416 datasets found

TechQA: a dataset for question answering on technical support articles

VAULT: a light-weight and parallel-efficient paragraph representation for Machine Reading Comprehension (MRC) based on contextualized representation from long document input

The MS MARCO Dev (small) dataset is a small version of the MS MARCO passage dev set.

The RetroMAE dataset is used for pre-training retrieval-oriented language models.

The TREC 2020 Deep Learning (Passage Subtask) dataset consists of 54 queries with manual judgments from NIST annotators (211 relevance assessments per query, on average).

The TREC 2019 Deep Learning (Passage Subtask) dataset consists of 43 manually-judged queries using four relevance grades (215 relevance assessments per query, on average).

The SemEval-2013 task 13 dataset, containing 20 nouns, 20 verbs, and 10 adjectives in WordNet-sense-tagged contexts.

The DSTC2 dialog dataset consists of 5 different tasks, each of which has 1,000 synthetically-generated goal-oriented dialogs between a user and the system in the domain of...

The bAbI dialog dataset consists of 5 different tasks, each of which has 1,000 synthetically-generated goal-oriented dialogs between a user and the system in the domain of...

The bAbI story-based QA dataset is composed of 20 different tasks, each of which has 1,000 synthetically-generated story-question pairs. A story can be as short as two sentences...

The dataset used in the paper to test the persuasion and susceptibility scores of language models.

The Emergent dataset is a dataset derived from a digital journalism project at Columbia University, containing 300 rumored claims and 2,595 news articles.

QALD-7: A dataset for question answering over linked data.

WikiWebQuestions: A dataset for semantic question answering over Wikidata.

COMPMIX: A benchmark for heterogeneous question answering.

SPAGHETTI: A hybrid open-domain question-answering system that combines semantic parsing and information retrieval to handle structured and unstructured data.

The LIPID dataset is a template-free dataset for probing models with prompts from the biomedical domain.

The Google-RE (Templates) dataset contains 6.11K template-based prompts from Wikipedia and 3 relations.

Template-based probing uses expert-made templates to create prompts, while template-free probing uses naturally-occurring text.

GeoQA is a geometric question answering benchmark.