Fig 1.
F1-F2 plot of /I/ and /E/ tokens produced by 40 speakers of North-Holland Dutch and West Flemish accents (two tokens per speaker) from Adank et al. (2004).
These tokens are the stimuli used in our experiments. Black = Female Dutch, grey = Female Flemish, Dark blue = Male Dutch, Light blue = Male Flemish. The average values for each group are circled. Note that there is variation between speakers of the same sex, speakers of the opposite sex, and speakers of the same accent (as well as between accents). The average acoustic distance between vowel category, sex, accent, and Accent+Sex is shown by lines connecting the respective symbols and listed in the left bottom corner. This shows that acoustic distance in F1-F2 space (computed as the Euclidean distance in Hz) does not reliably distinguish between different types of variability, especially sex and accent, as their distance is almost identical.
Fig 2.
(a) F1 and F2 values of the average Dutch (black symbols in white circles) and average Flemish (white symbols in black circles) vowel tokens used in the present study; (b) formant ratios of the stimuli in (a).
Female tokens are plotted in larger font and larger circles. The figure shows average formant values across several tokens that were used in the experiments (namely, across 8 tokens for Speaker change, and 12 tokens for each Sex, Accent, and Accent+Sex change).
Fig 3.
Stimulus sets used in Speaker, Sex, Accent and Accent+Sex conditions used in Experiment 1 and 2 (Dutch = black symbols in white circles, Flemish = white symbols in black circles).
Each participant was given one set per condition, and tested in two conditions (either Speaker and Sex, or Accent and Accent+Sex).
Fig 4.
Classification FAM and NEW tokens by AusE and Dutch listeners (combined) in Experiment 1 when no feedback was provided.
Error bars are SEM.
Table 1.
Categorization performance of AusE and Dutch listeners during test trials in Experiment 1.
Feedback was not provided for all trials with NEW tokens and for a subset of trials with FAM tokens (used in data analyses).
Fig 5.
Percent correct categorization responses of AusE listeners in Experiment 2 when feedback was provided.
(A) Proportion of correct responses pooled across all conditions. (B) Proportion of correct responses split by condition. When listeners were switched to NEW (Block 7), accuracy in Accent and Accent+Sex conditions was significantly lower than in Speaker and Sex conditions, as indicated by * (p < 0.05). Two blocks after the switch, accuracy in the Accent+Sex condition was no longer significantly different than Speaker, Sex (indicated by #); however, Accent remained significantly lower than Speaker and Sex in that block (p = 0.003). Accuracy in the Accent condition was no longer significantly different than Speaker or Accent+Sex in the last block (indicated by +), but was still significantly lower than Sex (p = 0.002). Error bars are SEM.
Table 2.
Proportion correct responses in the different conditions.