Fig. 3: Phonetic feature processing across the sequence.

A Temporal generalisation (TG) results superimposed for 10 phoneme positions. Results represent decoding performance averaged over the 14 phonetic features. From the first phoneme in a word (P1, dark red) moving forwards, and the last phoneme in a word (P-1, dark blue) moving backwards. The result for each phoneme position is shifted by the average duration from one phoneme to the next. The y-axis corresponds to the time that the decoder was trained, relative to phoneme onset. The x-axis corresponds to the time that the decoder was tested, relative to word onset. Contours represent a t-value threshold of 4 (darker lines) and 3.5 (lighter lines). Note that fewer trials are entered into the analysis at later phoneme positions because words contain different numbers of phonemes. This is what leads to the reduction of decoding strength at p5 and p-5, for example. B Decoding performance of just the diagonal axis of each phoneme position (where train times are equal to test times). The visualisation therefore represents when phonetic information is available, regardless of the topography which encodes it. We use a stack plot, such that the variance explained by different phoneme positions is summed along the y-axis. Source data are provided as a Source Data file.