Ram Frost's article is a valuable contribution to the debate about models of the reading process. His suggestion that different kinds of language/script pairing require different processing strategies, and that the strategies actually used differ as predicted, is plausible. And including structurally different languages is a commendable advance on earlier, unduly Eurocentric theorizing about reading.
Nevertheless, I would question several points. These all relate to Frost's claim that “the main goal of reading research is to develop theories that describe the fundamental and invariant phenomena of reading across orthographies” (target article, Abstract, original emphasis). Surely, given the facts about the five languages as Frost presents them, one would expect diverse orthographies to require diverse reading systems. Further, one might also wonder whether Japanese (or Chinese) store words the same way English speakers do. Given that kanji are rich visual images, visual recognition of Japanese (or Chinese) might result in a mapping to an inner representation that is very different from that of English. Frost acknowledges that some representation/strategy might be optimal in one dimension but not in another, which results in a large space of possible mappings. At the very least, in view of the data presented, one might expect the assumption of invariant reading universals to be defended. Yet, while Frost clearly believes that such universals exist, he never tells us why he holds this belief. One could argue that because reading and orthography are phenomena that depend on language, reading universals could be derived from language universals of the kind proposed by Chomsky (Reference Chomsky1965; Reference Chomsky1995). But in recent years the existence of language universals has been challenged (e.g., Evans & Levinson Reference Evans and Levinson2009; Everett Reference Everett2005). The debate on the issue is far from over, and Frost provides no independent reasons to assume there are any significant, specifically reading universals. Universals of reading need to be argued for rather than simply assumed.
Frost's detailed discussion of the role of letter position in English and Hebrew provides support for his critique of models that focus on a single dimension of processing. However, given the variation that Frost finds among only five languages, it is at least questionable whether the aim of “outlin[ing] the blueprint principles for a universal model of reading” (sect. 6, para. 7) is realistic. And, if one aims at such a universal model, it seems misguided to focus on differences between English and Hebrew (and the differences within Hebrew), which seem to suggest different processing mechanisms. Instead of arguing for either letter-position rigidity or flexibility, one might argue that these languages are located on (opposite ends of) a spectrum. It is probably true that the described strong priming effect holds for Frost's examples (e.g., SANDWICH/SNAWDCIH; sect. 6, para. 3) and many other polysyllabic English words. However, it hardly seems that “anything goes” (i.e., complete letter-position flexibility): Would WASHDINC still be recognized as SANDWICH? And, a badly scrambled UNDERDOG turns into GROUNDED with several indeterminate intermediaries.
Possibly, then, letter-position flexibility in English is a superficial phenomenon that masks the rigidity of some letters in the target; posing processing demands that are not fundamentally different from Hebrew. But this is only one aspect of the task a truly universal reading model has to account for. Whether it is possible to develop learning models that can satisfy Frost's universality requirement remains to be seen. Different models might be needed to explain the facts specific to reading in different languages. Recent cross-linguistic modeling work on word segmentation has shown that models that performed well for English failed miserably for Sesotho (Blanchard et al. Reference Blanchard, Heinz and Golinkoff2010). And, anyone attempting to reach Frost's ambitious goal of “full linguistic coherence” (sect. 7, para. 2) would also need to consider cognitive limitations that may be in place when children first learn to read, and the interactions between the different modalities (visual, auditory) that are involved in statistical learning (for recent findings on such interactions, see Emberson et al. Reference Emberson, Convey and Christiuancen2011). Models that can account for all aspects of the reading task are beyond the reach of available technology. Moreover, given the lack of evidence supporting the notion that significant universals exist specifically for reading, it would make more sense to start without preconceptions and generalize from a diverse set of specific cases. Postulating untestable universals at this stage seems at best to be premature. Frost is correct, of course, to insist that models need to be informed by facts from a wide range of reading phenomena and that the scope of any model needs to be clearly acknowledged.
Turning to the most serious problem with Frost's article: The claim that “every language gets the writing system it deserves” (sect. 3) is too strong, as it implies that scripts have evolved to be “perfectly” optimal for their respective spoken languages. I did not find convincing justification for the suggestion that all writing systems necessarily develop “optimization aimed at providing their readers with maximal phonological and semantic information by the use of minimal orthographic units for processing” (sect. 3.1, para. 1, emphasis in the original). Certainly, arbitrary historical facts about which scripts have been available when a language was first reduced to writing have also been important. Frost's discussion of Chinese and Japanese only hints at the complex interaction of arbitrary components (e.g., adoption of Chinese kanji by Japanese) and non-arbitrary ones (e.g., introduction of hiragana and katakana) affecting the evolution of reading systems. The examples provided are suggestive at best, and one is reminded of largely unsupported claims regarding optimal design of language (e.g., Berwick & Chomsky Reference Berwick, Chomsky, Di Sciullo and Aguero2010; Boeckx Reference Boeckx, Di Sciullo and Boeckx2010; Chomsky Reference Chomsky1995; Reference Chomsky2007; Reference Chomsky, Bricmont and Franck2010; Fitch Reference Fitch2007; Uriagereka Reference Uriagereka1998). These claims have been challenged by Johnson and Lappin (Reference Johnson and Lappin1997), Postal (Reference Postal2003; Reference Postal2004), and Seuren (Reference Seuren2004), and similar challenges would need to be addressed by Frost. The repetitive reference to terms like “optimization,” “optimal representation,” “optimal information,” “optimal solution,” “optimization of information,” “perfect match,” and “perfect example of optimization of information” (in sect. 3 and its subsections) is question-begging. And, from an evolutionary perspective, claims such as “all resulting linguistic developments are to some extent entirely predetermined” (sect. 3.2.1, para. 1), “this evolution was to some extent inevitable” (sect. 3.2.2, para. 3), “nothing is arbitrary” (sect. 3.2.3, para. 2), and “[reading systems] necessarily evolve to …” (sect. 3.2.4, para. 2) are either incorrect or virtually meaningless. Little would have been lost had Frost made appropriately measured claims, and I hope future discourse will be conducted with more attention to the limitations of natural evolution. Nevertheless, these criticisms do not undermine the validity of Frost's main points about the psychology of reading and the importance of cross-linguistic comparative work.
Ram Frost's article is a valuable contribution to the debate about models of the reading process. His suggestion that different kinds of language/script pairing require different processing strategies, and that the strategies actually used differ as predicted, is plausible. And including structurally different languages is a commendable advance on earlier, unduly Eurocentric theorizing about reading.
Nevertheless, I would question several points. These all relate to Frost's claim that “the main goal of reading research is to develop theories that describe the fundamental and invariant phenomena of reading across orthographies” (target article, Abstract, original emphasis). Surely, given the facts about the five languages as Frost presents them, one would expect diverse orthographies to require diverse reading systems. Further, one might also wonder whether Japanese (or Chinese) store words the same way English speakers do. Given that kanji are rich visual images, visual recognition of Japanese (or Chinese) might result in a mapping to an inner representation that is very different from that of English. Frost acknowledges that some representation/strategy might be optimal in one dimension but not in another, which results in a large space of possible mappings. At the very least, in view of the data presented, one might expect the assumption of invariant reading universals to be defended. Yet, while Frost clearly believes that such universals exist, he never tells us why he holds this belief. One could argue that because reading and orthography are phenomena that depend on language, reading universals could be derived from language universals of the kind proposed by Chomsky (Reference Chomsky1965; Reference Chomsky1995). But in recent years the existence of language universals has been challenged (e.g., Evans & Levinson Reference Evans and Levinson2009; Everett Reference Everett2005). The debate on the issue is far from over, and Frost provides no independent reasons to assume there are any significant, specifically reading universals. Universals of reading need to be argued for rather than simply assumed.
Frost's detailed discussion of the role of letter position in English and Hebrew provides support for his critique of models that focus on a single dimension of processing. However, given the variation that Frost finds among only five languages, it is at least questionable whether the aim of “outlin[ing] the blueprint principles for a universal model of reading” (sect. 6, para. 7) is realistic. And, if one aims at such a universal model, it seems misguided to focus on differences between English and Hebrew (and the differences within Hebrew), which seem to suggest different processing mechanisms. Instead of arguing for either letter-position rigidity or flexibility, one might argue that these languages are located on (opposite ends of) a spectrum. It is probably true that the described strong priming effect holds for Frost's examples (e.g., SANDWICH/SNAWDCIH; sect. 6, para. 3) and many other polysyllabic English words. However, it hardly seems that “anything goes” (i.e., complete letter-position flexibility): Would WASHDINC still be recognized as SANDWICH? And, a badly scrambled UNDERDOG turns into GROUNDED with several indeterminate intermediaries.
Possibly, then, letter-position flexibility in English is a superficial phenomenon that masks the rigidity of some letters in the target; posing processing demands that are not fundamentally different from Hebrew. But this is only one aspect of the task a truly universal reading model has to account for. Whether it is possible to develop learning models that can satisfy Frost's universality requirement remains to be seen. Different models might be needed to explain the facts specific to reading in different languages. Recent cross-linguistic modeling work on word segmentation has shown that models that performed well for English failed miserably for Sesotho (Blanchard et al. Reference Blanchard, Heinz and Golinkoff2010). And, anyone attempting to reach Frost's ambitious goal of “full linguistic coherence” (sect. 7, para. 2) would also need to consider cognitive limitations that may be in place when children first learn to read, and the interactions between the different modalities (visual, auditory) that are involved in statistical learning (for recent findings on such interactions, see Emberson et al. Reference Emberson, Convey and Christiuancen2011). Models that can account for all aspects of the reading task are beyond the reach of available technology. Moreover, given the lack of evidence supporting the notion that significant universals exist specifically for reading, it would make more sense to start without preconceptions and generalize from a diverse set of specific cases. Postulating untestable universals at this stage seems at best to be premature. Frost is correct, of course, to insist that models need to be informed by facts from a wide range of reading phenomena and that the scope of any model needs to be clearly acknowledged.
Turning to the most serious problem with Frost's article: The claim that “every language gets the writing system it deserves” (sect. 3) is too strong, as it implies that scripts have evolved to be “perfectly” optimal for their respective spoken languages. I did not find convincing justification for the suggestion that all writing systems necessarily develop “optimization aimed at providing their readers with maximal phonological and semantic information by the use of minimal orthographic units for processing” (sect. 3.1, para. 1, emphasis in the original). Certainly, arbitrary historical facts about which scripts have been available when a language was first reduced to writing have also been important. Frost's discussion of Chinese and Japanese only hints at the complex interaction of arbitrary components (e.g., adoption of Chinese kanji by Japanese) and non-arbitrary ones (e.g., introduction of hiragana and katakana) affecting the evolution of reading systems. The examples provided are suggestive at best, and one is reminded of largely unsupported claims regarding optimal design of language (e.g., Berwick & Chomsky Reference Berwick, Chomsky, Di Sciullo and Aguero2010; Boeckx Reference Boeckx, Di Sciullo and Boeckx2010; Chomsky Reference Chomsky1995; Reference Chomsky2007; Reference Chomsky, Bricmont and Franck2010; Fitch Reference Fitch2007; Uriagereka Reference Uriagereka1998). These claims have been challenged by Johnson and Lappin (Reference Johnson and Lappin1997), Postal (Reference Postal2003; Reference Postal2004), and Seuren (Reference Seuren2004), and similar challenges would need to be addressed by Frost. The repetitive reference to terms like “optimization,” “optimal representation,” “optimal information,” “optimal solution,” “optimization of information,” “perfect match,” and “perfect example of optimization of information” (in sect. 3 and its subsections) is question-begging. And, from an evolutionary perspective, claims such as “all resulting linguistic developments are to some extent entirely predetermined” (sect. 3.2.1, para. 1), “this evolution was to some extent inevitable” (sect. 3.2.2, para. 3), “nothing is arbitrary” (sect. 3.2.3, para. 2), and “[reading systems] necessarily evolve to …” (sect. 3.2.4, para. 2) are either incorrect or virtually meaningless. Little would have been lost had Frost made appropriately measured claims, and I hope future discourse will be conducted with more attention to the limitations of natural evolution. Nevertheless, these criticisms do not undermine the validity of Frost's main points about the psychology of reading and the importance of cross-linguistic comparative work.
ACKNOWLEDGMENTS
I thank David Johnson, Paul Postal, and Geoffrey Sampson for very helpful comments on earlier drafts and assume sole responsibility for remaining errors.