Distribution and trichotomic realization of voiced velars in Japanese - an experimental study

Shin-ichiro Sano, Tomohiko OoigawaInterspeech 2010

TL;DR

This paper challenges the traditional view of Japanese voiced velars as either plosives or nasals, demonstrating through acoustic experiments that they are realized in a trichotomy of plosives, nasals, and fricatives. The study reveals that the distribution of these allophones is influenced by phonological context, following vowels, and dialect, suggesting a more nuanced understanding of Japanese phonology and the ongoing decline of velar nasalization.

What Problem Does This Paper Address?

For decades, the phonological description of voiced velars in Japanese has largely relied on a dichotomy: they are either realized as plosives (like [g]) or nasals (like [ŋ]). This view suggests that plosives occur word-initially, while nasals appear word-internally, particularly after nasal sounds. However, some researchers have noted instances of velars being realized as fricatives (like [y]), a phenomenon termed "spirantization." This paper aims to systematically investigate the distribution of voiced velars, moving beyond the plosive/nasal dichotomy to explore the trichotomy of plosives, nasals, and fricatives. The authors seek to understand how phonological factors, such as word position, surrounding sounds, and dialectal variations, govern these realizations, and to re-evaluate claims about the decline of velar nasalization.

Methods

The researchers conducted speech production experiments involving three participants, carefully selected to represent different Japanese dialects (Tohoku, Kanto, Kansai) and to be younger speakers, thus focusing on contemporary language use. They designed specific target words across six phonological contexts (e.g., word-initial, post-nasal, sequential velars) and five vowel contexts (/i/, /e/, /a/, /o/, /u/). These words were embedded in carrier sentences and recorded.

The audio data was then analyzed using the Praat software. The key innovation here is the classification criterion for velar allophones, which explicitly includes fricatives. The classification was based on acoustic features:

  • Plosives ([g]): Characterized by an obvious burst (closure and release) in the waveform.
  • Nasals ([ŋ]): Identified by the presence of a nasal formant.
  • Fricatives ([y]): Classified when neither a burst nor a nasal formant was observed.

This systematic acoustic analysis allowed for a quantitative examination of the distribution of these three allophones across various linguistic contexts.

Datasets

The study utilized a custom dataset of recorded Japanese speech. Target words were specifically chosen and arranged to cover six phonological contexts and five vowel contexts. A total of 600 tokens were analyzed (6 phonological contexts × 5 vowels × 5 tokens × 3 speakers, with some contexts having slightly different token counts). This dataset was generated for the purpose of this study and is not a pre-existing benchmark.

Key Insights and Ideas

  • Beyond the Dichotomy: The most significant finding is the robust evidence for a trichotomy of voiced velar allophones ([g], [ŋ], [y]) in Japanese, directly challenging the traditional plosive/nasal binary. Fricatives are not rare exceptions but a significant part of the realization landscape.
  • Context is King: The distribution of these allophones is highly sensitive to phonological context.
  • Word-Initial: Velars are predominantly plosives, but a notable percentage (20%) are fricatives, contradicting strict positional rules.
  • Post-Nasal: Velars show a higher tendency towards nasal realization, though this is tempered by the overall decline in nasalization.
  • Sequential Contexts: Velars are most likely to be realized as fricatives, especially when adjacent to other velars (even across intervening vowels), suggesting a form of long-distance agreement.
  • Vowel Influence: The type of following vowel plays a role. Plosives are favored before high/front vowels (/i/, /e/), while fricatives and nasals are more common before low/back vowels (/a/, /o/), likely due to articulatory demands.
  • Dialectal Differences Persist: Despite the general decline in nasalization, dialectal differences remain. Tohoku speakers show a higher rate of nasalization compared to Kanto and Kansai speakers, confirming previous observations. Conversely, fricative realization shows an inverse pattern, being most frequent in Kansai.
  • Accelerated Decline of Nasalization: The study suggests that velar nasalization is declining more rapidly than previously thought, especially among younger speakers, with plosives and fricatives becoming more dominant even in contexts where nasals were traditionally expected.

Key Results

  • Overall Distribution: In the analyzed data, plosives ([g]) occurred with 46.17% frequency, fricatives ([y]) with 44.83%, and nasals ([ŋ]) with a significantly lower 9%.
  • Word-Initial: Utterance-initial velars were categorically plosives, while word-initial velars showed an 80% plosive rate and 20% fricative rate.
  • Post-Nasal: Nasals ([ŋ]) had a higher probability (35%) in post-nasal contexts compared to other positions, but this was still suppressed by the overall decline in nasalization.
  • Sequential Contexts: Fricative realization was high in sequential contexts (e.g., 64% for VgVg1st, 75% for NgVg2nd).
  • Dialectal Distribution: Nasal realization rates were approximately 20% for Tohoku, lower for Kanto, and minimal for Kansai. Fricative rates showed an opposite trend, being highest in Kansai.

Limitations and Future Work

Limitations: * Small Sample Size: The study involved only three participants, which limits the generalizability of the findings, especially concerning individual variation and broader dialectal patterns. * Focus on Younger Speakers: While beneficial for studying current trends, this focus might not fully capture the historical trajectory of velar nasalization. * Limited Lexical Scope: The target words were primarily Sino-Japanese, which might not fully represent the distribution in all types of Japanese vocabulary.

Future Work: The authors suggest that future research should explore the interaction between language change and phonological factors in more detail. They also propose that the nasal realization of velars might eventually disappear, leaving only plosives and fricatives. Further investigation into the phonetic mechanisms underlying spirantization and the role of prosodic structure in velar allophone distribution is also warranted. The comparison with Spanish velar allophones suggests potential cross-linguistic parallels in the plosive/fricative dichotomy, which could be a fruitful area for future study.

Summary updated

BibTeX
@inproceedings{sano2010distribution,
  author = {Shin-ichiro Sano and Tomohiko Ooigawa},
  title = {{Distribution and trichotomic realization of voiced velars in Japanese - an experimental study}},
  booktitle = {Interspeech},
  year = {2010},
}
Figure