Frost proposes two principles for universal reading models: They must apply to any writing system and accommodate findings related to all linguistic dimensions. These correspond to a central theme in developmental theory: Development is intimately bound with experience, particularly cultural experience.
In Frost's view, neurocognitive and cultural constraints shape languages to idiosyncratically maximize the information packed into each word. Complementarily, individuals' brains learn to extract information efficiently from their linguistic environment. While we strongly agree with this claim, our attention was drawn to Frost's contention that reading Hebrew is fundamentally about identifying morphemic roots, and that “the other letters of the word do not serve for lexical access, at least not initially” (sect. 4.1). Those other letters comprise the root's sibling in morphological decomposition: the word pattern (WP). We believe it is important to recognize WPs as a source of information, processed alongside their big brother, the root.
Frost suggests that the WP morpheme's primary role in visual word identification is to “flash out” (sect. 5, para. 4) the all-important root, due to the skewed frequency distribution of WP letters. In fact, as he has acknowledged elsewhere (Frost et al. Reference Frost, Kugler, Deutsch and Forster2005), the Semitic WP can independently pack a substantial volume of information – pronunciation, part of speech, person, number, gender, tense, voice, and reflexivity. Far from merely flashing out a root, word-pattern morphemes flesh out the meaning of a word.
Analyses of Hebrew print generally distinguish three groups of morphemes: some 3,000 roots, about 100 nominal WPs, and about 100 verbal WPs, the latter grouped into seven “constructions” (Shimron Reference Shimron2006). These distributions lead Frost to argue statistical invisibility for the WP. However, the forms are distinguished not only by number and linguistic class. They lie along a continuum of information volume. At one extreme, roots anchor mutually exclusive morphemic families. At the other, most nominal WPs carry little more information than “this word, pronounced this way, is a noun,” occasionally signaling narrower semantic categories, such as “professional,” “craftsman,” or “event.” Somewhere in the middle fall verbal WPs.
This notion of a continuum of information volume, from roots to verbal patterns to noun patterns, is not inconsistent with Frost's research. In a series of priming experiments, Frost and colleagues have demonstrated reaction-time and fixation-duration facilitation by shared roots (Deutsch et al. Reference Deutsch, Frost, Peleg, Pollatsek and Rayner2003; Frost et al. Reference Frost, Forster and Deutsch1997) or verbal word patterns (Deutsch et al. Reference Deutsch, Frost and Forster1998; Reference Deutsch, Frost, Pollatsek and Rayner2005). Although facilitation by nominal word patterns has not reached statistical significance (Deutsch et al. Reference Deutsch, Frost and Forster1998;), Frost et al. (Reference Frost, Kugler, Deutsch and Forster2005) found consistent benefits for shared nominal word patterns; these were especially large – up to 22 msec – in pseudo-words, where noun patterns, the sole information-bearing morphemes, reduced reaction time by 12 msec more than did identity primes.
Since roots and verbal patterns carry large volumes of information, it makes sense that priming them significantly shortens fixations and reaction time. And since nominal patterns carry little information, it makes sense that they would not produce as strong an effect. However, these findings do not imply that nominal WPs are unimportant in word identification; their contribution may be too small or too early to affect behavioral measures.
Electrophysiological research has found indexes of multiple types of information bound to single written words within 200 msec of exposure (Pulvermüller et al. Reference Pulvermüller, Shtyrov and Hauk2009), including morphological analysis as early as 180 msec (Kellenbach et al. Reference Kellenbach, Wijers, Hovius, Mulder and Multer2002). Pulvermüller and colleagues make the case for near-simultaneous processing of multiple, overlapping information components, a case strengthened with respect to morphemes by Kuperman et al. (Reference Kuperman, Schreuder, Bertram and Baayen2009). Event-related potentials (ERPs), a measure at the temporal scale of eye-tracking but with access to hidden neural events, might provide evidence of the role of nominal WPs in Semitic word identification.
Such findings would support the general principle that readers learn to use every scrap of information provided by written language. Evidence of word identification guided by the volumes of information in cognitively defined segments of a visual stimulus, whether single or multiple-letter, consecutive or interdigitated, would go beyond Frost's demonstration that the extraction of information from written words is not universally controlled at the level of letter identity. It would suggest that extraction tactics may not map to categories like orthography, morphology, phonology, and semantics at all, nor to theoretical timetables distinguishing pre-lexical perception from lexical cognition, but emerge in response to both linguistic and non-linguistic qualities of the stimuli, including information volume or real-time usefulness (Miller Reference Miller2012). For instance, Holcomb and colleagues, using ERPs to study repetition and associative priming, proposed that differences might derive not from the linguistic categories, form, and meaning, but from the processes evoked by relationships between neural representations of prime and target: activation of shared representations in the first case, connection across related representations in the second (Holcomb & Grainger Reference Holcomb and Grainger2009; Holcomb et al. Reference Holcomb, Reder, Misra and Grainger2005).
If minds extract all they can from every word, morpheme, and letter, guided by many characteristics of the segmented stimulus, then the role of formal patterns in reading Hebrew goes beyond probabilistic invisibility. In the same vein, developmental research indicates that experiences which enable reading extend far beyond the statistical learning invoked by Frost. Readers acquire cognitive tools through trial, error, and cumulative familiarity, to be sure, but also with adult guidance, during powerful one-trial events, and by forming their own insightful associations (Homer Reference Homer, Olson and Torrance2009). A universal learning model of reading will incorporate all these opportunities.
Frost does great service in going beyond models based rigidly on letter identity and position. He points toward a model that respects the phylogenetic capacity to devise symbolic systems that extend memory into non-biological space, as well as the ontogenetic capacity to develop culturally specific schemata that recapture embedded information. Our interpretation of the research suggests the need for an even more capacious explanation of how we develop and apply those schemata – associating each visual stimulus with prior knowledge, seeking and extracting the information available in each word.
You have
Access
Frost proposes two principles for universal reading models: They must apply to any writing system and accommodate findings related to all linguistic dimensions. These correspond to a central theme in developmental theory: Development is intimately bound with experience, particularly cultural experience.
In Frost's view, neurocognitive and cultural constraints shape languages to idiosyncratically maximize the information packed into each word. Complementarily, individuals' brains learn to extract information efficiently from their linguistic environment. While we strongly agree with this claim, our attention was drawn to Frost's contention that reading Hebrew is fundamentally about identifying morphemic roots, and that “the other letters of the word do not serve for lexical access, at least not initially” (sect. 4.1). Those other letters comprise the root's sibling in morphological decomposition: the word pattern (WP). We believe it is important to recognize WPs as a source of information, processed alongside their big brother, the root.
Frost suggests that the WP morpheme's primary role in visual word identification is to “flash out” (sect. 5, para. 4) the all-important root, due to the skewed frequency distribution of WP letters. In fact, as he has acknowledged elsewhere (Frost et al. Reference Frost, Kugler, Deutsch and Forster2005), the Semitic WP can independently pack a substantial volume of information – pronunciation, part of speech, person, number, gender, tense, voice, and reflexivity. Far from merely flashing out a root, word-pattern morphemes flesh out the meaning of a word.
Analyses of Hebrew print generally distinguish three groups of morphemes: some 3,000 roots, about 100 nominal WPs, and about 100 verbal WPs, the latter grouped into seven “constructions” (Shimron Reference Shimron2006). These distributions lead Frost to argue statistical invisibility for the WP. However, the forms are distinguished not only by number and linguistic class. They lie along a continuum of information volume. At one extreme, roots anchor mutually exclusive morphemic families. At the other, most nominal WPs carry little more information than “this word, pronounced this way, is a noun,” occasionally signaling narrower semantic categories, such as “professional,” “craftsman,” or “event.” Somewhere in the middle fall verbal WPs.
This notion of a continuum of information volume, from roots to verbal patterns to noun patterns, is not inconsistent with Frost's research. In a series of priming experiments, Frost and colleagues have demonstrated reaction-time and fixation-duration facilitation by shared roots (Deutsch et al. Reference Deutsch, Frost, Peleg, Pollatsek and Rayner2003; Frost et al. Reference Frost, Forster and Deutsch1997) or verbal word patterns (Deutsch et al. Reference Deutsch, Frost and Forster1998; Reference Deutsch, Frost, Pollatsek and Rayner2005). Although facilitation by nominal word patterns has not reached statistical significance (Deutsch et al. Reference Deutsch, Frost and Forster1998;), Frost et al. (Reference Frost, Kugler, Deutsch and Forster2005) found consistent benefits for shared nominal word patterns; these were especially large – up to 22 msec – in pseudo-words, where noun patterns, the sole information-bearing morphemes, reduced reaction time by 12 msec more than did identity primes.
Since roots and verbal patterns carry large volumes of information, it makes sense that priming them significantly shortens fixations and reaction time. And since nominal patterns carry little information, it makes sense that they would not produce as strong an effect. However, these findings do not imply that nominal WPs are unimportant in word identification; their contribution may be too small or too early to affect behavioral measures.
Electrophysiological research has found indexes of multiple types of information bound to single written words within 200 msec of exposure (Pulvermüller et al. Reference Pulvermüller, Shtyrov and Hauk2009), including morphological analysis as early as 180 msec (Kellenbach et al. Reference Kellenbach, Wijers, Hovius, Mulder and Multer2002). Pulvermüller and colleagues make the case for near-simultaneous processing of multiple, overlapping information components, a case strengthened with respect to morphemes by Kuperman et al. (Reference Kuperman, Schreuder, Bertram and Baayen2009). Event-related potentials (ERPs), a measure at the temporal scale of eye-tracking but with access to hidden neural events, might provide evidence of the role of nominal WPs in Semitic word identification.
Such findings would support the general principle that readers learn to use every scrap of information provided by written language. Evidence of word identification guided by the volumes of information in cognitively defined segments of a visual stimulus, whether single or multiple-letter, consecutive or interdigitated, would go beyond Frost's demonstration that the extraction of information from written words is not universally controlled at the level of letter identity. It would suggest that extraction tactics may not map to categories like orthography, morphology, phonology, and semantics at all, nor to theoretical timetables distinguishing pre-lexical perception from lexical cognition, but emerge in response to both linguistic and non-linguistic qualities of the stimuli, including information volume or real-time usefulness (Miller Reference Miller2012). For instance, Holcomb and colleagues, using ERPs to study repetition and associative priming, proposed that differences might derive not from the linguistic categories, form, and meaning, but from the processes evoked by relationships between neural representations of prime and target: activation of shared representations in the first case, connection across related representations in the second (Holcomb & Grainger Reference Holcomb and Grainger2009; Holcomb et al. Reference Holcomb, Reder, Misra and Grainger2005).
If minds extract all they can from every word, morpheme, and letter, guided by many characteristics of the segmented stimulus, then the role of formal patterns in reading Hebrew goes beyond probabilistic invisibility. In the same vein, developmental research indicates that experiences which enable reading extend far beyond the statistical learning invoked by Frost. Readers acquire cognitive tools through trial, error, and cumulative familiarity, to be sure, but also with adult guidance, during powerful one-trial events, and by forming their own insightful associations (Homer Reference Homer, Olson and Torrance2009). A universal learning model of reading will incorporate all these opportunities.
Frost does great service in going beyond models based rigidly on letter identity and position. He points toward a model that respects the phylogenetic capacity to devise symbolic systems that extend memory into non-biological space, as well as the ontogenetic capacity to develop culturally specific schemata that recapture embedded information. Our interpretation of the research suggests the need for an even more capacious explanation of how we develop and apply those schemata – associating each visual stimulus with prior knowledge, seeking and extracting the information available in each word.