suggests that a general-purpose inhibitory control system holds additional
responsibilities in the bilingual mind, enabling the speaker to ignore distractions from
the irrelevant language in order to effectively communicate. It is proposed that in the
bilingual brain, both languages are simultaneously activated in response to stimuli,
regardless of the relative relevance of these respective languages to the speaker’s
context; the inhibitory control system must react to this conflict by suppressing the
irrelevant information. Support for the hypothesis comes from studies indicating that
parallel language activation is a feature of the bilingual brain. For example, Thierry and
Wu (2007) found measured event related potentials when Chinese-English bilinguals
were asked to judge the semantic relatedness of English word pairs (e.g. train-ham); the
results suggested that participants unconsciously translated the words into Chinese in
cases where the Chinese translations of the two words shared a character (e.g. the
Chinese words for train and ham are Huo Che and Huo Tui). This parallel activation
appears to affect on performance in comprehension and language production tasks (e.g.
Phillipp & Koch, 2009).
There are two key flaws with the BICA hypothesis. First, the evidence for
reduced interference effects in bilinguals is equivocal. Several researchers (e.g.
Bialystok, Martin & Viswanathan, 2005; Bialystok, 2006; Colzato et al., 2008) have
struggled to reliably reproduce this reduced interference effect. Meanwhile, one of the
most commonly cited examples of reduced Simon effect in bilinguals – a study by
Bialystok, Craik, Klein, and Viswanathan (2004) – suffers from methodological flaws;
the monolingual participants were all Canadian residents, whilst the bilingual
participants all lived in India or Hong Kong, raising the strong possibility that social
and cultural differences in the sample influenced results. Hilchey and Klein’s (2011)
review concludes that the bilingual advantage in conflict resolution tasks is “sporadic at
best, and in some cases conspicuously absent”. Further, the BICA hypothesis alone

cannot explain the more reliable finding that bilinguals outperform monolinguals by
similar magnitudes in both conflict and non-conflict conditions. Whilst some
supporters continue to endorse the hypothesis (Philipp & Koch, 2009), others (e.g.
Bialystok, 2006; Costa, 2009; Hilchey & Klein, 2011; Bialystok, Craik & Luk, in press)
are increasingly identifying a domain-general conflict monitoring network as a
promising alternative candidate for the source of the bilingual advantage.
Botvinick and colleagues’ conflict monitoring theory uses evidence from
cognitive neuroscience to argue that a network of specific brain regions, particularly the
dorsal anterior cingulate cortex (ACC), play a key role in detecting conflict in
information processing; upon the detection of such conflicts, the network allocates
additional cognitive control resources to the task in order to protect against subsequent
conflict (Botvinick, Cohen, & Carter, 2004). This network has been implicated in the
monitoring of conflict between two competing languages (e.g. Hernandez, Dapretto,
Mazziotta, & Bookheimer, 2001) as well as more general conflict monitoring tasks.
Hilchey & Klein (2011) explain the plausible theory that it is bilinguals’ frequent use of
this domain-general system to monitor two conflicting languages, rather than a specific
inhibitory control mechanism, that is the source of the bilingual advantage in non-
linguistic interference tests. Costa et al. (2009) provided evidence to support this theory
by manipulating conflict resolution tasks with the aim of placing differential demands
on the conflict monitoring network. In the high conflict-monitoring condition, the task
involved rapid switching between congruent and incongruent trials; conversely, in the
low conflict-monitoring condition, most of the trials were of the same type (congruent
or incongruent). Bilinguals tended to perform faster in the high-monitoring condition
but there was little difference between the groups in the low-monitoring condition. The
authors argued that this was due to impact of bilingualism on the conflict monitoring
system.

A recent neuroimaging study observed differences in the brain activity of
bilinguals and monolinguals during a flanker task (Luk, Anderson, Craik, Grady, &
Bialystok, 2010). The study demonstrated that both monolinguals and bilinguals
showed increased activation in similar brain regions (bilateral middle occipital gyrus,
left fusiform gyrus, left lingual gyrus, bilateral cerebellum, and right caudate and IFG)
during successful performance on congruent tasks; however, whilst monolinguals also
activated those regions during successful incongruent trial responses, bilinguals showed
increased activation in different brain regions (bilateral cerebellum, bilateral superior
temporal gyri, left supramarginal gyri, bilateral postcentral gyri, and bilateral
precuneus). The authors interpret the findings as support for the bilingual inhibitory
control hypothesis, an interpretation that appears to be logical. However, Hilchey and
Klein (2011) suggest that instead, bilinguals may possess a greater ability to detect and
allocate inputs to different brain regions, based on the presence or absence of conflict;
this frees up valuable processing resources, thus increasing processing speed and
reducing global RT. This account sits more comfortably with the evidence from
behavioural studies, since the inhibitory control explanation does not explain the global
RT advantage. Bialystok et al. (in press) suggest that the two theories need not be
mutually exclusive; in fact, they argue an inhibition account is still required in addition
to the conflict monitoring theory in order to explain why bilinguals sometimes show a
reduced interference effect in addition to faster global reaction times.
Whatever the source of the bilingual advantage, the finding that bilinguals
perform better on non-verbal conflict tasks has been widely replicated. However, the
studies described above are limited in two important ways: first, they have generally
focussed on early proficiency bilinguals – those who have been exposed to two
languages from birth or early childhood – and therefore cannot be generalised to those
who learn second languages later in life; second, they exclusively focus on balanced

bilinguals – that is, bilinguals who are equally proficient in both languages – so it is not
possible to discern whether or not this bilingual advantage would apply to those with
one more dominant language.
Fortunately, a study from Tao et al. (2011) makes good progress in addressing
these gaps in the literature. The authors compared the efficiency of the attention
networks of young Australian adults who were English monolinguals or Chinese–
English bilinguals. The bilingual group was further divided according to age of
acquisition: early bilinguals, who had arrived in Australia before the age of 6, and had
received their formal education in English; and late proficiency bilinguals, who had
arrived in Australia between the ages of 12 and 19 years old. The efficiency of three
attention networks – alerting (achieving and maintaining an alert state), orienting
(selecting information from sensory input), and executive control (monitoring and
resolving conflict) – as well as the hemispheric symmetry of these networks, was
measured using Lateralized Attention Network Test (LANT), designed by Green et al.
(2008). (This is an adaptation of the Attention Network Test described in detail in the
methods section and Box B below.)
Both bilingual groups significantly outperformed the monolingual group,
suggesting that the mastery of two very different grammars and lexicons produces the
bilingual advantage found in bilinguals speaking two similar languages. The early
proficiency group demonstrated significantly faster reaction times in all conditions; the
authors suggested was due to the enhancement of monitoring systems (Costa et al.
2009) resulting from early bilingualism, though they also acknowledged that this could
also be simply attributed to greater vigilance. Fascinatingly, however, the late
proficiency bilingual showed the greatest advantage in conflict resolution conditions;
the authors attributed this to greater reliance on the executive network in later second
language acquisition, both in order to stave off greater interference from their “more

solidified” first language and to support the processing of the weaker second language.
Participants in both groups of the study were not considered balanced bilinguals; the
early proficiency group were considered strongly dominant in L2, and the late
proficiency group were considered moderately dominant in L1. (It should be noted that
proficiency were self-rated using a 7-point Likert scale; no proficiency tests were
conducted.)
In summary, this section has described evidence for a bilingual advantage in
non-verbal conflict tasks. Bilingual children, middle aged and older adults perform
these tasks faster than their monolingual peers; bilingual young adults also perform
faster in the more cognitively demanding of these tasks. This advantage is thought to
derive from a conflict monitoring network, which is enhanced in bilinguals due to its
role in preventing transfer between their two languages. Most of the evidence for the
global RT advantage derives from studies of balanced, early-proficiency bilinguals;
however, it has also been found in late-proficiency bilinguals and those with
asymmetric linguistic abilities. Some studies have also found that bilinguals
demonstrate a reduced interference effect during incongruent trials; however these
results have proved difficult to replicate, particularly in children and young adults.
Importantly, it is generally agreed that the reason why bilinguals perform better in tests
of executive control is the additional cognitive demands of managing two separate
language systems.
Section II: Computer programming and cognition
Papert’s (1980) influential work raised hopes for that the Logo programming
language could be used as a tool to aid the development of children’s abstract thinking.
To date, empirical studies have focused primarily on seeking to identify gains in general

problem-solving ability as the result of programming instruction, and the evidence has
been less than conclusive (Palumbo, 1990). Pea & Kurland’s (1984) critical review
concluded that claims of cognitive transfer could not be substantiated. Pea (1983)
likened Papert’s enthusiasm for the potential of Logo in education to “the overzealous
prescriptions for studying Latin in Victorian times”; his implication was that Logo, like
Latin, had been over-hyped and had under delivered. However, several studies have
reported evidence of positive transfer. In a meta-analysis of 65 studies on the effects of
computer programming on cognitive outcomes, Liao and Bright (1991) found that 89
percent reported positive results, but the overall effect size was moderate (0.41); the
power of the effect could not match the power of Papert’s ideas.
Both Papert’s powerful ideas and Pea’s criticism equate programming languages
with natural languages, but to what extent is this comparison justified? In the 1970s,
some American universities accepted computer programming languages in fulfilment of
their foreign language requirement (Norman, 2008). Papert (1980) explicitly compares
the learning of computational languages with the learning of natural languages, “one of
the things children do best”. More recently, Cohen and Haberman (2007) called for
recognition of computer science as a high-level scientific language, “the language of
technology”. Weinberg (1971), however, was more cautious about the comparison:
“Just calling it a language doesn’t make it one.”
If empirical evidence suggested that the brain utilised the natural language
system to store and process computer programming languages, it would be reasonable
to hypothesise than expert programmers have bilingual brains. Unfortunately the
relationship between natural language processing and computer programming languages
has merited little empirical attention (Murnane, 1993; 2006). Studies of programming
tend to view it as a problem solving – rather than a linguistic – activity, so there has
been little exploration of the relationship between the vocabulary, grammar and syntax

of programming languages and programmers’ semantic and conceptual knowledge
(Ormerod, 1990).
A computer program is a text combining a data structure and a set of instructions
that enable the computer to calculate desired functions, written according to strict
grammatical rules that can be interpreted by the computer (Détienne, 2002). A
programming language is a semiotic code comprising vocabulary and grammar used to
convey instructions to the computer but also to other programmers, many of whom may
contribute to a single piece of software. There are, of course, many differences between
natural languages and programming languages. One difference that is often cited as the
most important (e.g. Detienne 2002) is that while in natural languages statements may
be open to interpretation, in programming languages they must be clear and
unambiguous; computer programmers are strongly constrained by the syntactic rules of
the programming languages.
Hockett (1960) isolated 13 features that characterize human language and which
distinguish it from other communication systems. Weinberg (1971) noted several
discrepancies between these and the characteristics of programming languages. Unlike
spoken languages, programming languages contain no vocal-auditory channel; and
because they are written rather than spoken, there is no rapid fading – instead, they must
be explicitly erased. Spoken languages are characterized by broadcast transmission and
directional reception – that is, the signal is sent out in all directions (to anyone within
earshot) and the listener can identify the direction the sound is coming from and thus
identify the sender; in contrast, programming languages feature directional transmission
in the form of computational input, but produce broadcast reception. The feature of
interchangeability exists in human languages – a typically developing person could
both receive and broadcast the same message, whereas computers and their
programmers cannot switch roles. However, as mentioned, programming languages

also convey information between human beings. Tremblay and Sorenson (1985, p.74)
consider this to be “the most important goal of a programming language”; a survey of
over 780 programmers found that 95% agreed that understanding existing code is a
significant part of their job (Cherubini, Venolia, DeLine, & Ko, 2007). Like natural
languages, computer programming languages allow people to collaborate and share
ideas.
Although there are differences between programming languages and natural
languages, there are also many similarities. Weinberg (1971) concedes that computer
programming languages appear to “live up to or exceed natural language[s]’ in the
remaining nine areas identified by Hockett (1960). For example, like natural languages,
programming languages are arbitrary – there is no direct connection between the word
and its meaning (an aspect that proves confusing for novices, as discussed below); both
programming and natural languages also have duality of patterning (they can be broken
down into small parts which can be recombined to form new meaning).
Norman (2008) argues that programming languages technically meet “the three
most important criteria” for languages. First, a language must be meaningful; whilst
natural languages use words and sentences to convey meaning, programming languages
achieve this via objects, functions and relations. Second, a language should allow
displacement, that is, communication about things that are not immediately present;
natural language achieves this with concepts such as yesterday or somewhere else, and
programming languages systematically convey instructions for the future and reference
data from the past. Third, a language must be productive, allowing the expression of
ideas that have never been expressed before; computer programming languages are
particularly strong in this respect.
Few studies have systematically compared natural and programming languages.
Kokol, Podgorelec, Zorman, Kokol, and Njivar (1999) provided a rare exception. The

authors compared the complexity of natural language texts to computer programs using
long-range power law correlations (LRCs). The authors found that the mean α values –
which are thought to measure the complexity or information content of the text – were
significantly higher for programming languages. This difference was attributed to three
key differences: first, the formality of programming languages, which forces order and
reduces randomness; second, the ambiguity of natural languages, which creates
randomness and disorder; third, the significantly larger vocabularies that feature in
natural language contribute more to randomness than complexity. In short, computer
programming languages were found to be more efficient in communicating complexity.
However, the presence of LRC in both natural and programming languages indicates
that they are derived from common (as yet unidentified) laws.
We have established that there are shared features between natural languages
and programming languages; however, it would be ridiculous to suggest that the
acquisition of programming languages is similar to an infant’s acquisition of language.
Instead, we will consider similarities between the later acquisition of second languages
and the difficult task of acquiring the fluent use of computer programming languages.
One striking similarity is the problem of language transfer. This term is traditionally
used to describe the influence of the native language (L1) on the acquisition of a
subsequent language (L2); Pavlenko and Jarvis (2002) have argued that this effect is in
fact bidirectional. Cross-linguistic interference is the negative side of language transfer,
and can occur at the lexical, semantic and also at conceptual level (Pavlenko, 2009). At
the lexical level, errors may occur when speakers erroneously assume that similar-
sounding words share meaning (false cognates), when words from the wrong language
are unconsciously inserted into sentences (unintentional language switches), or when
speakers inadvertently blend words from two languages to create new words – for
example, a Swedish-English bilingual may create the word clothers from the English

clothes and the Swedish klä der (Jarvis, 2009). At the conceptual level, Boroditsky
(2001) found that Mandarin-English bilinguals showed a bias towards thinking about
time in terms of vertical metaphors (which is how time is conceptualised in the Chinese
language) even when completing a task in English (which conceptualises time in terms
of horizontal metaphors); the strength of this bias was positively correlated with age of
English acquisition and the extent to which participants reported thinking in English.
Studies of the errors made by novice programmers indicate that knowledge of
natural language influences the way programming languages are understood by novice
and intermediate programmers (Witschital, 1995). Soloway and Spohrer (1989)
videotaped interviews with novice programmers solving problems and documented
examples of bugs caused by natural language transfer. For example, in the English
language lexicon, instructions such as while imply continuous checking of the condition
(“while x is happening, do y”); the novices therefore erroneously expected similar
results from the while loop in Pascal (in which the condition is only checked at discrete
points of time). This supports the findings of an earlier study (Soloway et al., 1981) in
which 34% of the students in an introductory programming courses reported this “while
demon” misconception. Another example from Soloway and Spohrer (1989) concerned
if-then-else commands. In natural language, this implies a looping construct – for
example, “If you find a door unlocked then lock it”; the novice therefore confuses
lexical similarity between the two languages with functional similarity. Negative
transfer has also been identified in the naming of variables; Sleeman, Putnam, Baxter,
and Kuspa (1986) found that students frequently erroneously believed that assigning
meaningful names (in natural language) to variables in computer programming would
have somehow aid the computer’s understanding of the program. These errors are very
similar to the false cognates errors experienced by bilinguals and second language
learners.

Language transfer can have positive as well as negative effects, as knowledge
about the way language works can aid the acquisition of additional languages.
Metalinguistic awareness is knowledge of the structural components of language, for
example, understanding of the arbitrary connection between symbols and their referents
(Yelland, Pollard, & Mercuri, 1993). Childhood bilingualism has been linked to the
advanced development of metalinguistic awareness. For example, Ianco-Worrall (1972)
asked Afrikaans/English bilingual children whether it was possible to swap the names
for cow and dog; twice as many bilingual children understood that this was possible,
compared to their monolingual peers. Similarly, Ben-Zeev (1977) found that Hebrew-
English bilinguals performed better at symbol substitution tasks compared
monolinguals, suggesting that experience of more than one language system left
children “freer to abandon the rules of a particular language system for a different set of
rules where necessary”.
The relationship between computer programming and metalinguistic ability has
attracted less empirical attention. However, computer programming languages not only
provide a new lexicon, as natural language does, but also necessitate the programmer’s
active involvement in the construction of that lexicon via the naming of variables
(Détienne, 2002); therefore it is reasonable to predict that there is a relationship between
programming and metalinguistic ability. Some support for this assumption can be
found in studies of the effects of programming instruction on children with language or
literacy deficits. Lehrer and DeBernard (1987) found that language-impaired children
who received Logo programming instruction performed better in tests of perceptual-
language skills, compared to children who used commercially available learning
software or teacher-led instruction. Peppler and Warschauer (2011) observed how
“Brandy”, a nine-year-old girl with learning disabilities, learnt to program using Scratch
– an accessible programming language for children, in which programs are built from

Lego-inspired graphical programming blocks (Resnick et al., 2009) – before she could
read or write. The authors, who followed Brandy over a two-and-a-half year period,
linked Brandy’s developing programming literacies to marked improvement in her
traditional print literacy: “Working with Scratch seemed to illuminate for Brandy the
mechanics of language and stimulate her metalinguistic awareness of how language
operates, as she made connections between the Scratch programming language and her
spoken (and increasingly written) English language” (p.32).
In summary, descriptions of the learning experiences of novice programmers
indicate that they experience language transfer effects. The errors they make suggest
that in order to successfully master computer programming languages, novices must
develop the ability to successfully inhibit transfer from their spoken languages. In the
bilingual literature, it is generally agreed that the mechanism responsible for the
successful management of two or more languages – and thus the avoidance of negative
transfer – is some domain-general aspect of executive control. It is therefore logical to
hypothesise that this same mechanism is employed by computer programmers to
facilitate the successful management of two very different language systems. The exact
specification of this mechanism is currently unclear – a promising theory suggests a
conflict monitoring system, in which the dorsal ACC plays a key role – but its existence
is indicated by the presence of a bilingual advantage in executive control tasks. The
present study therefore investigates the performance of computer programmers on these
tasks in order to ascertain whether computer programmers benefit from the “bilingual
advantage”.

Section III: The present study
Literature searches indicate that to date, there has been little systematic
investigation of executive control abilities in computer programmers. Studies that have
considered the impact of computer programming on specific aspects of executive
function have focused on the role of working memory (e.g. Bergersen & Gustafsson,
2011). No studies comparing the cognitive effects of computer programming to the
cognitive effects of bilingualism have been found. The present study aimed to
contribute towards filling this gap in the literature.
The performance of computer programmers on two executive control tasks that
have been found to demonstrate the bilingual advantage was compared to that of
monolingual, age-matched controls. The aim was to demonstrate whether the computer
programmers showed a similar advantage. The Attention Networks Task (Fan,
McCandliss, Sommer, Raz, & Posner, 2002) is a computer-based task designed to
evaluate three attention networks: alerting (achieving and maintaining an alert state),
orienting (selecting information from sensory input), and executive attention
(monitoring and resolving conflict). Costa et al. (2008) compared the ANT
performance of Spanish-Catalan bilinguals (mean age 22 years) to monolingual
controls; the former recorded faster global reaction times (RTs) and exhibited less
interference in conflict conditions. Importantly, using a lateralised version of the ANT
(which was designed to provide additional information about the interhemispheric
organisation of the attention networks), Tao et al. (2011) found that early but less-
balanced bilinguals had significantly higher global RTs, but that late proficiency but
more balanced bilinguals showed an advantage in conflict resolution specifically. A
computer-based version of the Stroop colour-word test was also included to provide a
second measure of executive control; Bialystok et al. (2008) found that bilinguals

displayed smaller Stroop effects than monolinguals (though there was no difference in
reaction times).
The study involved two groups of programmers, all of whom (like Tao et al.’s
late proficiency bilinguals) acquired programming languages during adolescence. In
order to provide an indication of the length of programming experience required to
obtain any observed executive control advantages, two different age groups were used:
young adults, employed professionally as programmers, with at least five years’
experience of programming; and adolescents who programmed regularly but had less
programming experience. It was initially hoped to include a third group, children aged
8-11 who had 6 months’ programming experience, in order to provide further insight
into any cut-off point; however, last minute recruitment difficulties precluded this.
The hypotheses for the study were as follows: 1. Programmers would record
faster global RTs than controls on both executive control tasks. 2. Programmers would
be less affected by conflict conditions than controls in the executive control tasks. 3.
The difference between programmers and controls would be greater in the older age
group, due to greater experience of computer programming.
Method
Participants
A total of 40 monolingual, English-speaking adolescents and young adults
participated in the study. Participants were grouped by age (adolescent or young adult)
and programming experience (programmer or non-programmer). The mean age of the
young adults’ groups (programmers: N = 10, M = 24.13, SD = 0.82; non-programmers:
N = 10, M = 23.74, SD = .41) were not statistically different from each other; nor were

the adolescents’ groups (programmers: M = 16.05, SD = 1.17; non-programmers: M =
16.39, SD = 1.62).
The young adult programmers were recruited via contacts at digital agencies in
London. All were currently employed as professional computer programmers; they had
had an average of M = 7.8 years’ programming experience, SD = 2.23; 4 reported
coding “every day”, 6 coded “most days”. The group reported regularly using an
average of M = 3.34 programming languages; the most commonly used languages were
JavaScript (10), Ruby (8) and PHP (6). The group comprised 10 males and 1 female.
The young adult non-programmers were recruited via word of mouth. There were 10
males and one female; all were currently employed in office-based professions
involving daily computer use, but they had no experience of computer programming.
In the adolescent programmers’ group, 6 participants were recruited via an
advert in the newsletter of Young Rewired State, an organization that runs annual
events for young programmers. It had been hoped that all young programmers would
be recruited in this way; however the timing (summer term), the age group (GCSEs,
AS-levels) and the monolingual requirement (4 additional adolescents expressed interest
but were ineligible due to being bilingual) hindered recruitment. Therefore an
additional 4 adolescents were recruited via an IT teacher at a school in Kent, where
computing GCSE is taught. The group comprised 6 males and 4 females with an
average of M = 2.7 years’ programming experience, SD = 1.70; 4 reported coding “most
days”, 6 coded “weekly”. The group reported regularly using an average of M = 2.7
programming languages; the most commonly used languages were Javascript (6), Java
(4), C# (4) and Scratch (4). The control group comprised 10 adolescents, 6 males, 4
females, recruited via family and friends.

Ethics. All participants received information sheets, tailored to their age and
programming group, containing details of the study. This explained that people who
speak two or more languages perform better at some tasks that involve quickly deciding
what to pay attention to and what to ignore; the aim of the study was to find out whether
people who regularly use computer programming languages are also better at these
tasks. All participants gave written consent to participation prior to taking part. In
addition, written parent or guardian consent was obtained for all under 18s. The
research received ethical approval under the Institute of Education ethics procedures.
Participants were not paid for taking part.
Design
The research employed a quasi-experimental factorial design, with between-
subject factors of programming experience and age. Participants were first asked to
complete the Wechsler Abbreviated Scale of Intelligence (Wechsler, 1999; 2-subtest
version) in order to establish that the age-matched groups did not differ in IQ. In order
to test the experimental hypothesis, executive control was then tested by means of the
Attentional Networks Task (ANT, Fan et. al, 2002) and the Stroop colour-word task
(Stroop, 1935).
Materials.
Wechsler Abbreviated Scale of Intelligence (WASI). In order to compare IQ
across the groups, participants completed the 2-subtest version of the Wechsler
Abbreviated Scale of Intelligence (Wechsler, 1999). First, participants completed the
Vocabulary subtest, which assesses the vocabulary and lexical knowledge of the
participant. Second, participants completed the Matrix reasoning subtest, which

assesses the fluid intelligence of the participant. Raw scores were converted into t-
scores according to the participant’s age, and combined in order convert total t-scores
into FSIQ-2 score.
Attentional Networks Test (ANT). The first of two tasks selected to assess
executive control was the Attention Network Test, which was adapted from Fan et al.
(2002). As mentioned earlier, the ANT is a computer-based task designed to evaluate
three attention networks: alerting (achieving and maintaining an alert state), orienting
(selecting information from sensory input), and executive attention (monitoring and
resolving conflict).
After reading on-screen instructions, participants were presented with a series of
arrows on a screen, pointing either left or right. Their task was to identify the direction
in which the target arrow was pointing by pressing the corresponding arrow key on the
keyboard, as quickly and accurately as possible. In order to assess executive control,
target arrows were presented either alone (“neutral flanker type”) or flanked by arrows
pointing in the same direction (“congruent flanker type”) or in the opposite direction
(“incongruent flanker type”) (see Box 2, Step 3). Arrows were presented either above
or below a fixation point in the centre of the screen. Alerting and orientating networks
were assessed through the use of cues prior to the presentation of the stimulus: in order
to manipulate alertness, participants received either a cue comprising two stars (“double
cue”) or no cue; in order to assess orienting effects, participants received either a cue
indicating where on the screen the arrows would appear (“spatial cue”) or a cue in the
centre of the screen which gave no clues about where the arrows would appear (“central
cue”) (see Box 2, Step 2).

In total, there were 12 different types of trial: flanker type (congruent,
incongruent or neutral) x cue type (no cue, centre cue, double cue, spatial cue). Each
type of trial was repeated 4 times, creating a total of 48 trials. Trials were presented in a
random order. Each trial ran as follows: 1. A fixation point (+) appeared in the centre
Box 2: The adaption of the Attention Networks Task used in the present study
Step 1: fixation point. At the beginning of each trial, participants viewed a fixation point for 400 ms.
Step 2: cue. In no cue trials, no cue was presented. In centre cue and double cue trials, either one
or two cues (*) were presented; these did not indicate the location in which the arrows would appear.
In spatial cue trials, the location in which the cue appeared indicated where the target arrow would
appear.
Step 3: arrows. Target stimulus was then presented either above or below the fixation point for
1700ms or until a response was given. In neutral trials, the target arrow appeared alone. In
congruent trials, the target arrow was flanked by arrows facing in the same direction; in incongruent
trials, flanker arrows faced in the opposite direction.

of the screen for 400ms; 2. A cue (or no cue) was presented for 150ms; 3. The target
stimuli was presented (with or without flankers), disappearing either when the
participant responded or after 1700ms. (See Box 2, Step 3).
The ANT was programmed using Paradigm experimental software; the software
recorded response time and accuracy for each trial. All stimuli were presented in black
font, on a light blue background. The fixation point and cues were presented in
Microsoft Sans Serif font at 27.75pt. Target and flanker arrows were created using the
font Wingdings 3 at 36pt; these were presented either above, or below the fixation
point. Prior to the actual task, a practice run of 12 trials was conducted in order to
ensure that participants understood the task. There were no difficulties in task
comprehension
Stroop Colour-Word Task. The second task employed to assess executive
control was a computer-based version of the Stroop colour-word task (Stroop, 1935).
The task assesses participants’ ability to selectively attend to stimuli, filtering out
misleading information. On-screen instructions explained to participants that they
would see a series of words written in different colours. Their task was to identify the
colour in which the word was written by pressing one of two keys on the keyboard:
RED and BLUE mapped onto the left arrow key; GREEN and YELLOW mapped on
the right arrow key.
Each word belonged to one of three trial types: congruent, incongruent or
neutral. In congruent and incongruent trials, the words were RED, YELLOW, GREEN,
BLUE. In congruent trials, the colour of the words matched their semantic meaning:
RED, YELLOW, GREEN, BLUE. In incongruent trials, the meaning of the colour
word conflicted with the target colour, resulting in semantic conflict e.g. RED,
YELLOW, GREEN, BLUE. All semantically incongruent condition also required an
incongruent response; no incongruent colour-words were presented that mapped onto

the same response key as the target colour. The neutral condition comprised four non-
colour words: DOG, JUMP, KNIFE and FLOWER. These words are not semantically
associated with any of the target colours, but their character lengths correspond to the
character lengths of the four colour words. Each neutral word was presented three times,
in a different colour each time (red, blue, yellow or green, selected at random).
The Stroop colour-word task was programmed using Paradigm experimental
software; the software recorded response time and accuracy for each trial. Stimuli
comprised a series of words presented in one of four target colours – red, blue, green
and yellow – using Courier New font in bold at 36 point. These words were presented
in the centre of the screen, on a Gainsboro (light grey) background.
A total of 12 trials were conducted in each condition, making a total of 36 trials.
Trials were presented in a random order. Each word disappeared after 2000ms or as
soon as a response was given. Prior to the actual task, a practice run of 8 trials was
conducted in order to ensure that participants understood the task. There were no
difficulties in task comprehension.
Procedure
Due to the disparate geographic location of participants in the experimental
groups (Greater London, Birmingham, St Albans, Milton Keynes, Manchester,
Cheltenham, Kent, West Sussex), the researcher visited participants at their homes
(adolescents), work places (young adults) or alternative suitable locations as required,
and the testing took place in a quiet, well-lit room. For practical reasons, there were
differences in the times and days that the different age groups were seen: adolescents
were seen at weekends or holidays, whereas young adults were seen towards the end of
a working day. The order of testing was: WASI Vocabulary, WASI Matrix Reasoning,
ANT, Stroop colour-word task. The executive function trials were all presented on a

Toshiba Equium L40 156 laptop computer with a 15 inch screen, with a resolution of
1280 x 800 pixels. At the end of the experiment, participants were debriefed and
offered the opportunity to ask questions.
Results
Results from the experiment were analysed as follows. WASI: In order to
identify any significant IQ differences that might influence results, a two-way ANOVA
was first conducted on FSIQ-scores, with age and programming experience as the
independent variables. ANT: Group performances on the ANT were then compared.
Response times analysed via a three-way mixed ANOVA with a within-participant
factor of congruency (3 levels: congruent, incongruent, neutral) and two between-
participant factors: age group (2 levels: adolescent or young adult) and programming
experience (programmer or non-programmer). Conflict effect was calculated
(incongruent trials RT – congruent trials RT) and a two-way ANOVA performed with
age and programming experience as the independent variables. Kruskal-Wallace tests
were used to compare overall % errors and conflict effect ERR (errors on incongruent
trials – errors on congruent trials). Stroop colour-word task: Group differenced in
performance on the Stroop colour-word task were then analysed. A Kruskal-Wallis test
was used to compare response times of the four groups in each of the three trial types
(congruent, incongruent and neutral). The Stroop effect was calculated (incongruent RT
– congruent RT); a Kruskal-Wallis test was used to compare group differences in the
size of the Stroop effect. A MANOVA was used to compare differences in facilitation
effect (congruent RT – neutral RT) and cost (incongruent RT – neutral RT). Finally,
Kruskal-Wallis tests were used to compare group differences in global errors and in
Stroop error effects (incongruent ERR% - congruent ERR%).

WASI
Mean FSIQ-2 scores for each of the four groups are presented in Table 1.
Exploratory data analysis indicated that the data met the assumptions required for
parametric testing; there was one extreme (low) outlier in the adolescent programmers
group but this had a minimal effect on the mean for the group so it was not removed
from the sample.
Table 1.
Mean FSIQ-2 Scores And Standard Deviations For Adolescent and Young Adult
Programmers And Non-Programmers.
Age group Programming experience Mean FSIQ-2 SD
Adolescents Programmer 112.30 9.44
Control 113.40 6.93
Young adults Programmer 119.50 7.68
Control 118.30 5.23
A two-way ANOVA was conducted with FSIQ-2 score as the dependent
variable; the factors were age and programming experience. This revealed a significant
difference in age, F(1, 36) = 6.28, p < .05, partial η2
= .15; the young adults had higher
FSIQ-2s scores than the adolescents. There was no significant difference in

programming experience, F < 1, and no significant interaction between age and
programming experience, F(1, 36) = .23, p > .05.
Attentional Networks Test
Executive control network.
RT analysis. Response time (RT) analysis focused primarily on participants’
overall scores in each flanker condition – congruent, incongruent and neutral – as this
monitors the executive control network, in which bilingual advantage has been found.
Response times (RTs) were analysed for each participant’s responses to each of
the 48 trials. RTs for error trials and no response trials were excluded (3.70% of all
responses). Median RTs were then calculated across each flanker condition (congruent,
incongruent, neutral).
First, exploratory data analysis was performed. The Kolmogorov-Smirnov test
indicated that the data for the adolescent non-programmers group in the congruent
condition, D(10) = .276, p <. 05, were not normally distributed; however, the more
sensitive Shapiro-Wilk test was not significant: D(10) = .883, p > .05. Levene’s Test of
Homogeneity of Variance was significant in the incongruent condition, F(3, 36) = 3.39,
p < .05. Examination of the boxplots revealed extreme scores for case 25 in both
congruent and incongruent conditions; these scores had a substantial effect on the mean
RT of the adolescent non-programmers’ group and were therefore removed from the
analysis. Other outliers (cases 14, 19 and 29 in the congruent condition; 14, 19 in the
incongruent condition; 3, 5 in the neutral condition) did not substantially affect the
mean and were therefore retained. Table 2 presents median RTs and SDs by group and
flanker condition (congruent, incongruent, neutral) following the removal of case 25.

A three-way mixed ANOVA was conducted with a within-participant factor of
congruency (3 levels: congruent, incongruent, neutral) and two between-participant
factors: age group (2 levels: adolescent or young adult) and programming experience
(programmer or non-programmer). This showed a main effect of congruency, F (2, 70)
= 93.56, p < .001, partial η2
= .73. This was due to all groups responding more slowly
in the incongruent condition than in the neutral and congruent conditions, as illustrated
in Fig. 1. The main effect of programming experience was also significant, F (1, 35) =
8.01, p < .01, partial η2
= .19. This was as a result of programmers responding more
quickly than non-programmers in all conditions. The effect of age was not significant,
F < 1; as Fig. 1 illustrates, young adult programmers had the global RTs, and young
adult non-programmers had the slowest global RTs. None of the interactions were
significant (congruency and age; congruency and programming; and congruency, age
and programming), F < 1.
Table 2.
Mean Response Times and Standard Deviations for Adolescent and Young Adult
Programmers and Controls in the Attention Networks Task
Adolescents Young Adults
Trial type
Programmers Controls Programmers Controls
RT (ms) SD RT (ms) SD RT (ms) SD RT (ms) SD
Congruent 483.9 65.0 520.1 50.8 474.1 43.6 551.4 87.1
Incongruent 563.5 49.5 611.4 109.8 524.4 33.7 613.1 51.4
Neutral 456.0 466.5 503.2 62.7 454.3 32.1 528.8 57.1

Figure 1. Line graph showing how age and programming experience affected response
times in congruent, incongruent and neutral trials in the Attention Networks Task. Both
groups of programmers were faster than non-programmers.
Conflict effect RT was calculated by subtracting RT in the congruent condition
from RT in the incongruent condition. A two-way ANOVA was performed with
conflict effect RT as the dependent variable; age and programming experience were the
independent variables. The results were not significant, F < 1; there was no significant
difference in the degree to which the groups were disadvantaged in the incongruent
conditions.
Error analysis. Table 3 shows the mean % error rates for all four groups in
congruent, incongruent and neutral conditions, as well as the total error rates. Three
Adolescent Programmers Adolescent Controls
Young Adult Programmers Young Adult Controls
Congruent
Incongruent
Neutral
0 100 200 300 400 500 600 700
TrialType
Mean RT (ms)

participants were excluded from the analysis due to extreme high scores (cases 7, 19,
and 25) that had a substantial effect on group means.
Table 3.
Exploratory data analysis indicated that the distribution of the total error rates
deviated significantly from normality in the programming, F(18) = .28, p < .01, and
non-programming, F(19) = .25, p < .05, groups. Therefore, a Kruskal-Wallace test was
used to compare all four groups: adolescent programmers, adolescent non-programmers,
young adult programmers and young adult non-programmers; the difference between
the groups was not significant, H(3) = 1.06, p > .05.
Conflict effect ERR was calculated by subtracting ERR% in the congruent
condition from ERR% in the incongruent condition. The Kolmogorov-Smirnov test
indicated that although the data were normally distributed in the professional
programmers’ group, D(10) = .18, p > .05, they were not normally distributed in the
ANT Task: Mean Percentage Error Rates By Group and Flanker Type.
Trial type
ERR% SD ERR% SD ERR% SD ERR% SD
Congruent 0 0 0 0 0.7 2.1 0 0
Incongruent 8.3 8.3 7.6 7.1 9.7 9.2 5.0 4.9
Neutral 0 0 0.6 2.0 1.3 4.0 3.1 1.4
Overall 2.8 2.8 2.8 2.6 3.9 3.2 2.7 2.0

remaining three groups: adolescent programmers, D(10) = .36, p < . 05, adolescent non-
programmers D(10) = .35, p < . 05 and professional non-programmers, D(10) = .25, p <
. 05. A non-parametric Kruskal-Wallis test was used to compare the means; the results
were not significant, H (3) = 2.01, p = .56.
Stroop Colour-Word Task
Response times (RTs) were analysed for 12 trials of each condition (congruent,
incongruent and neutral). RTs for error trials and no response trials were excluded
(3.06% of all responses). Median RTs were then calculated for each participant in each
condition. Initial examination of the revealed the presence of a small Stroop effect for
the adolescent controls and young adult programmers, and a small negative Stroop
effect for adolescent programmers and young adult controls; a surprising result. It was
hypothesized that, due to the random presentation of the stimulus, there was an uneven
distribution of trials in which the target colour was the same as the proceeding trial.
Further examination of the data revealed that this was indeed the case. Participants
varied in the number of these types of trial they experienced, and in whether these trials
occurred on congruent, incongruent or neutral trials; this negated the Stroop effect.
Therefore, RTs for these trials were excluded (23.06% of all responses) and median RTs
recalculated for each participant. Mean RTs and SDs are presented in Table 4.
RT analysis. As Table 4 illustrates, young adult programmers displayed the
slowest RTs across all three conditions. Exploratory data analysis indicated that the data
violated the assumption of homogeneity in congruent, F(3, 35) = 10.09, p < .001,
incongruent, F(3, 35) = 3.52, p < .05, and neutral trials, F(3, 35) = 5.55, p < .05. A
Kruskal-Wallis test was used to compare the four groups’ RTs in each of the three
conditions. This revealed a significant difference in reaction times between groups in

the congruent condition, H(39) = 12.45, p < .05; the young adult programmers (M =
781.3, SD = 344.7) were much slower than the young adult controls (M = 591.1, SD =
114.3), while the adolescent programmers (M = 546.4, SD = 64.2) were slower than the
adolescent controls (M = 471.7, SD = 55.5), who were the fastest group overall. The
professional programmers were also the slowest group in the incongruent and neutral
conditions, but this difference was not significant, p > .05.
The Stroop effect is calculated by subtracting the participant’s median score in
the congruent trials from their median score in the incongruent trials. Due to the
presence of interference in the incongruent trials, we would expect participants to record
slower response times, thus we would predict a positive Stroop effect score for all
groups. As Fig X shows, adolescent programmers (M = 23.42, SD = 82.14), adolescent
non-programmers (M = 73.45, SD = 51.48) and professional non-programmers (M =
Table 4.
Response Times For Adolescent and Young Adult Programmers and Non-Programmers in the
Stroop Colour-Word Task.
Trial type
RT (ms) SD RT (ms) SD RT (ms) SD RT (ms) SD
Congruent
546.4 64.2 471.7 55.5 781.3 344.7 591.1 114.3
Incongruent
569.9 108.9 545.1 80.7 739.0 265.0 642.9 179.2
Neutral
577.9 108.9 529.5 58.5 715.2 243.8 656.6 177.9

51.82, SD = 89.51) all demonstrated a positive Stroop effect; this reflects higher RTs in
incongruent trials than in congruent trials. However, professional programmers
demonstrated a negative Stroop effect (M = -11.57, SD = 76.27); contrary to
predictions, they responded more quickly in incongruent than congruent trials.
Figure 2. Bar chart showing the Stroop effect by group. Stroop effect was calculated by
subtracting mean RT in congruent trials from mean RT in incongruent trials.
Exploratory data analysis on Stroop effect RT by group indicated that data in the
adolescent programmers’ group, D(10) = .266, p <. 05, and the professional
programmers’ group, D(10) = .317, p < .05, were not normally distributed. Therefore,
the groups were compared using a Kruskal-Wallis test; the difference between the
groups was not significant, H(39) = 7.244, p > .05.
-60
-40
-20
0
20
40
60
80
Adolescent programmer Adolescent control
Young adult programmer Young adult control
StroopEffectRT(ms)
Group

Another way of analysing the data is in terms of facilitation (the use of the
additional helpful information in the congruent trials) and cost (the extra effort required
to suppress the misleading information in the incongruent trials) relative to scores in the
neutral trials. Mean RTs for Stroop facilitation were calculated by subtracting scores in
the congruent trials from scores in the neutral trials. Since we would expect the
congruent RTs to be faster than neutral RTs, due to the presence of helpful information,
we would expect Stroop facilitation scores to be positive. Mean RTs for Stroop cost
were calculated by subtracting incongruent trial RTs from neutral trial RTs. Since we
would expect the incongruent RTs to be slower than neutral RTs, we would expect this
score to be negative. 1 extreme outlier in the non-programmers’ group (case 38), which
was significantly affecting the mean and standard deviation in that group (with case 38,
M = 13.70, SD = 161.20, without case 38, M = -29.16, SD = 86.58), was removed from
the analysis. The resulting RTs and standard errors for facilitation and cost by group
are presented in Fig 3. As we can see, there were two surprising trends in the data.
First, the adolescent programmers group recorded a positive Stroop cost score (M = 8.0,
SD = 80.03), which means that they responded more quickly in incongruent trials than
in neutral trials; the adolescent controls (M -15.65, SD = 69.65), young adult
programmers (M = -23.87, SD = 85.81) and young adult controls (M = -29.20, SD =
84.57) all recorded negative scores, as would be expected. Second, and even more
surprisingly, the young adult programmers recorded a negative Stroop facilitation score
(M = -66.10, SD = 131.57), which means that they were negatively affected by
congruent trials in comparison to neutral trials. The young adult controls (M = 35.08,
SD = 83.33), adolescent programmers (M = 31.48, SD = 110.60), and adolescent
controls (M = 57.80, SD = 70.74) all recorded positive facilitation scores, as expected.

Fig 3. Bar chart showing Stroop facilitation effect (Congruent RT – Neutral RT) and
Stroop cost (Neutral RT – Incongruent RT) by group.
Kolmogorov-Smirnov and Shapiro-Wilk tests confirmed that data was normally
distributed across all groups, Levene’s Test of Homogeneity of Variance was not
significant, and there were no outliers; the data met the criteria for parametric testing. A
MANOVA was conducted with Stroop facilitation effect RT and cost effect RT as the
dependent variables and group as the independent variable (4 levels: adolescent
programmer, adolescent control, young adult programmer, young adult control). Using
the Roy’s largest root statistic, this effect was significant, F(3, 34) = 3.71, p < .05, Θ =
.33; however Pillai’s trace, Wilks’ Lambda and Hotelling’s trace returned non-
significant results, p > .05. Univariate analyses indicated that the effect of group on
Stroop facilitation effect was approaching significance, F (3, 34) = 2.84, p = .053,
Adolescent programmer Adolescent control
Young adult programmer Young adult control
-70
-60
-50
-40
-30
-20
-10
0
10
20
30
40
50
60
Stroop facilitation effect Stroop cost
MeanRT(ms)

partial η2
= .200, reflecting atypical direction of the effect in young adult programmers;
the effect of group on Stroop cost was not significant, F < 1.
Error analysis Percentage error rates in congruent, incongruent and neutral
trials, overall percentage errors, and Stroop error effect (ERR% congruent trials –
ERR% incongruent trials) were calculated and are presented in Table 5. Participants 8
and 12 were excluded from the analysis due to extremely high error rates. As Table 5
shows, young adult controls (M = 2.0, SD = 2.0) followed by adolescent programmers
(M = 2.5, SD = 1.7) made the fewest errors; adolescent controls (M = 5.8, SD = 4.0)
made the most errors, followed by young adult programmers (M = 5.2, SD = 4.0).
Table 5.
Error Rates and Stroop Error Effect for Adolescent and Young Adult Programmers and Non-
Programmers in the Stroop Colour-Word Task
Trial type ERR% SD ERR% SD ERR% SD ERR% SD
Congruent 1.8 3.6 5.8 6.9 7.4 8.8 1.9 3.7
Incongruent 2.8 4.2 6.7 5.3 4.6 8.4 3.7 6.1
Neutral 2.8 4.2 5.8 6.9 3.7 6.1 0 0
All 2.5 1.7 5.8 4.0 5.2 4.0 2.0 2.0
Stroop effect 4.6 8.5 0.8 8.3 -2. 8 12.5 1.9 8.1

Global errors. Exploratory data analysis of total error rates indicated that the
data were not normally distributed in three of the four groups: adolescent programmers,
D(8) = .327, p <. 05, adolescent controls, D(10) = ..276, p <. 05, and young adult
controls, D(9) = .272, p <. 05. A Kruskal-Wallis test indicated that there was a
significant difference between the groups, H(36) = 8.51, p < .05; however post-hoc
Mann Whitney tests with Bonferroni corrections were not significant, p >.008.
Stroop effect errors. Exploratory data analysis of the Stroop error effect
indicated that the data were not normally distributed in the adolescent programmers’
group, D(9) = .375, p = .001. A Kruskal-Wallis test indicated that there was a
significant difference between the groups, H(36) = 8.51, p < .05. This reflected the
unusual negative Stroop error effect recorded by the young adult programmers (M = -
2.78, SD = 12.5), indicating that they made more errors in congruent trials than
incongruent trials; in contrast, the adolescent programmers (M = 4.63 SD = 8.45),
adolescent controls (M = 0.84, SD = 8.29) and young adult controls (M = 1.85, SD =
8.10) made more errors in incongruent trials.
Discussion
The goal of the present study was to investigate whether the “bilingual
advantage” was also found in computer programmers. The study was motivated by a
hypothesis that the that same mechanism used by bilinguals to manage two separate
languages may also be utilised by computer programmers, in order to prevent negative
transfer between computer programming languages and natural languages. The study
used two tests of executive control in which bilinguals have been found to outperform
their monolingual peers: the Attention Networks Task (ANT) and the Stroop colour-

word task. It was predicted that computer programmers would respond to these tasks
more quickly than their monolingual peers. More tentatively, it was predicted that a
reduced conflict effect might also be found in the computer programming groups.
All of the computer programmers who participated in the study acquired their
computer programming languages during adolescence; importantly, bilinguals who
acquired their second language during adolescence have been found to exhibit the
bilingual advantage. Two groups of programmers, with differing levels of experience,
were studied in order to provide an indication of the extent of programming language
experience required to produce an advantage. In order to provide a baseline comparison,
two groups of age-matched and IQ-matched controls also took part.
The results of the ANT supported the first hypothesis: programmers recorded
faster global RTs than monolingual controls, and this difference was significant. Both
adolescent and young adult programmers’ groups recorded faster global RTs than both
of the control groups. The young adult professionals recorded the fastest RTs, while the
young adult controls recorded the slowest RTs; higher levels of experience and
expertise were associated with greater executive control advantages over age-matched
controls. Contrary to the second hypothesis, there was no significant difference in
conflict effect. The results suggest that computer programmers may benefit from
greater efficiencies in conflict monitoring network, the same mechanism thought to be
responsible for the bilingual advantage.
The results of the Stroop test are more difficult to interpret. Contrary to
predictions and to the ANT results, young adult programmers recorded the slowest
response times; the difference was significant in the congruent condition. Largely as a
result of this slow response in congruent conditions, the young adult programmers’
group recorded a positive Stroop effect; they responded more quickly in incongruent
than congruent conditions. A significantly slower global RT for programmers does not

fit with the predictions of the study. One study from the bilingual literature is relevant:
using the (admittedly quite different) spatial Stroop test, Bialystok, Craik and Luk
(2008) found that bilinguals outperformed monolinguals on interference effect but not
on global RT. This was the result of bilinguals responding 50ms more slowly than
monolinguals on congruent trials (they were just 10ms faster than monolinguals on
incongruent trials). So, whilst the slow reaction times by programmers in the
congruent condition were not predicted, they are not entirely at odds with findings
relating to the bilingual advantage. It should also be noted that in a neuroimaging
study, Waldie, Badzakova-Trajkov, Milivojevic, and Kirk (2009) failed to find a
significant difference in the performance of young adult monolinguals and bilinguals in
a very similar Stroop colour-word task; however, they did find evidence that different
areas of the brain were active during this task.
Somewhat surprisingly, three of the groups – both adolescent and young adult
controls as well as the young adult programmers – were negatively affected by the
presence of helpful cues (the word spelling out the correct answer) compared to neutral
conditions; only the adolescent programmers made efficient use of these cues – they
also showed less conflict cost than both control groups. It could be argued there are
differences in the way computer programmers process text-based versus symbolic
information – perhaps the Stroop activity is more like bug-fixing, and therefore experts
access some automatic process to spot mistakes, but have no such practiced skill in
utilizing useful cues – and that this difference only emerges following extensive
practice, hence the contrast with the less experienced group. However, the unexpected
pattern of these results, coupled with observed weaknesses in the experiment as the
result of random presentation of the stimulus, raise the strong possibility that something
other than the Stroop effect may have influenced the results. The results should
therefore be interpreted with caution.

Overall, the present study has put forward a strong argument for the possibility
that the “bilingual advantage” in executive control may also be found in computer
programmers. The results of the study provide some evidence in support of that
argument – there was a statistically significant advantage for the programming group in
the ANT. However, the results of this small study are far from conclusive, and further
research is needed before clear evidence of an association between computer
programming and executive control can be claimed. The annual Young Rewired State
event, which occurs in August, offers the opportunity to access large numbers of young
coders; over 500 under 19s participated this year. There are also ample opportunities
for prospective studies, for example, a new computer science GCSE will be available in
schools from September and, in the younger age group, the organization Code Club
aims to run computer programming courses in half of the UK’s primary schools.
It should be noted that the due to the quasi-experimental design, we cannot be
certain that programming experience caused the differences in performance on the
tasks; it may be that people with better executive control are more likely to persist with
the development of programming language expertise because this advantage eases the
learning process. A prospective study could help to clarify the direction of causation.
However, if programming language acquisition is a benefit, rather than a cause, of
enhanced executive control, this would not make the association any less interesting; in
fact, it may make it more so. Learning to program is notoriously difficult – it is
estimated that between 30% and 60% of university students fail the first programming
course (Dehnadi & Bornat, 2006) and predicting which students are likely to make the
grade has proved equally difficult (Bornat, Dehnadi, & Simon, 2008). If other studies
replicate an association between executive control and computer programming, it would
be interesting to compare the performance of bilinguals and monolinguals on
introductory programming courses. It has been argued that since most bilinguals

acquire two languages as a result of circumstance, rather than as a result of social
advantages or cognitive abilities, the bilingual advantage is likely to be causal (though
this argument is not yet been supported by prospective studies). If second language
acquisition could help to develop the cognitive prerequisites of learning to program, this
could have important implications for computer science education.
[Words: 11,341, excluding references and diagrams]

References
Ben-Zeev, S. (1977). The Influence of Bilingualism on Cognitive Strategy and
Cognitive Development. Child Development, 48(3), 1009–1018.
Bergersen, G. R., & Gustafsson, J. E. (2011). Programming skill, knowledge and
working memory among professional software developers from an investment
theory perspective. Journal of Individual Differences, 32(4), 201–209.
Bialystok, E. (2005). Consequences of bilingualism for cognitive development. In J. F.
Kroll & A. M. B. De Groot (Eds.), Handbook of bilingualism: psycholinguistic
approaches (pp. 417–453). New York, NY: Oxford University Press.
Bialystok, E. (2006). Effect of Bilingualism and Computer Video Game Experience on
the Simon Task. Canadian Journal of Experimental Psychology/Revue
canadienne de psychologie expérimentale, 60(1), 68-79.
Bialystok, E. (2008). Bilingualism: The good, the bad, and the indifferent. Bilingualism:
Language and Cognition, 12(01), 3.
Bialystok, E., Craik, F., & Luk, G. (2008). Cognitive control and lexical access in
younger and older bilinguals. J Exp Psychol Learn Mem Cogn, 34(4), 859-873.
Bialystok, E., Craik, F. I., & Freedman, M. (2007). Bilingualism as a protection against
the onset of symptoms of dementia. Neuropsychologia, 45(2), 459-464.
Bialystok, E., Craik, F. I., Klein, R., & Viswanathan, M. (2004). Bilingualism, aging,
and cognitive control: evidence from the Simon task. Psychol Aging, 19(2), 290-
303.
Bialystok, E., Craik, F. I. M., Green, D. W., & Gollan, T. H. (2009). Bilingual Minds.
Psychological Science in the Public Interest, 10(3), 89-129.
Bialystok, E., Craik, F. I. M., & Luk, G. (in press). Bilingualism: consequences for
mind and brain. Trends in Cognitive Sciences.

Bialystok, E., & DePape, A. M. (2009). Musical expertise, bilingualism, and executive
functioning. J Exp Psychol Hum Percept Perform, 35(2), 565-574.
Bialystok, E., Martin, M. M., & Viswanathan, M. (2005). Bilingualism across the
lifespan: The rise and fall of inhibitory control. International Journal of
Bilingualism, 9(1), 103-119.
Bornat, R., Dehnadi, S., & Simon. (2008). Mental models, Consistency and
Programming Aptitude. Paper presented at the Proceedings of the tenth
conference on Australasian computing education, Wollongong, NSW, Australia.
Boroditsky, L. (2001). Does language shape thought? Mandarin and English speakers'
conceptions of time. [Research Support, U.S. Gov't, Non-P.H.S.]. Cogn Psychol,
43(1), 1-22. doi: 10.1006/cogp.2001.0748
Botvinick, M. M., Cohen, J. D., & Carter, C. S. (2004). Conflict monitoring and anterior
cingulate cortex: an update. Trends Cogn Sci, 8(12), 539-546.
Brefczynski-Lewis, J. A., Lutz, A., Schaefer, H. S., Levinson, D. B., & Davidson, R. J.
(2007). Neural correlates of attentional expertise in long-term meditation
practitioners. Proc. Nat. Acad. Sci., 104(27), 11483–11488.
Carlson, S. M., & Meltzoff, A. N. (2008). Bilingual experience and executive
functioning in young children. Dev Sci, 11(2), 282-298.
Cherubini, M., Venolia, G., DeLine, R., & Ko, A. J. (2007). Let's go to the whiteboard:
how and why software developers use drawings. Paper presented at the
Proceedings of the SIGCHI conference on Human factors in computing systems,
San Jose, California, USA.
Cohen, A., & Haberman, B. (2007). Computer Science: A Language of Technology.
Inroads — SIGCSE Bulletin, 39(4).
Colzato, L. S., Bajo, M. T., van den Wildenberg, W., Paolieri, D., Nieuwenhuis, S., La
Heij, W., & Hommel, B. (2008). How does bilingualism improve executive

control? A comparison of active and reactive inhibition mechanisms. J Exp
Psychol Learn Mem Cogn, 34(2), 302-312.
Costa, A., Hernandez, M., Costa-Faidella, J., & Sebastian-Galles, N. (2009). On the
bilingual advantage in conflict processing: now you see it, now you don't.
Cognition, 113(2), 135-149.
Costa, A., Hernandez, M., & Sebastian-Galles, N. (2008). Bilingualism aids conflict
resolution: evidence from the ANT task. Cognition, 106(1), 59-86.
Crystal, D. (1997). English as a global language. Cambridge, UK: Cambridge
University Press.
Dehnadi, S., & Bornat, R. (2006). The camel has two humps (working title). School of
Computing, Middlesex University, UK.
Détienne, F. (2002). Software design: cognitive aspects. New York, NY, USA.:
Springer-Verlag New York, Inc.
Emmorey, K., Luk, G., Pyers, J. E., & Bialystok, e. (2008). The Source of Enhanced
Cognitive Control in Bilinguals: Evidence From Bimodal Bilinguals.
Psychological Science in the Public Interest, 19(12), 1201-1206.
Fan, J., McCandliss, B. D., Sommer, T., Raz, A., & Posner, M. I. (2002). Testing the
Efficiency and Independence of Attentional Networks. Journal of Cognitive
Neuroscience, 14(3), 340–347.
Green, D. W. (1998). Mental control of the bilingual lexico-semantic system.
Bilingualism: Language and Cognition, 1, 67-81
Grosjean, F. (1992). Another view of bilingualism. In R. Harris (Ed.), Cognitive
Processing in Bilinguals. Amsterdam: North Holland.
Hernandez, A. E., Dapretto, M., Mazziotta, J., & Bookheimer, S. (2001). Language
switching and language representation in Spanish-English bilinguals: an fMRI
study. Neuroimage, 14(2), 510-520.

Hernandez, A. E., Li, P., & MacWhinney, B. (2005). The emergence of competing
modules in bilingualism. Trends Cogn. Sci. , 9(5), 220–225.
Hilchey, M. D., & Klein, R. M. (2011). Are there bilingual advantages on nonlinguistic
interference tasks? Implications for the plasticity of executive control processes.
Psychon Bull Rev, 18(4), 625-658.
Hockett, C. F. (1960). The origin of speech. Scientific American, 203, 88-96.
Huttenlocher, P. R. (2002). Neural plasticity: The effects of environment on the
development of the cerebral cortex. Cambridge, MA: Harvard University Press.
Ianco-Worrall, A. D. (1972). Bilingualism and Cognitive Development. Child
Development, 43(4), 1390-1400.
Jarvis, S. (2009). Lexical transfer. In A. Pavlenko (Ed.), The bilingual mental lexicon:
interdisciplinary approaches. Clevedon, UK: Multilingual Matters.
Kessler, C., & Quinn, M. E. (1980). Positive effects of bilingualism on science
problem-solving abilities. In J. E. Alatis (Ed.), Current issues in bilingual
education: Proceedings of the Georgetown Roundtable on Languages and
Linguistics. Washington, DC: Georgetown University Press.
Kessler, C., & Quinn, M. E. (1987). Language minority children’s linguistic and
cognitive creativity. Journal of Multilingual and Multicultural Development, 8,
173–186.
Kokol, P., Podgorelec, V., Zorman, M., Kokol, T., & Njivar, T. (1999). Computer and
Natural Language Texts - A Comparison Based on Long-Range Correlations.
Journal of the American Society for Information Science, 50(14), 1295-1301.
Kuhl, P. K. (2004). Early language acquisition: cracking the speech code. Nat Rev
Neurosci, 5(11), 831-843.

Lehrer, R., & DeBernard, A. (1987). Language of learning and language of computing:
The perceptual-language model. Journal of Educational Psychology, 79(1), 41–
48.
Liao, Y. K. C., & Bright, G. W. (1991). Effects of computer programming on cognitive
outcomes: A meta-analysis. Journal of Educational Computing Research, 7(3),
251–268.
Luk, G., Anderson, J. A., Craik, F. I., Grady, C., & Bialystok, E. (2010). Distinct neural
correlates for two types of inhibition in bilinguals: response inhibition versus
interference suppression. Brain Cogn, 74(3), 347-357.
Martin-Rhee, M. M., & Bialystok, E. (2008). The development of two types of
inhibitory control in monolingual and bilingual children. Bilingualism:
Language and Cognition, 11(01).
Meisel, J. (1989). Early differentialation of languages in bilingual children. In K.
Hyltenstam & L. Obler (Eds.), Bilingualism across the lifespan. Aspects of
aquisition, maturity and loss. Cambridge, UK.: Cambridge University Press.
Murnane, J. (2006). Programming Languages for Beginners: A Comparison of Textual
and Graphic Programming Environments for Novice Programmers. University
of Melbourne, Melbourne.
Murnane, J. S. (1993). The psychology of computer languages for introductory
programming courses. New ideas in psychology, 11(2), 213–228.
Norman, K. L. (2008). Cyberpsychology: An Introduction to Human-Computer
Interaction. New York: Cambridge University Press
Ormerod, T. (1990). Human Cognition and Programming. In J. M. Hoc, T. R. G. Green,
R. Samurçay & D. J. Gilmore (Eds.), Psychology of Programming. (pp. 63–82).
London: The European Association of Cognitive Ergonomics and Academic
Press.

Palumbo, D. B. (1990). Programming Language/Problem-Solving Research: A Review
of Relevant Issues. Review of Educational Research, 60(1), 65-89.
Papert, S. (1980). Mindstorms: Children, Computers, and Powerful Ideas. New York,
NY: Basic Books.
Pascual-Leone, A. (2001). The Brain That Plays Music and Is Changed by It. Annals of
the New York Academy of Sciences, 930: 315–329., 930, 315–329.
Pavlenko, A., & Jarvis, S. (2002). Bidirectional Transfer. Applied Linguistics, 23(2),
190–214.
Pea, R. D. (1983). Logo Programming and Problem Solving. [Technical Report No.
12.]. New York: Center for Children and Technology Bank Street College of
Education.
Pea, R. D., & Kurland, D. M. (1984). On the Cognitive Effects of Learning Computer
Programming. New Ideas Psychol, 2, 137–168.
Peal, E., & Lambert, W. (1962). The relation of bilingualism to intelligence.
Psychological Monographs, 76(546), 1–23.
Peppler, K. A., & Warschauer, M. (2011). Uncovering Literacies, Disrupting
Stereotypes: Examining the (Dis)Abilities of a Child Learning to Computer
Program and Read. International Journal of Learning and Media, 3(3), 15-41.
Philipp, A. M., & Koch, I. (2009). Inhibition in language switching: what is inhibited
when switching between languages in naming tasks? J Exp Psychol Learn Mem
Cogn, 35(5), 1187-1195.
Resnick, M., Silverman, B., Kafai, Y., Maloney, J., Monroy-Hernández, A., Rusk, N., .
. . Silver, J. (2009). Scratch. Communications of the ACM, 52(11), 60.
Ricciardelli, L. A. (1992). Bilingualism and cognitive development in relation to
threshold theory. Journal of Psycholinguistic Research, 21, 301–316.

Sleeman, D., Putnam, R. T., Baxter, J., & Kuspa, L. (1986). Pascal and High-School
Students: A study of errors. Journal of Educational Computing Research.
Special issue on Novice Programming, 5–23.
Soloway, E., & Spohrer, J. (1989). Studying the Novice Programmer. Hillsdale, NJ:
Lawrence Erlbaum Associates.
Spelke, E. S. (2003). What makes us smart? Core knowledge and natural language. In
D. Gentner & S. Goldin-Meadow (Eds.), Language in Mind: Advances in the
Investigation of Language and Thought. Cambridge, MA: MIT Press.
Stern, Y. (2002). What is cognitive reserve? Theory and research application of the
reserve concept. J. Int. Neuropsychol. Soc., 8, 448–460.
Tao, L., Marzecova, A., Taft, M., Asanowicz, D., & Wodniecka, Z. (2011). The
efficiency of attentional networks in early and late bilinguals: the role of age of
acquisition. Front Psychol, 2, 123.
Thierry, G., & Wu, Y. J. (2007). Brain potentials reveal unconscious translation during
foreign-language comprehension. Proc Natl Acad Sci U S A, 104(30), 12530-
12535.
Tremblay, G. P., & Sorenson, P. G. (1986). The Theory and Practice of Compiler
Writing. New York: McGraw-Hill.
Waldie, K., Badzakova-Trajkov, G., Milivojevic, B., & Kirk, I. (2009). Neural activity
during Stroop colour-word task performance in late proficient bilinguals: a
functional Magnetic Resonance Imaging study. Psychology and Neuroscience,
2(2), 125-136.
Wechsler, D. (1999). Wechsler Abbreviated Scale of Intelligence (WASI). San Antonio,
TX: Harcourt Assessment.
Weinberg, G. M. (1971). The Psychology of computer programming: Van Nostrand
Reinhold.

Witschital, P. (1995). TRAPS – An intelligent tutoring environment for novice
programmers. In K. F. Wender, F. Schmalhofer & H.-D. Böcker (Eds.),
Cognition and computer programming. Norwood, New Jersey: Ablex
Publishing Corporation.
Woollett, K., & Maguire, E. A. (2011). Acquiring “the Knowledge” of London's Layout
Drives Structural Brain Changes. Current Biology, 21(24), 2109-2114.
Yelland, G. W., Pollard, J., & Mercuri, A. (1993). The metalinguistic benefits of limited
contact with a second language. Applied Psycholinguistics, 14, 423-444.

Young Minds and Computer Coding

More Related Content

Similar to Young Minds and Computer Coding (20)

More from Michael Newbold (20)

Recently uploaded (20)

Young Minds and Computer Coding