Bias Benchmark for Danish language models

[KennethEnevoldsen]

Statement of need

Currently no benchmarks exist for evaluating biases (e.g. based on genders or minorities) in Danish NLP models. This might limit our understanding of the safe application of these models.

Current status

Project such as DaCy and GenDa Lens have already implemented some outline for bias evaluations, but the benchmark within dacy only test fine-tuned model and gendalens does not maintain and active benchmark.

Contributing

If you wish to contribute to this feel free to start a discussion here.