The acquisition of a sound system is an integral component of second language (L2) communication, yet it is one of the most difficult skills to teach and is therefore largely ignored in L2 classrooms (Derwing, 2010). In laboratory settings, phonetic training studies have typically examined syllables, rather than words, with no referential meaning. Support for this decontextualization of the stimuli and the subsequent distributional learning that takes place has come from findings that word learning impedes phonetic learning, especially when involving minimal pairs (Feldman, Griffiths, Goldwater, & Morgan, 2013; Hayes-Harb & Masuda, 2008). This advantage for distributional learning over minimal pair learning, however, has not been demonstrated with regard to generalizing to untrained, analogous contexts.
To investigate the role of the lexicon in L2 phonetic learning, the experiment in this dissertation trained participants on an artificial language with a different VOT category boundary than that of their first language, English. The experiment featured a between-subjects design in which participants were exposed to one of two training conditions. In the explicit, minimal pair-based condition, participants learned fine-grained VOT differences through voiced-voiceless minimal pairs (bilsu/pilsu) that illustrated a VOT category boundary of around 0 ms, providing explicit evidence for the phones’ contrasting phonological statuses. The implicit condition served to determine whether distributional learning could take place without explicit information from minimal pairs. The implicit condition’s lexicon contained no minimal pairs to illustrate a direct meaningful relationship between voiced/voiceless pairs of phones (binsu/pilsu). Lexical learning was measured by accuracy scores from the training phases, which took place across five different days. Participants’ approximation to the new category boundary was measured by a discrimination task.
Most of the predictors examined showed similar effects in both groups, but the explicit condition group alone improved in perceiving prevoicing, providing support for the minimal pair (Maye & Gerken, 2000) and noticing (Schmidt, 2012) hypotheses. Both groups successfully generalized to analogous contexts, even outperforming the contexts on which they were trained. This may be attributable to overtraining and novelty effects. The data suggest that, when invoking the lexicon, minimal pairs can help, rather than hinder, L2 phonetic learning.
Congratulations, Dr. Neumann!