How abstract is phonetics?

Osamu FujimuraInterspeech 2010

TL;DR

This paper argues that phonetics, while dealing with physical speech signals, requires abstract phonological representations to be fully understood and modeled. The authors propose the C/D model as a framework for phonetic implementation, demonstrating how it can account for complex articulatory variations influenced by prosody and language-specific syllable structures, particularly in Kaingang.

What Problem Does This Paper Address?

The paper tackles the fundamental question of how abstract phonetic representations should be, especially when mapping them to concrete speech signals. It highlights the "opaqueness" of the relationship between phonological features and actual acoustic properties, suggesting that current models might be too simplistic. The authors aim to bridge the gap between abstract linguistic units (phonemes, features) and the continuous, multidimensional nature of speech production, emphasizing the need for powerful phonetic implementation algorithms that can handle language-specific variations and prosodic influences.

Methods

The core of the paper's methodology revolves around the C/D model (Control/Display model) for phonetic implementation. This model views phonetic implementation as a system that superimposes various phonetic variables, with a final, non-linear "signal generator" process that converts articulatory control variables into physico-acoustic variables. The C/D model uses "CD diagrams" to visually represent articulatory control actions, showing how control variables change over time and interact with threshold values to trigger events like articulation release or voice onset. The authors also employ a syllable-based feature analysis to represent phonological information, moving beyond traditional linear phoneme sequences. They use examples from English and Kaingang to illustrate how this model can capture complex articulatory phenomena.

Datasets

Not discussed in the paper. The paper relies on existing linguistic analyses and phonetic transcriptions of specific words and sentences in English and Kaingang.

Key Insights and Ideas

  • Phonetics is inherently abstract: The paper challenges the notion of phonetics being purely concrete and objective, arguing that abstract phonological representations are crucial for understanding speech.
  • The C/D model as a unifying framework: The C/D model provides a powerful way to describe phonetic implementation, integrating symbolic phonological information with continuous articulatory and acoustic variables.
  • Prosody significantly impacts articulation: Even in the same segmental context, vowel articulation can vary dramatically based on prosodic factors like contrastive emphasis, as demonstrated with the English example.
  • Syllable structure is key to phonetic implementation: The paper advocates for syllable-based phonological representations, showing how this approach simplifies the description of language-specific phonetic patterns, particularly in languages like Kaingang.
  • Kaingang's unique nasal default: The analysis of Kaingang highlights a fascinating linguistic feature where nasal syllables are the default, and oralization is a marked feature that affects not only vowels but also surrounding consonants.

Key Results

The paper doesn't present quantitative results in terms of accuracy or performance metrics. Instead, its key results are qualitative and descriptive:

  • Demonstration of the C/D model's explanatory power: The paper successfully uses CD diagrams to illustrate how the C/D model can represent complex articulatory gestures and their timing, such as the ambisyllabic /t/ in "splatter" and the articulatory movements in Kaingang words.
  • Explanation of Kaingang's nasal/oral distinction: The syllable-based feature analysis, within the C/D model framework, provides a coherent explanation for the default nasal nature of Kaingang syllables and how the {oral} feature influences vowel and consonant articulation.
  • Illustration of prosodic influence: Figure 1 visually demonstrates how contrastive emphasis on "six" and "America" leads to distinct tongue blade height variations for the same vowel sounds.

Limitations and Future Work

  • Limited scope of languages: While examples from English and Kaingang are used, the paper acknowledges that a more comprehensive understanding would require analyzing a wider range of languages.
  • Reliance on existing analyses: The Kaingang analysis is based on reports and personal communication, suggesting that direct empirical investigation by the authors would be beneficial.
  • Computational complexity: The paper mentions that the signal generator component of the C/D model is "exceedingly complex and computationally heavy," implying that practical implementation might be challenging.
  • Future work: The authors suggest that further exploration of syllable-based features and their implementation within the C/D model can lead to more effective descriptive frameworks for phonetics and potentially improve automatic speech recognition systems by exploiting linguistic redundancy.

Summary updated

BibTeX
@inproceedings{fujimura2010how,
  author = {Osamu Fujimura},
  title = {{How abstract is phonetics?}},
  booktitle = {Interspeech},
  year = {2010},
}
Figure