"Sink positive": Experience with foreign accents influences native and non-native word recognition

Spoken language contains extensive variability in pronunciation. Effects of mispronunciations (e.g., /s/ instead of / production in theft) on word recognition are often explained in terms of perceptual similarity: the less similar mispronunciations are to target words, the more lexical activation is disturbed (Connine, Blasko, & Titone, 1993). Recent research suggests that recognition of phonological variants is also modulated by production frequency (e.g., the nasal flapped variant in a production of gentle in American English): frequent variant forms are recognized more easily than infrequent ones (Ranbom & Connine, 2008). The aim of the present study was to compare the relative influence of perceptual similarity with that of production frequency on native and non-native recognition of mispronounced words in foreign -accented speech. We used th-mispronunciations that occur regularly in foreign-accented English with varying frequency for different speaker groups. In a production study, we showed that while Dutch speakers of English frequently substitute voiceless th with /t/ (e.g., /tεft/ for theft, 80% of the mispronunciations), German speakers prefer /s/ (e.g., /sεft/, 71% of the mispronunciations); the perceptually close /f/ occurs infrequently in both groups (10% and 12%). Contrary to non-native speakers, native English speakers have mainly experience with the /f/ substitute, which occurs in some English dialects. Five eye-tracking experiments then examined whether frequent or perceptually similar substitutions cause stronger lexical activation during word recognition than infrequent or less similar ones. Advanced German and Dutch learners of English as well as native English participants listened to English sentences spoken with a German or Dutch accent (e.g., "Now you will hear /tεft/"), while they were looking at a display with four printed words; displays showed a potential match/target for the mispronunciation (theft), a competitor that also differed from the speech input only in onset (left), and two unrelated distractors (kiss, mask). The time course of lexical activation was measured as a function of amount of looks to printed th-words after hearing mispronounced words with a /t/, /s/, or /f/ substitute. Irrespective of the heard accent, all three substitutes always led to significantly more fixations to the th-words than to any of the competitors or distractors. This suggests that listeners can recognize mispronounced words, even when the substitutes occur infrequently in their own non-native speech. Furthermore, for Dutch listeners th-words were fixated over time most often when hearing /t/-substitutes, and for German listeners when hearing /s/-substitutes, while /f/ never outperformed the accent-specific dominant substitute. Native listeners, however, preferably fixated th-words after the perceptually similar substitute /f/. A correlation analysis with fixation probabilities and participants' production data did not show any effects, suggesting that individual productions do not necessarily drive the observed effects. The results show that word recognition is influenced by linguistic experience with mispronounced instances of words that vary for different foreign-accents; and effects of production frequency can outweigh perceptual similarity in the recognition of non-native mispronounced words