Study the questions the field is actually asking, in a platform built for it
Hardest constructs
by class error rate- Numeral67% error · 3 students
- Subjunctive67% error · 2 students
- Relative56% error · 3 students
By grammar area
- Degree41%
- Gender38%
- Case37%
The ideal platform for studying human and AI interaction directly
Learners move between human partners and the AI in the same environment, and every conversation logs who the partners were. Questions that other platforms can only approximate become directly measurable, like this one:
Does the number of human conversation partners predict student engagement?
Open the “AI” black box and test cross-LLM variability
Compare the effect of using different leading AI providers or test your university's sanctioned (or in-development) model.
Conversation model
- OpenAIhosted API
- Anthropichosted API
- Google Geminihosted API
- your-lab/model-v0.2self-hosted
Deterministic replies: on
Independent Fall 2026 studies
Willingness to communicate for newcomers
A within-student design: every learner is measured against their own average across more- and less-practised tasks, with pre- and post-study Willingness to Communicate questionnaires. Fall 2026 to Spring 2027.
GPT chatbots vs. VR vs. Pangea Chat
A within-subjects comparison in a university German program: every student uses all three modalities weekly in a third asynchronous contact hour. Fall 2026 to Spring 2027.
Conversation practice and writing outcomes
A controlled two-arm study: paired sections, one control and one treatment, replicated across two semesters and measured on departmental exam grammar and writing sections.
Peer vs. bot vocabulary use
How does vocabulary use differ between student-to-student and student-to-bot conversation? Both happen in one interface under one analytics layer, so the contexts can be isolated per learner. From Fall 2026.
What do you want to study?
Get in touch. We'll collaborate on study design, data access, and technical configuration to support your research goals.
Research stipends and free app use are available for qualifying studies; mention your interest below.
The messaging client is open source, so you can audit exactly what your participants' data does. Read our Privacy Policy.
