0 50 100 150 200 250 300 350 Voiced Voiceless Average Vowel Length (ms) Initial Stop Consonant Vowel Length High Engagement Task Low Engagement Task 0 100 200 300 400 500 600 700 800 900 1000 Voiced Voiceless Average F1 (Hz) Initial Stop Consonant F1 Onset High Engagement Task Low Engagement Task 0 50 100 150 200 250 300 350 Voiced Voiceless Average f0 (Hz) Initial Stop Consonant F0 Onset High Engagement … Low Engagement … High Engagement Low Engagement ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 80 100 120 20 25 30 35 Voiceless Voiced 0.25 0.5 0.75 1 0.25 0.5 0.75 1 Normalized Trial Number (Proportion of total trials) Voice onset time (ms) Stimulus Voicing: ● Voiceless Voiced Initial VOT: ● ● Long Short Voice Onset Time β = 3.55, SE = 1.86, χ 2 (1) = 3.37, p = 0.07 3 β = 1.62, SE = 0.85, χ 2 (1) = 3.31, p = 0.07 3 High Engagement Low Engagement ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 150 200 250 150 200 250 300 Voiceless Voiced 0.25 0.5 0.75 1 0.25 0.5 0.75 1 Normalized Trial Number (Proportion of total trials) Vowel length (ms) Stimulus Voicing: ● Voiceless Voiced Initial VL: ● ● Long Short Vowel Length β = -8.78, SE = 2.72, χ 2 (1) = 7.82, p = .005 3 β = -6.17, SE = 3.18, χ 2 (1) = 3.51, p = .06 3 β = -3.72, SE = 1.76, χ 2 (1) = 4.43, p = .04 2 High Engagement Low Engagement ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 150 175 200 225 150 175 200 225 Voiceless Voiced 0.25 0.5 0.75 1 0.25 0.5 0.75 1 Normalized Trial Number (Proportion of total trials) f0 onset (Hz) Stimulus Voicing: ● Voiceless Voiced Initial f0 onset: ● ● High Low Phonetic convergence in an immersive game-based task INTRODUCTION METHOD RESULTS DISCUSSION ACKNOWLEDGEMENTS & REFERENCES Tifani Biro ([email protected]) Department of Speech-Language-Hearing: Sciences and Disorders University of Kansas Joe Toscano ([email protected]) Department of Psychology Villanova University Navin Viswanathan ([email protected]) Department of Speech-Language-Hearing: Sciences and Disorders University of Kansas Statistical Analyses Three linear mixed effects models were used to examine: The effect of trial number, task engagement, initial VOT, and their interactions on the produced VOTs for voiced or voiceless tokens separately. A trial number x task engagement x initial VOT interaction would indicate convergence The effect of initial VOT, trial number, and their interactions on VOTs of voiced or voiceless tokens separately in each individual task. A initial VOT x trial number interaction would indicate convergence The effect of trial number on the VOTs of voiced or voiceless tokens for each of the participants in each task Similar models examined these effects on VL, F1, and f0 Summary of Results Elicited mean VOT were longer than what had been previously reported (voiced VOT: 25 ms; voiceless VOT: 101 ms; Lisker & Abramson, 1964). Lexical factors (Baese-Berk & Goldrick, 2009) and task difficulty (Schertz, 2013) could contribute to these differences However, participants’ productions were not significantly different between the low and high engagement tasks Convergence was observed along certain acoustic dimensions (VOT, VL, F1 onset), but not others (f0). Interestingly, the extent of convergence was affected by the level of task engagement. Overall, subjects in the high engagement task converged more, whereas subjects in the low engagement task were less likely to converge. This finding was especially the case for VL and F1 onset These preliminary results suggest that engaging, naturalistic tasks may yield results that more accurately reflect real-world phonetic variation than traditional laboratory experiments Future Directions Future studies will use these communicative tasks to look at convergence among speakers of different native languages Participants completed either a high- or low-engagement task 30 word-initial voicing minimal pairs provided key information that interlocutors had to provide to each other Phonetic convergence was tracked for 4 acoustic dimensions (VOT, VL, F1, f 0) over the course of the one hour experiment Acknowledgements Thank y ou to our c oders: Anne Marie Crinnion, Sarah Welsh, J acklyn Coelho, N ic ole J ohnson, J ohn Mic hael Kay, Rakshana Selvarajan, and C hristopher Burley ; thanks to David Saltz man and Michael Phelan for help programming the Minec raft puz zles , and thanks to Emma Folk for as sistance with running s ubjec ts. TMB w as s upported by a Villanov a Graduate Student Fellows hip. References B abel, M. (2012). E vidence for phonetic and social selectivity in spontaneous phonetic imitation. Journal of P honetics, 40(1), 177-189. B aese-B erk, M., & Goldrick, M. (2009). Mechanisms of interaction in speech production. Language and cognitive processes, 24(4), 527-554. Buxó-Lugo, A., Toscano, J. C., & Watson, D. G. (2016). Effects of participant engagement on prosodic prominence. Discourse Processes, (just-accepted). Lisker, L., & A bramson, A . S . (1964). A cross-language study of voicing in initial stops: A coustical measurements. Word, 20, 384-422. Olmstead, A. J., Viswanathan, N., Aivar, M. P., & Manuel, S. (2013). Comparison of native and non-native phone imitation by E nglish and S panish speakers. Frontiers in psychology, 4. P ardo, J. S . (2010). E xpressing oneself in conversational interaction. Expressing Oneself/E xpressing One's self: Communication, Cognition, Language, and Identity, 183-196. S chertz, J. (2013). E xaggeration of featural contrasts in clarifications of misheard speech in E nglish. Journal of P honetics, 41(3), 249-263. Mapping Acoustic Signals to Phonetic Categories Stimuli (subset) Phonetic category membership is indicated by multiple dimensions in the acoustic signal. However, acoustic dimensions are variable and context-dependent, which can lead to ambiguities between two speech sounds that only differ in one acoustic attribute. One phenomenon that may affect this variability is phonetic convergence (the observation that speech patterns of interlocutors become more similar during a conversation) Procedure 1. Production Measures The extent talkers can converge on productions outside of their native language range may be limited (Olmstead et al., 2013). What dimensions talkers converge on also varies across studies (Babel, 2012; Pardo, 2010). However, since convergence is a phenomenon occurring conversationally within the natural world, typical laboratory tasks may fail at fully eliciting it 2. Convergence Measures High Engagement Low Engagement ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 600 700 800 500 550 600 Voiceless Voiced 0.25 0.5 0.75 1 0.25 0.5 0.75 1 Normalized Trial Number (Proportion of total trials) F1 onset (Hz) Stimulus Voicing: ● Voiceless Voiced Initial F1 onset: ● ● High Low β = -25.56, SE = 12.63, χ 2 (1) = 3.99, p = .05 β = 14.18, SE = 6.01, χ 2 (1) = 4.48, p = .03 Factors such as engagement have been found to have an effect on language production (Buxó-Lugo et al., 2016) How might task engagement affect speech production and phonetic convergence? 0 20 40 60 80 100 120 140 160 b d g p t k Average VOT (ms) Stop Consonant VOT Distributions Lisker & Abramson (1964) High Engagement Task Low Engagement Task Natural Setting • Conversational partner • Less control • Highly engaging • Lack of interlocutor • More control • Typically boring Laboratory Setting Voice Onset Time (VOT) Release Burst f1 onset f0 onset Vowel Length /b/ Time Freq Amp Maze Participant 1 Participant 2 1 Listener Talker 2 Talker Listener 3 Listener Talker 4 Talker Listener 5 Listener Talker 6 Talker Listener Voiced Voiceless Velar got cot goat coat ghost coast Alveolar dent tent dip tip dart tart Bilabial bat pat bear pear bark park gap fig goat coat muck luck sigh high rig cap say rare done pat yard doe nun ten bet sad pay heart frown glass try goat muck fig high cap Start End VOT Conversation /b/1 /b/2 /p/1 /p/2 VOT /b/1 /b/2 /p/1/p/2 F1 Onset Fundamental Frequency Onset 1 2 3 3 3 β = -8.67, SE = 3.97, χ 2 (1) = 4.73, p = .03 1 β = -18.77, SE = 7.51, χ 2 (1) = 6.20, p = .01 2 Low Engagement Listener High Engagement Listener Low Engagement Talker High Engagement Talker