Abstract
This study investigates the mental representation of speech sounds, specifically focusing on whether the lexicon stores exhaustive phonetic detail or utilizes a sparse, underspecified representation. Using Bengali as a case study, researchers examined the processing of underlying nasal vowels (NV), oral vowels (OV), and nasalized oral vowels (NC). Results from a cross-modal forced-choice priming task support the Featurally Underspecified Lexicon (FUL) model, suggesting that predictable phonetic features, such as nasality induced by a neighboring consonant, are not stored in the mental lexicon.
Introduction
A central debate in phonology concerns the nature of lexical representations. The Surface Representation Hypothesis posits that every phonetic detail is stored (e.g., a nasalized vowel is stored as [+nasal]). Conversely, the Underspecification Hypothesis suggests that only distinctive or unpredictable features are stored.
Bengali provides a unique testing ground for these theories due to its three-way vowel distinction:
- Underlying Nasal (NV): Inherently nasal vowels (e.g., cãd 'moon').
- Oral (OV): Plain oral vowels (e.g., cal 'rice').
- Nasalized Oral (NC): Vowels nasalized predictably by a following nasal consonant (e.g., cãn 'bath').
Methods
The researchers conducted a cross-modal forced-choice priming experiment with native Bengali speakers. Participants heard a truncated CV fragment (the prime) and were immediately presented with two full words on a screen (the targets). They were tasked with selecting the word that matched the audio snippet as quickly as possible.
The stimuli consisted of 14 triplets and 28 doublets. According to the FUL model, a "Match" occurs when the acoustic signal aligns with a stored feature, while a "No-Mismatch" occurs when the signal contains a feature (like nasality) for which the lexical entry is underspecified (blank).
Results
The experimental data revealed significant asymmetries in reaction times (RT) and accuracy across vowel types. Nasal Vowel (NV) primes were recognized most efficiently, with the highest accuracy and fastest reaction times.
In the "Match Only" condition, NV primes yielded an average RT of 549 ms and an accuracy of 66.9%. In contrast, NC and OV primes were significantly slower, at 581 ms and 595 ms respectively, with lower accuracy rates (55.5% and 57.3%). These findings suggest that while the brain accepts nasalized-oral vowels as a "no-mismatch" for oral lexical entries, the lack of a specific [NASAL] feature in the lexicon for these vowels results in a processing delay.
In the "Neither Only" condition—where no perfect match was available—participants tended to follow the phonetic signal (e.g., choosing a nasal target for a nasalized prime), but the underlying latency patterns remained consistent with the underspecification hypothesis.
Discussion
The results provide strong evidence for the FUL model. The brain appears to prioritize efficiency by ignoring predictable phonetic features. Because nasality in NC vowels is predictable from the following consonant, it is not stored in the mental lexicon. This "sparse" representation allows for flexible word recognition but introduces measurable latencies when the acoustic signal contains features that are not explicitly specified in the stored representation.
Conclusion
This study demonstrates that the mental representation of Bengali vowels is underspecified for predictable nasality. The observed processing asymmetries between inherent and predictable nasalization highlight the sophisticated, efficient nature of human speech perception and lexical storage.
Summary updated