datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
qa_srlThe dataset contains question-answer pairs to model verbal predicate-argument structure. The questions start with wh-words (Who, What, Where, What, etc.) and contain a verb predicate in the sentence; the answers are phrases in the sentence.
There were 2 datsets used in the paper, newswire and wikipedia. Unfortunately the newswiredataset is built from CoNLL-2009 English training set that is covered under license
Thus, we are providing only Wikipedia training set here. Please check README.md for more details on newswire dataset.
For the Wikipedia domain, randomly sampled sentences from the English Wikipedia (excluding questions and sentences with fewer than 10 or more than 60 words) were taken.
This new dataset is designed to solve this great NLP task and is crafted with a lot of care.qa_srl2018The dataset contains question-answer pairs to model verbal predicate-argument structure. The questions start with wh-words (Who, What, Where, What, etc.) and contain a verb predicate in the sentence; the answers are phrases in the sentence.
This dataset, a.k.a "QASRL Bank", "QASRL-v2" or "QASRL-LS" (Large Scale), was constructed via crowdsourcing.qa_srl2020The dataset contains question-answer pairs to model verbal predicate-argument structure.
The questions start with wh-words (Who, What, Where, What, etc.) and contain a verb predicate in the sentence; the answers are phrases in the sentence.
This dataset, a.k.a "QASRL-GS" (Gold Standard) or "QASRL-2020", was constructed via controlled crowdsourcing.
See the paper for details: Controlled Crowdsourcing for High-Quality QA-SRL Annotation, Roit et. al., 2020qa_srlThe dataset contains question-answer pairs to model verbal predicate-argument structure.
The questions start with wh-words (Who, What, Where, What, etc.) and contain a verb predicate in the sentence; the answers are phrases in the sentence.
This dataset loads the train split from "QASRL Bank", a.k.a "QASRL-v2" or "QASRL-LS" (Large Scale),
which was constructed via crowdsourcing and presented at (FitzGeralds et. al., ACL 2018),
and the dev and test splits from QASRL-GS (Gold Standard), introduced in (Roit et. al., ACL 2020).multi_news_qasrlqa_srl_promptsource