An Experimental Study of the Relationship between the Semantic Priming Effect and Degrees of Autistic Traits in Japanese-speaking Adults
Ayuno Kawakami
A Master’s thesis submitted for the degree of Master of Linguistics Sciences, Linguistics and Language Technology
University of Eastern Finland February 2019
ITÄ-SUOMEN YLIOPISTO – UNIVERSITY OF EASTERN FINLAND Tiedekunta – Faculty
Linguistics Sciences Osasto – School
University of Eastern Finland Tekijät – Author
Ayuno Kawakami Työn nimi – Title
An Experimental Study of the Relationship between the Semantic Priming Effect and Degrees of Autistic Traits in Japanese-speaking Adults
Pääaine – Main subject Työn laji – Level Päivämäärä – Date Sivumäärä – Number of pages Linguistics and Language
Technology Pro gradu -tutkielma ✓ 25 April 2019 117
Sivuainetutkielma Kandidaatin tutkielma Aineopintojen tutkielma
Tiivistelmä – Abstract
This thesis presents the results of an experimental study examining the correlation between the degree of autistic traits in individuals and the semantic priming effect for Japanese-speaking adults not diagnosed with autism spectrum disorder (ASD). In the experiment, a semantic priming test using 60 Japanese Katakana noun word-pairs was administered to two groups: a low AQ score group and a high AQ score group. Although both groups showed faster response times in the word target condition than in the non-word target condition, the low AQ score group was consistently faster than the high AQ score group in close semantic pairs, far semantic pairs and unrelated semantic pairs. This result suggests possible similarities in automatic semantic processing for individuals with ASD and individuals with a high AQ score. Unexpectedly, the study observed similar reaction times (RTs) for three types of word conditions. Moreover, the close word pairs were responded to fastest in the low AQ score group, while the unrelated word pairs produced the fastest RTs in the high AQ score group. This finding may imply that the degree of relatedness between the prime and target words was too small (lower than 0.50) to produce reliable priming effects. A higher number of errors in the low AQ score group may demonstrate a detail-focused cognitive style similar to that of individuals with ASD. Overall, the slower semantic processing speeds and higher accuracy rates in the high AQ score group provide insights into the relationship between varying degrees of autistic traits and semantic priming among Japanese-speaking adults without ASD. Future research is expected to include a larger number of participants and to use all three Japanese scripts (Hiragana, Katakana and Kanji) with different degrees of relatedness between the prime and target words in order to better understand the cause for the differences in Japanese semantic priming processing.
Abstrakti
Tässä työssä esitellään kokeellisen tutkimuksen tulokset, jotka tutkivat autististen ominaisuuksien astetta yksilöissä ja semanttista alustavaa vaikutusta japanikielisille aikuisille, joille ei ole diagnosoitu autismi-taajuuksien häiriöitä (ASD). Kokeessa annettiin semanttinen alustava testi, jossa käytettiin 60 japanilaista Katakana-sanasparia, kahdelle ryhmälle: matalat AO-pisteet omaaville ja korkeat AO-pisteet omaaville. Vaikka molemmilla ryhmillä oli nopeampi vasteaikaa sanaobjektiivisessa kohdetilassa kuin ei-sanan kohdetilassa, alhaiset AO-pisteet omaava ryhmä oli jatkuvasti nopeampi kuin korkean AO-pisteet omaava läheisissä semanttisissa pareissa, pitkälti semanttisissa pareissa ja toisiinsa liittymättömissä semanttisissa pareissa. Tämä tulos osoittaa mahdollisia samankaltaisuuksia automaattisessa semanttisessa käsittelyssä yksilöille, joilla on ASD ja korkeat AO-pisteet. Yllättäen tutkimuksessa havaittiin samanlaisia reaktio-aikoja (RT) kolmille sanaolosuhteille. Lisäksi läheisiin sanapareihin vastattiin nopeimmin alhaiset AO-pisteet omaavassa ryhmässä, kun taas etuyhteydettömät sanaparit tuottivat nopeimmat RT- arvot korkeat AO-pisteet omaavassa ryhmässä. Tämä havainto voi tarkoittaa sitä, että alkusuhteiden ja
kohdesanojen välinen suhde on liian pieni (alle 0,50) luotettavien alustavien vaikutusten aikaansaamiseksi. Alhaisen AO-pisteet omaavan ryhmän virheiden määrä voi osoittaa yksityiskohtaisen tarkennetun kognitiotyypin, joka on samanlainen kuin ASD:n omaavien yksilöiden. Kokonaisuudessaan hitaammat semanttiset käsittelynopeudet ja korkeamman tarkkuuden nopeus korkeat AO-pisteet omaavassa ryhmässä antavat tietoa erilaisten autististen ominaisuuksien asteesta ja semanttisesta pohjustuksesta japania puhuvien aikuisten joukossa ilman ASD:ää.
Tulevan tutkimuksen odotetaan sisältävän suuremman määrän osallistujia ja käyttävän kaikkia kolmea japanilaista skriptiä (Hiragana, Katakana ja Kanji), joilla on eriasteisuus suhteessa pää- ja kohdesanoihin, jotta voidaan
Table of Contents
1. Introduction ... 1
1.1. Autism Spectrum Disorder ... 2
1.1.1. History of autism spectrum disorder ... 2
1.1.2. Autism spectrum disorder and communication impairments ... 5
1.1.3. Autism spectrum disorder and the semantic priming effect ... 8
1.2. Semantic Priming Effect ... 10
1.2.1 History of semantic priming experiments ... 10
1.2.2. History of Japanese semantic priming experiments ... 11
1.2.3. High-frequency script word processing and low-frequency script word processing ... 15
1.2.4. Semantic processing by skilled readers and less skilled readers ... 19
1.3. Current Study ... 22
1.3.1. The aim of the current study ... 22
2. Method ... 25
2.1. Pilot Study ... 25
2.1.1. Review of Original Study ... 25
2.1.2. Pilot Study Materials ... 26
2.1.3. Pilot Study Procedure ... 30
2.1.4. Results of Pilot Study ... 33
2.1.5. Problem Solving with the Pilot Study ... 37
2.2. Current Study ... 39
2.2.1 Participants ... 39
2.2.2. Stimuli ... 40
2.2.3. Stimuli Selection ... 42
2.2.4. Stimuli Selection Process ... 44
2.2.5. Selected Stimuli ... 47
2.2.6. The Autism-Spectrum Quotient (AQ) ... 52
2.2.7. The personal details questionnaire ... 54
2.2.8. Procedure ... 54
2.2.9. Hypotheses ... 56
3. Results ... 58
3.1. Summary of Personal Questionnaire ... 58
3.2. Item Analysis ... 60
3.2.1. Low AQ score group and High AQ score group ... 60
3.2.2. Analysis of the Overall Mean RTs ... 61
3.2.3. Analysis of the RTs of the Two Groups ... 67
3.2.4. RTs per Item Analysis ... 70
3.3. Comparison of Autism Quotient Score ... 77
4. Discussion ... 83
4.1. Hypotheses and findings ... 83
4.2. Limitations and future plans ... 92
4.3. Conclusions ... 95
References ... 97
Appendixes ... 109
Appendix 1 ... 109
Appendix 2 ... 111
Appendix 3 ... 113
Appendix 4 ... 114
Appendix 5 ... 115
1. Introduction
This thesis reports on an experimental study that examined the correlation between autistic traits and the Japanese semantic priming test in people not diagnosed with autism spectrum disorder (ASD). The ultimate aim of this study is to provide a possible objective measure to assess the relationship between the degrees of autistic traits and the Japanese semantic priming effect. There are three possible advantages to using autism quotient (AQ) scores and the Japanese semantic priming behavioural test. First, the combined use of the AQ score and semantic priming test enables an examination of the correlation between autistic traits and automatic semantic processing in people with no diagnosis for ASD or developmental disorders; that is, this combined method provides both a subjective measurement and an objective measurement when seeking to understand autistic traits.
Second, designing a Japanese semantic priming test allows for an examination of the methodological problems involved in developing Japanese semantic judgment tasks, as the complexity of the Japanese scripts may hinder the development of such tests. Third, being able to use a semantic priming test to identify communication impairments could save time and money, as the test could be a useful for ASD pre-diagnosis or to determine whether semantic deficits correlate with autistic traits, especially in people with no language delay history. In previous studies, while semantic priming tests have successfully distinguished ASD and control groups, the semantic priming effects have varied.
1.1. Autism Spectrum Disorder
1.1.1. History of autism spectrum disorder
Autism spectrum disorder (ASD) is a neurodevelopmental disorder that involves social and communication impairments, repetitive behaviours and stereotypical restricted interests. (DSM-5; American Psychiatric Association, 2013). Initially, Kanner (1943) described ‘early infantile autism’ (Kanner, 1944) in eleven children that had severe verbal and non-verbal communication deficits but good cognitive intelligence. They were also characterized by a social isolation from the outside world, a desire to stick to a set routine and a dislike of unfamiliar situations and surroundings. Of these eleven children, three never spoke while the other eight had echolalia and pronoun deficits. Asperger’s syndrome has been claimed to be a closely related disorder. Asperger (1944) examined children who appeared to have autistic behavioural features but found that they had higher verbal intelligence and were behaviourally less impaired. Asperger’s syndrome is also generally characterized by fluent speech with unique, abnormal prosody.
After autism was recognized as a disorder in Europe and the US, in the 1950s, it was introduced to Japan, the Japanese translation of which was made up of the three Kanji letters ‘ ’ (jihei-syo), which literally means ‘ =self’ ‘ =shutdown’ ‘ =syndrome’
(Hiraiwa, 2012). For the first few decades after Kanner (1943), ASD research mainly focused on the behavioural analysis of children and their parents, after which in the 1960s, research attention shifted to the neurological aspects. Oi (2013) reviewed public ASD
(1944) had initially proposed that the parents of autistic children were generally highly educated and cold, many people and medical professionals believed at the time that the cause of autism was parental problems, with ASD often being referred to as a
mother-caused-disorder; this particular bias toward the cause of ASD spread and was accepted in Japan. Eventually, Rutter (1968) established a new theory that contradicted the previously claimed acquired deficit theory. Rutter’s (1968) study revealed a relationship between ASD, specific medical conditions and other developmental disorders, from which it was concluded that autistic children had symptoms that were significantly different from other psychiatric disorders.
Autism was first recognized as a separate disorder in the Diagnostic and Statistical Manual of Mental Disorders, third edition in 1980 (DSM-3; American Psychiatric Association, 1980) Since that time, the diagnostic criteria have changed significantly because ASD symptoms were found to include a wide range of characteristics and the level of impairment was difficult to define. The most recent diagnostic criteria were published in 2013 in the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5;
American Psychiatric Association, 2013), in which the Asperger’s Disorder and pervasive developmental disorders (PDD-NOS) were omitted and unified into a single ASD category (Lord and Bishop, 2015). The DSM-5 states that the two main ASD features are as follows:
1. A constant deficit in communication and interpersonal relationships
2. Restricted and repetitive behaviour and interests.
Based on the published APA diagnostic criteria, the number of people diagnosed with ASD has increased dramatically, with the ratio rising from 4–5 people per 10,000 in the 1960s to the 1980s to more than 100 per 10,000 in the 2000s. However, the ASD prevalences differ widely; for example, in Taiwan, 34 per 10,000 are diagnosed, while in South Korea, the prevalence is reported to be 264 per 10,000 (Oi, 2013). A Japanese Ministry of Education (2012) survey claimed that 6.5% of the 53,882 students (excluding special needs students) attending Japanese public elementary school and secondary school had academic and behavioural developmental disorder characteristics, which indicated that the total number of students with developmental disorders could be even higher. Baio et al.
(2018) found that the ASD prevalence in the US was 13.1–29.3 per 1,000 children in a sample of 8 year olds, that ASD prevalence was four times higher in males, and that is was the highest in non-Hispanic whites. In sum, ASD prevalence appears to have increased dramatically in the past fifty years; however, there have been significant differences found for gender, race and ethnicity. It is also possible that the changes in the diagnostic criteria in the Diagnostic and Statistical Manual of Mental Disorders could have affected the
prevalence of ASD diagnoses.
1.1.2. Autism spectrum disorder and communication impairments
The unique and distinctively different communication impairments associated with ASD have been widely researched. However, there is no clear conclusion as to whether these communication impairments are a main issue or a secondary issue resulting from the primary deficit. Research has sought to clarify the particular features of language
development in ASD sufferers who have some verbal abilities, in which it was observed that not all language ability aspects appear to develop equivalently. For instance, Boucher (2003) observed that in general, the language ability in ASD individuals was impaired or absent; while larger deficits were associated with pragmatic failures, there were less
noticeable impairments in syntax and phonology. Non-verbal communication such as facial expressions, hand gestures and prosody were also impaired. Bogdashina (2005) also
pointed out individuals with ASD and those with average IQ were able to develop intact syntax, morphology and to demonstrate fluent speech; however, the non-verbal
communication was still impaired. It has also been proposed that the language spoken by non-ASD individuals sounds like a ‘foreign language’ to ASD individuals as they often develop their own unique languages with different internal and external speech patterns.
It was thought that the unbalanced language development in ASD individuals was because of their general lack of interaction (Charman, 2003) and lack of joint attention (Bono, 2004), or that they had a word association weakness in their mental lexicon. One of the more influential theories was the ‘weak central coherence theory’ suggested by Frith
(1989), which stated that the unique cognitive style seen in ASD individuals was because they tended to focus on detail and appeared unable to extend these concepts/images to wider information; that is, they had fluctuating cognitive styles that were less consistent than non-ASD individuals.
Neurological studies have also give plausible explanations for the communication impairments in individuals with ASD. It has been found that there are different activations in specific regions of the brain during phonological processing and semantic processing.
For example, the left inferior frontal gyrus is generally activated during semantic
processing in individuals without ASD (Poldrack et al., 1999). fMRI studies revealed that activation takes place in Broadmann areas 44 and 45 near the dorsal aspect of the left inferior frontal gyrus, especially when a non-word is presented, and that semantic tasks activated Broadmann areas 47 and 45 in the ventral aspect of the left inferior frontal gyrus.
Gaffrey et al. (2007) found that while the left inferior frontal gyrus (Brodmann areas 44 and 45) was activated during semantic decision tasks for non-ASD individuals, as found by earlier studies (Poldrack et al., 1999), the visual cortex (Broadmann areas 18 and 19) was activated during semantic decision tasks for individuals with ASD, which appeared to indicate that their lexical strategies relied on visual images and perceptual information. The fMRI image of this study indicated that the left inferior frontal activation was considerably smaller in individuals with ASD than in people without ASD. This study also clarified that the atypical lexical semantic performances associated with atypical brain activity correlated
with immature brain activity. In a similar experiment, Gaffrey et al.’s (2007) results were replicated by Harris et al. (2007), in which it was found that when processing concrete and abstract words, Broadmann area 45 and Broca’s area had smaller activations in individuals with ASD and that there was left temporal activation rather than left inferior frontal gyrus activation.
These fMRI studies have suggested, therefore, that immature brain development could possibly be a reason for ASD. Specifically, Brown et al. (2005) studied the
developmental changes in the brain cortex and found that there was a considerable increase and decrease in activity in Broadman area 44 during childhood and that there was only significant activation in this area in young children, which suggested that the maturation of area 44 during childhood strengthens semantic processing; however, this maturation does not appear to occur in individuals with ASD. In sum, the communication impairments in individuals with ASD may be because of a lack of semantic processing maturation and activation in Broadman areas 44 and 45.
Different right-left hemispheric synchronization was examined in children with and without ASD. Kurita et al. (2016) found a lower right-left hemispheric synchronization in children with ASD and a higher right-left hemispheric synchronization in children without ASD, which indicated that there was a lack of network maturation in the children with ASD.
Recent neural activity studies on children using MEG (Magnetoencephalography) have clarified that children with ASD have different brain activity during the
neurodevelopmental critical period. Takahashi et al. (2016) demonstrated that children with ASD have higher brain signal activity than children without ASD when watching mute DVDs. To be specific, an increase in brain signal variability was found in children without ASD and signal variability alteration was observed in children with ASD, which suggested that this lack of increase in brain signal variability resulted in a lower cognitive ability in children with ASD. To summarize the earlier findings, previous research on the brain activity associated with audio/visual lexical processing has found that there is atypical brain activation in individuals with ASD, which may be because of abnormalities in the
maturation process.
1.1.3. Autism spectrum disorder and the semantic priming effect
Although the fMRI and MEG studies verified that there was atypical brain activity for semantic processing in ASD individuals, the semantic priming test has also provided insights into ASD communication impairments, with different semantic priming effects being reported in previous studies. For example, Kamio et al. (2007) compared atypical semantic processing in individuals with ASD to typical semantic processing in individuals without ASD and found that while there were no differences in the error rates between the two groups in the lexical decision task, the RTs for the near semantic pairs were
significantly different, with the control group reacting significantly faster than the group with ASD (193.0ms: 845.9SD for the control group to 438.4ms: 1,194.6SD for the ASD
group); however, the RTs for the semantic pairs and phonological pairs were not significantly different between the two groups. The lack of a priming effect for the near semantic pairs in the ASD individuals was confirmed, which suggested that the different semantic processing was the cause of the communication impairments.
Different semantic priming experiments have identified short-term memory difficulties in individuals with ASD. Hala et al. (2007) studied the priming effect of homographs using a naming task and a lexical decision task in children with and without ASD, in which it was found that both the ASD group and the control group were able to pronounce the target homographs correctly and that the pronunciation error rates were similar in the first presentation of the prime and target pairs, which suggested that the children with ASD were as able to understand the sentence contexts as the children without ASD and that both groups had similar semantic processing during reading. However, in the second presentation of the homographs, the children with ASD had higher error rates from 25 times (46%) to 65 times (49%). Nevertheless, the results of the first presentation of the prime and target pairs indicated that the ASD group had the ability to use contextual information and that their pronunciation accuracy was even slightly higher than the control group. Hala et al. (2007) also investigated non-homograph semantic priming effects and also found similar effects in both groups; (related and unrelated words; 567ms: 182SD, 600ms: 152SD) for the ASD group: 551ms (147/SD), 602ms (158/SD) in the control group.
It was therefore surmised that the higher error rates observed in the second presentation
could have been associated with the observed short-term memory difficulties in children with ASD. The findings of both Kamio et al. (2007) and Hala et al. (2007) suggested that while individuals with ASD were able to demonstrate a semantic priming effect, it was different from individuals without ASD. While there were similar error rates in both experiments, the impairments in short-term memory were also confirmed.
1.2. Semantic Priming Effect
1.2.1 History of semantic priming experiments
Semantic priming is when a target word (e.g. nurse) is recognized faster when presented after a semantically related word (e.g. doctor) than when presented after a semantically unrelated word (e.g. apple). Semantic priming tests are applied to understand semantic processing, the structure of the metal lexicon and cognitive ability. Initially, Meyer and Schvaneveldt (1971) developed semantic priming effects using lexical decision tasks. In the experiment, it was found that semantically related word pairs (e.g.
bread-butter) were processed faster than semantically unrelated word pairs (e.g.
bread-doctor) when the two semantic stimuli were presented on the same screen. Mayer et al. (1975) than changed the semantic priming experimental methods from presenting the two stimuli on one screen to presenting each stimulus serially.
Semantic priming effects can be mainly divided into automatic spreading activation models and expectancy-based processing models. Posner and Snyder (1975) proposed that
automatic activation processing occurred without intent or conscious awareness, while controlled processing occurred under expectation, conscious awareness and attention. Later experiments showed that different SOA (Stimulus-onset asynchrony) could facilitate both automatic and expectancy based processing for semantic priming. Neely (1977) found an automatic priming effect when the SOA was less than 250 ms, whilst an expectancy-based semantic priming effect was found to occur when the SOA was longer than 700 ms. These test results indicated that the expectancy for the target word facilitated semantic priming effects at 700ms, but not at 250ms or 400ms; however, automatic semantic priming was exhibited only when a significantly shorter SOA was used. Wang and Kikuchi (1991) reported that there were no priming effects when participants were unable to visually recognize the prime word. Specifically, the presentation of the prime was shown for 5ms with pattern masking for 50ms, with the interval between the prime and the pattern masking being 10 to 100ms. Second, automatic semantic priming was not facilitated when the
visually recognizable prime and the visually unrecognizable prime were randomly presented. (Smith et al., 1994). Therefore, the facilitation of automatic semantic priming requires not only short SOAs but also visibility and the same SOAs.
1.2.2. History of Japanese semantic priming experiments
Semantic priming effects have been of interest to Japanese researchers since the 1970s.Traditionally, Japanese native speakers were assumed to have separate semantic
recognition systems to process the different types of scripts, and there have been many studies that have examined the script-specific semantic processing for Kanji, Hiragana and Katakana. Japanese researchers have been especially interested in the different semantic processing for logographic Kanji and the syllabic Hiragana and Katakana. However, since the 1980s, research has tended to suggest that semantic processing was more dependent on the script frequency rate than the type of scripts.
Most early experiments claimed that Kanji processing was faster than either Hiragana or Katakana processing; that is, a phonemic encoding in Hiragana and Katakana and an absence of phonemic encoding in Kanji was hypothesized. This was because several Japanese semantic experiments found that there was a shorter RT for Kanji words than for Hiragana words and Katakana words. For example, Kaiho (1975) found that the RTs for Kanji words were shorter than the corresponding Hiragana words in a category decision task. Further, in a naming task and a sentence judgment task, Saito (1982) observed that there was a faster recognition of Kanji than of Hiragana, and that increasing the number of mora had no effect on the recognition speed for Kanji, while longer mora slowed the speed of Hiragana recognition. Kaiho and Nomura (1983) identified that there was a syllable length effect in Hiragana but no significant effect in Kanji. Similarly, Kimura (1984) found that Kanji was recognized faster than Hiragana. Shinozuka and Kubota (2012) conducted a behavioural experiment using four-word compound words in Kanji, Hiragana and Katakana and compared the ease of comprehension, from which they found a significant difference;
Kanji words were the easiest (rated 100% for all four groups), followed by Hiragana (rated 84% to 97%), and Katakana (rated 3% to 16%); from which they concluded that Kanji was processed without the need for the phonological encoding needed to process Hiragana and Katakana.
These previous results indicate that phonemic encoding is required to comprehend Hiragana and Katakana but not Kanji. In other words, Hiragana requires phonemic encoding before lexical memory retrieval while lexical memory retrieval comes directly after seeing printed Kanji words as Kanji uses a visual code only before meaning retrieval in silent reading, whereas Hiragana is mediated by phonemic encoding before meaning processing. The opposite results, however, were also found. Feldman and Turvey (1980) demonstrated a faster processing of Kana than Kanji in a naming task using the names of colours; however, there were very few participants (n = 2) compared to the other studies.
There has been significant research evidence supporting the idea that semantic processing varies for different types of scripts. The ‘orthographic depth hypothesis’ proposed by Katz and Feldman (1983) classified languages as either ‘shallow orthography’ or ‘deep
orthography’. ‘Shallow orthography’ was defined as when the spelling and the sounds were consistent and the letter and phoneme correlation was one to one (e.g. Finnish), and ‘deep orthography' referred to languages in which the spelling and sounds were inconsistent and the pronunciation was arbitrary and irregular (e.g. English). While semantic priming effects were found for both English (deep orthography) and Serbo-Croatian (shallow orthography),
only English words were found to facilitate a priming effect in a naming task, which indicated that processing English words activated an internal lexicon, while processing Serbo-Croatian did not. It was therefore surmised that as languages with shallow
orthography are easily pronounceable, there is no need to refer to an internal lexicon, while languages with deep orthography require such an activation. Frost et al. (1987) examined English, Hebrew and Serbo-Croatian and also found that there was no priming effect in the naming task for Serbo-Croatian. When applying the ‘orthographic depth hypothesis’ to Japanese, however, there is possibly both ‘shallow orthography’ (Hiragana and Katakana) and ‘deep orthography’ (Kanji).
For a clearer understanding, Table 1, which was modified from Meyer and
Schvaneveldt (1974) and Shinozuka and Kuboa, (2012), shows the different hypothesized sematic processing routes for Kanji and Kana (Hiragana and Katakana). As shown, the phonological representations were mediated only for the Kana words.
Table 1
The different hypothesized sematic processing routes for Kanji and Kana (Hiragana and Katakana)
Input:
Kanji Words
⇒
Orthographic Representation
⇒
Semantic Memory Retrieval
⇒
Semantic Recognition
Input:
Kana Words
⇒
Orthographic Representation
⇒
Phonological Representation
⇒
Semantic Memory Retrieval
⇒
Semantic Recognition
1.2.3. High-frequency script word processing and low-frequency script word processing
In contrast to the hypothesis that Kanji does not require phonemic encoding and thus allows for faster semantic processing than Hiragana and Katakana, some research has suggested that the phonemic encoding for syllabic Hiragana and Katakana depends on the script frequency rate. It has been widely accepted that in alphabetic scripts high-frequency words are processed faster than low-frequency words. For example, McCusker et al. (1981) suggested that there was a dual access route model; the processing of high-frequency words directly accessed the semantic information from a visual representation, whereas
low-frequency words were mediated through the phonemic encoding process. In other words, word frequency influences whether semantic processing takes place, as shown in Table 2.
Table 2
The different hypothesized sematic processing routes for high frequency words and low frequency words
Input: High Frequency Words
⇒
Orthographic Representation
⇒
Semantic Memory Retrieval ⇒
Semantic Recognition
Input: Low Frequency Words
⇒
Orthographic Representation
⇒
Phonological Representation
⇒
Semantic Memory Retrieval
⇒
Semantic Recognition
However, controlling word frequency rates does not seem to lead to the same outcome in Japanese semantic priming tests. For example, Shinozuka and Kubota’ (2012) studied the word frequency rate for 4-word compound Kanji words (e.g. ),
Hiragana words (e.g. ) and Katakana words (e.g. ),
with the word frequency rates all being around 5 (Kanji words (mean) = 4.93, Hiragana words (mean) = 5.00 and Katakana words (mean) = 4.93). However, there was a major methodological problem in this study. As the script frequency rate was not accounted for, all participants found the Kanji to be the easiest to comprehend because the 4-word compound words used in the experiments were generally written in Kanji.
Besner and Hildebrandt (1987) suggested that the RTs needed for Japanese semantic processing were related to the degree of script familiarity; that is, phonological mediation was not required if the Hiragana or Katakana words were in a familiar script but
phonological mediation was required when the script was less familiar. Several previous studies have shown that there are script frequency effects in Japanese. For example, Hirose (1984) conducted a semantic priming experiment based on different script frequencies, in which three types of words were used; a ‘low-Katakana word’ (a word usually not written in Katakana), a ‘high-Katakana word’ (a word usually written in Katakana) and a
‘high-Kanji word’ (a word usually written in Kanji). It was found that the RTs were considerably longer for the ‘low-Katakana words’ than for the ‘high-Katakana words’ or
‘high-Kanji words’; however, no significant RT differences were found between the
‘high-Katakana words’ and the ‘high-Kanji words’. Further, while no priming effects for word pairs in the low script frequency words were found, there were priming effects for the high script frequency words. Hirose (1985) replicated these results in a further study to confirm that the RT for the ‘High-Katakana words’ was shorter than for the ‘low-Katakana words’, and that there was no repetition priming effect for the ‘high-Katakana words’ but there was for the ‘low-Katakana words’. Because a priming effect was only found for the
‘low-Katakana words’, it was concluded that the script familiarity for the ‘low-Katakana words’ was promoted by repeated visual input, while the script familiarity for the
‘high-Katakana words’ had not changed. Kawakami (1993) also conducted a semantic priming study that examined ‘high-Katakana words’, ‘high-Hiragana words’,
‘low-Katakana words’ and ‘low-Hiragana words’, and also found that the high frequency script words were processed faster than the low frequency script words in both Hiragana and Katakana. In addition, no effect for syllable length was found for the ‘high-Katakana words’ or ‘high-Hiragana words’, whereas the longer the number of syllables in the
‘low-Katakana words’ and ‘low-Hiragana words’, the longer the reaction times. A later study by Ukita (2000) also found the same script frequency effect for Hiragana words.
However, many Japanese words are a combination of Kanji, Hiragana and Katakana.
In particular, Japanese verbs are frequently written in a combination of Kanji and Hiragana, but rarely written in Kanji and Katakana. Fujita (1999) reported shorter reaction times for Kanji-Hiragana combination verbs (e.g. ) and longer reaction times for
Kanji-Katakana combination verbs (e.g. ). The effect of the script frequency rate has also been found in English studies. For example, McCelland (1976) found that mixed case words (e.g. bEeF ) were less accurately perceived than words using the same case (e.g.
beef/BEEF), which demonstrated that preliminary word identification was interrupted when less familiar letters were used. This orthographic structure effect for English words
appeared to confirm that word perception relies heavily on visual information.
In summary, the Japanese words in high frequency scripts were found to be
processed faster than the Japanese words in low frequency scripts regardless of the type of script. Unlike empirical results that have stated that logographic Kanji and syllabic
Hiragana/Katakana had different semantic cognition pathways, high frequency script words seemed to elicit semantic access without the need for phonemic encoding. As there also appears to be no syllable effect for RTs in high frequency scripts, it is surmised that words in high frequency scripts are visually stored in the mental lexicon. Taken together,
script-specific semantic processing ideas are questionable. Table 3 summarizes the different word processing required for high frequency script words and low frequency script words.
Table 3
The different word processing required for high frequency script words and low frequency script words
Input:
High Frequency Script Words
⇒
Orthographic
Representation
⇒
Semantic Memory Retrieval
⇒
Output:
Semantic Recognition
Input:
Low Frequency Script Words
⇒
Orthographic Representation
⇒
Phonological
Representation
⇒
Semantic Memory Retrieval
⇒
Output:
Semantic Recognition
1.2.4. Semantic processing by skilled readers and less skilled readers
In addition to the word frequency rates and script frequency rates, individual differences must also be considered. In English semantic priming studies, individual differences have been observed mainly in terms of vocabulary knowledge. Yap, Tse and Balota (2009) found that participants with lower vocabulary had higher word frequency effects than participants with higher vocabulary, from which it was concluded that people with higher vocabulary are able to similarly process both high and low frequency words without phonemic encoding, while people with lower vocabulary knowledge require phonemic encoding to process unfamiliar words.
Japanese studies have also found that the cognitive ability for the three Japanese scripts changes considerably during the compulsory education years from ages 6 to 15.
Tanaka (1977) demonstrated how the ease of script recognition changes, with the ease of
comprehension’ order being found to be Hiragana, Katakana and Kanji for children from 6 to 8 years old, and Kanji, Hiragana and Katakana for children older than 11. It was suggested that this order reflected the script learning order in Japanese schools;
that is, Japanese language learning starts with Hiragana at 6 years old, after which Katakana and Kanji are learnt; however, Tanaka (1977) did not consider script frequency rate in his experiment. Nonetheless, it seems logical that the Japanese script learning order and Kanji proficiency at a specific age affects semantic processing. Similar findings were reported by Shibazaki (2010), who examined the recognition speed for Kanji using various Kanji at different degrees of difficulty, from which it was clarified that the Kanji learnt at earlier ages was recognized faster than the Hiragana, while the Kanji and Hiragana learnt at junior high school level had similar processing speeds. More difficult and less common Kanji, however, had a slower response time than the Hiragana. Tables 4 and 5 give a summary of these results.
Table 4
Experiment on the Recognition of Japanese Letters (Tanaka, 1977)
Age 6–8 Hiragana > Katakana > Kanji
Age 9–10 Hiragana > Kanji > Katakana
Age 11 and over Kanji > Hiragana > Katakana
Table 5
Experiment on the Recognition Processing Speed for Difficulty Kanji Words (Shibazaki, 2010)
Kanji level for 2nd to 3rd grade in elementary school Kanji > Hiragana Kanji level for 1st to 3rd grade in Junior high school Hiragana = Kanji Kanji level for undergraduate in university to upper Hiragana > Kanji
These results are further summarized in Table 6; at elementary school, the Kanji were processed faster, and more difficult Kanji and low frequency Hiragana and Katakana were processed slower. In sum, Japanese semantic priming studies need to take account of the age of participants, the age-specific difficulty of the Kanji and the script frequency rate.
It is assumed that skilled Japanese readers and older native Japanese speakers are able to process words in levels to in Table 6 at a similar speed to less skilled Japanese readers and younger Japanese speakers as their Kanji knowledge increases with age.
Table 6
Ease of Word Recognition for Japanese Students, Summarized and Edited from (Tanaka, 1977) (Shibazaki, 2010)
Kanji words (level for elementary school) Kanji words (level for Junior High school) Hiragana words (usually written in Hiragana) Katakana words (usually written in Katakana)
Kanji words (level for undergraduate in university to upper) Hiragana words (usually not written in Hiragana)
Katakana words (usually not written in Katakana)
1.3. Current Study
1.3.1. The aim of the current study
The aim of present study is to examine the relationship between automatic semantic processing and autistic traits in individuals without ASD. This study is based on the lexical decision task study by Kamio et al. (2007), in which it was found that there was an
automatic semantic priming effect for near semantic prime and target pairs in a group
without ASD and an absence of the automatic semantic priming effect in a group with ASD.
While Kamio et al.’s (2007) study compared people with an ASD diagnosis and people without HFASD, this thesis focuses on the degree of autistic traits present in individuals without ASD. It is expected that people with a higher degree of autistic traits would have similar tendencies to individuals with ASD in terms of atypical automatic semantic processing. To be specific, if individuals with higher AQ scores show longer latencies in the priming effect compared with individuals with lower AQ scores, the automatic semantic processing aspect of language may be impaired in people with autistic traits. If the AQ score and the latencies in the semantic priming experiment do not correlate, the automatic semantic processing may not be impaired.
The first aim of this study is to test the predictions stated above; that is, is there a correlation between autistic traits and automatic semantic priming effects in people without HFASD. The second aim of this study is to examine semantic priming in Japanese with Japanese native speaker participants. While an English semantic priming study was
conducted by Kamio et al. (2007), a similar study has not been conducted in Japan using a Japanese semantic priming experiment, which may be because of the methodological complications in developing a Japanese semantic priming test. The third aim is to compare the overall AQ scores and the sub scales with the latencies in the semantic priming
experiment. If the subscales and semantic priming effect correlate, it may be because particular autistic characteristics are causing an automatic semantic processing deficit.
The novelties in this study are as follows. First, this research is the first to our knowledge to research the correlation between the semantic priming effect and AQ scores in people not diagnosed with ASD. Second, it is also the first experiment to use a Japanese semantic priming test and a Japanese AQ test on Japanese participants. The benefit of using Japanese university students without ASD as the participants is to assess the possibility of autistic traits. Takahashi and Uchino (2006) found that an increasing number of high school students not diagnosed with ASD or other developmental disorders had similar intellectual and developmental disabilities to individuals with an ASD diagnosis or other
developmental disorders. Sato and Tokunaga (2006) also found that 50% of the 132 universities surveyed had students who appeared to have learning disabilities, ADHD (attention deficit hyperactivity disorder) or HFASD but had no history of definitive diagnoses.
The results of this research could provide an objective communication measurement for people with no ASD diagnosis; that is, the AQ provides a subjective measurement of
possible autistic traits and the semantic priming experiment provides a more objective judgment. If a higher AQ score and the absence of a semantic priming effect are confirmed, the semantic priming test could be a useful tool for the analysis of autistic traits, which in turn could explain possible communication difficulties. The combination of the AQ test and the semantic priming test, therefore, could provide a more precise judgment regarding the presence of autistic traits and atypical semantic processing for people without ASD.
2. Method 2.1. Pilot Study
2.1.1. Review of Original Study
A pilot study was conducted in August 2017 to solve any methodological problems in using the Japanese semantic priming test. The primary aim of this pilot study was to determine the extent to which the English version of the semantic priming test by Kamio et al. (2017) would be replicated by the Japanese version. In the original experiment by Kamio et al. (2017), 11 individuals with ASD were recruited; 11 individuals without ASD, age and gender-age matched, served as the control group. The 11 participants with ASD were confirmed to have an IQ of more than 70 and to have no language delay in childhood.
A lexical decision task was used to test the priming effect. Fifty percent of the prime words were actual words, while fifty percent were non-words. All the words were written in the English alphabet. The non-words were formed by changing the spelling order. The average word length was 4.6 letters. The words had either one or two syllables. The majority of the words were nouns; the remainder were adjectives, verbs and prepositions. The degrees of relatedness of the prime and target words were divided into three categories: near-related far-related and unrelated. The semantically related pairs involving words in the same category (e.g., truck : car) were limited in order to have producible results from the priming effect. Results showed that the RTs for the far related semantic pairs and unrelated
semantic pairs for participants with ASD were not significantly different from the RTs of the participants without ASD. However, the individuals without ASD indicated significant
semantic priming effects for the near semantic pairs, whereas those with ASD did not.
Accordingly, the absence of automatic semantic priming processing for the near semantic pairs indicated atypical semantic processing in individuals with ASD.
2.1.2. Pilot Study Materials
The pilot study used Psychopy (Version 1.82.01) (Peirce, 2007). The word list
published by Kamio et al. (2007) was used as the basis for selecting the Japanese prime and target words used in the pilot study. Phonological prime and target words in the original study were omitted as no phonological priming effect would be tested in our experiment.
To create a suitable list of Japanese prime and target words, additional processing was required, as a simple translation of an English word list into Japanese was insufficient. The processing used here included (1) choosing script from Kanji, Hiragana and Katakana, and (2) checking the availability of non-words in the selected scripts.
The second column of the word list by Kamio et al. (2007) can be used to illustrate process (1). This second column contains ‘Bike’ and ‘Ship’ as prime words and ‘Boat’ as a target word, with ‘Soat’ as a target non-word. As a native Japanese speaker, we considered the script frequency rate for each of the words and used Kanji for the first two words (‘Bike’
was written as ‘ ’, ‘Ship’ was written as ‘ ’ ) and Katakana for the third word (‘Boat’ was written as ‘ ’ ) (see the Table 7).
Table 7
Japanese translations for ‘Bike’ and ‘Ship’ and ‘Boat’
Prime Target
Far semantic Close semantic Word Non-word
Bike Ship Boat Soat
N/A
As for process (2), creating a non-word in Japanese is not as straightforward as in English. For example, many Kanji words cannot produce a non-word merely by changing the order of the Kanji. For instance, ‘ ’ and ‘ ’ both mean ‘Beef’, regardless of the Kanji order. Moreover, Kanji words such as ‘ ’, shown in the second column of the table 7 above, cannot be used to create a non-word. On the other hand, with Hiragana and Katakana, changing the character order is sometimes an appropriate method to create non-words. However, we need to be careful when using these two scripts, as sometimes changing the order of the characters actually produces another word; for example, the noun
‘ ’ (shoes) becomes the verb ‘ ’when the order of the characters is reversed (attach).
Consequently, when attempting to create non-words in Japanese, one must consider whether changing the character order will accomplish the goal.
After applying process (1) and (2), we selected 15 words and 15 non-words for target words; the corresponding primes consisted of 10 far semantic word pairs, 10 close sematic word pairs and 10 control (xxxx) word pairs. In the table 8, ‘f’ signifies that ‘word’ is the correct answer; ‘j’ signifies that ‘non-word’ is the correct answer. Notably, the word pairs for the far semantic, close semantic and target items were not always written in the same
script.
Table 8
Japanese word list for the pilot study
Prime word Target word Correct Answer f
f f f f f f f f f
xxxx f
xxxx f
xxxx f
xxxx f
xxxx f
Japanese word list for the pilot study
Prime word Target word Correct Answer j
j j j j j j j j j
xxxx j
xxxx j
xxxx j
xxxx j
xxxx j
2.1.3. Pilot Study Procedure
Four females and one male, all between the ages of 28 and 36 years, participated in the pilot study. All were recruited individually and agreed with the purpose of the pilot study. MacBook Pro (Retina, 13-inch, Early 2015) was used for the experiment. Psychopy (Version 1.82.01) (Peirce, 2007) was used for stimuli presentation and data collection.
Each practice trial began with the typing of the participant’s initials on the computer screen. The participant then needed to press the space key to move to the next screen.
Instructions appeared to explain each procedure in the practice trial and main test. The prime words and target words in the practice test were different from the main test, but the process and time interval were the same in order to confirm the testing procedure. The original test started with the presentation of a fixation point ‘+’ for 500 ms, followed by the prime, which appeared for 250 ms; the target word was then displayed with no time limit.
In the pilot study, due to a mechanical problem with Psychopy, a 100-ms interval was unintentionally inserted between the prime word and the target word. Consequently, the SOA was effectively 350 ms, with the prime word presented for 250 ms, followed by 100 ms of blank screen with no stimuli. The practice trial did not record, and participants could ask questions about the procedure before the main test. The main test included 60 questions in total. The experiment took approximately 20 to 30 minutes for each participant, with no break time. The 30 prime and target words appeared randomly in the test and were repeated twice. Thus, each participant was given a different word order.
Screens of pilot study ( 1 )
( 2 )
( 3 )
( 4 )
( 5 )
( 6 )
Explanation of Screens
(1) Typing the participant’s initials.
(2) Test instructions explaining that the participant will see ‘+’ followed by a word or
‘xxxx’ and a second word or non-word in Japanese Kanji, Hiragana or Kanji. The participant will then need to decide by pressing ‘f’ or ‘j’ whether the second word appearing on the screen is a real word or a non-word as quickly as possible. Pressing the space key moves the participant to the second screen.
(3) Instructions explaining that the practice test would appear immediately after this screen and indicating to participants that their answer would not be recorded.
(4) The practice test includes five prime words and five target words with two repetitions,
which is different from the real experiment but has the same timing (0.5 ms for the fixation, 0.25 ms for the prime word and no limitation for the target word, with 0.1 ms waiting time between each screen to avoid overlapping the word on the screen).
(5) The first screen of main test.
(6) The main test, which has 60 questions in total, with no rest.
2.1.4. Results of Pilot Study
The mean RTs, error rates and priming effects for the five pilot study participants are summarised in the table 9. The table 10 and figure 1 indicate the RTs of the first and second responses.
Table 9
Mean Reaction Times, Error Rates, and Priming Effects Participants (N=5) Mean RT
(SD) (ms)
Mean % error Mean
Priming effect (SD) (ms)
Far semantic 1,634.5 (131.9) 0.6 169.1 (60)*
Near semantic 1,675.7 (130.9) 0.8 127.8 (61)*
Control 1,803.6 (192.0) 1.2
Total 1,704.60 0.87
*(RT of control) – (RT of far semantic) = priming effect (ms)
*(RT of control) – (RT of near semantic) = priming effect (ms)
Table 10
Comparison of 1st response and 2nd response Mean RT:
1st Response
Mean RT:
2nd Response
Difference (ms)*
Far semantic 1,650.5 1,653.6 -3.10
Near semantic 1,727.9 1,623.6 104.3
Control 1,881.1 1,726.0 155.1
Total 1,658.6 1,750.6 85.4
*(1st response) – (2nd response)
Figure 1. Mean RTs of 1st and 2nd response.
1650.5
1727.9
1881.1
1653.6 1623.6
1726
1450 1500 1550 1600 1650 1700 1750 1800 1850 1900 1950
Far semantic Near semantic Control
Mean RT: 1st Response Mean RT: 2nd Response
Error analysis
We first examined the error percentages for each prime and target condition. These values are presented in Table 1. Due to the small error percentages (0.6% to 1.2%), no further analysis was conducted.
Response time
The mean RT for the far semantic pairs was significantly shorter than the mean RT for the control pairs (t (9) = 2.69, p < .05, t-test). The mean RT for the near semantic pairs was also significantly shorter than the mean RT for the control pairs (t (9) = 3.70, p < .05, t-test). This indicated that both the far semantic pairs and the near semantic pairs activated semantic processing. The RTs for the near semantic pairs and the far semantic pairs were not significantly different (t (9) = 0.97, p >.05, t-test), suggesting that different degrees of relatedness did not facilitate the priming effect differently. The RTs for the 1st response and 2nd response were significantly different (t (29) = 2.28, p < .05, t-test), with the second response mean 85.4 ms faster than the mean of the first responses.
Priming Effects
Priming effects were found for both the far semantic and near sematic pairs. The RTs for the far semantic pairs and near semantic pairs were not significantly different; thus, the priming effects were also not significantly different. Although the far semantic pairs
showed a smaller priming effect in the 2nd response, the near semantic pairs showed a greater priming effect in the 2nd response. (see the table 11)
Table 11
Priming effects of 1st and 2nd response for far and near semantic pairs Priming Effect of 1st
response*1
Priming Effect of 2nd response*2
Far semantic 253 85
Near semantic 125 131
*1 (control) – (far semantic) in first response/ (control) – (near semantic) in first response
*2 (control) – (far semantic) in second response/ (control) – (near semantic) in second response
Comparison of the RTs, % error and priming effect between the original study and the pilot study
In comparing the results of the original study by Kamio et al. (2007) and the pilot study described here, we found that the RTs, error rates and priming effects showed different characteristics. First, the RTs for each condition were longer in the pilot study than the RTs for the control group in the original study (mean: 845.9 ms to 1,022.2 ms in the original study; 1,634.5 ms to 1,803.6 ms in the pilot study). Second, the mean
percentage error in the original study was 5.5% to 18.2% (mean: 10.9%), while the error rate in the pilot study was considerably smaller (mean: 0.87%). Although only the near semantic pairs showed a priming effect in the original study [17.0 (249.3) for near
semantic; 176.2 (237.6) for far semantic], the pilot study indicated a priming effect in both
sematic pairs, unlike the original study.
2.1.5. Problem Solving with the Pilot Study
The purpose of the pilot study was to examine whether the Japanese semantic
priming test produced different results from the original English version and to evaluate the use of Psychopy. The RTs, error rates and priming effect in the pilot study did not coincide with the results of the original study. The different RTs may have been induced by
mechanical differences (Multi-Stim for Windows in the original study and Psychopy in the pilot study). Differences in presentation colour and letter size, and the reaction key might also have affected results. In addition, the experimental environment in the pilot study was noisier than in the original study. The considerably lower error percentages in the pilot study might be closely related to Japanese script characteristics and word length. For example, Shark (prime) and Whale (target) were written as (sa/me) and
(ku/ji/ra) in the pilot study. In general, Japanese words are shorter than English words because one character is either a vowel or a combination of a consonant and a vowel.
Furthermore, the Kanji prime and target pairs tend to be more visually clear and more difficult to mistake, as the character are relatively complex. For instance, Skull (prime) and Brain (target) become (zu/ga/i/ko/tsu) and (no/u), and the target could be easily identified as ‘word’. In the pilot study, roughly half of the participants stated that identifying Kanji is easier than identifying Hiragana or Katakana. Therefore, the small number of errors in the pilot study might be because of the shorter Japanese words and the
visual effect of the script. A priming effect was found for both far semantic and near
semantic conditions in the pilot study, indicating that the relatedness of prime and target did not influence semantic processing in Japanese. This result may suggest that the Japanese translations might not have the same functionality as the original English in terms of relatedness of word/prime and word/non-word because Japanese script and English script are so visually different. Even the same word pairs in two different scripts can affect semantic processing differently.
2.2. Current Study 2.2.1 Participants
All participants in the full study were native Japanese speakers recruited from undergraduate and graduate classes at several private universities in Tokyo. They were recruited mainly in classrooms and on campus in agreement with an associate professor and school staff. The recruitment and experiment were conducted from 18 January 2017 to 23 March 2017. Participation was voluntary and no compensation was provided.
The table 12 shows participant characteristics. In all, there were 22 original
participants— four males and 18 females, ranging in age from 19 years to 68 years. Two of the 22 did not fully meet the experimental criteria and were excluded from the analysis, leaving 20 students ranging in age from 19 to 28 years (M = 22; SD = 2.9)—two males (10%) and 18 females (90%). All participants were right-handed and had no diagnosis of any developmental disorder and no delay in language development. All had normal or corrected normal vision. Humanities majors were in the majority (two males and 15 females). The remaining three students (all female) were Music, Economics and Law majors, respectively.
For purposes of analysis, the participants were divided into a low AQ score group and a high AQ score group based on their AQ scores. A mean AQ score of 18.2 was used as the separation point, placing 11 participants (one male and 10 females) in the low AQ group (M = 14.3, Min = 9, Max = 18) and nine participants (one male and eight females) in the high AQ group (M = 20.9, Min = 20, Max = 30). The Research Ethics Committee of the
University of Eastern Finland approved this research. The purpose of the experiment was not revealed to the participants.
Table 12
Participants’ information
Number Age mean (SD)
Gender:
Male/Female
AQ mean (SD)
All 20 22
(2.9)
2/18 18.2
(5.3)
AQ low 11 21.8
(3.1)
1/10 14.3
(2.7)
AQ high 9 22.1
(2.8)
1/8 20.9
(3)
2.2.2. Stimuli
As explained in the previous chapter, the script frequency rate of a word is as important as word frequency rate for Japanese semantic processing. Nevertheless, not all studies have used script frequency rate in selecting stimuli for their semantic priming experiment. In the present study, we considered the script frequency rate of each Japanese word to be a crucial factor in designing our experiment and avoided large differences in script frequency rate in our prime and target word lists. Ignoring script frequency rate in choosing the prime and target words would have run a high risk of an unintended script frequency effect whereby words in low frequency script would be processed more slowly than the same word written in high frequency script. Consequently, controlling and
calculating the script frequency rate for each word was crucial to the design of the prime and target word pairs.
Selecting stimuli in Japanese raises several other issues. First, there is no large-scale Japanese database of prime and target word pairs having differing degrees of relatedness.
Thus, we needed to consider a methodology for selecting suitable stimuli before conducting our study. While extensive data for semantic priming studies in English are available and detailed information regarding prime and target English words has been widely circulated (e.g. Hutchison et al., 2013), similar large scale research has not yet been done in Japanese to provide reliable prime and target word sets for designing new semantic priming experiments. Thus, there were no readily available resources to use in our selection of prime and target word pairs in Japanese.
Second, the number of word association list databases in Japanese is still quite limited. Although Japanese semantic priming studies have been conducted using word association list databases (Umemoto, 1969; Ogawa, 1972; Mizuno, 2011), little information on the script type used in these studies is available. For example, Umemoto (1969) and Ogawa (1972) provide no information regarding script type for answers given in a free recall task involving given stimuli. (To illustrate, the third most frequent answer
(apple) for the stimulus (red) from Umemoto (1969) can be written either in Hiragana or Katakana, but only Hiragana was used in writing the answer. Similarly, the most frequent answer (cleaning) for the stimulus (housework) from Ogawa
(1972) could be either in Kanji or Hiragana, although only Kanji was indicated.) Furthermore, the word association list that was used in both studies (Umemoto, 1969;
Ogawa, 1972) was so old that some words were already outdated (e.g. America-Soviet).
Although Mizuno et al. (2011) used three script types in their answers, the ranking included all three scripts for each word. In addition, a preliminary survey was conducted by Mizuno et al. (2011) to ask participants to rate relatedness in the given word list. In the end, we employed the word association database from Mizuno et al. (2011) because this was the only database that we could access to select stimuli for our experiment.
2.2.3. Stimuli Selection
As noted above, selecting prime and target pairs with consideration of the script frequency rate has several problems due to the absence of previous research and the lack of information regarding script frequency rates in possible references. The word association lists from Mizuno et al. (2011) chosen for the experiment included the newest information and indicated script type in the answers. Moreover, word frequency had already been considered by Mizuno et al. (2011), who used high frequency words selected from Amano and Kondo (2003). Although the stimuli were classified into three script types, the
corresponding answers accepted all three types of scripts.
The selection criteria in our experiment were set as follows (see figure 2):
A) To avoid cognitive load while processing two words in different script types, prime and target word pairs should be written in the same script type.
B) Prime and target word pairs should be similar with respect to word length and mora (syllables).
C) Relatedness of the prime and target words should be calculated based on the script type of the words and not just on the words themselves. For example, ‘niwatori’
(chicken) is the highest frequent word related to ‘tamago’ (egg) when all script types are included ( in Hiragana, in Kanji, in Katakana). However, when the same script type is used in the stimuli, the relatedness of the :
pair does not rank highest. In fact, the relatedness of : (Hiragana-Hiragana) is only 0.034, while the relatedness of : (Hiragana-Kanji) is 0.23.
The above data is a summary of ‘tamago’ (egg) : ‘niwatori’(chicken) pairs. The numbers after ‘niwatori’
indicate the number of participants who selected the word (from left to right, men, women and total). At
the top is the total number of participants who joined the experiment by Mizuno et al. (2011). The numbers on the right side indicate the relatedness of each word with the stimuli, based on the total
number of participants.
2.2.4. Stimuli Selection Process
Script Type of Stimuli Hiragana Katakana The Total Number of Stimuli 100 stimuli 100 stimuli After Process (A) to (C) 9 stimuli 64 stimuli
30 close semantic pairs 31 far semantic pairs relatedness
(mean 0.171) (median 0.118) (min. 0.05/max. 0.668)
relatedness (mean 0.019) (median 0.013)
(min. 0.007/max. 0.037)
20 close semantic pairs 20 far semantic pairs relatedness
(mean 0.227) (median 0.176) (min. 0.08/max. 0.668)
relatedness (mean 0.013) (median 0.012)
(min. 0.007/max. 0.023)
20 Unrelated semantic pairs 20 close semantic pairs 20 far semantic pairs
20 unrelated prime words
20 unrelated target words
20 close semantic prime words
20 close semantic target words
20 far semantic prime words
20 close semantic target words
10 unrelated target words
10 unrelated target non-words
10 close semantic target words
10 close semantic target non-words
10 far semantic target words
10 far semantic target
non-words
Figure 2. Chart of Selection Process.
