dataset_infos.json

{"all_languages": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "all_languages", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 162802556, "num_examples": 1926192, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 162802556, "size_in_bytes": 195015682}, "af": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "af", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 21219, "num_examples": 307, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 21219, "size_in_bytes": 32234345}, "ar": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "ar", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 546200, "num_examples": 6446, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 546200, "size_in_bytes": 32759326}, "az": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "az", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 44461, "num_examples": 624, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 44461, "size_in_bytes": 32257587}, "be": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "be", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 140376, "num_examples": 1512, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 140376, "size_in_bytes": 32353502}, "ber": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "ber", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 5118620, "num_examples": 67484, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 5118620, "size_in_bytes": 37331746}, "bg": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "bg", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 590535, "num_examples": 6324, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 590535, "size_in_bytes": 32803661}, "bn": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "bn", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 146654, "num_examples": 1440, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 146654, "size_in_bytes": 32359780}, "br": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "br", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 177919, "num_examples": 2536, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 177919, "size_in_bytes": 32391045}, "ca": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "ca", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 39404, "num_examples": 518, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 39404, "size_in_bytes": 32252530}, "cbk": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "cbk", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 19404, "num_examples": 262, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 19404, "size_in_bytes": 32232530}, "cmn": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "cmn", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 964514, "num_examples": 12549, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 964514, "size_in_bytes": 33177640}, "cs": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "cs", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 482292, "num_examples": 6659, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 482292, "size_in_bytes": 32695418}, "da": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "da", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 848886, "num_examples": 11220, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 848886, "size_in_bytes": 33062012}, "de": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "de", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 10593377, "num_examples": 125091, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 10593377, "size_in_bytes": 42806503}, "el": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "el", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 926054, "num_examples": 10072, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 926054, "size_in_bytes": 33139180}, "en": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "en", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 15070349, "num_examples": 158053, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 15070349, "size_in_bytes": 47283475}, "eo": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "eo", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 16810965, "num_examples": 207105, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 16810965, "size_in_bytes": 49024091}, "es": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "es", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 6851135, "num_examples": 85064, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 6851135, "size_in_bytes": 39064261}, "et": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "et", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 17127, "num_examples": 241, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 17127, "size_in_bytes": 32230253}, "eu": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "eu", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 42702, "num_examples": 573, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 42702, "size_in_bytes": 32255828}, "fi": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "fi", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 2520167, "num_examples": 31753, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 2520167, "size_in_bytes": 34733293}, "fr": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "fr", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 9481426, "num_examples": 116733, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 9481426, "size_in_bytes": 41694552}, "gl": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "gl", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 26551, "num_examples": 351, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 26551, "size_in_bytes": 32239677}, "gos": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "gos", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 18442, "num_examples": 279, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 18442, "size_in_bytes": 32231568}, "he": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "he", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 6024345, "num_examples": 68350, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 6024345, "size_in_bytes": 38237471}, "hi": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "hi", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 209382, "num_examples": 1913, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 209382, "size_in_bytes": 32422508}, "hr": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "hr", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 36638, "num_examples": 505, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 36638, "size_in_bytes": 32249764}, "hu": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "hu", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 5289610, "num_examples": 67964, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 5289610, "size_in_bytes": 37502736}, "hy": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "hy", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 49230, "num_examples": 603, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 49230, "size_in_bytes": 32262356}, "ia": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "ia", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 194035, "num_examples": 2548, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 194035, "size_in_bytes": 32407161}, "id": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "id", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 124568, "num_examples": 1602, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 124568, "size_in_bytes": 32337694}, "ie": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "ie", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 31956, "num_examples": 488, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 31956, "size_in_bytes": 32245082}, "io": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "io", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 33892, "num_examples": 480, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 33892, "size_in_bytes": 32247018}, "is": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "is", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 132062, "num_examples": 1641, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 132062, "size_in_bytes": 32345188}, "it": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "it", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 15073750, "num_examples": 198919, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 15073750, "size_in_bytes": 47286876}, "ja": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "ja", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 4314423, "num_examples": 44267, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 4314423, "size_in_bytes": 36527549}, "jbo": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "jbo", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 201564, "num_examples": 2704, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 201564, "size_in_bytes": 32414690}, "kab": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "kab", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 1211051, "num_examples": 15944, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 1211051, "size_in_bytes": 33424177}, "ko": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "ko", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 40458, "num_examples": 503, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 40458, "size_in_bytes": 32253584}, "kw": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "kw", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 88577, "num_examples": 1328, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 88577, "size_in_bytes": 32301703}, "la": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "la", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 485749, "num_examples": 6889, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 485749, "size_in_bytes": 32698875}, "lfn": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "lfn", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 203383, "num_examples": 2313, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 203383, "size_in_bytes": 32416509}, "lt": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "lt", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 599166, "num_examples": 8042, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 599166, "size_in_bytes": 32812292}, "mk": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "mk", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 1240185, "num_examples": 14678, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 1240185, "size_in_bytes": 33453311}, "mr": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "mr", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 1838921, "num_examples": 16413, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 1838921, "size_in_bytes": 34052047}, "nb": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "nb", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 85371, "num_examples": 1094, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 85371, "size_in_bytes": 32298497}, "nds": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "nds", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 195021, "num_examples": 2633, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 195021, "size_in_bytes": 32408147}, "nl": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "nl", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 1790975, "num_examples": 23561, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 1790975, "size_in_bytes": 34004101}, "orv": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "orv", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 40484, "num_examples": 471, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 40484, "size_in_bytes": 32253610}, "ota": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "ota", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 44996, "num_examples": 486, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 44996, "size_in_bytes": 32258122}, "pes": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "pes", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 433406, "num_examples": 4285, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 433406, "size_in_bytes": 32646532}, "pl": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "pl", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 1722188, "num_examples": 22391, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 1722188, "size_in_bytes": 33935314}, "pt": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "pt", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 6141178, "num_examples": 78430, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 6141178, "size_in_bytes": 38354304}, "rn": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "rn", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 47387, "num_examples": 648, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 47387, "size_in_bytes": 32260513}, "ro": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "ro", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 162955, "num_examples": 2092, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 162955, "size_in_bytes": 32376081}, "ru": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "ru", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 24540667, "num_examples": 251263, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 24540667, "size_in_bytes": 56753793}, "sl": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "sl", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 49610, "num_examples": 706, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 49610, "size_in_bytes": 32262736}, "sr": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "sr", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 667308, "num_examples": 8175, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 667308, "size_in_bytes": 32880434}, "sv": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "sv", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 509884, "num_examples": 7005, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 509884, "size_in_bytes": 32723010}, "tk": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "tk", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 95047, "num_examples": 1165, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 95047, "size_in_bytes": 32308173}, "tl": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "tl", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 76059, "num_examples": 1017, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 76059, "size_in_bytes": 32289185}, "tlh": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "tlh", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 185309, "num_examples": 2804, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 185309, "size_in_bytes": 32398435}, "toki": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "toki", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 310864, "num_examples": 3738, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 310864, "size_in_bytes": 32523990}, "tr": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "tr", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 11271158, "num_examples": 142088, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 11271158, "size_in_bytes": 43484284}, "tt": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "tt", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 277269, "num_examples": 2398, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 277269, "size_in_bytes": 32490395}, "ug": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "ug", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 118474, "num_examples": 1183, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 118474, "size_in_bytes": 32331600}, "uk": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "uk", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 4885677, "num_examples": 54431, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 4885677, "size_in_bytes": 37098803}, "ur": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "ur", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 24075, "num_examples": 252, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 24075, "size_in_bytes": 32237201}, "vi": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "vi", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 84773, "num_examples": 962, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 84773, "size_in_bytes": 32297899}, "vo": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "vo", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 22164, "num_examples": 328, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 22164, "size_in_bytes": 32235290}, "war": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "war", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 25759, "num_examples": 327, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 25759, "size_in_bytes": 32238885}, "wuu": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "wuu", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 31640, "num_examples": 408, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 31640, "size_in_bytes": 32244766}, "yue": {"description": "A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a\ncrowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular\nlinguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links\nbetween sentences \u201cmeaning the same thing\u201d. This graph is then traversed to extract sets of paraphrases. Several language-independent\nfilters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows\nthat between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or\nnear-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 \u2013 250 000\nsentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists. \n", "citation": "@dataset{scherrer_yves_2020_3707949,\n  author       = {Scherrer, Yves},\n  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 \n                   Languages}},\n  month        = mar,\n  year         = 2020,\n  publisher    = {Zenodo},\n  version      = {1.0},\n  doi          = {10.5281/zenodo.3707949},\n  url          = {https://doi.org/10.5281/zenodo.3707949}\n}\n", "homepage": "https://zenodo.org/record/3707949#.X9Dh0cYza3I", "license": "Creative Commons Attribution 2.0 Generic", "features": {"paraphrase_set_id": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_id": {"dtype": "string", "id": null, "_type": "Value"}, "paraphrase": {"dtype": "string", "id": null, "_type": "Value"}, "lists": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "tags": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "language": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "builder_name": "tapaco", "config_name": "yue", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 42766, "num_examples": 561, "dataset_name": "tapaco"}}, "download_checksums": {"https://zenodo.org/record/3707949/files/tapaco_v1.0.zip?download=1": {"num_bytes": 32213126, "checksum": "b8463e8087df02ad1a077a12b8316a29fb1e2f3a8dac328841044611e75c67ff"}}, "download_size": 32213126, "post_processing_size": null, "dataset_size": 42766, "size_in_bytes": 32255892}}