Acquisition process of typing skill using hierarchical materials in the Japanese language
In the present study, using a new keyboard layout with only eight keys, we conducted typing training for unskilled typists. In this task, Japanese college students received training in typing words consisting of a pair of hiragana characters with four keystrokes, using the alphabetic input method, while keeping the association between the keys and typists’ finger movements; the task was constructed so that chunking was readily available. We manipulated the association between the hiragana characters and alphabet letters (hierarchical materials: overlapped and nonoverlapped mappings). Our alphabet letter materials corresponded to the regular order within each hiragana word (within the four letters, the first and third referred to consonants, and the second and fourth referred to vowels). Only the interkeystroke intervals involved in the initiation of typing vowel letters showed an overlapping effect, which revealed that the effect was markedly large only during the early period of skill development (the effect for the overlapped mapping being larger than that for the nonoverlapped mapping), but that it had diminished by the time of late training. Conversely, the response time and the third interkeystroke interval, which are both involved in the latency of typing a consonant letter, did not reveal an overlapped effect, suggesting that chunking might be useful with hiragana characters rather than hiragana words. These results are discussed in terms of the fan effect and skill acquisition. Furthermore, we discuss whether there is a need for further research on unskilled and skilled Japanese typists.
KeywordsAcquisition of typing skill Priming Japanese language Typewriting Fan effect Chunking Mora
Typing skills have become ubiquitous worldwide, especially among youths in developed Western countries, because of the popularization of personal computers. Logan and his colleagues (e.g., Crump & Logan, 2010a, 2010b, 2010c; Liu, Crump, & Logan, 2010; Logan & Crump, 2009, 2010, 2011; Snyder, Ashitaka, Shimada, Ulrich, & Logan, 2014; Yamaguchi, Crump, & Logan, 2013; Yamaguchi & Logan, 2014; Yamaguchi, Logan, & Li, 2013) have extensively explored the control of cognitive processes involved in typewriting1 and have proposed the two-loop theory of skilled typewriting (see Logan & Crump, 2011, for a summary of this theory; see also Yamaguchi, Crump, & Logan, 2013; Yamaguchi, Logan, & Li, 2013, for an illustration of this theory). The theory proposes that, in skilled typists, the outer loop represents a higher-level control process involved in comprehending sentences, decomposing sentences into words, and submitting the words to the inner loop, whereas the inner loop represents a lower-level control process that is responsible for receiving the words from the outer loop, activating the keystrokes in parallel, and executing them in accurate order.
Although the typing skills in youths in developed Western countries are robust, several researchers have demonstrated that skilled performance can deteriorate by disabling one of several associations that support skilled typing: (a) the association between words and letters (Crump & Logan, 2010b; Logan & Crump, 2011); (b) the association between letters and keys (Liu et al., 2010; Logan, 2003); or (c) the association between keys and finger movements (Crump & Logan, 2010a). Recently, research conducted by Yamaguchi and Logan (2014) demonstrated that a manipulation preventing the skilled typists from chunking in perception, short-term memory, and motor planning could cause previously skilled typists to again be unskilled. The authors suggested that chunking played an important role in the processing of several letters and keystrokes in skilled typewriting. Skilled performance has been thought to develop with chunking, which allows performers to reduce cognitive load in action planning and to concentrate on higher-level action goals (Newell & Rosenbloom, 1981; Yamaguchi & Logan, 2014).
Typing Japanese words is more complicated than typing English words. In the Japanese language, the unit of processing is a syllable (especially a mora, which has a constant duration) consisting of either a single vowel (V) or a combination of a consonant and a vowel. Hiragana scripts directly represent these syllables. When copying Japanese words using an alphabetic input method through typing on a keyboard, if all of the words consist of hiragana characters representing syllables, these alphabet letters never appear as visual stimuli.
Typical pattern of alphabet letters corresponding to hiragana characters
We manipulated the new combination of associations (mapping) between the alphabet letter and the hiragana character for each hiragana character using the “overlapped” or “nonoverlapped” mapping. For overlapped mapping, the alphabet “m”5 was used when typing a hiragana character む (mu), which was included in the hiragana words むし (musi) and かむ (kamu), and when typing the hiragana character め (me), which was included in the hiragana words しめ (sime) and めし (mesi). For nonoverlapped mapping, the alphabet “k” was used only when typing a hiragana か (ka), which was included in the hiragana words かて (kate) and かむ (kamu).
In the actual experiment, the alphabet letters were not presented visually; only the hiragana words were presented. The hiragana words in this task had hierarchical structures in which each hiragana character at the higher level was composed of a pair of alphabet letters at the lower level. As was described above, the typists had to type the key corresponding to the alphabet letter associated with their finger movement. However, early in the skill development process and in typing training, the typists had to convert each hiragana character used in a hiragana word into a pair of alphabet letters. Since two hiragana characters are associated with a single key location in the overlapped mapping, this situation might produce an overlapping effect between the overlapped and nonoverlapped mappings, as a type of interference (a kind of “fan effect”; see the Discussion section below). Late in the training, chunking should modulate the overlapping effect, which would diminish with the acquisition of typing a series of four keystrokes as a single unit of response, because the association between the key to be pressed and the typists’ finger movements would be strong, and because the key layout in this task had only eight keys (see Yamaguchi & Logan, 2014, for a manipulation in the reverse direction). Furthermore, previous research (Crump & Logan, 2010a) had confirmed that recent experience could influence the degree of skilled typing performance acquired in an individual’s long life history and had considered such an interaction between the recent association and the stored association in long-term memory to be evidence for instance-based skill acquisition (Logan, 1988). Our manipulation (overlapping) of the recent (new) association between the hiragana character and the alphabet letter was anticipated to reveal that the overlapping effect could appear early during typing training and could be diminished by typing the four keystrokes in the hiragana word through chunking later in the training.
Specifically, we expected that the typing speed would be slower in the overlapped than in the nonoverlapped mapping early in skill development. The participants could have started learning associations between individual hiragana characters and key locations early in the training. Since two hiragana characters were associated with a single key location in the overlapped mapping, this would produce interference early in the training. The overlapping effect early in skill development would then decrease during the training, because the unskilled typists would be influenced by the alternative alphabet letter from the hiragana character and would be presented with the new alternative association in the overlapped mapping, whereas the typists late in skill development who had received typing training would have a tendency to type four keystrokes as a single unit through chunking, irrespective of the new association. Conversely, other accounts of the acquisition of skilled performance (Botvinick & Plaut, 2004; Cooper & Shallice, 2000; Lashley, 1951; Norman & Shallice, 1986), such as parallel distributed processing and the schema model, would produce no prediction such as that the overlapping effect would diminish during typing training.
The theory of skilled typing (e.g., Crump & Logan, 2010a, 2010b, 2010c; Liu et al., 2010; Logan & Crump, 2009, 2010; Yamaguchi & Logan, 2014) proposes that reaction time (RT; the interval between word onset and the first keystroke) measures the duration of both outer and inner loops, and that the interkeystroke interval (IKSI; the interval between successive keystrokes) measures the duration of inner-loop processes. By using unskilled Japanese typists, we could closely investigate RT and each IKSI, because each keystroke corresponded to a constant syllable consisting of alphabet letters (the first and third letters represented consonants, and the second and fourth letters represented vowels). This analysis might reveal that chunking is more readily available for a hiragana character (a pair of keystrokes) than for a hiragana word (four keystrokes). If this were the case, the change in the overlapping effect should appear more explicitly in the speed of the second and fourth keystrokes (typing the vowel letters).
Twelve Japanese college students participated in this experiment as part of a course requirement. All were native speakers of Japanese, and all reported normal or corrected-to-normal vision. They reported that they always used the alphabetic-input method when typing Japanese words, and they were non-touch-typists.
Stimuli and apparatus
Hiragana words used as visual stimuli in the present experiment
Consequently, the typists needed to type four keystrokes for a Japanese word consisting of a pair of hiragana characters (e.g., かて, kate). We controlled the familiarity of hiragana words and the frequency of occurrences, in accordance with the work of Amano and Kondo (1999). Half of the keys each (K, S, A, and I) corresponded to the alphabet letters on a one-to-one basis (nonoverlapped mapping). The remaining half of the keys (T, M, U, and E) each did not correspond to the specific alphabet letters on a one-to-one basis (overlapped mapping; Fig. 2).
For the overlapped mapping, the alphabet letter “m,” which was used when typing the hiragana character む (mu), was included in the hiragana words むし (musi) and かむ (kamu); it was also used when typing the hiragana character め (me) and was included in the hiragana words しめ (sime) and めし (mesi; see Table 2). For the nonoverlapped mapping, the alphabet “k,” which was used only when typing the hiragana か (ka), was included in the hiragana words かて (kate) and かむ (kamu).
Design and procedure
Error trials were noted whenever an erroneous keystroke appeared within a word,7 in accordance with previous research (Crump & Logan, 2010a). The error analysis revealed a significant learning effect of trial blocks using one-way repeated measures analysis of variance (ANOVA), F(7, 77) = 2.73, MSE = 33.331, p < .05, ηp2 = .20; the error percentages were 19.0 %, 13.3 %, 13.0 %, 11.3 %, 10.5 %, 11.1 %, 10.8 %, and 12.4 %, for the first, second, third, fourth, fifth, sixth, seventh, and eighth blocks, respectively. The overall error rate was 12.7 %. Finally, all of the typists recalled the keyboard layout immediately following the 1-h training period.
The following analyses included the correct trials by excluding the error trials. To construct a measure of central tendency, we calculated the median RT and IKSI for each condition in each block for each participant (see Altmann, 2007; Blais & Besner, 2007, for the measurement of medians). We collapsed the data across the two key layouts, because no significant effects emerged between the groups of participants using the two different key layouts (F < 1). The RT and IKSI data were separately subjected to a two-way repeated measures ANOVA using trial blocks and mapping as variables. The p values for all F tests were adjusted using the Greenhouse–Geisser correction for departures from sphericity.
As we described in the introduction, our material of the alphabet letters corresponded to the regular order within each hiragana word (the first and third letters referred to consonants, and the second and fourth letters referred to vowels within the four letters). We performed a two-way ANOVA with repeated measures applied to the trial blocks and the mapping design for each position of the keystroke (viz., the first, second, third, and fourth).
RT (L1) involved in initiating keystrokes of consonant alphabets
IKSI (L3) involved in initiating keystrokes of consonant letters
The third keystroke was involved in typing consonant letters. The interval between the relevant keystroke and the keystroke before this keystroke (i.e., the interval between the second and third keystrokes; L3) was involved in initiating the keystroke of consonant letters. This interval revealed no overlapping main effect, F(1, 11) = 1.34, MSE = 66,676, p = .26, ηp2 = .11, and a significant main effect of trial blocks, F(7, 77) = 45.6, MSE = 58,599, p < .001, ηp2 = .81. We found no interaction between overlapping and trial blocks, F < 1 (Fig. 4c presents L3). This pattern was similar to the pattern for RTs (L1), which were also involved in typing consonant letters (Fig. 4a).
IKSIs (L2 and L4) involved in initiating keystrokes of vowel letters
The second and fourth keystrokes were involved in typing vowel letters. The intervals between the relevant keystroke and the keystroke before these keystrokes (the intervals between the first and second keystrokes and between the third and fourth keystrokes; L2 and L4) were involved in initiating the keystrokes of vowel letters. These intervals revealed a significant main effect of overlapping, F(1, 11) = 15.1, MSE = 35,575, p < .01, ηp2 = .58; F(1, 11) = 17.5, MSE = 24,728, p < .01, ηp2 = .61, for L2 and L4, respectively. At the same time, these intervals revealed a significant main effect of trial blocks, F(7, 77) = 19.0, MSE = 61,363, p < .001, ηp2 = .63; F(7, 77) = 11.4, MSE = 56,273, p < .01, ηp2 = .51, for L2 and L4, respectively. Furthermore, the intervals revealed a significant interaction, F(7, 77) = 4.68, MSE = 8,187, p < .05, ηp2 = .30; F(7, 77) = 6.81, MSE = 8,121, p < .05, ηp2 = .38, for the second and fourth keystroke latencies, respectively (Fig. 4b and d present L2 and L4).
In this study, we investigated the skill acquisition process, during which unskilled typists were moving toward becoming skilled typists through typing training when typing Japanese hiragana words on an unfamiliar keyboard. The previous studies of typing skill had always been conducted with skilled typists (thereby leveraging the two-loop theory of the skilled typewriting). The manipulation of our study aimed to improve typing skill through chunking in the direction opposite that used by Yamaguchi and Logan (2014), who degraded skilled into unskilled typists on the learning curve by preventing the skilled typists from chunking. The hierarchical organization of skill has theoretically been thought to be developed through chunking (Newell & Rosenbloom, 1981). We investigated RTs and IKSIs separately regarding the effect of overlapped mapping between hiragana characters and alphabet letters, in the situation in which chunking was readily available. Although, as we described in the introduction, we predicted that typing training would reveal the effect of the overlapped mapping design early in skill training, and that this effect would diminish in late training, this prediction was the case for only L2 and L4, involved in typing vowel letters. However, for RT and L3, both involved in typing consonant letters, this prediction was not correct. These findings suggest that the typists used chunking to make two keystrokes per hiragana character rather than four keystrokes per whole hiragana word.
The effect of the overlapped or nonoverlapped mapping design between the alphabet letters and the hiragana characters, which we found in L2 and L4, might be associated with the fan effect (Anderson, 1974). The fan effect was identified in Anderson’s studies on the actualization of multiple concepts (Anderson, 1974, 1983; Anderson & Reder, 1999), followed by priming studies (Neumann & Deschepper, 1991, 1992), thus identifying the degree of the fan effect as the priming effect’s size (either positive or negative priming). The present experiment revealed that only the latencies of the keystrokes involved in typing vowel letters reflected the fan effect, whereas the latencies of the keystrokes (RT and L3) involved in typing consonant letters did not reflect this effect. As the typists became familiar with the typing materials late in the training, the typists might have learned to process the two keystrokes as a single unit of response to a hiragana character. That would diminish the fan effect, because all two-keystroke sequences would be uniquely associated with a particular hiragana character. Thus, the results could be interpreted as evidence supporting the application of chunking in skilled typewriting (Yamaguchi & Logan, 2014).8
However, the alternative possibility is that the reduction observed in the overlapping effect in L2 and L4 might reflect a mere floor effect, in which the typing speed in the nonoverlapped mapping design displayed limited improvement in the last trial blocks (Fig. 4b and d). To investigate the possibility of this interpretation, we conducted a follow-up experiment using a QWERTY keyboard with the same hiragana words and participants as in the present main experiment (see the Appendix for details). Furthermore, we compared the median latencies in the last trial block of the present main experiment with those obtained in the experiment using the QWERTY keyboard by collapsing the overlapped mapping design, because the overlapping effect could not be investigated in the follow-up experiment (see the Method section in the Appendix for further details). Consequently, the typing speeds in the present main experiment for all keystrokes were significantly longer than those in the follow-up experiment. Thus, this finding indicates that the improved typing speed in the last trial block of the present experiment was not limited, and that the typing speed might decrease through further training, suggesting that the diminished overlapping effect could not be attributed to a mere floor effect.
We inferred that we failed to find an overlapping effect for RT and L3 for the following reason. The typing speeds of these keystrokes in the last trial block (970 and 323 ms for RT and L3, respectively) were considerably slower in the present experiment than were those in the follow-up experiment (638 and 183 ms for RT and L3, respectively). Furthermore, the typing speeds for these keystrokes were remarkably slow in the early phase of training (2,099 and 1,300 ms for RT and L3, respectively). The typists might have had to convert a hiragana character into a pair of alphabet letters and look for the key corresponding to the initial consonant for each hiragana character. Thus, the typists might take more time to type the keys of consonant alphabets. The overlapping effect was considerably smaller than the typing speed for these keystrokes; the processing while converting a word into two letters and searching for the correct key required a large amount of time in RT and L3. Thus, the processing involved in the overlapping effect could progress in parallel (rather than additively) with converting and searching, especially early in typing training. The study of unskilled typists has only just begun. Future studies using unskilled typists will need to further investigate the relationship between the skill acquisition process and priming.
By manipulating the direction of learning through setting up a direction opposite to that in Yamaguchi and Logan’s (2014) study, the present experiment confirmed that chunking plays an important role in skill acquisition of typing. We investigated only unskilled typists who were moving toward becoming skilled typists. However, we could not confirm that typists had become skilled late in training. Thus, it appears prudent to exercise caution when comparing the present results to those of previous findings on skilled typists.
To do so, we would need to resolve some problems. First, a method to confirm the accurate assessments of typing speed and accuracy in the Japanese language—such as that observed in Logan and Zbrodoff’s (1998) study in the English language—has never been established. To the best of our knowledge, this study is the first to have investigated Japanese typists from a cognitive psychological perspective. Thus, we need to develop a method through which to assess skill level using hiragana words. Second, it might be difficult to find actual touch-typists in the Japanese population, because we did not find them in this study. As we described in the introduction, according to previous research based only on self-report, the rate of touch-typists is surprisingly low in Japan. Thus, it might be difficult to find touch-typists using a precise assessment method in the Japanese population.
One open question continues to demand further investigation and study. The analysis of the IKSIs revealed that the keystrokes required to type the alphabet letters denoting vowels were significantly shorter than the keystrokes denoting alphabetic consonants over all of the training. This effect was extremely stable. Note that the order of the alphabet letters denoting consonants and vowels within the hiragana character was constant.9 The previous studies had indicated that IKSIs were affected by the syllable boundaries within words when typing words in English (Weingarten, Nottbusch, & Will, 2004; Will, Nottbusch, & Weingarten, 2006). It is not clear whether this result involves a Japanese language-specific issue. Future research on Japanese touch typists will resolve this problem.
There is no explicit differentiation between typing and typewriting (Logan & Crump, 2011). However, “typing” seems to refer to keypresses, as in “typing skill” and “typing training”; “typewriting” seems to involve language production, such as control in typewriting and an expression of language (e.g., Yamaguchi & Logan, 2014). In the present article, we use these terms according this distinction.
This survey investigated 3,486 persons, including 823 youths (18- to 22-year-olds), in Japan. Unfortunately, no more recent survey has investigated touch-typing rates in Japan.
This article distinguishes hiragana characters from alphabet letters for descriptive purposes.
The alphabet input method is recommended in the Guidelines for Elementary Information Education in the Japanese Government (Ministry of Education, Culture, Sports, Science and Technology of Japan, 2010, p. 81). We investigated 88 students (18- to 21-year-olds) enrolled at Kobe University about their preference for using the alphabetic input method in October 2013 for unpublished studies. In all, 87.5 % of the students answered that they consistently used this method. We failed to find other surveys on the preferences of Japanese college students for the alphabet input method.
Hereafter, the letters in italics denote the overlapped mapping; the concerned letters are underlined.
These latencies refer to the intervals between the relevant keystrokes and the keystrokes before these keystrokes.
Error trials were defined on the basis of a word; if any symbols denoting error keystrokes (_) appeared within a word, the PC counted the word as producing an error response.
We thank Motonori Yamaguchi for these suggestions regarding the fan effect and chunking.
Several studies have investigated the stable and robust syllabic structure of the Japanese language. This syllabic structure has a constant duration of pronunciation as units of processing, which are called mora or morae (e.g., Hino, Kusunose, Lupker, & Jared, 2013; Hino, Lupker, & Taylor, 2012; Kinoshita, 1998; Tamaoka & Makioka, 2004; Perea, Nakatani, & van Leeuwen, 2011; Sakuma, Sasanuma, Tatsumi, & Masaki, 1998; Verdonschot et al., 2011; Witzel, Qiao, & Foster, 2011).
We are grateful to Motonori Yamaguchi for his very helpful comments and beneficial suggestions on this article. We are also grateful to Sakura Matsunami for conducting this experiment, and to Kentaro Tanaka, Akihiro Tsuru, Shizuka Iwata, Noritaka Tsujimoto, and Masaaki Kitaoka for their help in producing the figures, tables, and references in this article.
Open Access This article is distributed under the terms of the Creative Commons Attribution License which permits any use, distribution, and reproduction in any medium, provided the original author(s) and the source are credited.