Human-Computer Interaction Thesis Defense

  • Ph.D. Student
  • Human-Computer Interaction Institute
  • Carnegie Mellon University
Thesis Orals

"Alex speaks with my voice!" Promoting science discourse with bidialectal virtual peers

As the United States becomes increasingly diverse, classrooms become home to a wider variety of student backgrounds that may not always align with traditional educator expectations. Scholars argue these mis-alignments may contribute to the systematic under-performance of students of color. In response to this, many of these scholars have also identified that culturally aligned learning environments may be able to mitigate this problem and improve the performance of marginalized learners. Regardless, there remains much to understand about what specific interventions may best serve which student outcomes, and what the mechanisms are behind the success of these interventions.

At the same time, though educational technologies have the ability to personalized instruction, these systems have largely also treated culture as one-size-fits-all. In this thesis, we demonstrate how the design choice of one cultural factor in an educational technology, dialect, impacts the social behavior and science performance of language minority students. The students in our work, like many African American students, were fully or partially bidialectal, with African American English as their first dialect. We designed a virtual peer  who collaborated with students on open-ended science tasks and modeled the use of school-ratified science discourse. We then manipulated whether Alex was monodialectal (only speaking Standard English) or bidialectal, speaking both African American English and Standard English based on context.

In two studies, I found that students demonstrated greater post-test science discourse after working with a bidialectal agent than with an agent who only spoke using Standard English These results seem to be driven by students' social behavior: children with the monodialectal  Standard agent demonstrated significantly lower rapport and significantly more "social challenges," like insults, threats, and other outwardly negative resistance behaviors. These behaviors were associated with less science discourse during the intervention and ultimately less benefit from the intervention. Conversely, children with the  bidialectal agent not just demonstrated increased science discourse, but also demonstrated more positive attitudes about Standard English after a six week intervention.

We believe this work provides three interdisciplinary contributions:

  1. These are the first studies to my knowledge that perform a controlled manipulation of the dialect used within an educational technology to support the achievement of language minority African American children. To my knowledge, it is also the first set of experiments that have investigated the benefits of a bidialectal pedagogy approach towards a domain outside of Reading Language Arts. We find that bidialectal learning environments are consistently associated with improved science performance for African American students.
  2. We demonstrate that agent dialect has significant impacts on students' social behaviors with the agent. In turn, rapport is associated with improved science performance at post-test. This adduces evidence for the hypothesis that there are quantifiable social benefits of bidialectal instruction on language minority students. This serves counter to the folk theories held by many educators that, as one teacher told me, "students can't learn anything in broken English."
  3. We demonstrate one way in which technologies are not immune to some of the cultural critiques that have historically been given to brick-and-mortar school systems. Though perhaps unintentionally, unquestioned design choices in the systems we deploy may be unwittingly perpetuating cultural barriers that disproportionately impact marginalized students.

Thesis Committee:
Justine Cassell (Chair)
Amy Ogan
Marti Louw
Sandra Calvert (Georgetown University)