Procedural memory in the gifted child Sybren Spit
Abstract
Gifted children are identified by a wide range of characteristics that, depending on the theoretical framework, can vary from outstanding general intelligence to creativity, social adaptability, motivation, practi-cal intelligence, a good family environment and much more. However, independently from the theoretical framework one adopts, these chil-dren probably should have at least some sort of cognitive advantage that enables them to detect patterns and find creative solutions in differ-ent domains. A relatively better procedural memory might be responsi-ble for these skills. This type of memory is involved in the implicit recognition of abstract rules in several domains, including language. Nevertheless, results from an SRT and AGL task show no difference between gifted children and their non-gifted age mates in the visuo-mo-toric and auditory domains, whereas a relative clause comprehension task did show gifted children have better knowledge of object relative clauses. This is probably the case because gifted children have had more experience with this type of sentences compared to non-gifted children. Gifted children in this study do not distinguish themselves on the basis of their procedural memory, but possibly through the benefi-cial sobenefi-cial environment they grow up in. These findings might have broader consequences with regard to empirical findings in the field of language acquisition.
Key words: artificial grammar learning, giftedness, object relative clauses, procedural memory
Zadie Smith, White Teeth, p. 324
1. Introduction
Gifted children are often described as children who are very talented and who achieve or have the potential to achieve more than their age mates in one or more domains (Subotnik, Olszewski-Kubilius & Worrell 2011). Research in this area emerged in the early 20th century and was initially focussed on general intelligence, measured using standard IQ-tests (Terman 1922; 1925). Gifted people were those who scored above a certain threshold at such a test and distinguished themselves on the basis of their gen-eral intelligence. Gengen-eral intelligence in this sense was often understood as something like Spearman’s (1904) g factor, which is mostly defined as the single cognitive force that underlies a wide range of distinct cognitive functions (van der Maas, Dolan, Grasman, Wicherts, Huizenga & Raijmakers 2006). However, over the course of years, the definition of giftedness has undergone several changes. As a result, differ-ent models and definitions of the phenomenon exist. These models do not solely focus on general intelligence anymore, but take several other factors into account as well.
The most basic model is the triadic model put forward by Barbe & Renzulli (1975), which was further elaborated upon by Mönks (1985). Within the original model, general cognitive abilities, creativity and motivation are all important factors in being gifted. Children are regarded as gifted when they distinguish themselves in all three factors; for example, when they have a very high g factor, are well motivated and are creative above average. Mönks added some contextual factors to this model
Yes, but he’s had everything, and so much of it is nurture, isn’t it? I really believe that. We’ve just been lucky enough to give him so much and with a daddy like Marcus - it’s like having a strong sunbeam shin-ing on him twenty-four hours a day, isn’t it, darlshin-ing? He’s so fortunate to have that. Well, they all are.
that should, at least to some extent, be in favour of a gifted child too. These contextual factors could be the right school, a good family environment and supportive age mates. A gifted child is thus one that not only has the right traits, but also the right en-vironment in which these traits can flourish; if the cognitive capacities of a child nev-er get the opportunity to develop, the child will not be identified as being gifted.
This model forms the basis of different models of giftedness that have been formulated (Gagné 1999; Sternberg 2005), some of which make these personal traits more specific (Heller 2004), whereas others emphasise the negative impact that gift-edness might have on the social development of the child (Kieboom 2007). However, quite independently of the exact theoretical framework one likes to adopt, one thing transcends all models: a gifted child, at least to some extent, must possess certain cognitive abilities. Whatever the precise role of the environment is, both nurture ánd nature play an important role in any definition or model of giftedness. What these ex-act cognitive abilities are, remains unclear and is not very well defined in the litera-ture, but a general picture can be drawn. Gifted people all seem very well able to recognise patterns and come up with creative solutions in at least (!) one domain (Sternberg & Davidson 1985). Furthermore, recent psychological literature suggests procedural memory is responsible for, for example, pattern recognition in the auditory as well as in the visual domain (Packard 2009; Ullman & Pierpont 2005). Pattern recognition might be a skill that is necessary in order to come up with creative solu-tions in different domains (Hofstadter & Sander 2013; Hong 2013); thus, it might seem reasonable to assume gifted people distinguish themselves on the basis of their procedural memory capacity.
Procedural memory is the part of our cognitive system that is affiliated with the implicit learning and automatization of, amongst other things, abstract rules and habituated motor skills. This memory system relies, at least partly, on the basal gan-glia and, for example, people with Parkinson’s disease, who have problems with their habituated motor skills, show malfunctioning of these brain parts (Middleton & Strick 2000), whereas they are able to recall concrete facts and events. Procedural memory furthermore is not restricted to a certain domain, but it can operate on auditory, visual, as well as other domains.
Procedural memory is often opposed to declarative memory. This declarative memory is our storage of more arbitrary information, concrete facts and events. It is involved in our memory of specific spatial locations, but also of sequences (Conway & Pisoni 2008). Moreover, declarative memory seems to rely on medial temporal lobe structures and people with Alzheimer’s disease, who have medial temporal lobe deficits, often show problems remembering certain events, whereas they have a well functioning procedural memory (Squire & Zola 1996). Together with the people with Parkinson’s disease, these clinical populations show supporting evidence for the dis-tinction between the two memory systems, as they can be impaired independently of one another.
This declarative/procedural memory distinction plays an important role in lan-guage as well. Ullman (2004) suggests declarative memory can also be regarded as the lexical memory which stores the arbitrary relationships between sounds and their meanings. Procedural memory, on the other hand, can be seen as our knowledge of abstract rules and thus our grammar. It is even argued that SLI, which is often defined as an impairment that is specific to language without having any hearing impairments or general cognitive deficits (Bishop 1997; 2014), is caused by malfunctioning of pro-cedural memory (Ullman & Pierpont 2005; Lum, Conti-Ramsden, Page & Ullman 2012). Additionally, there have been studies suggesting domain-general procedural memory and linguistic processing are tightly related in non-clinical populations as well (Misyak, Christiansen & Tomblin 2010).
Thus, what becomes clear, is that there are two main memory systems, of which one is involved with memorising specific events, facts and also words: declara-tive memory. Procedural memory, on the other hand, stores abstract rules of all differ-ent kinds, for example in language. Furthermore, part of the characterisation of gifted children is that they excel in coming up with creative solutions and possibly do this in different domains. One might expect they need a certain ability of recognising pat-terns of all sorts and noticing abstract regularities in incoming information, as these abilities might be necessary in solving problems (Hong 2013). No prior research has investigated which cognitive capacities actually are responsible for these skills in gift-ed children, but it might be reasonable to assume these children distinguish
them-selves in the power of their procedural memory capacity. In this thesis, I investigate whether or not these children indeed distinguish themselves on this basis. Further-more, procedural memory also seems to subserve linguistic learning, and some sug-gest knowledge of abstract grammatical rules and procedural memory are correlated. Based on this idea, I will investigate if gifted children also distinguish themselves in their knowledge of abstract grammatical rules.
In the following sections, I will first elaborate further on giftedness. Subse-quently, I will discuss the distinction between procedural and declarative memory. Thereafter, the working hypotheses will be given. I will then describe the methods in more detail and present the results. In the last sections, I will discuss these results and draw a conclusion.
2. Giftedness
As mentioned in the introduction, different models of giftedness have been proposed in the literature. I will discuss these different models in more detail in this chapter, but I will start with a small history of the research that has been done in the field, to un-derstand why the models differ and what points they have in common.
2.1 A basic model of giftedness
Systematic research to the nature of giftedness started out with Terman’s (1922) New
Approaches to Genius. In his time, research on the subject of genius mainly
encom-passed anecdotal stories and biographical notes of the lives of adults who are consid-ered to be very talented, which were used to determine which factors might influence the emergence of genius. However, the reliability of these stories is questionable, be-cause of their very subjective nature. Secondly, even if these biographical notes are trustworthy, only people that actually became successful were investigated; thus ex-cluding all people who potentially could have become geniuses, but did not manage to do so.
influ-encing its development, Terman proposed to investigate gifted children, because this “opens the way to a more thoroughgoing study of the genetic aspects of the problem, of the environmental factors which affect genius, and of the exact nature of its devia-tions from average” (p.311). Whilst aiming at more systematic research, he chose the IQ-test as an objective measure of intellectual capacities. He investigated children who scored above a threshold of 140 on this test and gathered additional data on, for example, their general achievements in school, general knowledge, and their home and school situation. This information was used to investigate which factors might influence the potential of later becoming a genius and determine whether or not someone is gifted.
In his research, giftedness was thus measured using a standard IQ-test. This test is generally said to measure what is often called Spearman’s g factor (van der Maas et al. 2006). Spearman's g factor is a psychometric variable that refers to general intelligence, which has been proposed by Spearman (1904), when he noticed nearly half of the within-individual differences on different cognitive tasks could be ex-plained by a single factor. Because Terman (1922) used IQ-tests as his primary source of determining whether or not children are gifted, giftedness in his research can be regarded as possessing outstanding general intelligence. Research on the nature of genius and giftedness in this way rather resembled the investigation of the nature of general intelligence.
This methodological and theoretical approach has been the standard for nearly half a century. Throughout the years, most researchers used IQ-tests to determine whether or not children are gifted and to investigate which factors might contribute to this giftedness (Subotnik et al. 2011 and references therein). Nevertheless, several re-searchers came to notice that general intelligence alone is not indicative of giftedness. Holland and Astin (1962), for example, observed the following:
"Getting good grades in college has little connection with more remote and more socially relevant kinds of achievement; indeed in some col-leges, the higher the student’s grade, the less likely it is that he is a
person with creative potential. So it seems desirable to extend our criteria of talented performance.” (pp. 132-133)
The idea that outstanding general intelligence is identical to giftedness thus might be a simplification of what is actually going on. This conclusion led several researchers to extend these criteria and propose different models of giftedness.
A first model of giftedness that takes more into account than just general intel-ligence, is the three ring model by Renzulli (1978). Renzulli redefines giftedness and argues it should encompass at least three elements: above average intelligence, a high level of task commitment, and a high level of creativity. In this view, children that ex-cel in just one of these domains, for example when they have just above average intel-ligence, should not be regarded as gifted. As mentioned before, just above average intelligence might be related with good school performance, but there is more to gift-edness than just academic achievements. Something similar can be argued for the oth-er two components that play an important role in this model; just excellent creativity does not result in giftedness (MacKinnon 1964), nor does outstanding motivation (Roe 1952; MacKinnon 1965).
These ideas thus culminated in a model of giftedness in which creativity, intel-ligence and motivation all play an important role and gifted children are the ones that
Figure 1. The triadic model by Renzulli & Mönks.
excel in these different domains. If these domains are represented as circles, gifted-ness can be regarded as the place where the circles meet. This very basic model has been expanded with a couple of different contextual factors by Mönks & Ypenburg (1993), who draw a triangle around these three circles. This triangle represents the environment and incorporates a good home, school and social environment into the model. Creativity, motivation and intelligence are of course important, but these need to be stimulated by a facilitating family, proper education and supportive age mates. This new triadic model can be seen in Figure 1 and forms the basis of different, more elaborate models that I will discuss in the next section.
2.2 Some more elaborate models of giftedness
First of all, a multifactorial modal, or a model of multiple intelligences can be distin-guished (Heller 2004; Gardner 1995). Within this model, different cognitive factors determine whether or not someone is gifted. These factors can, for example, be one’s intellectual abilities, social abilities, psychomotor skills, musicality or creativity. These factors are a prerequisite, but can only flourish within the right context; namely, supportive family, good education and sufficient life experiences. Apart from the con-text, one also needs the right - as the authors call them - non-cognitive capacities, like motivation, stress-sensitivity and confidence. Giftedness can be recognised by the skills children have in different areas, such as abstract thinking, technical abilities, sportsmanship etc. These skills are the synthesis of these different contextual and cognitive factors, and the model is best characterised as a very detailed extension of the basic triadic model.
Another approach is the theory of successful intelligence by Sternberg (2005). In this view, gifted people achieve goals within a specific sociocultural context and are able to successfully exploit their (specific) intelligence. Intelligence in this sense must be understood as the competence to excel within a certain domain, and this con-sequently means the abilities a gifted person must have are partly dependent on the environment within one wants to excel. If you want to excel in mathematics, you are gifted when you possess the mental abilities to excel in mathematics. If one wants to
excel in sports, someone with good physical abilities is gifted. A gifted person thus is the one who has the domain-specific intelligence to become successful in the domain of interest. Nevertheless, independent of the domain of interest, someone needs a cer-tain cognitive competence in the first place to be able to excel in any given domain.
Gagné’s (1999) differentiated model of giftedness and talent distinguishes, as might be expected, between giftedness and talent. Giftedness can be regarded as the combination of natural abilities that are observable even before exposure to any struc-tured learning, practice or training. Talents, on the other hand, are systematically de-veloped skills that flourish due to learning, practice and training. This approach dis-tinguishes itself from the earlier mentioned approaches, because giftedness just refers to those mental abilities that do not require any contextual stimulation to flourish. There are also certain abilities that can flourish when the context is right, but these are defined as talents. The definition of giftedness in this sense is thus rather minimalistic. A last model is the one by Kieboom (2007), which not only focusses on cogni-tive abilities, but also incorporates non-cognicogni-tive traits, whilst emphasising the thin line between the positive and negative impact these non-cognitive traits could have on the gifted child. Kieboom argues the gifted child does not only have above average general abilities, as they are described in the different frameworks before, but also possesses certain non-cognitive traits like perfectionism, a sense of justice, hypersen-sitivity and a critical attitude. Furthermore, she points towards the possible downside these traits could have, as for example perfectionism could lead to performance anxi-ety and a critical attitude could scare other people off, which ultimately might lead to social misfitting. The negative impact of these personal characteristics makes Kieboom question whether or not giftedness indeed is a true gift or might be regarded as some sort of Trojan Horse.
2.3 Summary
I have discussed several different approaches in the characterisation of gifted children and sketched a short history of giftedness in which researchers originally identified gifted children using IQ-tests. Children who scored above a certain threshold fell into
the group. As shown, giftedness in this way became quite similar to general intelli-gence. As a result, the conception of giftedness has changed over the years and sever-al other factors, like creativity, motivation and multiple contextusever-al factors, have been incorporated in models of the phenomenon. Some theories emphasise personal traits, whereas for others giftedness is largely dependent on the domain within which a per-son wants to become talented. However, quite independently of the exact theoretical model one wants to adopt, some observations transcend the different models.
First of all, a gifted child has to possess at least some cognitive traits that make them gifted; the child needs to have at least some gifts. A child without suffi-cient intellectual abilities will still not be gifted when it grows up in a very advanta-geous environment. Without ignoring the benefits and disadvantages of the different models, some behavioural characteristics correspond with all different models. Gifted children quickly come up with solutions to problems of all different kinds. These children seem better at this, because they have outstanding motivation, intelligence and creativity. These observations thus seem to follow logically from the basic triadic model, independent of the specific theoretical position, and capture the main charac-teristic I would like to focus on in this study; outstanding problem solving in (differ-ent) domains.
This position is also taken because of the type of diagnostics that is used in education, when there is a need to identify gifted children. In the Netherlands, some primary schools have special classes, in which children who are not challenged enough in regular class come together, and treat specific subjects. Two main identifi-cation methods that are used in primary schools are Digitaal Handelingsprotocol
Hoogbegaafdheid (DHH ‘Digital Operational Protocol Giftedness, van Gerven &
Drent 2011) and SIDI-3 (Kuipers & de Bruin-de Boer 2010), mostly in combination with school specific norms (Verouden, personal contact). However, independent of the specific measurements, children are selected who outperform their age peers in abilities like problem solving, in which pattern recognition plays a substantial role.
Gifted children thus outperform their age mates in several domains and seem to posess some cognitive skills associated with gifted children, like better intelligence, creativity and motivation. However, which cognitive capacities are responsible for the
cognitive development these children undergo and for the better developed cognitive skills these children have, is an open question, as there are no studies that investigate the distinguishing capacities of this population (Hettinger Steiner & Carr 2003). What capacities could these children possess that make them able to outperform their age mates? Is there a specific cognitive capacity that can be held responsible for this? One capacity that could be responsible for these skills is a better developed procedural memory (Ullman 2004). This cognitive capacity is affiliated with abstract rule learn-ing and the recognition of patterns; skills that possibly support fast problem solvlearn-ing (Hong 2013). I will further elaborate on procedural memory in the next section.
3. Procedural memory
As mentioned in the introduction, a distinction is made between procedural memory and declarative memory (Ullman 2004). The former is associated with our knowledge of abstract rules and patterns, whereas the latter is more involved in knowledge of facts and events. The two are often argued to be represented in different areas of the brain (Middleton & Strick 2000; Conway & Bisoni 2008) and both seem to be related to different linguistic domains as well (e.g. Ullman & Pierpont 2005; Misyak et al. 2010). In the following section, I will discuss the distinction between the procedural and the declarative memory systems and in the subsequents section, I will show how they are related to our linguistic knowledge.
3.1 The declarative/procedural distinction
The distinction between declarative and procedural memory is rooted in the “funda-mental idea that memory is not a single entity but consists of several separate entities that depend on different brain systems” (Squire & Zola 1996, p. 13515), of which Figure 2 gives an impression. Squire & Zola argue amnesic patients are severely im-paired in declarative memory. Amnesia is best described as a psychological disorder resulting in total or partial loss of memory and caused by a psychological or physio-logical trauma. This kind of memory loss is not of the ordinary type - the forgetting of
telephone numbers - but has further reaching consequences. People with amnesia of-ten forget certain events that happened prior to the trauma and have problems remem-bering who they are (Minderovic 2008), but also when they have to recognise specific combinations of words that have been presented to them in an experimental setting. Neurologically, the disorder is caused by impairment in the medial temporal lobe le-sions. This is the part of the brain that is commonly associated with the storage of concrete facts and events (Squire, Stark & Clark 2004), with different parts of it being responsible for the storage of objects and locations (Buffalo, Bellgowan & Martin 2006).
These findings lead to the idea that there is a distinct part of the brain that is involved in the storage of these arbitrary relationships. This part of our memory is Figure 2. Brain scans showing the averaged activation of 14 participants when processing either syntax, which is considered a procedural process, or semantics, which is considered a declarative process, indicating several distinct areas involved in these processes (Ullman 2001).
called declarative memory, and thus is involved in the storage of, amongst other things, objects, locations and events. An important quality of declarative memory, of-ten noted in the literature, is that the information it acquires can be consciously re-called and some suggest this part of memory explicitly learns new information, as if it were textbook learning. However, one can question whether or not all our knowledge in this memory system indeed should be explicitly acquired. One can imagine learn-ing certain facts or other arbitrary information without belearn-ing explicitly aware of it (Batterink, Reber, Neville & Paller 2015). Thus, here I will not assume declarative memory equates explicit memory.
Declarative memory is often opposed to procedural memory, which can be located in the basal ganglia, but also in other areas (Packard 2009). The basal ganglia contain the neurotransmitter dopamine. People with Parkinson’s disease have prob-lems synthesising dopamine and are generally known for their shaking arms or legs. This can be explained by this lack of dopamine as it is one of the brain’s chemicals that is involved in physical movement. However, these people not only have shaking limbs, but also have more severe behavioural problems in for example habituated mo-tor learning (Richman 2008). In experimental settings, people with Parkinson’s dis-ease have difficulties when they have to learn a visuo-motoric procedure, whereas non-impaired controls are improving their performance (Harrington, Haaland, Yeo & Murder 1990). A malfunctioning of the basal ganglia, by a lack of dopamine, can thus lead to impaired behaviour in procedural motor learning, but also to other impair-ments in procedural memory.
Like there is a separate brain area involved in our memory of facts and events, there also seem to be distinct brain networks involved in our motor control and our recollection of more abstract rules and habitual knowledge: procedural memory. This type of memory is often measured using the Serial Reaction Time (SRT) paradigm (Lum, Conti-Ramsden, Morgan & Ullman 2014). In an SRT task, participants see a visual stimulus that repeatedly appears in one of four boxes on a screen. These stimuli appear in a certain pattern of which participants are unaware. Four buttons accompany these boxes and participants press the button that matches the box with the visual stimulus. Experimenters measure the speed with which participants press these
but-tons and show that participants tend to become faster on trials that behave according to the pattern, whereas they slow down on trials that lack the pattern.
As participants are unaware of the pattern they are learning in such tasks, pro-cedural memory is often equated with implicit learning. We are often not aware of the habits we have learned and the procedures we come to acquire are mostly not explicit-ly taught. However, just as not all declarative knowledge necessariexplicit-ly has to be ac-quired explicitly, not all procedural knowledge strictly has to be implicitly acac-quired. We can very well explicitly learn to acquire certain rules or patterns, even if these are incorporated in our memory as procedures (Batterink et al. 2015). I will thus not equate procedural memory with implicit learning.
In sum, there are several reasons to assume that we have two distinct memory systems that are involved in different types of behaviour and have different neurobio-logical roots. These combined perspectives emphasise the idea that a distinction should be made between on the one hand our declarative memory and on the other hand our procedural memory. The distinction between the declarative and procedural memory plays an important role in the acquisition and processing of language as well (Ullman 2001). I will turn to these issues in the next section.
3.2 Declarative vs. procedural memory in language
Several linguistic theories make a distinction between a grammar and a lexicon (Jack-endoff 2002; Halle & Marantz 1993). In these theories, the lexicon contains informa-tion about words and their meanings and stores the arbitrary relainforma-tionships between sounds and meanings. The grammar, on the other hand, is the mental device that oper-ates on these elements and combines them in order to generate utterances of all kinds. Pinker & Ullman (2002) call this the Words-and-Rules theory and combine it with the declarative/procedural distinction. In this approach, the lexicon is a subdivision of de-clarative memory, as it is concerned with the storage of single items, and grammar is a part of procedural memory, because it requires the use of abstract rules and patterns. This distinction, for example, can be a very helpful distinction in explaining the
sec-ond language acquisition of groups with different learning profiles (Blom, Polisenka & Weerman 2008).
Declarative memory thus might be represented as our mental lexicon, which is often assumed to be the dictionary we have in our minds. This is the place where we store all information that is arbitrarily related to sounds. The linguistic elements that are stored here might be roots and idioms, but also irregular forms (Pinker & Ullman 2002). One can imagine this is the case, because this linguistic information is not composed of separate elements that can be combined according to a rule, but needs to be stored as such; for each of these cases, we need to learn which specific meaning is associated with the linguistic form that accompanies it. In addition to meanings, de-clarative memory also stores more abstract features that are associated with these lin-guistic elements. Moreover, declarative memory might store morphological elements, like suffixes. In English, one has to learn for example that the morpheme -ed realises the past tense. As this is a specific linguistic form that is associated with a meaning too, some argue it finds a place in here as well, but this highly debated (McClelland & Patterson 2003).
In order to store elements in declarative memory, someone simply has to en-counter them a sufficient number of times. When someone enen-counters a certain ele-ment, this gets internally represented and when it gets encountered more, the repre-sentation will get strengthened. After a certain amount of exposure the element will have a sufficiently strong representation in order to become part of declarative memo-ry, a process which is very well simulated using computational models like PARSER (Perruchet & Vinter 1998). This model is also able to show how very frequently oc-curring regular forms might be stored in declarative memory as well. If a regular form is encountered frequently enough, it will not be regarded as a combination of separate elements, but as a single unified item, because it has a strong enough representation as such.
Procedural memory, on the other hand, stores all kind of rules that work on the elements that are stored in declarative memory and the categories that these elements might belong to (Pinker & Ullman 2002). It tells, for example, that roots that are stored as regulars should be combined with the suffix that realises past tense in order
to create a verb in the past tense. It also tells that when you want to create an active sentence in Dutch, you most often begin with the subject, followed by a finite verb and that then possible objects are used. However, in contrast to the content of declara-tive memory, these rules cannot be simply learnt by their frequency. One has to detect the abstract rules that combine certain elements with each other and make it impossi-ble to combine others. Furthermore, because these are rules that apply to all instances of a given category, specific elements on which these rules can work, might have nev-er been encountnev-ered before by a language learnnev-er.
3.3 Different kind of dependencies
However, despite this lack of exposure, language learners are perfectly able to gener-ate these abstract rules from the input and the question is how they do so. A compe-tence that is often suggested to account for this behaviour is the ability to detect the statistical regularities the are present in linguistic input (Erickson & Thiessen 2015). In order to do so, people should have the ability to remark that a certain element be-longing to class X often co-occurs with a element bebe-longing to a certain class Y, inde-pendently from the realisation of these exact forms. For example, one has to know that lexical entries belonging to the class of verbs (class X) often are accompanied by a tense suffix (which belongs to class Y). In this sense, procedural memory might be regarded as a device that applies statistical learning to several domains.
To gain knowledge of this kind of statistical regularities, it is often argued people should have sensibility to transitional probabilities (Saffran, Johnson & Aslin 1996). A transitional probability is the probability that a certain syllable Y co-occurs with a certain other syllable X and can be estimated as in (1).
(1) frequency of pair XY frequency of X
This formula determines how likely it is that Y occurs once an X is encountered. If this results in a high probability, it means it is very likely that an X is followed by a Y.
Contrary, if the probability is low, the dependency between X and Y is less strong. More concrete, this means that if people are sensitive to this feature of language, they will notice that the transitional probability of a tense suffix (Y) following a verb (X) is high and consequently learn this abstract linguistic rule. These transitional probabili-ties can be estimated for different types of dependencies. One can for example esti-mate the transitional probability between two elements that directly follow each other; a so called adjacent dependency. However, one can also estimate a transitional proba-bility between elements that not do not directly follow each other, but have one or more intervening element(s) between them; so called non-adjacent dependencies.
Several studies show that people indeed are sensitive to these transitional probabilities of adjacent dependencies. This is typically observed using the artificial grammar learning (AGL) paradigm. In these experiments, material is created that fol-lows some sort of pattern or is constructed according to a certain rule. Participants are then presented with this input and at test have to distinguish between different kinds of stimuli in a forced-choice test; for example, between stimuli that are present in the input and stimuli that are not present in there, or between stimuli that are fully present in the input and stimuli that are just partly present in the input. In these tasks, the ex-perimenter can manipulate the input in such a way that, for example, the transitional probabilities within words are higher than between words. In order to pass the test, participants need to be able to recognise these transitional probabilities and this is ex-actly what they are capable of (Saffran et al. 1996; Aslin, Saffran & Newport 1998; Endress & Bonatti 2007).
Many of these studies used tasks that showed higher transitional probabilities between syllables that directly followed each other. However, one could argue that in order to know which words are really words and which are not, participants do not need to discover rules, but just a very good declarative memory to store very frequent-ly occurring chunks as a whole, without keeping track of the transitional probabilities between syllables. As there are no real rules that need to be inferred, there is no need for procedural memory in this test; if sufficient input is presented, participants can simply remember parts of the stream by heart. In such a case declarative memory would suffice in learning the task and procedural memory is not necessary.
However, natural language also shows high transitional probabilities between elements that do not directly follow each other, but have some intervening element between them, for example in the progressive tense in English, where the suffix -ing is dependent on the finite inflection of to be (I am writing a sentence.). In such cases, one has to detect a certain dependency between non-adjacent elements. Gómez (2002) shows that adults and infants are both able to detect this kind of regularities as well. In this AGL task, material either has the form AXC or BXD, where A and C or B and D are always the same, but the X in between can vary. Afterwards, participants have to distinguish between stimuli that are in accordance with the rule, having the form AXC or BXD, and stimuli that are not. Results show that they are able to correctly make such a decision, which they can only do because of their sensitivity to the transitional probabilities of non-adjacent dependencies. This supports the idea that this type of ability is one we posses and might explain our linguistic competence to detect abstract rules in natural speech.
Endress & Bonatti (2007) furthermore show that both declarative and pro-cedural memory are involved in such tests. When participants are presented with more linguistic input, they rely more on their declarative memory. In such cases, partici-pants can learn all frequently occurring combinations by heart and use this declarative knowledge to pass the test. On the other hand, when less input is available, partici-pants cannot learn frequently occurring combinations by heart. Moreover, when in-coming chunks are too big to be remembered as one unit, participants cannot learn them by heart even if they occur frequently enough. In these cases, they have to ex-tract some rule that can be used to generalise over all input; their choices at test thus are dependent on procedural memory and on some statistical regularity that is present. Consequentially, when participants learn non-adjacent dependencies the amount of intervening X’s between the dependent elements (A and C) will influence which way they will learn these dependencies. A low amount of X’s leads to only a few fixed combinations of AXC words, which can be learnt by heart individually. However, if the amount of X’s grows, there is not enough input to store each AXC combination separately and it is more efficient to learn the dependency between A and C. A higher variance of elements that intervene the non-adjacent dependency thus
makes it more likely procedural memory will be involved in remembering the rule that is present in the input, whereas a low amount of intervening X’s stimulates the usage of declarative memory in remembering individual forms in the input.
3.4 Dependencies in natural language
The scope of such tasks is of course limited to the laboratory setting in which these experiments are conducted and to the artificial input participants receive. However, there are several reasons to assume there is a tight relationship between procedural memory, statistical learning abilities and natural language processing. One of the main arguments comes from the population of people with specific language impairment (SLI). Individuals with SLI are generally said to have problems acquiring and using language, without this being caused by general cognitive delay or hearing impairment (Bishop 1997; 2014). The impairment tends to occur within families, which gives rise to the thought that it is are heritable disorder (Lenneberg 1967) with its basis in genet-ics. In line with Chomsky (1965), who assumes grammar has a genetic base, it can reasonably be argued that problems occur within this faculty of language and SLI is modality specific impairment (van der Lely 2005). Contrary to this position, others argue SLI is a domain general processing deficit (Joanisse & Seidenberg 1998).
Taking the latter position, Ullman & Pierpont (2005) formulated their Pro-cedural Defecit Hypothesis, claiming SLI is caused by an impairment in proPro-cedural memory. Evidence for this hypothesis is often gathered using the earlier mentioned Serial Reaction Time (SRT) paradigm (Lum et al. 2014), in which participants see a visual stimulus repeatedly appearing according to a pattern, in one of four boxes on a screen. Four buttons accompany these boxes and participants, while unaware of the pattern, press the button matching the box with the visual stimulus. Experimenters measure the speed with which participants press these buttons. Normally, people be-come faster at pressing these buttons, but Lum et al. show that children with SLI do not behave accordingly on this SRT test, as they do not perform better on pattern like blocks than on non-pattern like blocks. This suggests they are unable to extract the pattern from the input and reveals a deficit in their procedural memory.
More supporting evidence for a procedural memory deficit in the SLI popula-tion comes from Evans et al. (2009), who use an AGL task to investigate if children with SLI have problems extracting statistical regularities from given input. They show that children with SLI cannot discriminate between words and non-words at test in an AGL task similar to the earlier mentioned test by Saffran et al. (1996). When syllables are replaced by tones to create melodies, children with SLI are unable to distinguish real melodies from non-melodies. In addition to these findings, Hsu, Tomblin & Christiansen (2014) use a different type of AGL to show that children with SLI are unable to extract statistical regularities in the case of non-adjacent dependencies as well. These findings contribute to the idea that patients with SLI have problems with their procedural memory. The ability to extract statistical regularities in an experimen-tal setting thus seems tightly related to more general linguistic competences, which is also found when individual differences in procedural memory are compared to natural language processing in a typical developing population (Misyak et al. 2010).
What has become clear so far is that we have two different memory systems that are responsible for different types of cognitive skills. On the one hand, declara-tive memory, located in the medial temporal lobe, is responsible for factual knowl-edge and arbitrary relationships. Procedural memory, on the other hand, is located in the basal ganglia areas and is involved in our knowledge of abstract patterns and rules, in which statistical learning plays a prominent role. Both types of memory are involved in natural language as well. The classical distinction between lexicon and grammar can be reduced to these memory systems. Procedural memory in this case is involved in grammar, whereas declarative memory plays an important role in the emergence of our lexicon. In the next section, I will discuss how these two types of memory might relate to the population of gifted children and I will formulate the working hypotheses that were investigated in this study precisely.
4. Hypotheses and predictions
As discussed, gifted children are children who are performing above average in vari-ous domains and are associated with outstanding cognitive capacities. Nevertheless,
only a few studies have been conducted which explored the intellectual development of these children and their cognitive capacities. Hettinger Steiner & Carr (2003) give an overview of the few studies that investigated these issues and noticed that “despite similar interests […] the fields of gifted education and cognitive development have had little communication” (p.216). Still, some topics have been investigated in this specific population; namely their speed of processing, metacognition and problem solving abilities. The authors remark that in most of these studies gifted children out-perform their non-gifted peers. Gifted children are faster decision makers and they need less time to process incoming information. They often make plans deliberately in order to solve problems, which indicates more metacognitive awareness. Finally, gift-ed children also tend to choose the right strategy, when confrontgift-ed with problems that have multiple solutions.
Even though several cognitive behavioural traits have been investigated, these studies do not make a real connection with particular cognitive capacities that are necessary for this distinct behaviour. Hettinger Steiner & Carr try to explain the ob-served results from the different studies from a single mechanism. They argue gifted children develop and expand the multiple strategies in their repertoire very quickly. This not only leads to a vast array of strategies to chose from, when they have to solve problems, but it also gives them the opportunity to pick the strategy that makes the best use of the limited set of available cognitive resources. The fact that these children possess a large set of strategies to choose from, at the same time explains why they show above average metacognitive awareness. These children actually have multiple different strategies available between which they can make a choice. The importance of such a choice can, for example, be observed from the fact that gifted children often take longer to determine with which strategy they are going to solve a problem, but once they have chosen the right strategy, they solve the problem faster than their peers. This shows that having the possibility to chose the right strategy is very impor-tant in order to solve a problem.
However, in order to acquire a repertoire of strategies, Hettinger Steiner & Carr remark that gifted children require a certain level of abstract thinking and also the ability to quickly recognise the similarity between superficially different things. It
is not very surprising the fast recognition of analogies and patterns plays an important role in the development of gifted children, as some claim these are fundamental prin-ciples of our cognition and that great thinkers and innovators distinguish themselves by noticing analogies that others do not grasp (Hofstadter & Sander 2013).
In the previous sections, I have also concluded that it is procedural memory that is involved in the recognition of such patterns, which might necessary to solve problems of different kinds (Hong 2013). Although the reviewed studies did not actu-ally make a connection between procedural memory on the one hand and gifted chil-dren on the other, the two could be naturally connected. It could well be argued that gifted children have a better developed procedural memory. The main hypothesis I will investigate in the remainder of this thesis is thus the following:
Hypothesis: Gifted children have better developed procedural memory
skills than non-gifted age mates in different modalities.
A consequence of this hypothesis is that a better developed procedural memory en-ables gifted children to be better achievers in different domains. This is not to claim that contextual factors do not play a role in the cognitive development of these chil-dren anymore; procedural memory is of course very dependent on the amount and type of input it gets.
With this main hypothesis in mind, I will formulate a sub-hypothesis that will be investigated as well. As mentioned before, procedural memory and grammatical knowledge are tightly connected with one another. It might thus be reasonable to as-sume that the linguistic knowledge of gifted children is better developed as well, be-cause this correlates with a better developed procedural memory. This sub-hypothesis is thus the following.
Sub-hypothesis: Gifted children have better knowledge of abstract grammatical rules than non-gifted age mates.
These hypotheses were tested using a couple different tasks that measured procedural memory in different domains. These tests provided excellent material to investigate whether or not gifted children have a better procedural memory and whether this ap-plies to the linguistic domains too. I will discuss these methods in more detail in the next section.
5. Methods
5.1 Participants
25 Dutch speaking gifted children (9 males, 16 females, M = 10;6, SD = 1;2) were recruited in primary schools Rotterdam and Utrecht. Children were in grade 5 to grade 8, as this is when schools start to identify gifted children in their classes. All children participated in a so-called Plusklas (‘Plus-class’) or Wetenschapsklas (‘Sci-ence class’) in their schools. Children participate in these classes when regular educ-tion is not challenging enough. Teachers of the regular class decide who qualify to participate in this educational program. Teachers take both test results of these chil-dren in regular education and their own insight and personal experiences with the children into consideration when making this decision. A control group consisted of 25 Dutch speaking typically developing children (10 males, 15 females, M = 10;9, SD
= 1;2). These children were recruited at primary schools in Rotterdam and
Bode-graven. Children were from the same age (t(48) = -.66, p = 0.51) and matched on grade.
5.2 Materials
All children were assessed with tasks that measure procedural memory in different domains. An SRT task was used to measure procedural memory in the visuo-motor domain. An AGL task was used to investigate the auditory domain. A relative clause comprehension (RCC) task measured competence in the natural language domain. Furthermore, a digit span (DS) task tested their working memory capacity.
5.2.1 SRT
The SRT task is developed to measure implicit visuo-spatial sequence learning in procedural memory (Lum et al. 2012). Participants were presented a tablet with four boxes on the screen and received a game controller with four buttons that matched these boxes. A smiley appeared in each of the boxes and participants had to press the button that matches the box in which the smiley appears. Unknown to the participants, these smileys appeared in a pattern that started over after every tenth smiley. Reaction times (RTs) of pressing these buttons were measured. Normally, participants tend to press the buttons faster, when they (unconsciously) recognise the pattern is underlying the appearance of the smileys. Furthermore, when a test block does not follow the pat-tern, RTs consequently drop down.
This task consisted of a practice session, in which children could practice with pressing the right buttons on the controller. The practice session finished with 10 ran-dom appearances of the smiley. If children would press the wrong button 5 or more times, the practice session had to be done once more. After practice, four patterned blocks of 60 smileys were presented, with every patterned sequence consisting of 10 smileys. During these blocks, participants were expected to become faster at pressing the button, because they got accustomed to the pattern and the task in general. These four blocks were followed by a random block consisting of 60 smileys, in which it is expected that participants become slower. This random block was inserted, because participants might become faster only because they get used to the task and not be-cause they recognise the specific pattern. If participants indeed recognise a pattern and this causes them to become faster a the task, they should become slower at a dom block, as it does not follow the pattern. If participants are not slower at this ran-dom block, they only get faster because they accustom themselves to pressing the but-tons as fast as possible, but not because they recognise an underlying pattern. After the random block, a final pattern-like block was presented to participants. The pattern in this block was similar to the pattern in the first four blocks.
The task was presented on a tablet (Microsoft Surface 3), with a wired con-troller (Trust wired gamepad GXT540) attached to it. The explanation was
pre-recorded. Results were registered in Eprime (Psychology Software Tools, Inc.). Af-terwards an exit interview was held, in which the experimenter asked whether the par-ticipants knew a pattern was underlying the appearance of the smileys and if they could reproduce this pattern.
5.2.2 AGL
The AGL task measures procedural memory as well, but in the auditory domain. This design (Lammertink, van Witteloostuijn & Rispens in progress) combined the classi-cal forced-choice test after exposure, with an online processing measure during expo-sure to investigate whether a learning effect occurs during the test. Participants played a game in which they had to help Appie de Aap (‘Molly the Monkey’) pick bananas from a tree. Participants could help the monkey by pressing a button when they heard a specific word in a monkey language. The word in the monkey language that trig-gered a participant to press the button was either the non-existing syllable lut or the non-existing syllable jik. Two versions of this test were created. In one lut was the tar-get word and jik was used as a filler, in the other version this was vice versa. The monkey language was constructed such that a ‘monkey sentence’ consisted of four non-existing syllables, or ‘monkey words’. Sentences in the monkey language either followed the structure sot-X-jik or tep-X-lut, where the presence of jik depended on
sot and the presence of lut depended on tep. Between these two syllables another
monkey word consisting of two syllables, like wadim or domo, occurred. Pauses be-tween monkey words were 250 ms, and pauses bebe-tween monkey sentences were 750 ms. Participants could (unconsciously) predict the appearance of the target syllable by recognising the high transitional probability between non-adjacent dependencies.
A practice session consisted of several isolated syllables - thus not constructed as monkey sentences - within which there were six target stimuli. Participants had to press the button to pick a banana from the tree. They had to redo the practice session when they picked three bananas or less in order to continue with the training phase. In the training phase, participants first underwent four blocks that consisted of monkey sentences that followed the pattern. Each block consisted of 24 monkey sentences
fol-lowing the pattern sot-X-jik and 24 monkey sentences folfol-lowing the pattern tep-X-lut. 24 different monkey words could take the place of the X. A fifth block again consisted of 24 monkey sentences that ended with the target syllable. However, the presence of this syllable could not be predicted from the first syllable of the monkey sentence. Such a monkey sentence could for example be wadim-domo-jik or wadim-domo-lut. Participants still had to press when they heard the target syllable, but could not predict this from the first syllable of a monkey sentence anymore. Consequently, RTs were expected to drop in these blocks. The other 24 monkey sentences in this block were randomly constructed, not necessarily beginning or ending with a specific syllable. After this random block, a last block was presented that followed the pattern from the first four blocks.
After all blocks were presented to a participant, they were submitted to a two-alternative forced-choice test. In this test, they heard two possible sentences from the monkey language and had to decide which of the two sounded more familiar. Deci-sions had to be made between either a non-existing monkey sentence and a real mon-key sentence; for example, sot-X-lut and sot-X-jik. A total of eight of these choices had to be made by each participant. For four of these decisions, a generalisation had to be made, as the X was one that did not already occur as an X during training. This was not the case for the other four decisions. There, the intervening X did already oc-cur as an intervening X during training.
The task was presented on the same tablet (Microsoft Surface 3) and all audi-tory input was presented via headphones (Sennheiser HD 201). Instruction was pre-recorded as well and was the following: “Druk zo snel mogelijk als lut/jik komt” (‘Press as fast as possible when lut/jik comes’). This should have made it pos-sible for participants to press even before the target syllable was presented, if they predicted it was coming on the basis of the first syllable of a word. However, this in-struction did not make them explicitly aware of the fact there was a pattern behind these appearances. Results were again registered in Eprime (Psychology Software Tools, Inc.). Afterwards an exit interview was held, in which the experimenter asked the participants if they could complete a couple of monkey sentences in which the last
word was missing and if they knew when lut and jik were coming.
5.2.3 RCC
An RCC task, adapted from Duinmeijer (in progress), was conducted to tap syntactic knowledge of the children. In this test, participants were presented with a sentence and two pictures. The sentence described only one of these two pictures and partici-pants had to decide which picture matched the heard sentence. Choosing the right pic-ture could only be achieved when a participant understood the sentence correctly. Sentences could for example be as in (2):
(2a) Dit is de piraat die de clowns slaat. ‘This is the pirate who hits the clowns.’
(2b) Dit is piraat die de clowns slaan.
‘This is the pirate whom the clowns hit.’
These sentences always started with “Dit is de … die…” (‘This is the … who…’), where a person was introduced, for example piraat ‘pirate’. The relative clause was either a subject relative clause, where the relative pronoun was the subject in the rela-tive clause (2a), or an object relarela-tive clause, where the relarela-tive pronoun was the object in the relative clause (2b). The verb in the relative clause was always a transitive verb and could have both an animate subject as well as an animate object, like hit ‘slaan’. The interpretation of these sentences crucially depended on the agreement of the verb with a subject, as the orders of the two sentences were identical. In (2a), the verb is singular and thus has to agree with the relative pronoun resulting in a subject relative clause. In (2b), the verb is plural and has to agree with the plural clowns. The relative clause consequently is an object relative clause. Participants heard only one of these two sentences, but were presented with two pictures that each showed one of these two meanings, as can be seen in Figure 3.
under-standing that interpretation of the sentences is dependent on subject verb agreement. This type of sentence typically does not occur that frequently in spoken language, when both arguments of the verb in the relative clause are animate lexical items (Reali & Christiansen 2007). Therefore, it is not so straightforward to come to the right in-terpretation of such sentences in natural language. Subject verb agreement normally is not necessary for the interpretation of such sentences, because speakers can rely on word order cues. However, to rightly interpret these sentences, participants had to rely on their syntactic knowledge.
The test consisted of twelve sentences with a subject relative clause (2a) and twelve sentences with an object relative clause (2b). Additionally, six fillers had a subject relative clause in which both arguments of the verb were singular. These sen-tences were thus ambiguous as two arguments could possibly agree with the main verb in the relative clause (3a). Furthermore, six fillers contained a relative clause with a passive (3b).
(3a) Dit is de piraat die de clown slaat.
‘This is the pirate who hits the clown.’ / ‘This is the pirate whom the clown hits.’
(3b) Dit is piraat die door de clown wordt geslagen. ‘This is the pirate who is hit by the clown.’
Figure 3. The two pictures between which a participant had to choose, when s/he heard the sentence. Dit is de piraat die de
clowns slaan. ‘This is the pirate who is hit by the clowns.’ The right picture was the target.
All sentences were pre-recorded and played on a laptop. Pictures were shown on the screen of the laptop. Participants could either listen to the sentence via headphones or not, depending on their preference. Sentences might be repeated and all results were administered using pencil and paper.
5.2.4 DS
As a background measure for working memory capacity, a forward and backward dig-it span (WISC-III) were administered. Participants heard a numerical sequence and had to repeat these numbers either in forward direction or in the backward direction. The forward digit span started with three numbers that had to be repeated and after each two stimuli one number was added to the sequence. The backwards digit span started with two numbers. A number was added to the sequence after every two stim-uli as well. If a participant failed at two sequences of the same length, the task was terminated.
All sequences were pre-recorded by a female native speaker of Dutch. Results were administered with pencil and paper. Each correctly repeated sequence was worth one point and in the end scores of the forward and backward digit span were com-bined.
5.3 Procedure
The test battery was administered to participants during one session at their schools in a quiet room by one experimenter. The tasks were presented in two different orders. Participants either started out with the SRT task, followed by the RCC task, the digit span and the AGL task or exactly the other way around. This was done for attentional reasons, as the SRT and AGL task took the most time and children easily could get bored if they followed each other directly, but also because a possible learning effect on the SRT task might have influenced a possible learning effect on the AGL task or vice versa. An exit interview for the AGL and SRT task was only administered for one of these tasks, when it was the final task of the session. Ethical approval for this study
was obtained from the University of Amsterdam and informed passive consent was gained from parents or legal guardians of the children.
5.4 Analyses
All analyses were done using R (R Core Team 2015). Repeated measures ANOVA’s were executed to investigate an effect of learning and group differences in the AGL and SRT task. Independent samples t-tests investigated group differences in the RCC and DS task. Partial Pearson’s correlations were used to investigate possible relation-ships between different subtests.
6. Results
6.1 SRT
As said before, the SRT task consisted of 6 blocks of 60 trials. The first 4 blocks and block 6 followed the pattern, whereas block 5 was the random block. RTs were mea-sured from the appearance of a stimulus. As a first step, the raw data were inspected on outliers. Participants would be excluded if they were outliers based on their amount of incorrect presses (>3 SD from the group mean) or their mean RT (> 3SD than the group mean). No participants had to be excluded on these grounds. Further-more, all incorrect responses and responses that followed an incorrect response were excluded from the analysis. Additionally, every response by an individual participant that was an outlier (> 3SD) when compared to the mean RT of that same participant was excluded too. A total of 7.05% of all trials was removed from the data this way.
As the test battery was administered to participants in two orders - they either started their session with the SRT task and finished with the AGL task or vice versa - a first analysis investigated whether this order of tasks during a test session influenced RTs of all six blocks. A two-way repeated-measures ANOVA, with block as the within subjects factor and the order of tasks during a session as the between subjects factor, indicated a significant main effect of the order of the tasks during a session on RTs
(F(1, 288) = 21.402, p < .001 ηp2 = .069), but showed there was no significant inter-action between this order of testing and block number on RTs (F(5, 288) = 0.225, p = 0.952, ηp2 = .004). This indicates that participants who started a test session with the SRT task were significantly faster than participants who ended their session with the SRT task. However, this order did not influence the possible differences in RTs be-tween the different blocks of the test. Results from participants who started with the SRT and the results from those who finished with the SRT could thus be analysed to-gether. These results for the two groups can be seen in Figure 4.
Further analyses investigated whether there was a main effect of block on RTs in the last 3 blocks, whether there was an effect of group on these same RTs and if there was an interaction between the two. Table 1 shows the results from this analysis. A two-way repeated-measures ANOVA, with block as the within subjects factor and group as the between subjects factor, showed the effect of block nearly reached signif-icance (F(2, 144) = 2.985, p = .054, ηp2 = .040), but the medium effect size does indi-cate a tendency. No significant effect for group (F(1, 144) = .427, p = .515, ηp2 = . 003) and also no interaction between group and block was found (F(2, 144) = .084, p = .920, ηp2 = .001). A post-hoc test using the Bonferroni correction revealed that the
difference between block 4 and block 5 nearly reaches significance (p = .055) , but 1 that there is no difference between block 5 and 6 (p = .256).
6.2 AGL
The AGL task consisted of 6 blocks too, but here each block consisted of 24 trials. The first 4 blocks and block 6 again followed the pattern, whereas block 5 was the random block. RTs were measured from the onset of the target syllable. As a first step, the raw data were inspected on outliers. Participants would be excluded if they were outliers based on their amount of incorrect presses (<75%) or if their mean RT dif-fered > 3SD from the group mean. However, no participants had to be excluded on these grounds. Furthermore, all incorrect responses and responses that followed an incorrect response were excluded from the analysis. Additionally, every response by an individual participant that was an outlier (> 3SD) when compared to the mean RT of that same participant was excluded too. A total of 4.4% of all trials was removed from the data this way.
As mentioned before, the test battery was administered to participants in two orders. A first analysis investigated whether this order of tasks during a test session influenced RTs. A two-way repeated-measures ANOVA, with block as the within sub-jects factor and the order of tasks during a session as the between subsub-jects factor,
These results might in some sense be misleading and be due to the removal of outliers. When removing responses
1
from participants that were outliers (> 3SD) when compared to the mean RT of that same participant, 30% of the responses that were removed this way came from the random block. It might be dubious whether it was convenient to remove these items, as the high percentage of outliers in this block can be regarded as a sign of a distortion fect as well. Without these outliers removed, a two-way repeated-measures ANOVA did indicate a significant ef-fect for block type (F(2, 144) = 3.295, p = .040, ηp2 = .044), no effect for group type (F(1, 144) = .346, p = .558,
ηp2 = .002), no interaction between group type and block (F(2, 144) = .098, p = .907, ηp2 = .001) and a post-hoc
test using the Bonferroni correction revealed a significant difference between block 4 and block 5 (p = .038), but
Table 1 – Outcome of ANOVA’s examining gifted-control group differences on the final three blocks for the different tasks that measure RT's
Task F p ηp2
SRT .084 .920 .001
AGL .765 .511 .009
dicated no significant main effect of the order of the tests during a session on RTs (F(1, 288) = .592, p = .442, ηp2 = .002), and also no significant interaction between this order and block number on RTs (F(5, 288) = .232, p = .948, ηp2 = .004). This in-dicates that whether or not a participant started with AGL task did not influence the possible differences in RTs between the different blocks of the test.
Additionally, participants could have either jik or lut as a target syllable and this might have influenced RTs too. A two-way repeated-measures ANOVA with block as a within subjects factor and target syllable as the between subjects factor in-dicated a significant effect for syllable type (F(1, 288) = 6.964, p = .009, ηp2 = .024), participants with lut as a target were faster than participants with jik as a target. How-ever, there was no interaction between syllable type and block (F(5, 288) = 1.080, p = .371, ηp2 = .018) Consequentially, the results from participants who started with the AGL and the results from those who finished with the AGL and participants who had
lut as a target and participants who had jik as a target could be analysed altogether.
Results for the two groups can be seen in Figure 5.
Further analyses investigated whether there was a main effect of block on RTs in the last 3 blocks, whether there was an effect of group on these same RTs and if there was an interaction between the two. Results from these analyses can be found in
