English words, when adapted into Japanese, show interesting asymmetries in the occurrence and non-occurrence of geminate consonants. One asymmetry can be found in minimal pairs like 'stuff' [stEOEf] and 'tough' [tEOEf]. Both have a lax vowel followed by /f/, but the former (with a CC onset) is borrowed into Japanese with a geminate consonant while the latter (with a C onset) is borrowed with no geminate consonant. The question is whether this difference is due to phonetic properties of the English coda consonant, particularly its durational properties, or to the different phonological onset structures of the source forms. To test these hypotheses, three judgment experiments were conducted. Experiment 1 used stimuli with excised onsets, created by deleting the initial consonant of CC-onset forms. The percentage of geminate judgments by Japanese listeners (hereafter, geminate judgment percentage) was significantly lower for the excised stimuli than for the original stimuli. Statistical analysis showed that the duration of the coda consonant relative to the preceding vowel (C/V) was not related to the geminate judgments. CC onsets elicited more than eight times as many geminate judgments as C onsets and excised onsets. Experiment 2 tested geminate judgments with four onset types: CCC, CC, C, and no onset. We found that, across all four onset types, none of the durational factors we tested accounted for the geminate judgment percentages, but the different onset structures did influence geminate judgments. One interesting result from Experiment 2 was that the geminate judgment percentage for CCC onsets was smaller than that for CC onsets. When CCC onsets are adapted, the loanword has either four or five moras, depending on whether or not a geminate occurs. Assuming the perceptual boundary hypothesis proposed by Kubozono and Ogawa (in: T. Oishi et al. (eds) Gendaikeitairon no choryu [Trends in modern morphology], 2004), we claim that native Japanese listeners prefer four-mora forms to five-mora forms when considering them as single words and that this preference influences the gemination patterns. Experiment 3 was conducted to test whether a tendency toward geminate judgments would be observed with stimuli in which the initial C of a complex onset was replaced by a CV sequence, and the results demonstrated such a tendency. We claim this is further evidence that phonological factors and not durational properties affect Japanese listeners' geminate judgments.
