Purpose: This study compares goodness and accentedness ratings of speech tokens rated by listeners who are naive to and aware of speakers' native language backgrounds. Listener responses to open-ended questions regarding goodness and accentedness ratings are also examined. Method: Twenty-eight monolingual speakers of U.S. English served as listeners. Listeners were presented with 5 blocks of /hVt/ tokens. Each block represented a different vowel target and comprised correct and incorrect productions from English, Spanish, Korean, and Japanese speakers. Listeners rated goodness and accentedness using a 9-point Likert scale and explained their decision-making criteria when judging goodness versus accentedness. Results: There is a high positive correlation between goodness and accentedness. Both naive and aware listeners assigned poorer ratings when judging goodness compared with accentedness, but results varied on the basis of target accuracy. Aware listeners assigned better goodness and accentedness ratings compared with naive listeners. This difference was highly statistically significant. Both accentedness and goodness ratings are susceptible to listener bias. Conclusions: Goodness and accentedness are highly correlated yet distinct measures. Goodness is more reflective of target accuracy than is accentedness. Native English tokens were affected by listener bias to a greater extent than nonnative English tokens.
Purpose: This study compares goodness and accentedness ratings of speech tokens rated by listeners who are naive to and aware of speakers' native language backgrounds. Listener responses to open-ended questions regarding goodness and accentedness ratings are also examined. Method: Twenty-eight monolingual speakers of U.S. English served as listeners. Listeners were presented with 5 blocks of /hVt/ tokens. Each block represented a different vowel target and comprised correct and incorrect productions from English, Spanish, Korean, and Japanese speakers. Listeners rated goodness and accentedness using a 9-point Likert scale and explained their decision-making criteria when judging goodness versus accentedness. Results: There is a high positive correlation between goodness and accentedness. Both naive and aware listeners assigned poorer ratings when judging goodness compared with accentedness, but results varied on the basis of target accuracy. Aware listeners assigned better goodness and accentedness ratings compared with naive listeners. This difference was highly statistically significant. Both accentedness and goodness ratings are susceptible to listener bias. Conclusions: Goodness and accentedness are highly correlated yet distinct measures. Goodness is more reflective of target accuracy than is accentedness. Native English tokens were affected by listener bias to a greater extent than nonnative English tokens.