(syllabic stress)
What have we learned about word reading from neuroimaging?
We conclude our review of word reading with a few highlights of what neuroimaging research has added to understanding in this area (see Yeatman, this volume, for comprehensive review).
Two landmark papers in 1988 reported positron emission tomographic (PET) studies in Science (Posner et al., 1988) and Nature (Petersen et al., 1988). From a vantage point years later, the results seem modest. Petersen et al. (1988) concluded the results “favor the idea of separate brain areas…(for) separate visual and auditory coding of words, each with independent access to supramodal articulatory and semantic systems” (p. 585). More interesting for models of word identification was their conclusion that the results argued against “obligatory visual‐to‐auditory recoding.” If we understand “auditory” as phonological, this conclusion was at odds with the behavioral data just starting to emerge around that time.
Later research with fMRI confirmed the role of visual left posterior areas while modifying the earlier conclusion about phonology. The identification of brain networks that connect visual areas to phonological and meaning areas has been a major achievement of cognitive neuroscience. Studies found that increases in reading skill are associated with increased activation in left‐hemisphere areas in both temporal and frontal brain areas (Turkeltaub et al., 2003) and identified the left posterior (occipital‐temporal) region as the site of orthographic processing or the visual word form area (Cohen et al., 2000; McCandliss et al., 2003). Additional areas in the temporal, parietal, and frontal lobes support meaning, memory, and attentional functions. It is the interconnections among specific areas that comprise the multiple subcircuits that make up the larger reading network, as synthesized by Dehaene (2009).
An important question is how this reading network develops. A general answer is that the basic areas – the posterior visual areas and the left hemisphere language areas – get connected through experience in reading. Additionally, built‐in potentials may support this development. Saygin et al. (2016) found that the connectivity pattern within left hemisphere visual areas observed in individual children at age 8 could be predicted by connectivity “fingerprints” that were observed, but not functional at age 5, prior to reading instruction.
A more specific question is the development of the visual word form area, which becomes tuned through experience to respond to word forms (McCandliss et al., 2003). On the neuronal recycling hypothesis (Dehaene, 2009), this adaptation reflects a general principle, that neural circuits originally functioning for one purpose (recognizing objects) are redeployed for another (recognizing words). As to its location in the left hemisphere, a functional perspective suggests an additional consideration – that this allows word form perception to be near to left hemisphere language areas. This possibility was explored by Fiez and colleagues (Moore et al., 2014), who trained adults to associate phonemes with faces, a “face font” that was then used in text reading. Following training, reading the face font produced significant activation in a left hemisphere region close to the visual word form area. This suggests that the left‐hemisphere location of orthographic processing may serve the interconnections between the visual system and left‐lateralized language areas.
Does identifying the brain’s reading network add something to models of word identification and the behavioral data supporting them? Other than required connections between visual and language areas, there is little to constrain the neural implementation of reading processes. Additionally, results from imaging do not automatically align with behavioral results. For example, finding a brain area in an fMRI study that responds more to words than pseudowords reflects the cumulative effects of processing that extends over time intervals that greatly exceed the time course of the processes involved in word identification (though note that MEG can expose these short intervals). However, brain‐behavior model comparisons and theoretical syntheses are helpful, as Taylor et al. (2013) demonstrated. They concluded that both the DRC (Coltheart et al., 2001) and the Triangle PDP model (Plaut et al., 1996) could predict activation patterns during word and pseudoword reading. In fact, all components of the finer‐grain DRC model could be observed in brain data.
Disruptions in the word‐identification system.
Disruptions to the development and operation of visual word identification arise from inadequate orthographic‐phonological knowledge sources and/or the processes that use these knowledge sources, as shown in Figure 1.2. Because visual input initiates word identification, hypotheses about its disruption (in visual processes) emerged in the earliest observations of acquired dyslexia (Pringle‐Morgan, 1896; see Woollams et al., this volume), then developmental dyslexia (Orton, 1925; see Wagner et al., this volume) and in later work inspired by visual neuroscience (Lovegrove et al. 1986; Facoetti et al, 2019). However, it is now well established that the primary causes of disruption lie in the orthographic‐phonological phases of identification.
Figure 1.2 The word‐identification system of the Reading Systems Framework. “Phonological units” rather than “language units” are highlighted to reflect their specific importance in dyslexia.
The link between skilled and disrupted word identification processes is made explicit in dual‐route models, which postulate selective disruption to either the direct (lexical) route or the indirect (sublexical) route to word identity. Castles and Coltheart (1993) established the existence of each type of disruption by testing children’s performance on both irregular, exception words (requiring the lexical route) and pseudowords (requiring the sublexical route). Although a problem with both kinds of words was most common, disassociations between exception word and pseudoword performance appeared for some children. Most showed a phonological dyslexia profile (more difficulty with pseudowords), while others showed a surface dyslexia pattern (more difficulty with exception words). In the Reading Systems Framework, the surface dyslexic is impaired in visual‐orthographic memory, whereas the phonological dyslexic is impaired in the linkage between sublexical orthographic strings and pronunciations.
These two different manifestations of disruption to the word reading system might, however, arise from a single problem in the phonological system. Difficulty reading exception words might be recast not as surface dyslexia but as due to a developmental delay (Manis et al., 1996) – that is, reading experience that is insufficient to acquire high‐quality word representations of exceptionally spelled words. A connectionist model by Seidenberg and McClelland (1989) showed the plausibility of a key assumption: A serious problem in phonological representations can lead to a “deficit” in reading exception words. Other studies – a review by Rack et al. (1992), a critique of visual deficit hypotheses (Vellutino, 1981), demonstrations of phonological processing and memory deficits (Brady & Shankweiler, 1991; Snowling et al., 1986) and a review of acquired dyslexia cases (Ramus, 2003) – added to the persuasiveness of the phonological deficit hypothesis. Imaging results converged to show associations between reading problems and failures to engage left hemisphere language areas (Shaywitz et al., 2004; Simos et al., 2007; Turkeltaub et al., 2003). Although the cause of phonological problems is uncertain, there is evidence that they originate prior to literacy: Among children at risk for dyslexia, preliterate language skills predict their phonological skills and subsequent reading skills (Hulme et al., 2015; Snowling et al., 2003). Moreover, interventions can improve children’s oral language skills and their prospects for reading (Hulme et al., 2020).
Other work has suggested that phonological problems may reflect lower level deficits in, for example, temporal coding in the auditory system (Tallal,
