Nasal coarticulation and contrastive stress

Georgia Zellou, Rebecca ScarboroughInterspeech 2012

TL;DR

This study investigates how contrastive stress affects nasal coarticulation in English, revealing that stressed words exhibit less nasal coarticulation but maintain neighborhood density effects. The findings suggest that contrastive stress primarily influences speech through duration, while other forms of hyperarticulation, like those driven by lexical neighborhood density, involve more explicit spectral reorganization.

What Problem Does This Paper Address?

The paper tackles a seemingly contradictory set of findings in speech production research. On one hand, contrastive stress (emphasizing a word) is often associated with hyperarticulation – clearer, more distinct speech, which can lead to reduced coarticulation (where sounds influence each other). On the other hand, increased coarticulation, particularly nasal coarticulation, has been observed in words from phonologically dense neighborhoods (words with many similar-sounding competitors). This raises a key question: how do these two forces – contrastive stress and lexical neighborhood density – interact to influence nasal coarticulation? Do they work together, or does one override the other? The authors aim to disentangle these effects to understand the distinct mechanisms underlying different types of speech adjustments.

Methods

The researchers employed a controlled experimental design to elicit and measure speech production. They selected English words with specific phonological structures (Vowel-Nasal sequence and Nasal-Vowel sequence) and categorized them by their phonological neighborhood density (high vs. low). These words were embedded in carrier sentences designed to elicit both contrastive stress (indicated by capitalization and context) and unstressed conditions.

Key measurements included:

  • Vowel Duration: The length of the vowel segment.
  • Nasal Consonant Duration: The length of the nasal consonant segment.
  • Nasality (A1-P0): A measure quantifying the degree of nasalization. This is derived from the amplitude difference between the first formant (F1) and the first nasal formant (P0), where a smaller A1-P0 value indicates greater nasalization.

Statistical analysis involved fitting linear mixed-effects models to the acoustic data, accounting for individual speaker variability and other potential influences.

Datasets

The study used a custom corpus of 32 monosyllabic real English words with a VN sequence and 26 monosyllabic real English words with an NV sequence. These words were selected from the Hoosier Mental Lexicon and were balanced for familiarity, word frequency, vowel height, coda size, and adjacent sonorants. Neighborhood density was calculated using frequency-weighted neighbor counts from the Hoosier Mental Lexicon. No new benchmark datasets were introduced; rather, existing lexical resources were utilized to categorize stimuli.

Key Insights and Ideas

  • Stress vs. Neighborhood Density: The study highlights a potential divergence in how contrastive stress and lexical neighborhood density influence speech. While both can lead to hyperarticulation, they might operate through different acoustic mechanisms.
  • Duration as the Hallmark of Stress: Contrastive stress appears to primarily manifest as increased duration (longer vowels and consonants), leading to reduced coarticulation.
  • Spectral Reorganization for Neighborhoods: Lexical neighborhood density, on the other hand, seems to be associated with more explicit spectral adjustments and increased coarticulation, even when duration is controlled.
  • Coarticulation's Role in Lexical Access: The paper suggests that increased coarticulation in high-density neighborhoods might not be a sign of "sloppy" speech but rather a strategy to enhance intelligibility and aid lexical access by providing predictable acoustic cues.
  • Subconscious vs. Conscious Adjustments: The findings are interpreted in light of conscious vs. subconscious speech adjustments, drawing parallels between contrastive stress and simulated "clear speech" where duration increases and nasality decreases, versus more subconscious, listener-directed clear speech where nasality might increase.

Key Results

  • Vowel Duration: Stressed vowels were significantly longer (approx. 25 ms longer) than unstressed vowels. High neighborhood density (Hi ND) vowels were also longer (approx. 27 ms longer) than low neighborhood density (Lo ND) vowels.
  • Nasal Consonant Duration: Stressed nasal consonants were longer (approx. 21 ms longer) than unstressed ones. Hi ND nasal consonants were also longer (approx. 23 ms longer) than Lo ND ones.
  • Nasality (A1-P0): Stressed vowels were less nasal (higher A1-P0) than unstressed vowels. Hi ND words were more nasal (lower A1-P0) than Lo ND words. Anticipatory coarticulation (VN) was more nasal than carryover coarticulation (NV).

Crucially, while stress reduced nasality, the effect of neighborhood density on nasality was maintained.

Limitations and Future Work

  • Limited Stimuli: The study used monosyllabic words, which may not fully represent the complexities of polysyllabic words or natural speech.
  • Specific Type of Stress: The focus was on sentence-level contrastive stress. Future work could explore the interaction of lexical stress (stress within a word, independent of sentence context) with neighborhood density.
  • Interaction of Stress and ND: While the study found that stress reduced nasality and ND increased it, the interaction between these two factors on nasality was not significant. Further research could explore how these effects might combine or compete in more nuanced ways.
  • Future Directions: The authors suggest investigating the interaction between neighborhood density and lexical stress to further differentiate between durational and spectral reorganization as markers of different speech production strategies. They also propose exploring the claim that contrastive stress involves durational reorganization while lexically-specific facts involve spectral reorganization.

Summary updated

BibTeX
@inproceedings{zellou2012nasal,
  author = {Georgia Zellou and Rebecca Scarborough},
  title = {{Nasal coarticulation and contrastive stress}},
  booktitle = {Interspeech},
  year = {2012},
}

Loading related content...

Figure