TL;DR
This paper investigates how non-native English listeners learn to adapt to a vowel shift in an interactive conversation. They found that even with limited exposure and in a cognitively demanding L2 setting, listeners can adjust their perception based on lexical information, demonstrating the robustness of lexically guided perceptual learning.
What Problem Does This Paper Address?
The ability of our speech perception system to adapt to the vast variability in spoken language is crucial. A key area of research is "lexically guided perceptual learning," where listeners use word meaning to adjust how they interpret ambiguous speech sounds. While this has been well-studied for consonants in native speakers using passive listening tasks, it's less clear how it applies to vowels, in non-native (L2) listeners, or in more dynamic, interactive settings. Vowel perception is particularly interesting because subtle differences often distinguish dialects, making adaptation vital for communication. This paper aims to bridge this gap by examining whether L2 listeners can learn a vowel shift in an interactive dialogue, extending previous findings to a new domain and context.
Methods
The researchers designed an interactive computer game called "Code Breaker," where participants played a puzzle-solving game with a pre-recorded interlocutor. This allowed them to control the phonetic input.
- Interactive Game: Participants played a two-player game in English. The game involved solving pattern recognition puzzles and following oral instructions.
- Controlled Vowel Shift: The interlocutor's speech featured a specific vowel shift: the /e/ sound was consistently pronounced as /1/. This shift was embedded in critical target words.
- Lexical Constraint: Two conditions were used:
- Lexically Constraining Condition: The target words (with the /1/ pronunciation of /e/) were presented alongside minimal pair competitors that differed only in a consonant (e.g., "lesson" vs. "lemon"). This forced participants to interpret the /1/ sound as representing the /e/ sound to make sense of the words.
- Control Condition: The target words were presented with competitors that differed in the vowel itself (e.g., "lesson" vs. "listen"). This provided no lexical pressure to reinterpret the vowel.
- Phonetic Categorization Task: A pre-test and post-test were administered using a 12-step continuum of synthesized vowels ranging from the speaker's standard /e/ to /1/. Participants had to categorize these sounds to measure shifts in their phonemic boundaries.
Datasets
Not discussed in the paper. The stimuli were synthesized for the experiment.
Key Insights and Ideas
- L2 Listeners Adapt to Vowel Shifts: The study demonstrates that non-native English listeners can indeed learn to adapt their perception of vowels in response to a systematic shift, even when the shift is subtle and presented in an interactive L2 context.
- Lexical Guidance is Powerful: The "lexically constraining" condition, where word meaning forced an interpretation of the shifted vowel, led to significantly more adaptation compared to the control group. This highlights the crucial role of lexical information in recalibrating phonetic perception.
- Interaction is Not a Barrier: Performing this learning within an interactive dialogue, which involves speech production and higher cognitive load, did not prevent lexically guided perceptual learning. This suggests the learning mechanism is robust.
- Subtlety of Vowel Perception: The observed adaptation was relatively subtle. This is attributed to the continuous nature of vowel perception (less categorical than consonants) and a general bias towards perceiving sounds closer to the /1/ end of the continuum.
- Unexpected General Bias: Both groups showed a slight shift towards interpreting sounds as /1/ in the post-test, regardless of the condition. This unexpected finding suggests that exposure to the speaker's voice or the interaction itself might have influenced vowel perception more broadly, potentially due to asymmetries in vowel perception.
Key Results
- Code Breaker Game: Participants in the lexically constraining condition almost always chose the critical target word (98.8% accuracy), indicating they were successfully interpreting the /1/ pronunciation as /e/. In contrast, the control group rarely chose the target word (4.6% accuracy).
- Phonetic Categorization:
- Both groups showed an overall tendency to shift towards /1/ responses in the post-test.
- Crucially, the lexically constraining group showed less of this /1/-shift compared to the control group. This means they were more likely to label the ambiguous sounds as /e/ in the post-test, indicating successful adaptation to the interlocutor's vowel shift.
- The three-way interaction between vowel step, test time, and condition was significant, confirming that the lexically constraining condition attenuated the post-test shift towards /1/ responses, particularly for vowels closer to the /1/ end of the continuum.
Limitations and Future Work
- Limited Exposure: The study used only 16 critical words to demonstrate the vowel shift. The authors suggest that increasing the amount of evidence for the shift could lead to stronger adaptation.
- Vowel vs. Consonant: The subtle nature of the adaptation might be due to vowels being less categorical than consonants.
- General Bias: The unexpected general shift towards /1/ responses in both groups needs further investigation. The authors speculate it could be due to exposure to the speaker's voice or inherent asymmetries in vowel perception.
- Future Directions: The authors propose future research could compare the effect size for native vs. non-native listeners, investigate different types of sounds within the same speaker, and explore listening conditions with varying cognitive loads. They also suggest that using a "ventriloquist paradigm" (as in Felker et al., 2019) could allow for more experimental control over phonetic input in naturalistic settings.
Summary updated