Paper short abstract

VOT data from 103 young speakers across eastern Japan reveal three regional laryngeal patterns, including an aspiration-like system in Tohoku. These findings suggest that voicing- and aspiration-based contrasts can coexist and may be derived from a unified representation.

Paper long abstract

This study investigates the phonological representation of laryngeal source contrasts in Japanese and the regional variation found among younger speakers in eastern Japan. Cross-linguistically, word-initial obstruents are often used to diagnose laryngeal contrasts because this position is prosodically strong and exhibits relatively stable phonetic cues compared with weaker environments such as intervocalic or word-final positions. Previous typological research has examined these contrasts using a range of acoustic parameters, including voice onset time (VOT) (Lisker & Abramson 1964), low-frequency energy reduction during closure, and the presence or absence of F1 cutback.

Within Element Theory (Harris 1994; Backley 2011), stop contrasts are represented with combinations of |ʔ| (closure), |H| (frication/aspiration), and |L| (voicing). Voiceless unaspirated stops (0 VOT) correspond to |ʔH|, voiced stops (−VOT) to |ʔHL|, and voiceless aspirated stops (+VOT) to |ʔHH|. Two-way laryngeal systems are therefore classified either as voicing languages, contrasting |ʔH| and |ʔHL|, or aspiration languages, contrasting |ʔH| and |ʔHH|. Although Japanese has long been analysed as a voicing language (Shimizu 1996), the classification has not been systematically re-examined for younger speakers across different regions.

To address this gap, we measured word-initial VOT values for /b d g/ and /p t k/ produced by 103 native speakers (mean age 20.8 ± 2.4) from Tohoku, Kanto, Chubu, and neighbouring regions. A hierarchical cluster analysis based on mean VOT values revealed three major patterns: Cluster 1 (primarily Kanto–Chubu) showed −17.7 ms for voiced stops and 42.2 ms for voiceless stops; Cluster 2 (Kanto–Hokuriku–Tohoku) showed slightly positive VOT for voiced stops (8.9 ms) with slightly longer voiceless values (42.2 ms); and Cluster 3 (mainly Tohoku) showed consistently positive VOT for voiced stops (15.1 ms) and markedly longer VOT for voiceless stops (59.3 ms). The third pattern points toward an aspiration-type contrast, partly consistent with Takada (2011).

To capture this internally conditioned variation, we propose a unified underlying representation combining properties of |H| and |L|, with regional outcomes derived through selective suppression of one element. Suppressing |L| yields |ʔH| (/d/-like), while suppressing |H| yields |ʔH| (/t/-like). This model predicts that aspiration- and voicing-based systems may coexist within Japanese.

Paper short abstract

This study examines how vowel duration affects the intelligibility and accentedness of Japanese produced by Polish speakers. Using PSOLA manipulations and native listener evaluations, results show that correcting vowel length significantly improves both intelligibility and accent ratings.

Paper long abstract

This study investigates the role of vowel duration in the phonemic contrast between short and long vowels in Japanese. Nine native Polish speakers produced minimal pairs differing only in vowel length. Using the PSOLA function in Praat, these recordings were manipulated to match native Japanese vowel durations. Both unmanipulated and manipulated stimuli were then used in a perception experiment hosted on the Gorilla Experiment Builder.

Twenty-four native Japanese listeners completed two tasks: an intelligibility task (four-alternative forced-choice) and an accentedness task (7-point Likert scale). The experiment included 120 test trials and 30 filler items from native Japanese speakers.

Initial acoustic analysis revealed that Polish speakers generally produced both short and long vowels with durations exceeding native norms. While most preserved the short-to-long ratio, some speakers exhibited ratios that were either too small or too large.

Statistical analysis using a logistic regression model for the intelligibility task showed that adjusting vowel durations to native-like realizations significantly increased accuracy from 72% in the unmanipulated condition to 89% in the manipulated condition. For accentedness, a linear mixed-effects model revealed that duration manipulation predicted a significant 0.5-point increase on the Likert scale.

These findings suggest that while vowel duration is a primary cue for word recognition, it also serves as a significant marker of nativeness. The study contributes to the literature on L2 speech perception and highlights the necessity of prioritizing temporal accuracy in Japanese pronunciation pedagogy for Polish learners.

Paper short abstract

Polish learners reliably distinguished Japanese word-medial plosives, but their phonetic cue use only partly matched native Japanese patterns. They showed similar closure-duration differences, stronger prevoicing in voiced stops, more frequent releases in voiced stops, and longer releases in voiceless stops.

Paper long abstract

Speech contrasts are rarely realized through a single acoustic cue, and similarities between a first and second language may facilitate learning without necessarily leading to fully native-like production. Japanese word-medial voiced and voiceless plosives differ in closure duration, closure voicing, and release realization. Polish shows similar differences in closure voicing and duration, although release patterns are not identical. From the perspective of the revised Speech Learning Model (SLM-r), this overlap may facilitate the production of Japanese medial voicing contrasts while allowing language-specific patterns to persist.

A controlled production experiment was conducted with 8 Polish learners of Japanese and 10 native Japanese speakers. Participants produced nine Japanese real-word minimal pairs contrasting in word-medial plosive voicing, with three pairs at each place of articulation and three repetitions per word. In total, 970 productions were analyzed. We examined closure duration, the proportion of the closure containing voicing, and release occurrence and duration using mixed-effects models.

Both groups produced longer closures for voiceless than for voiced plosives. The raw voiced–voiceless differences were similar, at approximately 35 ms for Japanese speakers and 31 ms for Polish learners, although the proportional contrast was smaller in the Polish group because their closures were longer overall. Closure voicing in intended voiced plosives was also broadly similar: voicing occupied 73% of the closure in Japanese and 76% in Polish, with no clear group difference in devoicing. A clearer difference emerged in release realization. Japanese voiced plosives frequently had no identifiable release, with an adjusted no-release probability of 67%, compared with 23% for Polish learners. Voiceless-plosive releases were also descriptively longer in the Polish group (34 vs. 17 ms), although this pattern did not remain reliable after correction.

Overall, Polish learners produced a clear Japanese voiced–voiceless contrast and showed broadly similar closure timing and closure voicing to native speakers, but differed in release realization. The findings suggest that cross-language similarity can support second language production while language-specific phonetic patterns remain.

Paper short abstract

Slovenian has largely lost phonemic vowel length and use duration suprasegmentally to mark stress. Slovenian learners of Japanese master long–short contrasts but systematically reinterpret certain trimoraic CV–CV–R words as CV–R–CV, demonstrating prosodic transfer affecting L2 segmental structure.

Paper long abstract

In most Slovene dialects, vowel length no longer functions as a phonemic feature at the segmental level. The historical long–short contrast has been lost, and vowel duration is instead used at the suprasegmental level to cue stress placement and accentual prominence.

In contrast, Japanese encodes vowel length phonemically, and minimal pairs such as 通る tōru ‘pass by’ and 取る toru ‘take’ illustrate that vowel duration is contrastive at the moraic level and independent of stress or accent placement. Slovenian learners of Japanese generally acquire this contrast successfully and produce phonemic vowel length accurately in most lexical contexts.

However, a systematic deviation has been observed at the beginner level in the production of certain disyllabic, trimoraic Japanese words containing a sequence of identical vowels in the second syllable (CV–CV–R). Based on both controlled and spontaneous speech data, these forms are occasionally realized as CV–R–CV, yielding outputs that preserve the overall mora count but alter the mora–segment association. Crucially, this reordering occurs only under specific prosodic conditions, namely when the initial mora lacks lexical pitch accent. Thus, the target form 旅行 ryokō ‘a trip’ may surface as [rjoːko], rendering it homophonous with the female given name Ryōko.

This pattern suggests a reanalysis of L2 Japanese vowel length as a suprasegmental rather than a segmental property. The study systematically examines the influence of Slovene accentual representations on this non-target-like pronunciation and argues that learners reinterpret the long vowel not as a bimoraic vowel linked to a single syllabic nucleus, but as a prosodic lengthening effect associated with the most prominent unit in the word. This reinterpretation reflects transfer from the Slovene prosodic system, in which durational cues are systematically tied to stress rather than lexically specified at the segmental level.

The findings demonstrate that suprasegmental interference may give rise to non-target-like segmental outputs, even in cases where learners appear to have acquired the relevant contrast. Pedagogically, these findings underscore the need to explicitly represent moraic structure in teaching Japanese pronunciation to Slovene learners.

Keywords: vowel length, moraic structure, suprasegmental transfer, L2 phonological acquisition, Slovene–Japanese prosodic interference