chunked/nltk_chunking/model_doc_splinter/chunk_3.txt

We explore the more realistic few-shot setting, where only a few hundred
training examples are available, and observe that standard models perform poorly, highlighting the discrepancy between
current pretraining objectives and question answering.