Supervised and unsupervised learning of multidimensionally varying non-native speech categories

Abstract

Learning to recognize the contrasts of a language-specific phonemic repertoire can be viewed as forming categories in a multidimensional psychophysical space. Research on the learning of distributionally defined visual categories has shown that categories defined over 1 dimension are easy to learn and that learning multidimensional categories is more difficult but tractable under specific task conditions. In 2 experiments, adult participants learned either a unidimensional or a multidimensional category distinction with or without supervision (feedback) during learning. The unidimensional distinctions were readily learned and supervision proved beneficial, especially in maintaining category learning beyond the learning phase. Learning the multidimensional category distinction proved to be much more difficult and supervision was not nearly as beneficial as with unidimensionally defined categories. Maintaining a learned multidimensional category distinction was only possible when the distributional information that identified the categories remained present throughout the testing phase. We conclude that listeners are sensitive to both trial-by-trial feedback and the distributional information in the stimuli. Even given limited exposure, listeners learned to use 2 relevant dimensions, albeit with considerable difficulty.

Automatic pronunciation error detection in non-native speech: The case of vowel errors in Dutch

“…Vowel production data analyzed in Neri et al (2006) are in line with the vowel perception data reported on in Goudbeek et al (2008) in suggesting that distinctions based on two dimensions are problematic for DL2 learners. For example, in Neri et al (2006), DL2 learners with different language background appeared to confuse /A/ with /a+/, /I/ with /i/ and /O/ with /o+/.…”

“…The difficulties experienced by DL2 learners in perceiving Dutch vowels appear to be connected to the relation between the Dutch vowel system and that of their L1 and in particular to how L2 vowels map on to vowels in the native phonology (Escudero and Boersma, 2004;Goudbeek et al, 2008). In general, distinctions based on dimensions that are not relevant in the L1 are likely to be more difficult than distinctions that hinge on cues that are exploited in the native phonology.…”

“…For example, in their study on Dutch vowels, Goudbeek et al (2008) found that Spanish listeners had more difficulties in acquiring duration based distinctions, which are not exploited in their native phonology, than English listeners who are familiar with such distinctions. In addition, Goudbeek et al (2008) found that learning a unidimensional distinction like the one between the Dutch vowels /Y/ and /ø+/ as in the Dutch words "fut" and "feut," which differ essentially along the duration dimension, was easier for nonnative listeners than acquiring a distinction based on two dimensions, like that between the Dutch vowels /y/ and /ø+/ as in the Dutch words "fuut" (/fyt/) and "feut" (/fø+t/), which differ with respect to F1 and duration, even if both dimensions are employed in the native phonology.…”

“…In addition, Goudbeek et al (2008) found that learning a unidimensional distinction like the one between the Dutch vowels /Y/ and /ø+/ as in the Dutch words "fut" and "feut," which differ essentially along the duration dimension, was easier for nonnative listeners than acquiring a distinction based on two dimensions, like that between the Dutch vowels /y/ and /ø+/ as in the Dutch words "fuut" (/fyt/) and "feut" (/fø+t/), which differ with respect to F1 and duration, even if both dimensions are employed in the native phonology.…”

Auditory categories with separable decision boundaries are learned faster with full feedback than with minimal feedback

“…helps the learner evaluate alternative hypotheses that could have led to a correct response (Ashby and Maddox, 2011). In the auditory domain, the presence of feedback has been shown to enhance speech category learning in two investigations, but full and minimal feedback conditions could not be compared, as there were only two possible categories (Goudbeek et al, 2008;McCandliss et al, 2002). Full and minimal feedback conditions have been compared in speech category learning with four categories, but generalizability was limited, as the relevant dimensions had not been constrained by the experimenters (Chandrasekaran et al, 2014b).…”