A Computational Investigation on the Perception and Acquisition of Aspiration

Tej K. BhatiaCOLING 1973

TL;DR

This paper investigates how English speakers perceive and distinguish aspirated vs. unaspirated Hindi consonants, revealing significant perceptual interference. The study highlights that aspiration is a crucial feature for distinguishing Hindi consonants, and its misperception by English speakers can be explained by contrastive analysis and the specific phonetic features involved.

What Problem Does This Paper Address?

The paper addresses the phenomenon of aspiration in Hindi, a feature that has long intrigued phoneticians but lacked comprehensive investigation. Specifically, it aims to understand how English speakers, whose native language has a different system of aspiration, perceive and potentially confuse Hindi consonants. This is important for understanding cross-linguistic phonetic interference and for developing effective language training programs. The authors also question the absolute predictive power of contrastive analysis and explore the acquisition of perceptual cues over time.

Methods

The study employed a computational approach to analyze the perception of Hindi consonants. A context-free dataset of minimal pairs of 22 Hindi consonants was collected, covering unvoiced unaspirated vs. aspirated and voiced unaspirated vs. aspirated pairs, in initial, medial, and final positions. This data was recorded by three native Hindi speakers. A perception test tape (T2) was created, featuring randomized items and instructions. This tape was then presented to 25 English-speaking subjects in a sound-isolated laboratory. The subjects' responses were quantitatively analyzed using an IBM 360-25 computer to generate error matrices. These matrices detailed the confusion patterns, allowing for the identification of probable errors and the ranking of perceptual difficulties.

Datasets

The study utilized a custom-created dataset of 62 minimal pairs of Hindi consonants. These pairs were designed to contrast aspirated and unaspirated sounds in initial, medial, and final positions. The dataset included both meaningful and phonologically possible, but non-sensical, word pairs. The recordings were made by three native Hindi speakers. The evaluation was performed on 25 English-speaking subjects.

Key Insights and Ideas

  • Aspiration as a Key Distinguishing Feature: The study strongly suggests that aspiration is a primary feature for distinguishing Hindi consonants, and its presence or absence significantly impacts perception.
  • Perceptual Interference is Significant: English speakers exhibit considerable confusion between aspirated and unaspirated Hindi consonants, indicating a notable perceptual interference.
  • Contrastive Analysis is Predictive but Not Absolute: While contrastive analysis can predict some of the difficulties English speakers face with Hindi sounds, it doesn't fully explain all observed patterns, suggesting other factors are at play.
  • Feature Hierarchy Matters: The confusion patterns reveal a hierarchy of perceptual importance for phonetic features, with aspiration and voicing playing crucial roles.
  • Positional Differences in Perception: The perception of voiced aspirated consonants differs significantly between initial/medial positions and final positions, suggesting context-dependent perceptual strategies.
  • Palatal Series is More Confusing: The palatal series of consonants appears to be more prone to confusion for English speakers compared to other series.

Key Results

The study found that unvoiced unaspirated consonants were more frequently confused than unvoiced aspirated consonants across all positions. Voiced aspirated consonants were more confused in initial and medial positions than in final positions, where they were better recognized. The primary confusions occurred between consonant classes distinguished by a single feature (aspiration or voicing). The palatal series showed a higher rate of confusion. The analysis of probable errors indicated that confusions most frequently occurred between consonants differing by a single feature, and rarely between those differing by two features (voicing and aspiration).

Limitations and Future Work

The authors acknowledge that further research with acoustic instruments is needed to definitively support some of their hypotheses, particularly regarding the behavior of the velar voiced aspirated consonant 'gh' in final position. The study's findings on the "overcompensation" phenomenon are tentative and require more empirical support. The authors also suggest that the language training of the subjects might influence the results, and that the exact nature of the data used in previous studies makes direct comparison difficult. Future work could explore the underlying psychological processes and the role of acoustic instruments in confirming these perceptual findings.

Summary updated

BibTeX
@inproceedings{bhatia1973computational,
  author = {Tej K. Bhatia},
  title = {{A Computational Investigation on the Perception and Acquisition of Aspiration}},
  booktitle = {International Conference on Computational Linguistics},
  year = {1973},
}
Figure