In his target article, Ram Frost argues that current models of word recognition are “ill-advised” (sect. 1, para. 5) because they (1) focus mainly on orthographic processing, and (2) incorporate flexible letter-position coding as a general property of the cognitive system. We warmly welcome his criticism. As noted by Frost, in many languages the morphological structure of words has a key role to play in word recognition. Finnish is a prime example of a morphologically rich language. Although the crucial experiments have not been done, it appears likely that letter-position coding in Finnish is not as flexible as in some other alphabetic languages. When words as a rule contain multiple morphemes, letter positions become more constrained than in languages with restricted morphology (e.g., English). In Finnish, transposing two letters at morpheme boundary can completely change the word meaning. For example, transposing two letters in valalta, an inflected form of vala (oath), results in vallata, which has a completely different meaning (to conquer). Thus, although not yet demonstrated, it is highly likely that in these kind of instances Finnish would not show transposed letter priming (i.e., vallata would not prime valalta), similarly to Hebrew.
Although the relevant empirical evidence concerning transposed letter priming is still missing for Finnish, there is supportive evidence (Duñabeitia et al. Reference Duñabeitia, Perea and Carreiras2007) from a morphologically poorer language, Spanish, and a morphologically rich language, Basque, for the claim that in morphologically productive languages flexible letter-position coding across morphemes may lead to insurmountable problems for the processing system. This study was mentioned only in passing in Frost's target article. In a priming study, Duñabeitia et al. (Reference Duñabeitia, Perea and Carreiras2007) obtained no transposed letter effect for letter transpositions spanning the morpheme boundary either for suffixed Basque words, or for prefixed or suffixed (mesoenro→mesonero = landlord) Spanish words. Similar results have been obtained by Christianson et al. (Reference Christianson, Johnson and Rayner2005) in English for suffixed words (see, however, Rueckl & Rimzhim Reference Rueckl and Rimzhim2011) and compound words (susnhine→sunshine). If the compound word experiment is replicable in Finnish (as argued above, it should be) and other languages where compounding is highly productive (e.g., Dutch and German), it would further undermine the notion of flexible letter-position coding as a universal feature of written-word recognition. In Finnish, for example, more than 60% of all word entries in the dictionary are compound words.
In addition to morphological complexity, Finnish has other features that stress the significance of individual letters (and their positions). It is not infrequent that doubling a letter (either consonant or vowel) creates a new word, as in tuli (fire) versus tulli (customs) or sika (pig) versus siika (white fish). In other words, vowel and consonant length bears significance to word meaning. As phoneme–grapheme correspondence is nearly complete, short and long phonemes are transparently marked in written Finnish.
The second issue we would like to comment on is Frost's argument about languages possessing the most efficient representational system given the “ecological environment” of the language (sect. 3.1, para. 2). For instance, he argues that the extreme phonological transparency of Finnish has made it suitable for the creation of long words with morphologically densely packed information. This is to some extent true. However, the unpacking of morphological information during recognition–due to lack of clear segmentation cues and due to visual acuity limitations of the eye–is by no means straightforward. This is demonstrated by Bertram et al. (Reference Bertram, Pollatsek and Hyönä2004), who found that the identification of long compound words is slowed down when the letters spanning the morpheme boundary could create a frequent within-syllable bigram (i.e., the possibility for a misparse is created). On the other hand, when the morphological structure in long, three-constituent compound words is signaled by a hyphen positioned at a morpheme boundary, recognition is facilitated (Bertram et al. Reference Bertram, Kuperman, Baayen and Hyönä2011). These data cast doubt on whether a language with such an abundance of long, multimorphemic words like Finnish would indeed have created the most efficient representational system possible. In reading, a price has to be paid for dense packaging of information in long words, as unpacking the information is not straightforward. The eye is simply not equipped to extract all letter information from long words; hence, as long as morpheme boundaries are not clearly marked, the identification of sublexical units seems problematic.
Finally, Frost criticizes dual-route models like the one of Grainger and Ziegler (Reference Grainger and Ziegler2011) by claiming that they can explain any possible finding (sect. 8, para. 2). We think this is not the case. In the Grainger and Ziegler model, the full-form route is assumed to be coarse-grained, as it allows flexibility in letter positions. On the other hand, the decomposition route is more fine-grained, meaning that letter position coding is rigid. This has two consequences for the present discussion. First, this type of model is not overly flexible. It predicts, for example, that the coarse-grained route would be much more likely to be successful with frequent rather than infrequent morphologically complex words. However, morphological structure would be more readily detected by the fine-grained route in infrequent words and in words containing productive (rather than unproductive) sublexical units. A pattern of results pointing to the opposite would be hard to explain by such model. Second, such a model is able to capture the kind of morphological constraints discussed earlier in this commentary. It can explain a transposed-letter effect within a morpheme as well as a failure to find such an effect across a morphological boundary. By allowing both free and rigid letter coding, it reflects the strategy of optimizing encoding resources–a modeling feature pleaded for by Frost. Generally speaking, dual-route models are well equipped to adapt to the linguistic environment at hand. As Frost argues, this is one of the requirements of a universal model of reading and cannot thus be held against them. Given these considerations, we would like to inspire the author to come up with more solid arguments to discard these models.
In his target article, Ram Frost argues that current models of word recognition are “ill-advised” (sect. 1, para. 5) because they (1) focus mainly on orthographic processing, and (2) incorporate flexible letter-position coding as a general property of the cognitive system. We warmly welcome his criticism. As noted by Frost, in many languages the morphological structure of words has a key role to play in word recognition. Finnish is a prime example of a morphologically rich language. Although the crucial experiments have not been done, it appears likely that letter-position coding in Finnish is not as flexible as in some other alphabetic languages. When words as a rule contain multiple morphemes, letter positions become more constrained than in languages with restricted morphology (e.g., English). In Finnish, transposing two letters at morpheme boundary can completely change the word meaning. For example, transposing two letters in valalta, an inflected form of vala (oath), results in vallata, which has a completely different meaning (to conquer). Thus, although not yet demonstrated, it is highly likely that in these kind of instances Finnish would not show transposed letter priming (i.e., vallata would not prime valalta), similarly to Hebrew.
Although the relevant empirical evidence concerning transposed letter priming is still missing for Finnish, there is supportive evidence (Duñabeitia et al. Reference Duñabeitia, Perea and Carreiras2007) from a morphologically poorer language, Spanish, and a morphologically rich language, Basque, for the claim that in morphologically productive languages flexible letter-position coding across morphemes may lead to insurmountable problems for the processing system. This study was mentioned only in passing in Frost's target article. In a priming study, Duñabeitia et al. (Reference Duñabeitia, Perea and Carreiras2007) obtained no transposed letter effect for letter transpositions spanning the morpheme boundary either for suffixed Basque words, or for prefixed or suffixed (mesoenro→mesonero = landlord) Spanish words. Similar results have been obtained by Christianson et al. (Reference Christianson, Johnson and Rayner2005) in English for suffixed words (see, however, Rueckl & Rimzhim Reference Rueckl and Rimzhim2011) and compound words (susnhine→sunshine). If the compound word experiment is replicable in Finnish (as argued above, it should be) and other languages where compounding is highly productive (e.g., Dutch and German), it would further undermine the notion of flexible letter-position coding as a universal feature of written-word recognition. In Finnish, for example, more than 60% of all word entries in the dictionary are compound words.
In addition to morphological complexity, Finnish has other features that stress the significance of individual letters (and their positions). It is not infrequent that doubling a letter (either consonant or vowel) creates a new word, as in tuli (fire) versus tulli (customs) or sika (pig) versus siika (white fish). In other words, vowel and consonant length bears significance to word meaning. As phoneme–grapheme correspondence is nearly complete, short and long phonemes are transparently marked in written Finnish.
The second issue we would like to comment on is Frost's argument about languages possessing the most efficient representational system given the “ecological environment” of the language (sect. 3.1, para. 2). For instance, he argues that the extreme phonological transparency of Finnish has made it suitable for the creation of long words with morphologically densely packed information. This is to some extent true. However, the unpacking of morphological information during recognition–due to lack of clear segmentation cues and due to visual acuity limitations of the eye–is by no means straightforward. This is demonstrated by Bertram et al. (Reference Bertram, Pollatsek and Hyönä2004), who found that the identification of long compound words is slowed down when the letters spanning the morpheme boundary could create a frequent within-syllable bigram (i.e., the possibility for a misparse is created). On the other hand, when the morphological structure in long, three-constituent compound words is signaled by a hyphen positioned at a morpheme boundary, recognition is facilitated (Bertram et al. Reference Bertram, Kuperman, Baayen and Hyönä2011). These data cast doubt on whether a language with such an abundance of long, multimorphemic words like Finnish would indeed have created the most efficient representational system possible. In reading, a price has to be paid for dense packaging of information in long words, as unpacking the information is not straightforward. The eye is simply not equipped to extract all letter information from long words; hence, as long as morpheme boundaries are not clearly marked, the identification of sublexical units seems problematic.
Finally, Frost criticizes dual-route models like the one of Grainger and Ziegler (Reference Grainger and Ziegler2011) by claiming that they can explain any possible finding (sect. 8, para. 2). We think this is not the case. In the Grainger and Ziegler model, the full-form route is assumed to be coarse-grained, as it allows flexibility in letter positions. On the other hand, the decomposition route is more fine-grained, meaning that letter position coding is rigid. This has two consequences for the present discussion. First, this type of model is not overly flexible. It predicts, for example, that the coarse-grained route would be much more likely to be successful with frequent rather than infrequent morphologically complex words. However, morphological structure would be more readily detected by the fine-grained route in infrequent words and in words containing productive (rather than unproductive) sublexical units. A pattern of results pointing to the opposite would be hard to explain by such model. Second, such a model is able to capture the kind of morphological constraints discussed earlier in this commentary. It can explain a transposed-letter effect within a morpheme as well as a failure to find such an effect across a morphological boundary. By allowing both free and rigid letter coding, it reflects the strategy of optimizing encoding resources–a modeling feature pleaded for by Frost. Generally speaking, dual-route models are well equipped to adapt to the linguistic environment at hand. As Frost argues, this is one of the requirements of a universal model of reading and cannot thus be held against them. Given these considerations, we would like to inspire the author to come up with more solid arguments to discard these models.