Lexical competition in a non-Roman, syllabic script: An inhibitory neighbour priming effect in Japanese Katakana

Abstract

Previous masked priming studies have reported that lexical decision latencies are slower when a word target is primed by a higher-frequency neighbour (e.g., blue-BLUR) than when it is primed by an unrelated word of equivalent frequency (e.g., care-BLUR). These results suggest that lexical competition plays an important role in visual word identification in Indo-European languages such as English, French, and Dutch, consistent with activation-based accounts of lexical processing. The present research, using Japanese Katakana script, a syllabic script, demonstrates that lexical decision latencies were slower when targets were primed by word neighbour primes but not when targets were primed by nonword neighbour primes. Both results have clear parallels with previous research using Indo-European languages and therefore suggest that lexical competition is also an important component of word recognition processes in languages that do not employ the Roman alphabet.

Keywords:

• Masked priming
• Neighbour priming
• Word recognition
• Orthographic neighbours

Acknowledgements

This research was supported by a grant from the Natural Sciences and Engineering Research Council (NSERC) of Canada to Christopher R. Sears. We thank Carol Whitney and two anonymous reviewers for their valuable feedback and advice.

Notes

¹Although predicted by activation-based models, inhibitory priming effects are difficult for parallel distributed processing (PDP) models to accommodate because these models do not incorporate discrete lexical representations (e.g., Seidenberg & McClelland, 1989; Plaut, McClelland, Seidenberg, & Patterson, 1996). That is, because, in PDP models, there are no abstract units corresponding to words, there are no lexical representations for a prime to preactivate and, hence, there would be no competition among activated lexical representations. Thus, there is no obvious mechanism by which a word prime could produce delayed responding to an orthographically similar target. Indeed, the most straightforward prediction of PDP models is that neighbour primes will produce facilitory priming by activating sets of units that the prime and target share.

²Normative frequencies were based on the NTT database (Amano & Kondo, 2000), which provides frequency counts based on a corpus of approximately 300 million words. The normative frequencies reported here are per million words, created by dividing the reported frequencies by 300.

³The replaced pairs were and These pairs were replaced by and

⁴The four prime–target pairs listed in Footnote 3 were excluded from all analyses due to high error rates (greater than 60% for the prime or the target).

⁵For the nonword targets primed by words (see Table 3), the ANOVA factors were prime type (neighbour prime and unrelated prime) and prime frequency (high-frequency prime and low-frequency prime). Both factors were within-subject factors in the subject analysis; in the item analysis prime type was a within-item factor and prime frequency was a between-item factor. The only significant result was in the analysis of response latencies, with a significant effect of prime type in the subject analysis, F _s (1, 55) = 4.54, p<.05, MSE=1,275.6, partial η²=0.08; F _i (1, 38) = 3.72, p=.06, MSE=616.4, partial η²=0.09. Targets primed by neighbours were responded to faster (624 ms) than targets primed by unrelated words (634 ms).

⁶Two low-frequency targets (and ) had high error rates (greater than 60%). These targets were excluded from all analyses to be consistent with the treatment of targets with high error rates in Experiment 1A.

⁷For nonword targets (see Table 4), the data were analysed with single factor ANOVAs with two levels (prime type: neighbour vs. unrelated). The effect of prime type was significant in the response latency analysis, F _s (1, 55) = 12.98, p<.001, MSE=668.1, partial η²=0.19; F _i (1, 39) = 8.72, p<.01, MSE=763.9, partial η²=0.18. Targets were rejected as nonwords significantly faster when a nonword neighbour preceded them (614 ms) than when an unrelated nonword did (631 ms). In the error analysis the effect of prime type was not significant (both Fs < 1).

⁸The analyses were based on the items that were analysed both in Experiments 1A and 1B (34 low-frequency items and 36 high-frequency items).

⁹Three targets (and ) were excluded from all analyses because of high error rates (greater than 60%).

¹⁰For nonword targets (see Table 6), the effect of prime type was significant in the response latency analysis, F _s (1, 35) = 24.78, p<.001, MSE=1,281.7, partial η²=0.42; F _i (1, 59) = 25.69, p<.001, MSE=2,384.1, partial η²=0.30, with faster responses to nonwords primed by neighbours (621 ms) than to nonwords primed by unrelated primes (651 ms). The effect of prime lexicality was marginally significant in the subject analysis, F _s (1, 35) = 3.71, p=.06, MSE=980.0, partial η²=0.10; F _i (1, 59) = 2.71, p=.11, MSE=1,572.8, partial η²=0.04. No other effects were significant (all ps>.10).

¹¹In our experiments the primes and targets were always matched for number of characters. Note that matching for number of characters does not necessarily match for number of syllables; in fact, for approximately 35% of the prime–target pairs, the prime and target differed in the number of syllables, though in almost all cases (91%) this was a one-syllable difference (e.g., the prime had two syllables and the target had three). Note that this situation is common in the masked neighbour priming studies using English stimuli as well (e.g., Davis & Lupker, 2006; Forster, Davis, Schoknecht, & Carter, 1987; Nakayama et al., 2008). Our post-hoc analyses indicated that there were no differences in the priming effects for the prime–target pairs that differed in the number of syllables and for those that did not.

¹²We carried out a post-hoc analysis to determine if the magnitude of the inhibitory priming effect varied significantly depending on the position of the replaced character in the neighbour pair. The stimuli were divided into two groups: (1) neighbour pairs where the initial character was replaced (e.g., /re.be.ru/ and /no.be.ru/), and (2) neighbour pairs where another character position was replaced (e.g., /se.e.ta.a/ and /se.n.ta.a/, /ke.e.su/ and /ke.e.ki/). For Experiment 1A, the priming effect for low-frequency targets was 29 ms when the initial character was replaced and 20 ms when another character was replaced; for the high-frequency targets the priming effects were 11 ms and 9 ms, respectively. (These analyses were based on the item means.) For Experiment 2 the two priming effects were identical (36 ms). These analyses indicate that the magnitude of the inhibitory priming effect does not change depending on the position of the replaced character in the neighbour pair. This outcome is consistent with results reported by Janack et al. (2004), who found that the size of the inhibition effect from neighbour primes was not significantly different for neighbour pairs that differed in the initial letter (e.g., “mast-cast”) and neighbour pairs that differed in the last letter (“cash-cast”).