Semantic facilitation in bilingual everyday speech comprehension

Marco van de Ven, Benjamin V. Tucker, Mirjam ErnestusInterspeech 2010

TL;DR

This study investigates how bilinguals process spoken English, specifically their use of subtle semantic cues in both clear and acoustically reduced speech. The findings reveal that while bilinguals can use these cues, they are generally less sensitive to them than native speakers, particularly in reduced speech, suggesting a potential challenge in everyday conversational comprehension.

What Problem Does This Paper Address?

Everyday speech is often acoustically messy, with words being shortened or segments missing. Native listeners can leverage both acoustic and semantic information to understand these reduced forms, but they rely more heavily on semantics when speech is clear. However, how non-native speakers, particularly bilinguals, navigate this complex landscape of reduced speech and subtle semantic cues remains less understood. This paper specifically addresses the question of whether late Asian-English bilinguals can effectively use subtle semantic contextual information to aid their comprehension of both clear and acoustically reduced English speech, and how this compares to native English speakers. The authors highlight that previous studies often used highly predictable sentences, whereas everyday speech relies on more nuanced semantic relationships, which this study aims to investigate.

Methods

The researchers employed two auditory lexical decision experiments with implicit semantic priming. In these tasks, participants heard word pairs and had to decide if the second word was a real word. The key manipulation was the semantic relatedness between the words in a pair, measured using Latent Semantic Analysis (LSA). LSA quantifies semantic similarity based on word co-occurrences in large text corpora, providing a fine-grained measure of relatedness.

  • Experiment 1: Replicated a previous study using clear, unreduced speech.
  • Experiment 2: Used acoustically reduced speech, where words were shorter and had missing segments.

The study compared the reaction times (RTs) of late Asian-English bilinguals to those of native English speakers. Linear mixed-effects models were used to analyze the RT data, accounting for various predictors like word frequency, LSA scores, and participant characteristics.

Datasets

The study utilized existing datasets and materials:

  • Stimuli: Word pairs were drawn from the materials used in a previous study ([2]). These consisted of 154 target nouns forming 77 word pairs with varying LSA scores (mildly to highly related). The target nouns also had a wide range of word frequencies.
  • Word Frequencies: Estimated from the spoken section of the Corpus of Contemporary American English (COCA) [13].
  • Recordings: Speech stimuli were recorded by a male speaker of Canadian English.

No new datasets were introduced in this paper.

Key Insights and Ideas

  • Bilinguals can use subtle semantic cues: Contrary to some prior assumptions, bilinguals in this study demonstrated sensitivity to semantic relatedness, particularly for lower-frequency words in clear speech. This suggests their L2 comprehension isn't solely reliant on direct translation or high-level syntactic cues.
  • Reduced speech poses a greater challenge for bilinguals: The most striking finding is that bilinguals showed significantly less semantic facilitation in reduced speech compared to native speakers. This indicates that the acoustic degradation of speech disproportionately impacts their ability to leverage semantic information.
  • LSA as a robust measure of semantic priming: The study reinforces the utility of Latent Semantic Analysis (LSA) in predicting semantic priming effects, even for subtle relationships, in both native and bilingual listeners.
  • Word frequency matters: Similar to native speakers, bilinguals benefited more from semantic cues for low-frequency words. This suggests that when words are less familiar, semantic context becomes a more crucial aid for recognition.
  • Potential for L1-L2 transfer or distinct networks: The fact that bilinguals showed similar patterns to native speakers in clear speech, despite typologically distinct languages and few cognates, raises questions about whether this sensitivity is due to direct L1-L2 transfer or the maintenance of separate semantic networks.

Key Results

  • Experiment 1 (Clear Speech): Bilinguals showed semantic facilitation, similar to native listeners, especially for low-frequency words. However, a three-way interaction indicated that bilinguals' sensitivity to semantic cues diminished towards the end of the experiment, suggesting a potential fatigue or processing load effect.
  • Experiment 2 (Reduced Speech): Bilinguals showed smaller semantic facilitation compared to native listeners. While this facilitation was not statistically significant for native listeners in this experiment, the difference between the groups suggests bilinguals struggled more with reduced speech.
  • Combined Analysis: When data from both experiments were combined, a significant interaction between native language and LSA confirmed that bilingual listeners, overall, exhibited significantly smaller semantic facilitation than native listeners. This effect was more pronounced in reduced speech.
  • General Effects: Across both experiments, listeners responded faster to unreduced than to reduced words. Word frequency and previous response speed also consistently influenced reaction times.

Limitations and Future Work

  • Generalizability of Bilingual Group: The study focused on "late Asian-English bilinguals." The authors acknowledge that other bilingual groups (e.g., early bilinguals, or those with typologically closer languages) might exhibit different patterns.
  • "Subtle" Semantic Cues: While LSA provides a fine-grained measure, the definition of "subtle" semantic cues could be further explored.
  • Face-to-Face Interaction: The current study used isolated words. Future research should investigate how bilinguals utilize semantic cues in more naturalistic, face-to-face conversations where additional contextual information (visual, discourse) is available. The authors question whether these higher-level information sources can fully compensate for reduced sensitivity to semantic cues in L2.
  • L1-L2 Transfer vs. Separate Networks: Further investigation is needed to determine the underlying mechanisms driving the observed semantic processing in bilinguals.

Summary updated

BibTeX
@inproceedings{ven2010semantic,
  author = {Marco van de Ven and Benjamin V. Tucker and Mirjam Ernestus},
  title = {{Semantic facilitation in bilingual everyday speech comprehension}},
  booktitle = {Interspeech},
  year = {2010},
}
Figure