Haruo Kubozono - Handbook of Japanese Phonetics and Phonology

Before proceeding to the discussion, there is one remark about what is meant by a particular acoustic correlate being primary. The concept of being primary can mean several different things. A primary acoustic correlate can be used to mean an acoustic parameter that is invariant across speakers, speech styles, phonological contexts, or even across languages; a primary cue is also used to mean that it constitutes the most important perceptual cue for listeners, one that dominates other secondary cues (Lahiri and Hankamer 1988) so that secondary cues are only exploited when the target stimuli are ambiguous in terms of the primary cue, distributing around a range that is not found in natural speech (Hankamer, Lahiri, and Koreman 1989; Picket, Blumstein, and Burton 1999). For a general discussion on primacy of cues, see Abramson and Lisker (1985); Stevens and Blumstein (1981); Stevens and Keyser (1989); Whalen et al. (1993) and others; for a discussion of primacy in the context of length distinctions, see Abramson (1992); Hankamer, Lahiri, and Koreman (1989); Idemaru and Guion (2008); Lahiri and Hankamer (1988); Picket, Blumstein, and Burton (1999) and Ridouane (2010). Ridouane (2010) argues that cross-linguistically, differences in constriction duration are the most consistent acoustic correlates of singleton-geminate contrasts.

Figure 1: A singleton [t] in Japanese. Produced by a female native speaker of Japanese. The time scale is 300ms

With this said, the primary acoustic correlate of Japanese geminates is greater duration compared to singletons: geminate consonants are characteristically longer than singleton consonants. show illustrative waveforms and spectrograms of a singleton [t] and a geminate [tt] in Japanese (with the same time scale of 300ms). As we can see, the geminate [tt] has a longer closure than the singleton [t].

Many acoustic studies have investigated the durational properties of singletongeminate contrasts in Japanese, and summarizes their findings. This summary shows that geminate stops are generally at least twice as long as corresponding singleton stops, and can sometimes be as three times as long, regardless of the place of articulation or duration).

: A geminate [tt]. The time scale is 300ms

As with many other phonological contrasts, a singleton-geminate contrast is acoustically manifested not only by constriction duration, but by multiple other acoustic properties as well. (Multiplicity of acoustic correlates for phonological contrasts has been an important topic throughout the history of the phonetic theory; see, for example, Abramson 1998; Kingston and Diehl 1994; Lisker 1986 and references cited therein.)

2.2.1 Other durational correlates

In Japanese, vowels are longer before geminates than before singletons (Campbell 1999; Fukui 1978; Han 1994; Hirata 2007; Hirose and Ashby 2007; Idemaru and Guion 2008; Kawahara 2006a, 2013b; Kawahara and Braver 2014; Ofuka 2003; Port, Dalby, and ODell 1987; Takeyasu 2012). Port, Dalby, and ODell (1987) found, for example, that [] is on average 68ms before singleton [k] and 86ms before geminate [kk]; i.e. that [] is 18ms longer on average before geminates. Kawahara (2006a) found similarly that vowels before voiceless singletons are on average 36.9ms while those before voiceless geminates are 53.4ms. Furthermore, some studies even found that in C1VC2V contexts, C1 is longer when C2 is a geminate than when C2 is a singleton (Han 1994; Port, Dalby, and ODell 1987) (cf. Takeyasu 2012 who found the opposite, shortening pattern; Hindi shows the same lengthening pattern: Ohala 2007).

Table 1: Summary of the previous studies on closure duration of singleton and geminate stops and their ratios in Japanese. Duration measures are in miliseconds. SD = standard deviation; MoE = margin of error for 95% confidence intervals. Sing = singleton; Gem = geminate; VOT = Voice Onset Time; vls = voiceless; vcd = voiced

On the other hand, vowels that follow geminate/singletons show the reverse pattern: those that follow geminate consonants are shorter than those that follow singleton consonants (Campbell 1999; Han 1994; Hirata 2007; Idemaru and Guion 2008; Ofuka 2003). Han (1994) found the shortening of post-geminate vowels (and sometimes also the following word-final moraic nasals) by 9ms. In an acoustic study reported in Idemaru and Guion (2008), the mean duration of the following vowel is 63ms after geminates and 76ms after singletons. As explicitly noted by Hirata (2007), however, this difference in duration of the following vowels is less substantial and less consistent than the difference in the preceding vowel.

Finally, one may expect that Voice Onset Time (VOT) an interval between the release of the closure and the onset of voicing of the following vowel would be longer for geminate stops than for singleton stops, because longer closure would result in higher pressure build-up behind the stop occlusion. However, this expectation does not seem to hold: in Han (1994), VOT is slightly shorter for geminates than for singletons; in other studies (Hirata and Whiton 2005; Homma 1981), the relationship is inconsistent. See Kokuritsu Kokugo Kenkyjo (1990) for the data on the intraoral air pressure rise in Japanese consonants, which indeed shows that geminates do not involve higher intraoral air pressure rise.

2.2.2 Other non-durational, acoustic correlates

.

As observed in , Japanese geminates are associated with various non-durational cues. Given that, in addition to the primary acoustic correlate of constriction duration, there are a number of acoustic cues that are associated with Japanese geminates, they cannot be merely characterized as long consonants.