Place Shift as an Autonomous Process: Evidence from Japanese Listeners

Yuriko YokoeInterspeech 2019

TL;DR

This paper investigates why listeners sometimes misperceive consonant clusters that don't exist in their native language. By testing Japanese listeners with Russian and Hebrew consonant clusters, the study finds that the "place shift" phenomenon (e.g., hearing /tl/ as /pl/) isn't driven by learned phonotactic rules, but rather by an autonomous, signal-driven process related to how listeners weigh acoustic cues. This suggests language-specific speech perception arises from individual cue weighting strategies, not innate linguistic knowledge.

What Problem Does This Paper Address?

Human speech perception is remarkably adept at processing the sounds of our native language. However, when encountering sounds or sound combinations not present in our linguistic environment, listeners can exhibit systematic misperceptions. One such phenomenon, termed "place shift," occurs when unattested consonant clusters are perceived as phonotactically legal clusters in the listener's native language. For example, French and English listeners have been shown to perceive the unattested cluster /tl/ as /kl/ (a "velar shift"). This has led to the hypothesis that this "phonological filter" is driven by top-down, language-specific phonotactic knowledge.

This paper questions this prevailing view. Specifically, it asks:

  1. Can the "place shift" phenomenon be reproduced in Japanese listeners, whose native language lacks onset consonant clusters altogether?
  2. If so, does this shift align with the "velar shift" observed in French and English, or is there a different pattern?
  3. Crucially, is this phenomenon driven by learned phonotactic rules, or by a more fundamental, signal-driven perceptual process?

Methods

The study employed a perception experiment with monolingual Japanese listeners. Participants were presented with non-word stimuli containing unattested consonant clusters, specifically coronal-lateral clusters (/tl/, /dl/) from Russian and Hebrew. These stimuli were embedded in carrier sentences and presented via computer. Listeners were instructed to identify the initial plosive consonant by selecting one of three buttons corresponding to labial (/p/, /b/), coronal (/t/, /d/), or velar (/k/, /g/) place of articulation.

To further investigate the underlying perceptual mechanisms, decision tree analyses were conducted. These analyses aimed to determine which acoustic cues (e.g., plosive duration, voice onset time, spectral properties) Japanese listeners relied on when identifying the place of articulation for these challenging stimuli.

Datasets

The study utilized two sets of stimuli:

  • Russian nonwords: Recorded by native Russian speakers, these included various consonant-vowel combinations designed to test unattested clusters.
  • Hebrew nonwords: These were identical to those used in a previous study by Hallé & Best (2007), allowing for direct comparison.

No new benchmark datasets were introduced; rather, existing stimuli were adapted for the experimental design.

Key Insights and Ideas

  • The "Labial Shift" in Japanese Listeners: Unlike the "velar shift" observed in French and English, Japanese listeners showed a strong tendency to misperceive coronal-lateral clusters as labial (/pl/, /bl/) rather than velar (/kl/, /gl/). This is a novel finding that challenges a universal explanation for the place shift.
  • Beyond Phonotactics: The consistent misperception of coronal-laterals, even by listeners with no native onset clusters, suggests that the place shift is not simply about enforcing learned phonotactic rules. If it were, listeners might be expected to default to more common or "safer" clusters, or perhaps show less systematic errors.
  • Autonomous, Signal-Driven Process: The authors argue that the place shift is an autonomous process driven by the acoustic signal itself and the listener's inherent cue-weighting strategies. When faced with ambiguous acoustic information, listeners rely on their learned weighting of different acoustic cues to make a decision.
  • Language-Specific Cue Weighting: The differences in misperception between Russian and Hebrew stimuli (more labial shifts for Russian, more velar shifts for Hebrew) suggest that while the underlying process is signal-driven, the way listeners weigh cues is influenced by their native language experience. Japanese listeners, for instance, appear to weigh certain acoustic features differently when processing Russian vs. Hebrew sounds.
  • Reaction Time as an Indicator: Longer reaction times for shifted responses compared to correctly identified responses indicate that the misperception process requires additional cognitive effort, consistent with resolving acoustic ambiguity.

Key Results

  • Japanese listeners exhibited a significant tendency to misperceive coronal-lateral clusters.
  • For Russian coronal-lateral stimuli, Japanese listeners showed a strong "labial shift" (e.g., /tl/ perceived as /pl/).
  • This labial shift was less pronounced for Hebrew coronal-lateral stimuli, with a weaker tendency towards velar shifts.
  • Decision tree analyses revealed that Japanese listeners primarily relied on formant-related acoustic cues for Russian stimuli, and durational and burst-related cues for Hebrew stimuli, suggesting flexible cue weighting strategies.
  • Shifted responses were associated with significantly longer reaction times than non-shifted responses.

Limitations and Future Work

  • Limited Stimulus Set: While the study used Russian and Hebrew stimuli, the range of unattested clusters tested was specific. Future work could explore a wider variety of phonotactically illegal clusters across more languages.
  • Focus on Japanese Listeners: The study concentrated on Japanese listeners. Expanding the research to other language groups with different phonotactic inventories would provide a more comprehensive understanding of cross-linguistic differences in cue weighting.
  • Nature of "Ambiguity": The paper posits that the place shift arises from acoustic ambiguity. Further research could delve deeper into precisely which acoustic features contribute to this ambiguity for different language pairs and listener groups.
  • Integration with Production: The study focused solely on perception. Investigating how these perceptual biases influence speech production in non-native speakers would be a valuable next step.

Summary updated

BibTeX
@inproceedings{yokoe2019place,
  author = {Yuriko Yokoe},
  title = {{Place Shift as an Autonomous Process: Evidence from Japanese Listeners}},
  booktitle = {Interspeech},
  year = {2019},
}
Figure