KI-Starter - Explaining AI Predictions of Semantic Relationships

Explaining AI is a CATALPA project.

Artificial intelligence as a black box? Automatic assessment algorithms that can reliably evaluate learner responses create important capacities for teachers in the education sector. But what if learners can't do anything with the feedback? Explaining AI explores methods to generate helpful feedback using AI in the future.


Project goals and research questions

The goal of this project is research on approaches for explainable AI algorithms in the field of language technology. A core method is the prediction of the semantic relation between two statements, such as equivalency or entailment. Existing AI methods, while being able to predict the relation itself, are not able to justify the prediction.

We work on this problem using the example of automatic scoring algorithms in the educational domain, where learner answers can be automatically scored by comparison with a sample solution, but existing methods cannot generate helpful feedback for the learner. We want to enable such feedback by the research proposed in this project using methods of natural language generation.

  • Dr. Andrea Horbach

  • The project is funded by the state of North Rhine-Westphalia as part of the "KI-Starter" funding program.

    • Viet Nguyen (student Assistant, start date: 01.02.2022)
    • Finn Brodmann (student Assistant, start date: 01.02.2022)
    • Joey Pehlke (student Assistant, start date: 01.02.2022)
  • First publications from the project are currently in progress.