The Voicing Contrast in Stops and Affricates in the Western Armenian of Lebanon

Niamh E. Kelly, Lara KeshishianInterspeech 2019

TL;DR

This paper investigates the acoustic realization of the voicing contrast in stops and affricates in Western Armenian spoken in Lebanon. The findings suggest that while voiced sounds are pre-voiced and voiceless affricates have aspiration, the voiceless stops exhibit surprisingly short Voice Onset Time (VOT), aligning more with unaspirated sounds than previously described for Western Armenian. This challenges existing descriptions and may indicate influence from Lebanese Arabic or a re-framing of the contrast.

What Problem Does This Paper Address?

The study tackles the acoustic correlates of the voicing contrast in stops and affricates in Western Armenian (WA), specifically the variety spoken by a large population in Lebanon. Previous research on Armenian phonology has described a three-way contrast in Classical Armenian (voiceless aspirated, plain voiceless, voiced) and a two-way contrast in modern WA (voiceless aspirated and voiced). However, the acoustic details of this contrast, particularly in the Lebanese diaspora, have not been thoroughly examined. The paper aims to fill this gap by investigating the phonetic features, such as Voice Onset Time (VOT), closure duration, frication, and aspiration, that distinguish voiced from voiceless stops and affricates in this specific WA dialect. Understanding these acoustic details is crucial for accurate phonetic descriptions and for understanding language variation and contact.

Methods

The researchers employed a standard phonetic methodology. They recorded eight native speakers of Western Armenian from Lebanon producing nonsense words designed to elicit the target sounds. These nonsense words were created to contrast voiced and voiceless affricates and stops in both alveolar and postalveolar positions, and in word-initial and word-medial contexts. Each speaker produced each word four times within a carrier sentence, resulting in a substantial dataset of recorded speech. Acoustic analysis was performed using the Praat software, where key phonetic features like closure duration, burst, frication, aspiration, and Voice Onset Time (VOT) were meticulously labeled. Additionally, the fundamental frequency (f0) and intensity of the vowel following the target consonants were measured. Statistical analysis involved linear mixed-effects regression models to determine the influence of voicing, consonant type (stop vs. affricate), and word position on these acoustic measures.

Datasets

The study utilized a newly created dataset of nonsense words specifically designed for this research. The dataset consisted of 384 segments (8 speakers * 4 repetitions * 6 sounds * 2 word positions), with 376 segments ultimately used for analysis after excluding those with noise or speech errors. The nonsense words were: /thatha, dada, tshatsha, tſhatſha, dzadza, zadza/. These words were presented in the Armenian script.

Key Insights and Ideas

  • Voiceless Stops are Surprisingly "Unaspirated": The most striking finding is that the voiceless stops in this Lebanese WA dialect exhibit very short VOT, averaging around 20ms. This is significantly shorter than what is typically associated with aspirated stops and aligns more closely with unaspirated stops, challenging existing descriptions of WA phonology.
  • Re-framing the Contrast: The short VOT in voiceless stops suggests that the two-way contrast in WA might be achieved not through aspiration in stops, but perhaps through other cues or a re-framing of the contrast, potentially influenced by the phonology of Lebanese Arabic.
  • Affricates vs. Stops: Affricates consistently showed frication after the closure, a characteristic absent in stops. Stops also exhibited longer closure durations than affricates.
  • Voicing is Prevalent: As expected, voiced sounds were pre-voiced (negative VOT), indicating vocal fold vibration begins before the release of the consonant.
  • Word Position Doesn't Matter (for duration): Contrary to findings in some other languages, word position (initial vs. medial) did not significantly affect the duration of closure, frication, or aspiration in this study.
  • F0 and Intensity Follow Expectations: Voiceless sounds were associated with higher f0 and lower intensity in the following vowel, consistent with general phonetic tendencies.

Key Results

  • VOT: Voiceless stops had significantly higher VOT than voiced stops (p < 0.001), with an average of 20ms for voiceless stops.
  • Frication Duration: Frication was significantly affected by consonant type (longer for affricates, p < 0.001) and voicing (longer for voiceless affricates, p < 0.001). Alveolar affricates also had longer frication than postalveolar affricates (p < 0.001).
  • Closure Duration: Stops had significantly longer closure duration than affricates (p < 0.001). Voiced sounds had shorter closure duration than voiceless sounds (p = 0.002).
  • Word Position: Word position had no significant effect on any duration measures (VOT, Frication, Closure).
  • F0: F0 was significantly higher after voiceless sounds than voiced sounds (p < 0.001).
  • Intensity: Intensity was significantly lower after voiceless sounds than voiced sounds (p < 0.001).

Limitations and Future Work

The authors acknowledge that their study used nonsense words, and future research should examine natural words to see if these patterns hold. They also suggest investigating the influence of lexical stress. A key future direction is to examine heritage speakers of WA in English-speaking countries to determine if the short VOT in voiceless stops is indeed an effect of Lebanese Arabic influence, as English has a long VOT in voiceless sounds. The study also notes that while affricates had shorter aspiration than stops, further research could explore this difference more deeply.

Summary updated

BibTeX
@inproceedings{kelly2019voicing,
  author = {Niamh E. Kelly and Lara Keshishian},
  title = {{The Voicing Contrast in Stops and Affricates in the Western Armenian of Lebanon}},
  booktitle = {Interspeech},
  year = {2019},
}
Figure