TL;DR
This study investigates how English speakers perceive and discriminate Arabic voiceless fricatives, testing the Perceptual Assimilation Model (PAM). The findings largely support PAM's predictions about discrimination difficulty based on how non-native sounds are assimilated into native categories, but surprisingly, the time interval between sounds didn't impact discrimination as predicted by the Three Factor Model (3FM).
What Problem Does This Paper Address?
Understanding how humans learn to perceive sounds from a new language is crucial for fields like second language acquisition and speech technology. Previous research, particularly the Perceptual Assimilation Model (PAM), suggests that our native language shapes our ability to distinguish non-native speech sounds. PAM proposes three ways non-native sounds can be perceived: as belonging to two different native categories (Two-Category or TC), as belonging to the same native category but with different "goodness-of-fit" (Category-Goodness or CG), or as equally good/poor versions of the same native category (Single-Category or SC). PAM predicts discrimination accuracy will follow the pattern TC > CG > SC.
However, the Three Factor Model (3FM) adds another layer, suggesting that the time interval between sounds (Inter-Stimulus Interval or ISI) influences perception. 3FM posits that shorter ISIs allow for processing of finer acoustic details, while longer ISIs rely more on abstract phonemic information. This implies that discrimination of sounds that are very similar (like CG and SC) should be harder at longer ISIs. This paper aims to test both PAM and 3FM by examining how Australian English speakers, who don't have Arabic sounds in their native inventory, perceive and discriminate specific Arabic voiceless fricatives.
Methods
The researchers employed a two-part experimental design: a categorization task and an AXB discrimination task.
- Categorization Task: Native Arabic voiceless fricatives were presented to English monolingual participants. Participants were asked to assign each sound to an English consonant category and then rate how well it fit that category on a 7-point scale. This task is key to determining the PAM assimilation type (TC, CG, or SC) for different Arabic sound pairs.
- AXB Discrimination Task: Participants heard three sounds in sequence (A, X, B) and had to indicate if the middle sound (X) was more like the first (A) or the third (B) sound. This task was conducted with two different ISIs: 500 ms and 1500 ms. The use of varying vowel contexts within the discrimination trials was a novel approach to ensure participants were focusing on the consonant sounds themselves, rather than relying on vowel cues.
Datasets
The study utilized a custom dataset of Arabic voiceless fricatives.
- Stimuli: Six Arabic voiceless fricatives (/h/, /ħ/, /x/, /ʃ/, /s/, /sˤ/) were recorded in consonant-vowel (CV) syllables with three different vowels (/a/, /i/, /u/). These stimuli were produced by a native Arabic speaker.
- Participants: 40 monolingual Australian-English speaking students from the University of Western Sydney participated.
Key Insights and Ideas
- PAM Holds Up: The study strongly supports PAM's predictions. Discrimination accuracy was indeed highest for contrasts assimilated into different native categories (TC), followed by those assimilated into the same category with differing goodness-of-fit (CG), and lowest for those assimilated into the same category as equally good or poor versions (SC). This suggests that our native language's phonological system acts as a powerful filter for perceiving new sounds.
- ISI Doesn't Matter (Much): Contrary to the Three Factor Model (3FM), the time interval between sounds (500 ms vs. 1500 ms ISI) had little to no significant effect on discrimination accuracy. This is a surprising finding, as 3FM predicted that longer ISIs would impair discrimination, especially for more similar sound contrasts (CG and SC).
- Varying Vowels Didn't Hinder: The researchers successfully used varying vowel contexts within discrimination trials. This novel approach helped ensure participants were focusing on the consonant distinctions, and it didn't appear to negatively impact their ability to discriminate, even for the more challenging SC contrasts.
Key Results
- Categorization: Arabic glottal, pharyngeal, and velar fricatives were categorized as English /h/. The post-alveolar and alveolar fricatives were categorized as English /s/.
- Discrimination Accuracy:
- TC contrasts were discriminated with an average of 90.28% accuracy (500 ms ISI) and 91.32% accuracy (1500 ms ISI).
- CG contrasts were discriminated with an average of 86.54% accuracy (500 ms ISI) and 87.03% accuracy (1500 ms ISI).
- SC contrasts were discriminated with an average of 55.47% accuracy (500 ms ISI) and 53.07% accuracy (1500 ms ISI).
- ISI Effect: No significant main effect of ISI was found on discrimination accuracy. While there was a slight trend for performance to be better at 500 ms ISI, it was not statistically significant across the board.
Limitations and Future Work
- Task Demands: The authors suggest that the AXB discrimination task, while useful for PAM, might have higher memory demands than other tasks (like AX), potentially masking ISI effects. They propose future research using an AX task design to investigate this.
- ISI Effect Nuance: The lack of a clear ISI effect might be specific to the types of contrasts tested. The authors speculate that ISI effects might be more pronounced for SC assimilation types, particularly those that are very subtle, and that the near-chance performance on their SC contrasts might have obscured any ISI influence.
- Future Directions: The paper suggests further research to explore the impact of task design (AX vs. AXB) and to investigate different types of non-native contrasts to better understand the conditions under which ISI effects might emerge.
Summary updated