Letter perception is a critical step in reading because we cannot read any words without identifying the presented letters. How we represent letters in our mind and develop their representation during childhood are intriguing questions in the field of cognitive and developmental psychology. Letter similarity, which is the visual (but not phonological) similarity of letters, is a powerful tool for understanding these issues. Letter similarity, which has been extensively measured in alphabetic scripts over the past 130 years, has contributed to studies that are theoretically building cognitive models of letter perception and reading as well as empirically investigating the development of reading related skills (see Mueller & Weidemann, 2012 for a review). For example, Treiman et al. (2007) addressed the cross-linguistic issue of letter-naming acquisition by investigating the relationship between naming errors and letter properties, including letter similarity. The analysis of naming-error patterns showed that letter similarity had a similar effect in English- and Hebrew-speaking toddlers, suggesting that toddlers in both groups make visually-based confusions. In contrast, phonological similarity has a stronger effect on English-speaking toddlers than Hebrew-speaking toddlers because Hebrew has fewer phonologically similar letter pairs. These results imply a similar principle in letter-naming acquisition across different languages. Letter similarity is also useful for quantifying orthographic similarity. Although Coltheart et al. (1977) proposed using such orthographic neighbors as “time,” “lime,” and “mine” as a measure of orthographic similarity, they neglected a letter level of similarity. Despite the importance of letter similarity, less is known about it in non-alphabetic scripts. The present study provides new evidence for letter similarity from other languages, such as Japanese hiragana and katakana.
The Japanese language’s unique writing system consists of logographic kanji and two types of syllabic kana (hiragana and katakana), resulting in three types of writing systems in Japanese orthography (see Wydell & Butterworth, 1999 for details). While kanji characters are derived from Chinese characters, kana letters were created to simplify the surface forms of kanji characters. Kanji characters generally represent nouns (e.g., 会社 company), the root morphemes of inflected verbs (e.g., 行く to go), and adjectives (e.g., 楽しい fun). In contrast, hiragana letters represent the inflections of verbs, adjectives, and adverbs as well as function words. Katakana letters are mainly used for foreign words (e.g., インターネット internet). The Japanese kana system appears to resemble the upper and lower cases in the alphabet because hiragana and katakana letters represent identical sounds (e.g., hiragana あ and katakana ア represent /a/). However, the orthographic rules between kana and alphabetic scripts are completely different. For example, although a letter at the beginning of a sentence is capitalized in alphabetic script, this is not allowed in the Japanese kana writing system. Given these differences in writing systems, it might be fruitful to test whether cognitive models of letter perception and letter-knowledge development, which are established mainly based on studies of alphabetic scripts, are applicable to other language systems. Investigating similarity for Japanese kana (hiragana, katakana) letters is critical.
Despite the importance, however, very few studies have addressed letter similarity in Japanese kana compared with alphabetic script. In previous studies on hiragana, non-readers of Japanese performed a clustering task on 46 pairs of hiragana letters (Dunn-Rankin & Leton, 1973), and a subjective rating was conducted by Japanese-speaking adults on 71 pairs of hiragana letters (Kawakami & Tsuji, 2012). For katakana, the subjective letter similarity for 15 pairs of letters was investigated (Kaiho, 1970). Since the focus of this study was not to obtain a complete matrix of katakana letters, the evaluated pairs of letter similarity were limited. Other studies measured the letter similarity of every possible pair of the 71 katakana letters based on subjective ratings (Kawakami, 2002) and clustering tasks (Yamade & Haga, 2008). Given previous findings on Japanese kana, the subjective measurement of letter similarity for hiragana and katakana was obtained, although not the similarity across such script as hiragana お vs. katakana オ. Obtaining hiragana-katakana similarity across kana scripts is critical for two reasons. First, it can increase understanding of how children learn to name hiragana and katakana letters. In elementary school, Japanese-speaking children first learn hiragana and then katakana (Wydell & Butterworth, 1999). Perhaps a knowledge of hiragana influences the acquisition of katakana. For instance, hiragana-katakana letter pairs with a similar visual form (e.g., か-カ) are more easily acquired than those with a dissimilar form (e.g., あ-ア). Second, comprehensive data of letter similarity, including hiragana-katakana letter pairs, contribute to both theoretical and empirical research that is investigating how literate adults encode hiragana and katakana letters, especially whether the two types of kana letters are categorically represented.
Although the letter similarity of Japanese kana has been mainly measured by subjective methods, such as rating and clustering, previous studies on alphabetic script have measured it in other ways. The most common procedure is letter-naming tasks (e.g., Fisher et al., 1969; Gilmore et al., 1979), in which letter similarity is measured by counting how many times participants misnamed a presented letter (i.e., confusability). For example, when the letter O was presented, the participants misnamed 14.2% and 0.3% of their responses as Q and Z (van der Heijden et al., 1984). This error rate probably reflects the visual similarity of letters (Grainger et al., 2008). However, Mueller and Weidemann (2012) pointed out that these errors are not a direct measure of letter similarity because a pair of letters was not presented simultaneously (i.e., not compared directly); it also takes a tremendous amount of time to collect enough naming errors because participants correctly identified most of the trials. Therefore, although many studies tried to increase the naming errors by applying psychological techniques, such as noise, speed, or low-contrast images (e.g., Gervais et al., 1984; Gupta et al., 1983; Tinker, 1928), collecting enough naming errors remains time consuming. The second procedure to efficiently measure letter similarity uses letter-rating tasks to evaluate the visual similarity between two letters on a Likert scale (e.g., Boles & Clifford, 1989; Kuennapas & Janson, 1969). However, because this is a direct (but subjective) evaluation of letter pairs, it remains unclear whether the rating values substantially reflect a representation of the letters in the participants’ minds. The third procedure measures reaction times (RTs) in same-different judgments for letter pairs. Podgorny and Garner (1979) assumed that similar letter pairs take more time to be judged as the ‘same’ compared with dissimilar letter pairs, and Courrieu et al. (2004) proposed the inverse of RTs as a perceptual distance of letter pairs (i.e., an indicator of letter similarity). Unlike the number of errors in the letter-naming task, values based on RTs do not include zero elements, showing that a more useful indicator for investigating letter similarity should be free from struggling with zero elements. Thus, RT-based values are not subjective; they are an objective measure, obtained by directly comparing letter pairs. Furthermore, since this method does not require that letters be named, we can presumably purely measure the visual (but not phonological) similarity of letter pairs. In the present study, we applied this method to measure the letter similarity of Japanese kana.
In Experiment 1, we obtained the letter similarity of Japanese kana (hiragana and katakana) by measuring the discrimination RTs in a go/no-go task for presented letter pairs. Following a previous study (Courrieu et al., 2004), we asked the participants to only press a button as fast as possible if the letter pair was different (e.g., #あ#い#) and not it if the letter pair was identical (e.g., #あ#あ#). By analyzing RTs in different letter pair condition, we obtained perceptual distance as a measure of letter similarity. Note that similarity (as a general concept) can be equally measured by similarity (similar letter pairs have greater values) and dissimilarity (dissimilar letter pairs have greater values). Distance is a similar measure with dissimilarity (dissimilar letter pairs have greater distance) that is required to satisfy some requirement. Given the evidence that it took more time to discriminate between visually similar letter pairs (Podgorny & Garner, 1979), Courrieu et al. (2004) proposed an inversed value of discrimination RTs as an indicator of letter similarity (i.e., perceptual distance) that allows the application of more powerful data analysis methods, such as multidimensional scaling. Like dissimilarity, similar letter pairs have smaller perceptual distances. For example, Courrieu et al. (2004) reported that the perceptual distance for visually similar b-d pairs were 69, although it was 124 for visually dissimilar y-c pairs. By applying this method, we obtained the perceptual distances for 4278 pairs of Japanese kana.Footnote 1
Although we are trying to obtain the perceptual distances of kana letters to determine how a letter pair is visually similar, there are no guarantees that other factors do not influence them. Since a letter is a visual object that represents sounds, phonological information might be implicitly activated during our task. In addition, as in the case of alphabetic script (a = A), literate Japanese speakers know that letters are arbitrary signs, and a certain pair of hiragana and katakana letters refers to the same sound (え = エ ≠ お). That is, such knowledge-based factors as letter-sound knowledge (conceptual knowledge) might be induced whenever we perceive a letter. Therefore, such a confounding factor is likely to affect the perceptual distance in our task.
One possible way to confirm the contribution of conceptual knowledge is provided by Lupyan et al. (2010) who used a same-different judgment task of alphabetic letter pairs consisting of identical sounds (Bb) and different sounds (Bp). Both lower-case letters (b and p) consist of a straight line and a semicircle, and the visual similarity between Bb and Bp is almost identical. If the participants only respond by a visual comparison of the letter pairs, one would expect that the RTs for the identical (Bb) and different (Bp) condition do not produce significant differences. This effect was observed in the two-presentation condition. In the simultaneous presentation condition, both letters were presented at the same time, as in our Experiment 1. In the sequential presentation condition, however, the first letters were presented before the second letter appeared. Lupyan et al. (2010) predicted little or no difference in the simultaneous condition because their participants responded before the conceptual knowledge affected the ongoing visual processing. However, in the sequential presentation condition, Lupyan et al. (2010) predicted a greater difference between the same and different sound pairs because of the greater time that the conceptual knowledge interfered with the visual processing. As predicted, Lupyan et al. (2010) found a significant difference between the same and different sound pairs in the sequential condition but not in the simultaneous condition, suggesting a small amount of interference by conceptual knowledge in the simultaneous condition. Based on previous findings, we also simultaneously presented letter pairs in Experiment 1. Therefore, we confirmed that our obtained perceptual distance is mainly attributed by the visual comparison of letter pairs even if a similar phenomenon was observed in Japanese kana letters.
Experiment 2 examined the effect of conceptual knowledge in the same-different judgment based on the previous study (Lupyan et al., 2010). We presented participants with letter pairs of hiragana and katakana, such as え-エ (hiragana /e/ - katakana /e/) in the same-sound condition and え-コ (hiragana /e/ - katakana /ko/) in the different-sound condition. Although エ and コ are visually quite similar, they represent completely different sounds, and therefore different RTs between the same and different sound conditions can be interpreted as the effect of conceptual knowledge. Like in the previous study (Lupyan et al., 2010), we observed this effect when the letter pairs were only presented in the sequentially presented conditions. If there is little or no effect of conceptual knowledge in the simultaneously presented condition, our obtained data can be mainly attributed to the perceptual visual information of letters.