TL;DR
This study investigates why vowels in the middle of French sentences sometimes lose their voicing, a phenomenon previously under-documented. The authors found that this devoicing is common, particularly in fast speech and when vowels are temporally compressed, suggesting it's a natural phonetic consequence of speech production rather than a purely phonological rule.
What Problem Does This Paper Address?
While vowel devoicing at the end of phrases in French is a well-known phenomenon, its occurrence within phrases (phrase-medial) has been less studied. This paper aims to systematically investigate the frequency, characteristics, and conditioning factors of phrase-medial vowel devoicing in spontaneous European French. Understanding this phenomenon is crucial for a complete picture of French phonetics and phonology, potentially shedding light on how phonetic processes can evolve into phonological ones.
Methods
The researchers analyzed spontaneous speech data from a corpus of casual French. They focused on specific phonetic contexts known to favor devoicing: high vowels preceded by voiceless consonants. Measurements included:
- Vowel duration: The time the vocal tract was unobstructed.
- Voicing duration: The portion of the vowel with vocal fold vibration.
- Closure duration: The duration of consonant closure.
- Voice Onset Time (VOT): The time between consonant release and vocal fold vibration.
- Prosodic context: Annotation of phrase boundaries (Accentual Phrase and Intonational Phrase) and distance to these boundaries.
- Speech rate: Calculated based on syllable count and duration within an Accentual Phrase.
Statistical models (logistic and linear regression) were used to determine which of these factors influenced the likelihood and extent of complete and partial vowel devoicing. They also examined spectral characteristics of preceding consonants to assess potential increases in coarticulation.
Datasets
The study utilized the Nijmegen Corpus of Casual French [12]. This corpus contains spontaneous speech data. The authors randomly extracted tokens from this corpus, focusing on specific phonetic contexts (high vowels preceded by voiceless consonants) and ensuring a sufficient number of function and content words were analyzed. No new benchmark datasets were introduced.
Key Insights and Ideas
- Phrase-medial devoicing is frequent: Contrary to previous limited observations, this study reveals that phrase-medial vowel devoicing is a common occurrence in spontaneous French, affecting over a third of the target vowels.
- Temporal compression is key: Devoicing is strongly linked to reduced syllable and vowel durations, suggesting it's a natural consequence of speech being produced faster.
- Aerodynamics likely play a role: The authors propose that the aerodynamic conditions required for voicing are harder to maintain for high, short vowels, especially when combined with rapid speech.
- Not just phonology: While some evidence of increased consonant-vowel coarticulation was found (especially with fricatives), the primary drivers appear to be phonetic, related to temporal and aerodynamic constraints, rather than a strict phonological rule.
- Continuum of devoicing: Complete and partial devoicing seem to exist on a continuum, influenced by the same phonetic factors.
Key Results
- Frequency: 36.3% of target vowels were completely devoiced.
- Conditioning Factors: Complete devoicing was more probable in:
- Faster speech rates (significant effect).
- Vowels further from the end of an Accentual Phrase (significant effect).
- Vowels following fricatives compared to stops (significant effect).
- Temporal Compression: Complete devoicing was more likely in shorter syllables. Devoiced vowels were significantly shorter than voiced vowels, indicating temporal reduction primarily in the vocalic part of the syllable.
- Coarticulation: Some acoustic evidence of increased CV coarticulation was found, particularly for syllables with fricative onsets, suggesting that gestural overlap might contribute to devoicing in some cases.
- No difference between word types: Surprisingly, no significant difference was found in the likelihood or extent of devoicing between function and content words.
Limitations and Future Work
- Articulatory Data: The authors acknowledge that direct articulatory data (e.g., from EMA or ultrasound) is needed to definitively elucidate the precise articulatory mechanisms behind this phenomenon, particularly regarding gestural overlap and truncation.
- Coarticulation Evidence: While some spectral evidence for increased coarticulation was found, it was not consistently significant across all tested contexts (e.g., stops), suggesting that larger sample sizes might be needed for these specific analyses.
- Conventionalization: The paper raises the question of whether this frequent phonetic process might eventually become conventionalized into a phonological feature or allophone in French, a direction for future research.
- Cross-linguistic comparison: Further studies could explore how this phrase-medial devoicing pattern compares to similar phenomena in other languages.
Summary updated