All NLU tasks can be reframed as either text classification or text extraction. This repository gives access to zero-shot models that can be used for both of these tasks.
We propose to convert all NLU tasks as either text classification or text extraction. For example:
- text classification: sentiment analysis, textual entailment, nli, ...
- text extraction: ner, span detection, question-answering, ...
Our models are trained on a single task and expanded to any task in the given category.