Signal variability in accented speech listening: Effects of expected versus unexpected segmental errors

Although L2-accented speech can hamper word recognition, L1-listeners can adapt to variability in the acoustic signal of non-native speech. It is still unclear whether listeners facilitate lexical recognition during processing by making predictions about accent-specific error patterns. In this study, we manipulate the expectedness of a segmental error (insertion and deletion of word-initial /h/) by presenting the same error in French-accented English (where /h/ errors are expected) and Russian-accented English (unexpected). L1 English listeners assess the validity of a sentence that may contain an /h/-based error (e.g., ot dogs *[ɔtdɔgz] or heagles *[higlz]) by pressing True or False. Reaction times from target word unicity point (point of disambiguation of target lexical item) to keypress are taken as a measure of word recognition speed. We compare reaction times between accents: while for French-accented English /h/-based errors are predictable, and thus not expected to slow down listeners, this isn’t the case for Russian-accented English, for which we expect /h/-based errors to slow down listeners. Preliminary data show the expected differences in reaction times between stimulus accent (French vs. Russian) and error type (insertion versus deletion), particularly for participants with high exposure to accented speech, suggesting processing accommodation to accent over time.