TL;DR
This paper provides a preliminary acoustic analysis of the /r/ sound in the Tyrolean dialect, revealing significant intra-speaker variability in its production. The study highlights the prevalence of uvular fricatives and approximants, with acoustic measures like duration and Harmonics-to-Noise Ratio (HNR) helping to differentiate these variants.
What Problem Does This Paper Address?
The Tyrolean dialect, spoken in South Tyrol, Italy, is a fascinating linguistic borderland where Germanic and Romance languages meet. While previous research has noted the presence of various "r" sounds (allophones) in this dialect, much of this work has relied on auditory impressions rather than instrumental acoustic data. This lack of detailed phonetic evidence makes it difficult to fully understand the complex variations in how the "r" sound is produced, especially considering the high degree of dialectal variation and potential influence from Italian. This paper aims to fill this gap by providing a quantitative acoustic analysis of /r/ allophones in Tyrolean, focusing on their production by female speakers.
Methods
The researchers collected speech data from six native Tyrolean female speakers. They created a corpus of target words designed to elicit /r/ in various phonetic contexts: onset and coda positions, simple syllables and clusters, and word-initial, medial, and final positions. The recordings were made using a Sennheiser ME2 microphone and processed using the Articulate Assistant Advanced software.
A key methodological aspect was the detailed annotation process. After initial automatic forced alignment using the WebMAUS tool, one of the authors manually refined and classified each /r/ segment. This involved identifying the manner of articulation (trill, tap, approximant, fricative, vocalization, deletion), place of articulation (uvular, alveolar), and voicing. To capture finer details of articulation, an additional "ac_label" tier was created to segment the constriction and aperture gestures within each /r/ sound.
Acoustic analysis focused on two main measures: duration of the constriction phase and Harmonics-to-Noise Ratio (HNR). Linear mixed-effects models were employed to statistically analyze these measures, accounting for individual speaker differences and repetition effects.
Datasets
The study utilized a newly compiled corpus of 69 target words in the Tyrolean dialect, designed to elicit /r/ in a wide range of phonetic contexts. This corpus was double-checked by native Tyrolean speakers for accuracy and recognizability. The data was collected from six female Tyrolean speakers.
Key Insights and Ideas
- High Intra-Speaker Variability: The most striking finding is the significant variation in how individual speakers produce the /r/ sound, even within the same phonetic context. This suggests that simple auditory descriptions are insufficient to capture the full phonetic reality.
- Dominance of Uvular Fricatives: The uvular fricative (/f/) emerged as the most frequent and context-independent allophone, appearing in both onset and coda positions. This aligns with previous impressionistic reports.
- Acoustic Measures Differentiate Variants: Duration and HNR proved to be valuable acoustic correlates for distinguishing between different /r/ allophones, particularly between approximants and voiced fricatives.
- Vocalization and Deletion in Coda: The phenomena of /r/-vocalization and deletion were observed exclusively in coda positions, suggesting specific phonotactic constraints on these variants.
- Uvular Trills Avoid Word-Final Position: Uvular trills were notably absent in word-final positions, indicating another contextual restriction.
- Vocalic Fragments: The insertion of vocalic fragments (aperture gestures) was frequently observed, especially with taps, highlighting a common articulatory strategy.
Key Results
- The study confirmed the prevalence of uvular place of articulation for /r/ in Tyrolean, as expected.
- The uvular fricative was the most frequent allophone, appearing in both onset and coda positions.
- Fricatives showed a significantly longer constriction phase in coda position (67.83 ± 2.26 SE) compared to onset position (47.88 ± 1.99 SE).
- The approximant variant had the highest HNR.
- Voiced fricatives lowered HNR by approximately 6.44 dB compared to approximants, while voiceless fricatives lowered it by about 11.23 dB.
Limitations and Future Work
The authors acknowledge that this is a preliminary contribution and that the sample size (six speakers) is relatively small. They plan to address this by including more speakers to better capture intra- and inter-speaker variation. Future work will also involve analyzing additional acoustic features and comparing these findings with articulatory data obtained through ultrasound tongue imaging. The study also notes the potential influence of Italian on the dialect, particularly in one speaker's production, and plans to explore this further using sociolinguistic data from a questionnaire.
Summary updated