Investigating the influence of virtual peers as dialect models on students’ prosodic inventory

Children who speak non-standard dialects of English show reduced performance not just in language-oriented topics in school but also in math and science. Technolog-ical solutions have been rare exactly because of the non-mainstream nature of their talk, and hence the difficulty in automatically recognizing their speech and responding to it with, for example, computer tutors. In order to work to-wards overcoming this achievement gap, in this work we investigate African American students ’ prosodic invento-ries in different contexts as a first-step towards building a system that will be able to automatically recognize, and respond to, the dialect in which a child is speaking. We presented children with recordings of a peer (confeder-ate) speaking in either African American English (AAE) or Mainstream American English (MAE) during both a social task and a science task. We found that children showed decreased prosodic variation and peak slopes dur-ing speech segments which did not contain AAE features, resulting in more monotone and breathy utterances than when they are speaking in AAE. We also found that chil-dren who were speaking with a “peer ” who uses AAE have increased articulation rates, energy, and pitch varia-tion. We discuss potential interpretations of these results that are important to the design of a system to support linguistic diversity and decrease the achievement gap