ABSTRACT
Listeners can adapt their perceptual categories for speech sounds in response to speakers’ unusual pronunciations. The present study tested whether this generalisation is sensitive to surface or underlying properties of speech by exploiting the devoicing of voiced stops in German. This allows us to assess whether learning on phonetically voiceless stops that are underlyingly voiced generalises to stops that share the same surface form (i.e. voiceless) or the same underlying representation (i.e. voiced). Our results showed only minimal generalisation: learning for (surface) voiceless stops in offset position that are underlyingly voiced generalises to surface and underlying voiceless stops in the same position but neither to voiced nor voiceless stops in intervocalic position. This suggests that listeners extract segments of sufficient acoustic similarity from the input and use them for generalisation of learning in speech perception. The units of perception thereby appear context sensitive rather than abstract phonemes or phonological/articulatory features.
Acknowledgements
The data for Experiment 2 were gathered while the first author was a visiting professor at the University of Tübingen. We would like to thank Anna-Maria Meck and Elaine Stix for help with collecting data.
Disclosure statement
No potential conflict of interest was reported by the authors.
Notes
1. We indicate morph ratios only with the percentage of the alveolar part, the amount of velar signal follows from that, so that a morph that is using the alveolar signal for 60% uses the velar signal to 40%.
2. The r-command was: allBFs = anovaBF(logitAlv ∼ continuum * Group + participant, data = aggregatedLogOdds, whichRandom = “participant”)