Reduction of unstressed English vowels by L2 speakers with different language backgrounds
Heini Kallio & Kamil Kaźmierski
The 36th Finnic Phonetics Symposium 25-26.04.2024 Tallinn
English stress system :: Difficulties for learners
- position of stress, strength of stress contrasts & manner of production
- acoustic realizations of stress contrasts under inspection here
- reduction of unstressed vowels in terms of shrinking the vowel space area (VSA) and relative shortening
The L1 effect
English
- strong contrasts between stressed & unstressed syllables
- duration & vowel quality primary acoustic markers
Polish, Hungarian & Slovak
- relatively weak stress realizations
- little or no reduction in unstressed vowels
- some centralization effect found, but depends on vowel category (Rojczyk 2019, Markó et al. 2018, Beňuš & Mády 2010)
HU & SK use duration primarily for producing phonemic quantity distinctions → weaker role in producing prominence?
Why we do this?
Continuation of a previous study: acoustic syllable prominence estimates as prosodic proficiency predictors (Kallio et al. 2022)
- vowel quality disregarded
→ a pilot study to investigate the reduction of unstressed vowels in terms of centralization & relative duration
RQs
Does the degree of VSA shrinkage differ between English speaker groups with respect to their L1?
Does the degree of vowel shortening differ between English speaker groups with respect to their L1?
EFL groups expected to differ from native speakers in both features related to vowel reduction
Differences expected between EFL groups, e.g. in terms of vowel shortening
Speech data
- 16 speakers; 4 speakers from 4 L1s: English (Br), Hungarian, Polish, Slovak
- L2 speakers at B1 level proficiency
- Each read a narrative text consisting of 16 sentences (productions ~1.5 min)
- Read speech → more or less comparable productions of vowels in same contexts
Data preparation
Vowel segments annotated using a forced aligner, onsets & offsets corrected manually in Praat
Binary stress annotation based on standard British English transcription of the narrative text (weak forms ignored, e.g. that always transcribed ðæt not ðət)
Nasalized vowels & vowels in context with rhotics and approximants excluded
Number of vowels selected for analysis per speaker:
Stressed: 448 – 567
Unstressed: 686 – 819
Analysis
- Formant measurements extracted at 7 time points in vowel intervals with a Praat script
- Midpoint measures of F1 & F2 used here to determine VSAs
densityarea(Fruehwald 2023); density polygons for 0.8 probabilities- Formant measurements normalized using the Nearey method (vowel-extrinsic)
- Comparison of VSAs & relative duration differences of stressed & unstressed vowel productions
Results: VSAs
Degree of VSA shrinkage
English
54%
Polish
37%
Hungarian
41%
Slovak
33%
vowel_area ~ stress * l1
Degree of vowel shortening
English
30%
Polish
7.5%
Hungarian
4.5%
Slovak
7.2%
log10(duration) ~ stress * l1
Discussion
✔️ Yes, we have differences in the VSAs
- degree of VSA shrinkage larger for L1 than L2 speakers
- differences between L2 groups: shrinkage largest for Hungarian and smallest for Slovaks
- also differences in the size of stressed VSAs
→ effect of L1-specific vowel system?
- inter-speaker variation evident
Discussion
✔️ Yes, we have differences in reducing vowel durations
- all L2 speakers differ from natives
→ insufficient reduction of unstressed vowels or misplacement of word stress?
- average vowel shortening notably smaller for Hungarian EFL speakers than for the others
→ possible effect of quantity system?
Next steps
- More speakers per group to analyze
- How about other L1s such as Finnish/Estonian EFL speakers?
- Looking at vowel categories separately
- Including dynamic formant movements
Thank you!
Heini Kallio & Kamil Kaźmierski
Audio samples
“Instead, we were very surprised to see lots of different types of vegetables.”
English
Polish
Hungarian
Slovak