Pronunciation errors and variation in second language speech:
a question of lexical representation?

Our study focuses on Quebec francophone pronunciation errors involving English consonant phonemes, the tendency being: i) to substitute /t d/ for /θ ð/ (I think that → I tink dat) (Brannen, 2011; John, 2011); and ii) to delete or even epenthesize /h/ (happy → _appy; ankle → hankle) (Mah, 2011; John, 2009). First, we investigated whether francophones can in fact perceive /θ ð/ and /h/. If not, it follows that their lexical representations lack these phonemes. In this case, a second problem concerns variation: how is it that francophones are at times able to produce targetlike /θ ð/ and /h/? We hypothesize that variable production stems from the development in the second language lexicon of dual representations for individual items (e.g., both accurate think and inaccurate tink), which then compete for selection at the moment of speaking.

50 adult francophones and 11 anglophones completed two auditory tasks. The first task tested perception of /θ ð/ and /h/ using 450 trials in an oddball paradigm (e.g, think-think-think-tink vs think-think-think-think). Participants indicated whether the final word was the ‘same’ or ‘different’. The task included easy (fan-pan) and difficult (thank-tank, eat-heat) contrasts to verify perceptual assimilation (Best, 1994), according to which /θ ð/ should be perceived and mentally represented as /t d/; /h/ should simply not be detected. Under this view, inaccurate pronunciations are in fact accurate realizations of the stored form. While the view accounts for pronunciation errors, it encounters difficulty with the phenomenon of variation (e.g., alternation between output tink and think for target ‘think’).

In the second task (150 sentences with or without easy/difficult substitutions), participants indicated whether the sentence made sense or not (e.g., My legs are very thin versus very tin). The aim was to test our proposal (John, 2017) that variation in pronunciation errors is due to competition between dual lexical representations which arise during the course of acquisition. According to this hypothesis, participants should have greater difficulty detecting substitutions they themselves actually make (thin → tin) than the reverse substitutions (tie → thigh).

For both tasks, ANOVA results revealed significant differences in accuracy between easy and difficult contrasts for francophone but not anglophone participants. However, considerable cross-participant variation among francophones is hard to reconcile with perceptual assimilation. Also, contrary to our dual representations proposal, differences between /θ ð/ conditions in the second task were not significant. Alternate explanations will be proposed and discussed.