Datasets:

Modalities:
Text
Formats:
parquet
Languages:
English
ArXiv:
Libraries:
Datasets
pandas
License:
ambig_qa / dataset_infos.json
albertvillanova's picture
Add light data files
788845f
raw
history blame
7.6 kB
{
"light": {
"description": "AmbigNQ, a dataset covering 14,042 questions from NQ-open, an existing open-domain QA benchmark. We find that over half of the questions in NQ-open are ambiguous. The types of ambiguity are diverse and sometimes subtle, many of which are only apparent after examining evidence provided by a very large text corpus. AMBIGNQ, a dataset with\n14,042 annotations on NQ-OPEN questions containing diverse types of ambiguity.\nWe provide two distributions of our new dataset AmbigNQ: a full version with all annotation metadata and a light version with only inputs and outputs.\n",
"citation": "@inproceedings{ min2020ambigqa,\n title={ {A}mbig{QA}: Answering Ambiguous Open-domain Questions },\n author={ Min, Sewon and Michael, Julian and Hajishirzi, Hannaneh and Zettlemoyer, Luke },\n booktitle={ EMNLP },\n year={2020}\n}\n",
"homepage": "https://nlp.cs.washington.edu/ambigqa/",
"license": "CC BY-SA 3.0",
"features": {
"id": {
"dtype": "string",
"_type": "Value"
},
"question": {
"dtype": "string",
"_type": "Value"
},
"annotations": {
"feature": {
"type": {
"dtype": "string",
"_type": "Value"
},
"answer": {
"feature": {
"dtype": "string",
"_type": "Value"
},
"_type": "Sequence"
},
"qaPairs": {
"feature": {
"question": {
"dtype": "string",
"_type": "Value"
},
"answer": {
"feature": {
"dtype": "string",
"_type": "Value"
},
"_type": "Sequence"
}
},
"_type": "Sequence"
}
},
"_type": "Sequence"
}
},
"builder_name": "parquet",
"dataset_name": "ambig_qa",
"config_name": "light",
"version": {
"version_str": "1.0.0",
"major": 1,
"minor": 0,
"patch": 0
},
"splits": {
"train": {
"name": "train",
"num_bytes": 2739628,
"num_examples": 10036,
"dataset_name": null
},
"validation": {
"name": "validation",
"num_bytes": 805756,
"num_examples": 2002,
"dataset_name": null
}
},
"download_size": 1777867,
"dataset_size": 3545384,
"size_in_bytes": 5323251
},
"full": {
"description": "AmbigNQ, a dataset covering 14,042 questions from NQ-open, an existing open-domain QA benchmark. We find that over half of the questions in NQ-open are ambiguous. The types of ambiguity are diverse and sometimes subtle, many of which are only apparent after examining evidence provided by a very large text corpus. AMBIGNQ, a dataset with\n14,042 annotations on NQ-OPEN questions containing diverse types of ambiguity.\nWe provide two distributions of our new dataset AmbigNQ: a full version with all annotation metadata and a light version with only inputs and outputs.\n",
"citation": "@inproceedings{ min2020ambigqa,\n title={ {A}mbig{QA}: Answering Ambiguous Open-domain Questions },\n author={ Min, Sewon and Michael, Julian and Hajishirzi, Hannaneh and Zettlemoyer, Luke },\n booktitle={ EMNLP },\n year={2020}\n}\n",
"homepage": "https://nlp.cs.washington.edu/ambigqa/",
"license": "CC BY-SA 3.0",
"features": {
"id": {
"dtype": "string",
"_type": "Value"
},
"question": {
"dtype": "string",
"_type": "Value"
},
"annotations": {
"feature": {
"type": {
"dtype": "string",
"_type": "Value"
},
"answer": {
"feature": {
"dtype": "string",
"_type": "Value"
},
"_type": "Sequence"
},
"qaPairs": {
"feature": {
"question": {
"dtype": "string",
"_type": "Value"
},
"answer": {
"feature": {
"dtype": "string",
"_type": "Value"
},
"_type": "Sequence"
}
},
"_type": "Sequence"
}
},
"_type": "Sequence"
},
"viewed_doc_titles": {
"feature": {
"dtype": "string",
"_type": "Value"
},
"_type": "Sequence"
},
"used_queries": {
"feature": {
"query": {
"dtype": "string",
"_type": "Value"
},
"results": {
"feature": {
"title": {
"dtype": "string",
"_type": "Value"
},
"snippet": {
"dtype": "string",
"_type": "Value"
}
},
"_type": "Sequence"
}
},
"_type": "Sequence"
},
"nq_answer": {
"feature": {
"dtype": "string",
"_type": "Value"
},
"_type": "Sequence"
},
"nq_doc_title": {
"dtype": "string",
"_type": "Value"
}
},
"builder_name": "parquet",
"dataset_name": "ambig_qa",
"config_name": "full",
"version": {
"version_str": "1.0.0",
"major": 1,
"minor": 0,
"patch": 0
},
"splits": {
"train": {
"name": "train",
"num_bytes": 43538533,
"num_examples": 10036,
"dataset_name": null
},
"validation": {
"name": "validation",
"num_bytes": 15383268,
"num_examples": 2002,
"dataset_name": null
}
},
"download_size": 30674462,
"dataset_size": 58921801,
"size_in_bytes": 89596263
}
}