TL;DR
This study reveals that the speed at which someone speaks significantly influences how listeners interpret ambiguous sounds in spoken words. Even with identical acoustic input, listeners' real-time processing of words is affected by the surrounding sentence's speech rate, suggesting that current spoken word recognition models need to incorporate this factor.
What Problem Does This Paper Address?
The field of spoken word recognition has largely overlooked the impact of speech rate, despite its known effects on phonetic production (like coarticulation and assimilation). While listeners are adept at adjusting for different speaking speeds, existing computational models often don't account for this. This paper investigates whether speech rate influences the perceptual interpretation of lexically ambiguous forms that arise from phonetic assimilation, specifically coronal place assimilation, where a sound like /n/ can sound like /m/ when followed by a labial sound (e.g., "run picks" sounding like "rum picks"). The authors aim to see if listeners can compensate for this assimilation based on the speech rate of the carrier sentence, similar to how they compensate for coarticulation.
Methods
The researchers employed a visual-world eye-tracking paradigm, a standard technique in psycholinguistics for studying real-time language processing. Participants listened to pre-recorded sentences and looked at a display of four pictures. Their eye movements were tracked to infer which word or meaning they were attending to.
The core manipulation involved cross-splicing auditory stimuli. They created target sentences where a noun ending in a coronal nasal was followed by a verb starting with a labial (e.g., "Vone peeks"). The crucial aspect was that the carrier phrases surrounding these target sequences were manipulated to have different speech rates (fast, medium, slow), while the acoustic material of the target sequence itself remained identical across conditions. This allowed them to isolate the effect of the surrounding speech rate.
Participants were asked to click on the picture that best matched the sentence they heard. The stimuli included nonword noun-verb pairs (e.g., Vone/Vome, Kine/Kime, Shoon/Shoom) to control for lexical neighborhood effects and plausibility.
Datasets
The study used a set of 36 target sequences and 50 filler items. The nonword stimuli were drawn from the ARC Nonword Database [10]. The authors controlled for various linguistic properties of these nonwords, such as neighborhood density and phonological neighbors. No new datasets were introduced; existing resources were utilized and manipulated.
Key Insights and Ideas
- Speech Rate as a Perceptual Cue: The most striking finding is that speech rate isn't just about timing; it actively influences how listeners interpret phonetically ambiguous sounds.
- Compensation for Assimilation: Listeners appear to use speech rate as a cue to compensate for assimilation. When speech is faster, they are more likely to interpret the sound as the unassimilated (coronal) form, and when speech is slower, they are more likely to interpret it as the assimilated (labial) form.
- Implications for Models: This work directly challenges existing models of spoken word recognition that do not explicitly account for speech rate, highlighting the need for their revision.
- Beyond Coarticulation: While previous work showed listeners compensate for coarticulation based on speech rate, this study extends that finding to assimilation, suggesting a broader role for speech rate in phonetic perception.
Key Results
- Overall Trend: In the fast and medium speech rate conditions, participants were more likely to look at the picture representing the unassimilated form (e.g., "Vone") from the onset of the alien name. In contrast, in the slow rate condition, participants increasingly looked at the picture representing the assimilated form (e.g., "Vome").
- Statistical Significance: ANOVAs revealed significant effects of speech rate on eye movements, particularly in later time windows (1100-1400ms), where the slow rate condition showed a significantly higher proportion of looks to the assimilated form compared to the medium rate.
- Individual Pair Consistency: While there were minor variations across the different nonword pairs (Vone/Vome, Kine/Kime, Shoon/Shoom), the general pattern held: faster speech rates favored the unassimilated form, and slower rates favored the assimilated form.
Limitations and Future Work
- Determining the Mechanism: The authors acknowledge that their study doesn't definitively determine which underlying mechanism (e.g., gesture-based theories, general perceptual learning, exemplar models, or TRACE model modifications) best explains these findings. This remains an important question for future research.
- Modest Aims: The study's primary goal was to demonstrate the need for including speech rate in models, rather than proposing a specific new model.
- Future Directions: Future work could focus on differentiating between various theoretical accounts and exploring how speech rate interacts with other phonetic and contextual factors in spoken word recognition.
Summary updated