4.c A tale from the development of language in children

In the early 1970s, 10-year-old children on the Isle of Wight (UK) were part of a study in developmental psychology. The study was intended to measure ‘underachievement’, defined as ‘the ratio between the child’s mental age and his achievement age’ (Rutter and Yule 1975, p. 183). It was common in the 1970s to use IQ tests as a measure of innate ability and a predictor of literacy. Underachievement was identified as a discrepancy between verbal and non-verbal IQ tests and age-appropriate reading and literacy attainment. The distribution of the latter in a population was assumed to be normal, with a Gaussian curve and under- and overachievers at each end.

The Isle of Wight study questioned the reliability of this method, which gave rise to ‘misleading statistics’. It overestimated the number of underachievers in children with high IQ but underestimated it in other cases (Rutter and Yule 1975, p. 183; see also Yule et al. 1974).¹The study went further in comparing groups of 10-year-old children on the Isle of Wight and in London. Comparing data about reading accuracy and comprehension, Rutter and Yule were hoping to find a statistically significant distinction between children with specific reading difficulties and what might be called the ‘garden variety’ of children experiencing reading difficulties. They were looking for a ‘hump’ in the lower-end distribution of learners: a specific group of underachievers with specific reading difficulties.

The ‘hump’ was indeed evident from the data. But, crucially, the study was not able to establish any specific neurological pattern responsible for the ‘hump’ and the group it identified. Children with specific reading difficulties displayed no ‘overt neurological disorder’ and had ‘delays in the development of speech and language . . . no more frequent [than] in those from families of very low social status’ (Rutter and Yule 1975, p. 190, emphasis in original). From the behavioural point of view, the study concluded that, yes, there was a statistically identifiable group of learners within the lower end of the normal distribution; but it was unable to identify any meaningful pattern behind it. The authors declared that there was ‘no evidence for the validity of a single special syndrome of dyslexia. . . . Some kind of biological “marker” would be needed and so far none has been found’ (p. 194).

At the behavioural level, traditionally dyslexia has been identified with an ‘unexpected’ gap between verbal and non-verbal IQ, on the one hand, and reading and literacy skills, on the other. Although there are clearly identifiable symptoms available from Wechsler tests (e.g. low scores on the Digit Span subtest) to help with diagnosis, this discrepancy approach to defining dyslexia has been heavily criticized since the Isle of Wight study (see Elliott and Grigorenko 2014, and Siegel 1992). First, IQ does not correlate with the specific subskills involved in reading and writing (e.g. phonological awareness, or word recognition—see Stanovich 2005). Second, discrepancies of this nature are developmental and tend to change depending on whether the tests take place at the age of, say, 7, 10, or 13. Third, the threshold for the discrepancy to count as ‘unexpected’ has to be set high to identify children with reading difficulties. But in so doing it excludes many children who might also experience reading difficulties.

Reliance on the ‘unexpected discrepancy’ approach highlights also the socioeconomic inequalities in access to early diagnosis and support for dyslexia. As the historian Philip Kirby underlines, ‘[D]yslexia (then as now) was being diagnosed in higher proportions in children from wealthier socio-economic groups. Differential access to dyslexia specialists and their tests was a reason for this, sparking accusations that dyslexia was curiously prevalent in Surrey. . . . Parents with higher educational levels were also more likely to be aware of the condition, and earlier’ (Kirby 2018, p. 58). The 2019 All-Party Parliamentary Group report on dyslexia has underlined the ongoing high costs (over £1,000 extra per year) for families supporting children with dyslexia, which once again point to socioeconomic disparities in the ability to offer timely diagnoses and interventions available to all children (Hodgson 2019).

The Isle of Wight study raised awareness about the necessity for remedial education tailored to the specific needs of different groups of children with reading difficulties (see Vellutino et al. 1996). Later studies by educationalists like Marie Clay (1987) in New Zealand showed how the attainment gap for 7-year-old underachievers could be significantly improved by intensive weekly remedial teaching. And the trend among educationalists continues with some researchers studying what is now called ‘child characteristic-by-instruction (C-I) interactions’ (see Connor 2010). The idea is that ‘the effect of literacy instruction strategies does indeed appear to depend on students’ characteristics’ (Connor 2010, p. 256).

New longitudinal studies and randomized control trials have helped in identifying possible relevant C-I interactions. Students with word reading difficulties benefit from teaching that emphasizes word recognition (see Juel and Minden-Cupp 2000). Moreover, reading skills seem to improve visibly in teacher–child-managed instruction rather than child-managed instruction settings (see Connor et al. 2004b), preferably if the teacher reads to a small group rather than to the entire class (see Connor 2010, p. 259).

While these studies survey ways of catering for the educational needs associated with dyslexia, they have also highlighted a methodological gap in teachers’ training. A recent online survey of teachers in England and Wales revealed that the majority of them (79.5%) mentioned behavioural descriptors and ‘visual stress’ rather than cognitive descriptors such as ‘phonological awareness deficit’ (39.3%). This imbalance suggests ‘a “stereotypical” view of dyslexia’ as mainly attributed to the ‘singular category of the behavioural level’ (see Knight 2017, pp. 216 and 211). Even more problematic is the association with visual stress ‘despite research being inconclusive about this relationship’ (p. 216).

This gap between educational studies and the reality of classroom teaching shows the risks of a one-sided (mostly behavioural) understanding of dyslexia. It may prevent schools from identifying children with special educational needs and disabilities (SEND) at an early age. It may also reinforce socioeconomic inequalities in early diagnosis. The Children and Families Act 2014 in the UK has allowed parents to request funding from local councils to cater to the special needs of their children in specialized private schools. Yet there is still a long way to go in accessing timely diagnoses and interventions for children from socially disadvantaged backgrounds.

Going beyond behavioural and educational-psychological studies, significant research has also been conducted on the neurobiological basis of dyslexia. The failure to identify a ‘single special syndrome of dyslexia’ in the Isle of Wight study did not deter neuroscientists from looking into the possible neurobiological mechanisms behind it. And they suggest today that dyslexia is a neuro-developmental disorder of genetic origin with a neurobiological basis (see Frith 2002b, p. 51).

Neuroimaging studies using data from CT and fMRI scans started in the late 1970s. The goal was to find possible patterns of cerebral asymmetry or symmetry that could be related to language development. These studies (as well as post-mortem anatomical studies) have revealed significant differences in areas of brain activation for patients with and without dyslexia, with the former sometimes showing a more prominent activation of the right side of the brain (see Maisog et al. 2008; Richlan et al. 2009). Other studies using PET scans have related phonological short-term memory tasks to the concerted activation of the relevant areas of the brain (especially Broca’s area, involved in segmented phonology, and the superior temporal and inferior parietal cortex—see Paulesu et al. 1996).

One fMRI study (Olulade et al. 2013) has suggested that dyslexia is associated with a deficit in the magnocellular system, which is involved in human vision and the ability to detect edges, positions, and orientations of objects. This fits with the suggestion that decoding difficulties might be the product of a magnocellular deficit. But this hypothesis is just one among others at the same neurobiological level. Among them, the so-called cerebellar abnormality traces decoding difficulties back to a disconnection between right and left hemispheres.

Yet an explanatory gap inevitably remains between neurobiology and observed behaviour. These studies have enhanced the understanding of the possible neurobiological basis for dyslexia, but no genetic test can be performed as of today to secure an early diagnosis. Nor has a genetic marker for dyslexia been found. Things get more complicated. Not only is there no genotype for dyslexia. There is no phenotype either.

Indeed, the absence of a phenotype was a significant stumbling block in early neuroimaging studies, which often assumed the existence of a ‘dyslexic phenotype’ vis-à-vis a control group. R. H. Haslam et al. (1981) criticized early neuroimaging studies for ‘questionable dyslexic subtyping typologies in examining for possible interactions between subtypes and brain asymmetry’ (Hynd and Semrud-Clikeman 1989, p. 463). Further problems arose from the choice of control groups. Doubts were raised about how representative of the typical population the control groups might be given the available psychometric data (Hynd and Semrud-Clikeman 1989, p. 449).

This was an instance of Simpson’s paradox: namely, in order to find statistically relevant differences in brain morphology, one needs to know already who is dyslexic and who is not to partition the groups correctly. This in turn would require some control over a rather complex and wide-ranging set of behavioural, neurological, and cognitive variables that might lead to reliably identifying a prototypical control group. However, the authors of the study concluded, often ‘one must accept on faith the notion that these control subjects were indeed free of other behavioural, neurological or psychiatric disorders’ (Hynd and Semrud-Clikeman 1989, p. 449).

In addition to behavioural descriptions, educational studies, and neurobiological research, environment and culture play their role in understanding dyslexia. Learning how to read and write is an artificial skill that human beings acquire over time, across different cultures and languages, with huge variations among them. It takes years for any child to master. Developmental psychologists have been studying the stages in this process as a way of pinning down key junctures at which setbacks might take place. Expecting to understand dyslexia in light of a single one-size-fits-all approach cannot do justice to the great variety of cases.

Uta Frith at the University College London Institute of Cognitive Neurosciences has been a pioneer in the study of dyslexia and other neurodevelopmental disorders such as autism. Very early on, at a time when dyslexia was still being studied and understood primarily in terms of an information processing model—how fast the brain can process, store, and retrieve information concerning how letters represent phonemes—she pointed out the need to pay more attention to developmental change. She has identified four developmental stages—symbolic, logographic, alphabetic, and orthographic—that need to be mastered for a child to become fully literate (Frith 1986).

At the symbolic and then logographic stage, children acquire the ability to recognize symbols and then words on the basis of some salient graphic feature: for example, a child might be able to recognize the word ‘McDonald’s’ from the yellow M symbol. At the alphabetic stage, the child goes beyond symbols, associates letters with sounds, and blends sounds into words. This is the most demanding stage, which usually children acquire over a period of time at the start of primary school. And the degree of automaticity and fluency in blending sounds varies considerably from one natural language to another.

In transparent languages like Italian, for example, where the association between letter and sound is fairly stable and there is not much variability in the pronunciation of the same sounds, children on average acquire this skill by the end of primary one. But in a non-transparent language like English, where the same letter (say, the letter a) is associated with different sounds depending on the word it is in (think of the sound a in the two words: nature vs natural), acquiring such a skill takes on average two years (longer in the case of dyslexia, see Frith 2002a).

This has led to some statistically surprising results as the prevalence of dyslexia (measured in behavioural terms by the aforementioned ‘unexpected discrepancy’) has been estimated to be half in Italy what it is in the United States, for example (see Lindgren et al. 1985). Studies of Italian-speaking and English-speaking dyslexics with carefully chosen control groups—sharing the same age, levels of tertiary education and so on—have revealed that the former tend to perform better than the latter in tests, even if they perform in fact as the English-speaking ones compared with their respective control groups. At the neurological level, PET scans reveal similarly reduced areas of activation for the language-related left hemisphere (Broca’s area and Wernicke’s area) in both the Italian-speaking and the English-speaking groups (see Paulesu et al. 2001). Studies like this have corroborated the view concerning the common neurobiological basis of dyslexia while also drawing attention to the remarkable variability in its manifestation across different natural languages (and associated degrees of compensatory strategies available in each one).

In the final orthographic stage, the child instantly recognizes morphemic parts of words or the whole word without the need for blending. The child who experiences difficulties with phonological awareness and sound blending at the alphabetic stage may adapt and compensate for the deficit by overdeveloping something similar to the orthographic strategy (e.g. guessing words from the initial morphemes). Children with dysgraphia, on the other hand, tend to master the alphabetic phase and produce accurate blending but find the orthographic phase more challenging. A supporting strategy that might work for dyslexia might therefore not work for dysgraphia. And a remedial learning strategy for Italian-speaking children may not necessarily work for English-speaking children.

The debate on the nature and definition of dyslexia remains highly contentious among specialists, parents, teachers, and policymakers (see, e.g., Elliott and Grigorenko 2014, Ch. 1).² Dyslexia is a life-long condition with early onset in pre-school or school years. There have been historical difficulties with defining dyslexia, with the term often used as an umbrella to describe a range of symptoms from learning disability to specific reading difficulties in relation to fluency, automaticity, and spelling accuracy. The DSM-5 (Diagnostic and Statistical Manual of Mental Disorders), for example, treats dyslexia as part of a larger family of ‘specific learning disabilities’ (SLD).

A study for the British Dyslexia Association (Crisfield 1996) estimated that up to 10% of the population might have symptoms of dyslexia, ranging from mild to severe; while the National Institute of Child Health and Development (2007) estimated that up to 20% of the US population has some kind of language-based disability (for this and other statistics, see Elliott and Grigorenko 2014, p. 32). The figure of 20% appeared also in the Connecticut Longitudinal Study of children from kindergarten to secondary school run by Sally Shaywitz and colleagues at the Yale Centre for the Study of Learning and Attention (Shaywitz et al. 1999). And the aforementioned 2019 All-Party Parliamentary Group report (Hodgson 2019) gives a figure of 10%–15% for the UK population (i.e. affecting an estimated 6.6 to 9.9 million people, including up to 1.3 million of young people in education).

The absence of a clear cut-off point in these statistics shows the complexity of understanding the multifactorial nature of dyslexia and the challenge of timely diagnosis and effective educational interventions. A child undiagnosed during primary school—maybe because the symptoms are read as ‘laziness’ or ‘daydreaming’ or ‘inattentiveness’—is likely to be adversely affected in secondary and higher education. This is a reminder of the importance of securing timely diagnosis, and school support for these children and their educational needs.

The philosophically interesting question on which cognitive psychologists, educationalists, neurobiologists, and developmental psychologists focus today is not, then, whether there is dyslexia,³ or ‘what dyslexia really is’ (Morton 2004, p. 162, emphasis in original). As the 2012 Dyslexia Action Report says of the condition, ‘[T]here is no longer controversy about whether it exists and how to define it’ (Dyslexia Action 2012, p. 7). The debate is on how to identify the symptoms for individual children, offer timely diagnoses, and put in place effective interventions so as to give better educational prospects.

Cognitive psychologists, educationalists, neurobiologists, and developmental psychologists face the need to understand the behavioural phenomenon of ‘difficulties with reading’.⁴

That a child might experience difficulties with learning how to read is relatively easy to spot. Much more difficult is to ascertain whether the difficulties with reading are the tail end of a normal distribution or the symptom of a life-long condition such as dyslexia. A number of data-to-phenomena inferences are required to tease out these very different conclusions. These data-to-phenomena inferences are perspectival in the same ways exemplified by my other case studies: (1) the data in each case are sourced from experimental, theoretical, and technological resources available to distinct epistemic communities to reliably advance their knowledge claims; and (2) the methodological-epistemic principles at play to justify the reliability of their knowledge claims also pertain to distinct epistemic communities.

For example, the educationalists’ data may include reading and comprehension tests from sampled pupils of different ages and geographical locations. These statistical data may be used to identify school attainment gaps and underperformances in the student population (including the phenomenon of ‘difficulties with reading’). Educationalists use resources at their disposal to monitor (often through longitudinal studies) how effective particular remedial strategies might be (e.g. teacher–child-managed instructions).

Neurobiologists use data from CT/fMRI/PET scans, as well as post-mortem anatomical studies, as evidence for a range of other phenomena such as symmetry/asymmetry in brain morphology, lesions in the brain, possible abnormalities in the cerebellum, or possible magnocellular dysfunction. This neurobiological evidence is in turn used to infer the possible comorbidity of the phenomenon of difficulties with reading with other phenomena such as difficulties with motor development, or difficulties with motion detection. ‘Difficulties with reading’ in this case is part of a wider spectrum of co-occurring phenomena for which a neurobiological basis is sought. Being able to tease out these data-to-phenomena inferences is diagnostically important to help children whose difficulties with reading might be downstream from slow processing speed, or may be a consequence of ADHD, for example.

Developmental and cognitive psychologists use data from cognitive tests (e.g. slow naming speed, difficulties in letter–sound decoding) as evidence for a phonological deficit in the ability to associate phonemes with graphemes. The consensus view these days is that dyslexia has to do with some kind of phonological deficit,⁵ namely a defect in the representation of speech sounds which leads to difficulties with phonological awareness, slow naming speed, difficulties with letter–sound decoding, and hence non-fluent reading and difficulties with spelling.

The phenomenon ‘difficulties with reading’ in this case is the behavioural manifestation of a developmental-cognitive problem concerning the representation of phonemes and the ability to segment and blend them. This perspectival data-to-phenomena inference has been very important, among other things, in establishing the most effective pedagogical method for literacy in the so-called reading wars—whether it is a phonic approach (learning one letter–sound at a time, as is now believed to be the preferred method) or a whole-language approach (see Connor et al. 2004a).⁶

‘Difficulties with reading’ is, then, what I call a ‘modally robust’ phenomenon. For it robustly can happen in very many different ways. And, typically, it is the job of different epistemic communities to explore the network of perspectival inferences from specific data to the correct diagnostic profile in each case. Let us briefly take a closer look, first, at the semantic nature of those inferences, and, second, at the perspectival modelling that enables them.

Perspectival data-to-phenomena inferences have to be reliable to advance bona fide knowledge claims (rather than spurious claims). What is to be said about these knowledge claims? Consider, for example, the following claim:

(i) If a child has difficulties with reading, then they are dyslexic.

This claim rests on an indicative conditional ‘if . . . then’ with present tense in the antecedent and consequent. It is clear from the discussion so far that an unqualified claim of this nature is useless in diagnosing children with dyslexia from their non-dyslexic peers who might still experience difficulties with reading. As the Isle of Wight study revealed, behavioural data about reading accuracy and comprehension are not—in and of themselves—unequivocally reliable evidence for dyslexia. A more reliable diagnosis depends on how one understands the antecedent of this indicative conditional.

This in turn involves uncovering a number of additional perspectival data-to-phenomena inferences behind the phenomenon ‘difficulties with reading’. These inferences may again take the form of further indicative conditionals where the phenomenon features this time in the consequent. Here are two examples:

(ii) If a child experiences difficulties with schooling, they will have difficulties with reading.

(iii) If a child has a phonological deficit, they will have difficulties with reading.

Reliably diagnosing children with dyslexia among children without dyslexia who also experience learning difficulties depends on teasing apart (iii) from (ii). But even after screening to rule out difficulties with schooling as a potential cause, the reliability of the diagnosis depends on telling apart more indicative conditionals, such as the following:

(iv) If a child has a phonological deficit and an attention deficit, they will have difficulties with reading and with planning.

(v) If a child has a timing/sequence deficit, they will have a phonological deficit and a motor control deficit, and as a result difficulties with reading and with balance.

(vi) If a child has slow temporal processing, they will have a visual deficit as well as a phonological deficit, and as a result difficulties with reading and with motion detection.

(vii) If a child has a cerebellar abnormality, they will have a timing/sequence deficit, and as a result a phonological deficit and a motor control deficit with difficulties with reading and with balance.

These knowledge claims in the dress of indicative conditionals belong to different scientific perspectives. They are advanced by communities as diverse as cognitive psychologists (iv), vis-à-vis neuroscientists (e.g. vii). Each community relies on its own experimental, theoretical, and technological resources to source the relevant data and to reliably make these claims. Moreover, each community uses second-order (methodological-epistemic) principles to justify their reliability.

For example, from the neurobiological perspective, scans showing anomalies in brain morphology can be used as evidence for inferring the presence of both a visual deficit and a phonological deficit. From this perspective, the phenomenon of difficulties with reading goes hand in hand with others: say, difficulties with motion detection as in (vi). This could form the basis for a possible diagnosis of slow processing speed, for example. The validity of the diagnosis depends on the reliability of the relevant data-to-phenomena inference, which is in turn justified by methodological-epistemic principles adopted in neurobiology. Among them: that in screening brain images the relevant (non-biased) control group has been correctly identified; that there are functionally relevant pathways from the brain to the relevant behavioural phenomena; that there are ‘dyslexia candidate susceptibility genes’ (Fisher and Francks 2006) implicated in the relevant neurobiology. Each of these methodological-epistemic principles can of course be challenged and are typically called into question as new evidence and new studies come to the fore. Indeed, there are communities within communities where, for example, colleagues performing fMRI imaging do not typically make any assumption about possible ‘susceptibility genes’.

Consider now the perspective of cognitive psychologists, who use data from cognitive tests and reading tests as evidence for inferring specific reading difficulties. The reliability of the inference and associated knowledge claims is also in this case justified by methodological-epistemic principles internal to the discipline. One of these, as already mentioned, is, for example, the IQ-achievement test often used as an indicator of learning potential, with unexpected discrepancy from it being used as a diagnostic tool.

Sometimes the phenomenon inferred is the same (difficulties with reading in my example). But the perspectival nature of the inferences means that the phenomenon in question is each time differently located in a space of possibilities. Sometimes it is comorbid with other phenomena and symptomatic of a broader phenomenology (e.g. slow processing speed). At other times, it is continuous with garden variety reading difficulties that call for remedial teaching strategies of wider benefit for larger portions of children, as Marie Clay (1987) originally argued for.

It is in this specific sense that the modally robust phenomenon of ‘difficulties with reading’ lies at the intersection of a plurality of scientific perspectives. Evidence for it does not accrue by accumulation of more data of the same type(more reading tests, or more brain images). Nor is the phenomenon the manifestation of some hidden dispositional property. The reliability of the inference from data to the phenomenon cannot be justified by appealing to a genetic marker in neurobiology. Nor can it be justified by generically invoking underachievement, because there is no prototypical phenotype either.

This does not make the phenomenon any less real. On the contrary. It is very much real and can happen in a variety of different ways. But the reliability of the knowledge claims advanced through data-to-phenomena inferences within each scientific perspective needs to be cross-checked and cross-validated. This is usually done by bringing one perspective to bear on the other and vice versa. For example, one can bring the cognitive perspective to bear on the neurobiological one and the educationalist perspective to bear on the cognitive one. Moreover, one needs to take into account a number of other considerations about the environment and the transparent/non-transparent nature of the language in question.

This is perspectival pluralism in action. The pluralism of scientific perspectives is not just a desirable methodological feature of science. It is not just a way of offering a menu of different explanations for the same phenomenon. It is a way of checking the reliability of each data-to-phenomena inference within its own perspective in light of other scientific perspectives. Only in this way can key justificatory principles of each perspective be monitored, cross-checked, and held accountable.

No scientific perspective can sanction the reliability of its own inferences alone. One needs to ask: reliable with respect to what and to what degree? What is it that researchers are trying to achieve each in their own scientific perspective? And how successful are they in their inferences? Perspectival pluralism is required to improve the open-ended network of inferences (i)–(vii) for a variety of purposes (diagnostic, educational, screening, etc.). Most importantly, perspectival pluralism is required to maintain checks on the justificatory (methodological-epistemic) principles of each scientific perspective. In so doing, it allows different epistemic communities to have a debate about dyslexia.

Being located in a space of possibilities means that there is an element of contingency in what the data may reliably provide evidence for each time. It is not a necessary truth that if someone has a phonological deficit, they will also display difficulties with reading. They may in fact successfully develop compensatory strategies during the early years, maybe thanks to timely interventions and appropriate teaching support, or thanks to the transparent nature of their native language. Therefore the consequents of the indicative conditionals are best read as hiding a modal verb:

(iii*) If a child has a phonological deficit, they may have difficulties with reading.

(iv*) If a child has a phonological deficit and an attention deficit, they may have difficulties with reading and with planning.

I will have more to say about the semantic nature of these conditionals in Chapter 5. But, next, I want to illustrate how perspectival modelling understood as exploring the space of possibilities finds a natural expression in what is known as developmental contingency modelling for dyslexia, championed by Uta Frith and John Morton.

The Developmental Contingency Modelling (DCM) of Uta Frith and John Morton (Morton 1986, 2004, Ch. 8; Morton and Frith 1995) perfectly illustrates the cross-perspectival process of refining the reliability of knowledge claims in perspectival modelling. In continuity with the other two case studies, the term ‘perspectival modelling’ is again used in a broad sense and not to refer exclusively to the causal models within DCM. Or better, the individual causal models within DCM are an integral part of how the intersecting scientific perspectives of the educationalists, developmental psychologists and neurobiologists jointly deliver modal knowledge of the relevant phenomena over time. What is unique and particularly interesting about DCM is that it builds in enough modularity and contingencies⁷ to allow a variety of researchers—neurobiologists, cognitive psychologists, educationalists, among others—to differentiate similar learning difficulties by tracing and retracing them back to specific contingent points where breakdown might have occurred. It includes two main components: (1) a number of distinctive levels (biological, cognitive, behavioural, and environmental); and (2) a number of distinctive temporal stages in the acquisition of reading and writing skills.

In the words of Uta Frith:

A great challenge for cognitive theories is that they have to explain the diversity of dyslexia as it manifests itself in different people. Most cognitive theories are not designed to cope with individual variation. They address the prototypical case instead. The behaviour patterns characteristic of the prototypical case are distilled from many individual cases, and it is this distilled information that is usually the target of explanation. (Frith 2002b, p. 53)

DCM accommodates a plurality of causal models that aim to disentangle similar phenomena at the behavioural level and explore the variety of contingent pathways that might be at play in each case, as the following rival causal models show (Figures 4.c.1, 4.c.2, 4.c.3, and 4.c.4):

Prima facie similar behavioural phenomena (e.g. difficulties with phonological awareness and in general difficulties with reading) can hide diverse cognitive deficits. In some cases, the phonological deficit that is primarily responsible for difficulties with reading is the direct consequence of some neurobiological anomaly (like the left-hemisphere disconnection in Figure 4.c.3). In other cases, the phonological deficit is the joint effect of a common cause such as slow temporal processing in Figure 4.c.4 that manifests itself in a number of other symptoms such as visual deficit and difficulties with motion detection. Or a timing/sequence deficit, as in Figure 4.c.1, that results in motor control deficit and difficulties with motor balance. Once again, the search for a one-size-fits-all phenotype would be misguided.

As Frith and Morton have long been arguing, dyslexia is about individuals. Effective remedial strategies should address specific individual needs. The dyslexic child who also has an attention deficit problem (inattentive ADHD) needs learning support strategies different from those of a child with slow processing speed. Identifying the possible pathways within the developmental framework is key to go from ‘won’t read’ to ‘can’t read’ (because of visual deficit, motor control deficit, slow processing speed, . . .). And this in turn would allow appropriate educational support, which is necessary to transform ‘can’t read’ into ‘can read’.

Causal models within DCM take different forms (X-shape, V-shape, A-shape) depending on the number of factors and their causal relations across the three levels. In every case, a single causal nexus is implied. This can be at the cognitive level with multiple causes and multiple behavioural manifestations (X-shape, e.g. dyslexia). It can be at the behavioural level with multiple causes (V shape). It can be at the brain level, with a single known cause and a variety of behavioural consequences (A-shape, e.g. a rare single-gene defect). Thinking in terms of these shapes for different conditions allows practitioners to identify robust (almost lawlike) dependencies among relevant features of the phenomena of interest across different levels. For example, the ability to decode letter strings causally depends on both unimpaired vision that allows the child to discriminate visual features of letters at the alphabetic stage and the phonological ability to sequence phonemes in a particular order.

In turn, the phonological ability that is critical to identifying and sequencing phonemes may also causally determine the speed of object naming. Object naming tests are thought to be sensitive tests for diagnosing dyslexia, regardless of impairments in other phonological tests. Likewise, there might be children who do not have any problem with object naming and yet experience difficulties with decoding skills. All else being equal, causal graphs like Figure 4.c.5 allow practitioners to conclude (e.g. via object naming tests) that the setback might be at the alphabetic level of knowledge of the letters (maybe because of some vision deficit or delay in the transition from the logographic to the alphabetic phase) and ‘tell the difference between a child who cannot decode simply because letter knowledge is absent and a child who lacks the requisite phonological skills’ (Morton and Frith 1995, p. 378).

Let us draw some philosophical conclusions. DCM is a perfect illustration of what I call ‘perspectival modelling’ in that the model pluralism here at stake is exploratory. It enables a particular kind of inferential reasoning necessary to explore how learning to read and write might be affected by a number of contingent setbacks during the early years. Charting this space of possibilities and being able to correctly locate the modally robust phenomenon of ‘difficulties with reading’ is key to understanding all the possible routes through which a child becomes fully literate, and the adaptive strategies for each possible setback. Thus, the causal models are perspectival in that they chart the possible routes through which language skills can be learnt and relearnt over a period of time in response to a variety of conceivable neurobiological, cognitive, or environmental stumbling-blocks. Perspectival modelling in this case is pivotal to effective diagnostics and educational interventions tailored to the specific needs of individual children. Because every child is unique, so is every developmental setback.