The present study investigated the production of lexical stress by native speakers of English (NE), Arabic learners of English (ALE), and native speakers of Arabic (NA). In the first experiment, minimal pairs (e.g., ‘con.flict, con.’flict) were recorded by 8 native speakers of English and 16 (8 advanced and 8 beginning) learners. For comparison, a second experiment examined acoustic cues used to indicate stress in 8 native Arabic speakers. In both experiments, four acoustic cues were examined: duration, fundamental frequency, amplitude, and second formant frequency. Results showed that NE consistently used all four cues to signal stress, with longer duration, higher fundamental frequency, higher amplitude, and less reduced vowel quality for stressed syllables. Advanced ALE, but not the beginning ALE, made distinctions in duration and amplitude similar to the duration and amplitude cues used by NE. For fundamental frequency, both advanced and beginning ALE produced even higher fundamental frequency values for stressed syllables than NE and both learner groups produced full (unreduced) vowels in stressed and unstressed syllables. Implications for acoustic cues to lexical stress in English as a second language are discussed.

In this study, segmental and prosodic properties of word-length stimuli were assessed together. Six talkers from 5 L1 backgrounds (American English, Hindi, Korean, Mandarin, and Spanish) were recorded reading English stop-initial trochaic words. The productions were played for 20 monolingual American English-speaking listeners rated the accentedness of each talker. For each token, the deviation from native English productions was determined for segmental (VOT, vowel quality) and three prosodic properties (ratios of duration, intensity, f0 across the two syllables). For each non-native language background, a linear mixed-effects regression model was created to predict accentedness ratings from the phonetic deviations, and the significance of each fixed effect was examined. In each model, the significant predictors included both segmental and prosodic properties. For Hindi and Spanish talkers, the single best predictor was segmental; however, for Korean and Mandarin talkers, the single best predictor was prosodic. Thus, even for short stimuli, both segmental and prosodic information must be considered in accounting for accentedness judgments. We conclude that listeners are sensitive to the different ways that foreign accent may be manifested across different non-native backgrounds.