Question-Answer Pairs for Verbal Predicate-Argument Structure

Encompassing question-answer pairs designed to model verbal predicate-argument structure. The train split originates from the QASRL Bank (QASRL-v2/LS), constructed via crowdsourcing, while the dev and test splits are from QASRL-GS (Gold Standard).

Use Cases

• Train a model to generate wh-word questions (e.g., Who, What, Where) based on a sentence's verb predicate.
• Develop an answer extraction system to identify phrases corresponding to verbal arguments in a sentence.
• Analyze the relationship between verb predicates and their associated wh-word questions in the QASRL Bank data.
• Benchmark semantic role labeling systems using the gold-standard dev and test splits from QASRL-GS.

Strengths

• The train split is sourced from the large-scale QASRL Bank, indicating a substantial collection of annotated data.
• The dev and test splits are derived from the QASRL-GS (Gold Standard), providing a high-quality benchmark.
• The dataset is tagged with 'Size Categories100 Kn1 M', suggesting it contains at least 100,000 entries.

Limitations

• Specific row counts, column structures, and file formats are unknown, complicating integration planning.
• The dataset was last updated in December 2021, which may limit its relevance for contemporary NLP model training.
• The license and detailed provenance information are not provided in the input.

Provenance

Source

Train split from QASRL Bank (QASRL-v2/LS); dev and test splits from QASRL-GS (Gold Standard).

Collection Method

Constructed via crowdsourcing (QASRL Bank) and presented in academic papers (FitzGerald et al., ACL 2018; Roit et al., ACL 2020).

Freshness

Last updated on 2021-12-02.

Geography

Tagged with 'Regionus', suggesting a potential US focus, but not explicitly confirmed.