Some benchmarks imported from PWC are mapped to tasks that contain multiple concepts

[kathrinblagec]

There are at least two cases were this applies:

• Process "Question answering" encompasses both extractive (information retrieval) and abstractive question-answering (language generation) benchmarks
• Process "Summarisation" encompasses both extractive (information retrieval) and abstractive summarisaton (language generation) benchmarks

---

Potential solutions:

• Divide into subclassses and map benchmarks accordingly (e.g., for question answering: 'extractive question answering' and 'abstractive question' answering)
• Add question answering as a separate NLP category next to the existing hierarchy (natural language generation, natural language anaylsis...)