Datasets:
Update README.md
Browse files
README.md
CHANGED
@@ -796,8 +796,214 @@ configs:
|
|
796 |
# Portuguese Benchmark
|
797 |
|
798 |
This a collection of datasets in Portuguese initially meant to train and evaluate supervised language models such as BERT, RoBERTa, etc...
|
799 |
-
It contains
|
800 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
801 |
|
802 |
## Tasks:
|
803 |
|
|
|
796 |
# Portuguese Benchmark
|
797 |
|
798 |
This a collection of datasets in Portuguese initially meant to train and evaluate supervised language models such as BERT, RoBERTa, etc...
|
799 |
+
It contains 10 datasets and 18 Tasks for Classification (CLS), NLI, Semantic Similarity Scoring (STS) and Named-Entity Recognition (NER).
|
800 |
|
801 |
+
| NER | Classification | NLI | STS |
|
802 |
+
|---|---|---|---|
|
803 |
+
| [LeNER-Br](https://teodecampos.github.io/LeNER-Br/) | [HateBR_offensive_binary](https://github.com/franciellevargas/HateBR) | [assin2-rte](https://sites.google.com/view/assin2) | [assin2-sts](https://sites.google.com/view/assin2) |
|
804 |
+
| [UlyssesNER-Br-PL-coarse](https://github.com/ulysses-camara/ulysses-ner-br) | [HateBR_offensive_level](https://github.com/franciellevargas/HateBR) | | |
|
805 |
+
| [UlyssesNER-Br-C-coarse](https://github.com/ulysses-camara/ulysses-ner-br) | [brazilian_court_decisions_judgment](https://github.com/lagefreitas/predicting-brazilian-court-decisions) | | |
|
806 |
+
| [UlyssesNER-Br-PL-fine](https://github.com/ulysses-camara/ulysses-ner-br) | [brazilian_court_decisions_unanimity](https://github.com/lagefreitas/predicting-brazilian-court-decisions) | | |
|
807 |
+
| [UlyssesNER-Br-C-fine](https://github.com/ulysses-camara/ulysses-ner-br) | [multi_eurlex_pt](https://github.com/nlpaueb/MultiEURLEX/) | | |
|
808 |
+
| [harem-default](https://www.linguateca.pt/primeiroHAREM/harem_coleccaodourada_en.html) | [Portuguese_Hate_Speech_binary](https://github.com/paulafortuna/Portuguese-Hate-Speech-Dataset) | | |
|
809 |
+
| [harem-selective](https://www.linguateca.pt/primeiroHAREM/harem_coleccaodourada_en.html) | | | |
|
810 |
+
| [mapa_pt_coarse](https://huggingface.co/datasets/joelniklaus/mapa) | | | |
|
811 |
+
| [mapa_pt_fine](https://huggingface.co/datasets/joelniklaus/mapa) | | | |
|
812 |
+
| [rrip](https://bit.ly/rhetoricalrole) | | | |
|
813 |
+
|
814 |
+
## Datasets Descriptions
|
815 |
+
|
816 |
+
**LeNER-Br** (NER) [\[Link\]](https://teodecampos.github.io/LeNER-Br/)
|
817 |
+
|
818 |
+
LeNER-Br is a Portuguese language dataset for named entity recognition applied to legal documents.
|
819 |
+
LeNER-Br consists entirely of manually annotated legislation and legal cases texts and contains tags
|
820 |
+
for persons, locations, time entities, organizations, legislation and legal cases. To compose the dataset,
|
821 |
+
66 legal documents from several Brazilian Courts were collected. Courts of superior and state levels were considered,
|
822 |
+
such as Supremo Tribunal Federal, Superior Tribunal de Justiça, Tribunal de Justiça de Minas Gerais and Tribunal de Contas da União.
|
823 |
+
In addition, four legislation documents were collected, such as "Lei Maria da Penha", giving a total of 70 documents.
|
824 |
+
|
825 |
+
**assin2-rte** (RTE) [\[Link\]](https://sites.google.com/view/assin2)
|
826 |
+
|
827 |
+
The ASSIN 2 corpus is composed of rather simple sentences. Following the procedures of SemEval 2014 Task 1.
|
828 |
+
The training and validation data are composed, respectively, of 6,500 and 500 sentence pairs in Brazilian Portuguese,
|
829 |
+
annotated for entailment and semantic similarity. Semantic similarity values range from 1 to 5, and text entailment
|
830 |
+
classes are either entailment or none. The test data are composed of approximately 3,000 sentence pairs with the same
|
831 |
+
annotation. All data were manually annotated.
|
832 |
+
|
833 |
+
**assin2-sts** (STS) [\[Link\]](https://sites.google.com/view/assin2)
|
834 |
+
|
835 |
+
The ASSIN 2 corpus is composed of rather simple sentences. Following the procedures of SemEval 2014 Task 1.
|
836 |
+
The training and validation data are composed, respectively, of 6,500 and 500 sentence pairs in Brazilian Portuguese,
|
837 |
+
annotated for entailment and semantic similarity. Semantic similarity values range from 1 to 5, and text entailment
|
838 |
+
classes are either entailment or none. The test data are composed of approximately 3,000 sentence pairs with the same
|
839 |
+
annotation. All data were manually annotated.
|
840 |
+
|
841 |
+
**HateBR_offensive_binary** (CLASSIFICATION) [\[Link\]](https://github.com/franciellevargas/HateBR)
|
842 |
+
|
843 |
+
HateBR is the first large-scale expert annotated dataset of Brazilian Instagram comments for abusive language detection
|
844 |
+
on the web and social media. The HateBR was collected from Brazilian Instagram comments of politicians and manually annotated
|
845 |
+
by specialists. It is composed of 7,000 documents annotated according to three different layers: a binary classification (offensive
|
846 |
+
versus non-offensive comments), offensiveness-level (highly, moderately, and slightly offensive messages), and nine hate speech
|
847 |
+
groups (xenophobia, racism, homophobia, sexism, religious intolerance, partyism, apology for the dictatorship, antisemitism,
|
848 |
+
and fatphobia). Each comment was annotated by three different annotators and achieved high inter-annotator agreement. Furthermore,
|
849 |
+
baseline experiments were implemented reaching 85% of F1-score outperforming the current literature dataset baselines for
|
850 |
+
the Portuguese language. We hope that the proposed expert annotated dataset may foster research on hate speech detection in the
|
851 |
+
Natural Language Processing area.
|
852 |
+
|
853 |
+
**HateBR_offensive_level** (CLASSIFICATION) [\[Link\]](https://github.com/franciellevargas/HateBR)
|
854 |
+
|
855 |
+
HateBR is the first large-scale expert annotated dataset of Brazilian Instagram comments for abusive language detection
|
856 |
+
on the web and social media. The HateBR was collected from Brazilian Instagram comments of politicians and manually annotated
|
857 |
+
by specialists. It is composed of 7,000 documents annotated according to three different layers: a binary classification (offensive
|
858 |
+
versus non-offensive comments), offensiveness-level (highly, moderately, and slightly offensive messages), and nine hate speech
|
859 |
+
groups (xenophobia, racism, homophobia, sexism, religious intolerance, partyism, apology for the dictatorship, antisemitism,
|
860 |
+
and fatphobia). Each comment was annotated by three different annotators and achieved high inter-annotator agreement. Furthermore,
|
861 |
+
baseline experiments were implemented reaching 85% of F1-score outperforming the current literature dataset baselines for
|
862 |
+
the Portuguese language. We hope that the proposed expert annotated dataset may foster research on hate speech detection in the
|
863 |
+
Natural Language Processing area.
|
864 |
+
|
865 |
+
**UlyssesNER-Br-PL-coarse** (NER) [\[Link\]](https://github.com/ulysses-camara/ulysses-ner-br)
|
866 |
+
|
867 |
+
UlyssesNER-Br is a corpus of Brazilian Legislative Documents for NER with quality baselines.
|
868 |
+
The presented corpus consists of bills and legislative consultations from Brazilian Chamber of Deputies.
|
869 |
+
UlyssesNER-Br has seven semantic classes or categories. Based on HAREM,
|
870 |
+
we defined five typical categories: person, location, organization, event and date.
|
871 |
+
In addition, we defined two specific semantic classes for the legislative domain:
|
872 |
+
law foundation and law product. The law foundation category makes reference to
|
873 |
+
entities related to laws, resolutions, decrees, as well as to domain-specific entities
|
874 |
+
such as bills, which are law proposals being discussed by the parliament, and legislative consultations,
|
875 |
+
also known as job requests made by the parliamentarians.
|
876 |
+
The law product entity refers to systems, programs, and other products created
|
877 |
+
from legislation.
|
878 |
+
|
879 |
+
**UlyssesNER-Br-C-coarse** (NER) [\[Link\]](https://github.com/ulysses-camara/ulysses-ner-br)
|
880 |
+
|
881 |
+
UlyssesNER-Br is a corpus of Brazilian Legislative Documents for NER with quality baselines.
|
882 |
+
The presented corpus consists of bills and legislative consultations from Brazilian Chamber of Deputies.
|
883 |
+
UlyssesNER-Br has seven semantic classes or categories. Based on HAREM,
|
884 |
+
we defined five typical categories: person, location, organization, event and date.
|
885 |
+
In addition, we defined two specific semantic classes for the legislative domain:
|
886 |
+
law foundation and law product. The law foundation category makes reference to
|
887 |
+
entities related to laws, resolutions, decrees, as well as to domain-specific entities
|
888 |
+
such as bills, which are law proposals being discussed by the parliament, and legislative consultations,
|
889 |
+
also known as job requests made by the parliamentarians.
|
890 |
+
The law product entity refers to systems, programs, and other products created
|
891 |
+
from legislation.
|
892 |
+
|
893 |
+
**UlyssesNER-Br-PL-fine** (NER) [\[Link\]](https://github.com/ulysses-camara/ulysses-ner-br)
|
894 |
+
|
895 |
+
UlyssesNER-Br is a corpus of Brazilian Legislative Documents for NER with quality baselines.
|
896 |
+
The presented corpus consists of bills and legislative consultations from Brazilian Chamber of Deputies.
|
897 |
+
UlyssesNER-Br has seven semantic classes or categories. Based on HAREM,
|
898 |
+
we defined five typical categories: person, location, organization, event and date.
|
899 |
+
In addition, we defined two specific semantic classes for the legislative domain:
|
900 |
+
law foundation and law product. The law foundation category makes reference to
|
901 |
+
entities related to laws, resolutions, decrees, as well as to domain-specific entities
|
902 |
+
such as bills, which are law proposals being discussed by the parliament, and legislative consultations,
|
903 |
+
also known as job requests made by the parliamentarians.
|
904 |
+
The law product entity refers to systems, programs, and other products created
|
905 |
+
from legislation.
|
906 |
+
|
907 |
+
**UlyssesNER-Br-C-fine** (NER) [\[Link\]](https://github.com/ulysses-camara/ulysses-ner-br)
|
908 |
+
|
909 |
+
UlyssesNER-Br is a corpus of Brazilian Legislative Documents for NER with quality baselines.
|
910 |
+
The presented corpus consists of bills and legislative consultations from Brazilian Chamber of Deputies.
|
911 |
+
UlyssesNER-Br has seven semantic classes or categories. Based on HAREM,
|
912 |
+
we defined five typical categories: person, location, organization, event and date.
|
913 |
+
In addition, we defined two specific semantic classes for the legislative domain:
|
914 |
+
law foundation and law product. The law foundation category makes reference to
|
915 |
+
entities related to laws, resolutions, decrees, as well as to domain-specific entities
|
916 |
+
such as bills, which are law proposals being discussed by the parliament, and legislative consultations,
|
917 |
+
also known as job requests made by the parliamentarians.
|
918 |
+
The law product entity refers to systems, programs, and other products created
|
919 |
+
from legislation.
|
920 |
+
|
921 |
+
**brazilian_court_decisions_judgment** (CLASSIFICATION) [\[Link\]](https://github.com/lagefreitas/predicting-brazilian-court-decisions)
|
922 |
+
|
923 |
+
The dataset is a collection of 4043 Ementa (summary) court decisions and their metadata from the Tribunal de
|
924 |
+
Justiça de Alagoas (TJAL, the State Supreme Court of Alagoas (Brazil). The court decisions are labeled according
|
925 |
+
to 7 categories and whether the decisions were unanimous on the part of the judges or not. The dataset
|
926 |
+
supports the task of Legal Judgment Prediction.
|
927 |
+
|
928 |
+
**brazilian_court_decisions_unanimity** (CLASSIFICATION) [\[Link\]](https://github.com/lagefreitas/predicting-brazilian-court-decisions)
|
929 |
+
|
930 |
+
The dataset is a collection of 4043 Ementa (summary) court decisions and their metadata from the Tribunal de
|
931 |
+
Justiça de Alagoas (TJAL, the State Supreme Court of Alagoas (Brazil). The court decisions are labeled according
|
932 |
+
to 7 categories and whether the decisions were unanimous on the part of the judges or not. The dataset
|
933 |
+
supports the task of Legal Judgment Prediction.
|
934 |
+
|
935 |
+
**harem-default** (NER) [\[Link\]](https://www.linguateca.pt/primeiroHAREM/harem_coleccaodourada_en.html)
|
936 |
+
|
937 |
+
The HAREM is a Portuguese language corpus commonly used for Named Entity Recognition tasks. It includes about 93k words, from 129 different texts,
|
938 |
+
from several genres, and language varieties. The split of this dataset version follows the division made by [1], where 7% HAREM
|
939 |
+
documents are the validation set and the miniHAREM corpus (with about 65k words) is the test set. There are two versions of the dataset set,
|
940 |
+
a version that has a total of 10 different named entity classes (Person, Organization, Location, Value, Date, Title, Thing, Event,
|
941 |
+
Abstraction, and Other) and a "selective" version with only 5 classes (Person, Organization, Location, Value, and Date).
|
942 |
+
It's important to note that the original version of the HAREM dataset has 2 levels of NER details, namely "Category" and "Sub-type".
|
943 |
+
The dataset version processed here ONLY USE the "Category" level of the original dataset.
|
944 |
+
[1] Souza, Fábio, Rodrigo Nogueira, and Roberto Lotufo. "BERTimbau: Pretrained BERT Models for Brazilian Portuguese."
|
945 |
+
Brazilian Conference on Intelligent Systems. Springer, Cham, 2020.
|
946 |
+
|
947 |
+
**harem-selective** (NER) [\[Link\]](https://www.linguateca.pt/primeiroHAREM/harem_coleccaodourada_en.html)
|
948 |
+
|
949 |
+
The HAREM is a Portuguese language corpus commonly used for Named Entity Recognition tasks. It includes about 93k words, from 129 different texts,
|
950 |
+
from several genres, and language varieties. The split of this dataset version follows the division made by [1], where 7% HAREM
|
951 |
+
documents are the validation set and the miniHAREM corpus (with about 65k words) is the test set. There are two versions of the dataset set,
|
952 |
+
a version that has a total of 10 different named entity classes (Person, Organization, Location, Value, Date, Title, Thing, Event,
|
953 |
+
Abstraction, and Other) and a "selective" version with only 5 classes (Person, Organization, Location, Value, and Date).
|
954 |
+
It's important to note that the original version of the HAREM dataset has 2 levels of NER details, namely "Category" and "Sub-type".
|
955 |
+
The dataset version processed here ONLY USE the "Category" level of the original dataset.
|
956 |
+
[1] Souza, Fábio, Rodrigo Nogueira, and Roberto Lotufo. "BERTimbau: Pretrained BERT Models for Brazilian Portuguese."
|
957 |
+
Brazilian Conference on Intelligent Systems. Springer, Cham, 2020.
|
958 |
+
|
959 |
+
**multi_eurlex_pt** (MULTILABEL CLASSIFICATION) [\[Link\]](https://github.com/nlpaueb/MultiEURLEX/)
|
960 |
+
|
961 |
+
MultiEURLEX comprises 65k EU laws in 23 official EU languages.
|
962 |
+
Each EU law has been annotated with EUROVOC concepts (labels) by the Publication Office of EU.
|
963 |
+
Each EUROVOC label ID is associated with a label descriptor, e.g., [60, agri-foodstuffs],
|
964 |
+
[6006, plant product], [1115, fruit]. The descriptors are also available in the 23 languages.
|
965 |
+
Chalkidis et al. (2019) published a monolingual (English) version of this dataset, called EUR-LEX,
|
966 |
+
comprising 57k EU laws with the originally assigned gold labels.
|
967 |
+
|
968 |
+
**mapa_pt_coarse** (NER) [\[Link\]]()
|
969 |
+
|
970 |
+
The dataset consists of 12 documents (9 for Spanish due to parsing errors) taken from EUR-Lex,
|
971 |
+
a multilingual corpus of court decisions and legal dispositions in the 24 official languages
|
972 |
+
of the European Union. The documents have been annotated for named entities following the
|
973 |
+
guidelines of the MAPA project which foresees two annotation level, a general and a more
|
974 |
+
fine-grained one. The annotated corpus can be used for named entity recognition/classification.
|
975 |
+
|
976 |
+
**mapa_pt_fine** (NER) [\[Link\]]()
|
977 |
+
|
978 |
+
The dataset consists of 12 documents (9 for Spanish due to parsing errors) taken from EUR-Lex,
|
979 |
+
a multilingual corpus of court decisions and legal dispositions in the 24 official languages
|
980 |
+
of the European Union. The documents have been annotated for named entities following the
|
981 |
+
guidelines of the MAPA project which foresees two annotation level, a general and a more
|
982 |
+
fine-grained one. The annotated corpus can be used for named entity recognition/classification.
|
983 |
+
|
984 |
+
**Portuguese_Hate_Speech_binary** (CLASSIFICATION) [\[Link\]](https://github.com/paulafortuna/Portuguese-Hate-Speech-Dataset)
|
985 |
+
|
986 |
+
The dataset is composed of 5,668 tweets. For its annotation, we defined two different schemes used by
|
987 |
+
annotators with different levels of expertise. Firstly, non-experts annotated the tweets with binary
|
988 |
+
labels (‘hate’ vs. ‘no-hate’). Secondly, expert annotators classified the tweets following a fine-grained
|
989 |
+
hierarchical multiple label scheme with 81 hate speech categories in total. The inter-annotator agreement
|
990 |
+
varied from category to category, which reflects the insight that some types of hate speech are more subtle
|
991 |
+
than others and that their detection depends on personal perception. This hierarchical annotation scheme is
|
992 |
+
the main contribution of the presented work, as it facilitates the identification of different types of
|
993 |
+
hate speech and their intersections.
|
994 |
+
|
995 |
+
|
996 |
+
| NER | Classification | NLI | STS |
|
997 |
+
|---|---|---|---|
|
998 |
+
| [LeNER-Br](https://teodecampos.github.io/LeNER-Br/) | [HateBR_offensive_binary](https://github.com/franciellevargas/HateBR) | [assin2-rte](https://sites.google.com/view/assin2) | [assin2-sts](https://sites.google.com/view/assin2) |
|
999 |
+
| [UlyssesNER-Br-PL-coarse](https://github.com/ulysses-camara/ulysses-ner-br) | [HateBR_offensive_level](https://github.com/franciellevargas/HateBR) | | |
|
1000 |
+
| [UlyssesNER-Br-C-coarse](https://github.com/ulysses-camara/ulysses-ner-br) | [brazilian_court_decisions_judgment](https://github.com/lagefreitas/predicting-brazilian-court-decisions) | | |
|
1001 |
+
| [UlyssesNER-Br-PL-fine](https://github.com/ulysses-camara/ulysses-ner-br) | [brazilian_court_decisions_unanimity](https://github.com/lagefreitas/predicting-brazilian-court-decisions) | | |
|
1002 |
+
| [UlyssesNER-Br-C-fine](https://github.com/ulysses-camara/ulysses-ner-br) | [multi_eurlex_pt](https://github.com/nlpaueb/MultiEURLEX/) | | |
|
1003 |
+
| [harem-default](https://www.linguateca.pt/primeiroHAREM/harem_coleccaodourada_en.html) | [Portuguese_Hate_Speech_binary](https://github.com/paulafortuna/Portuguese-Hate-Speech-Dataset) | | |
|
1004 |
+
| [harem-selective](https://www.linguateca.pt/primeiroHAREM/harem_coleccaodourada_en.html) | | | |
|
1005 |
+
| mapa_pt_coarse | | | |
|
1006 |
+
| mapa_pt_fine | | | |
|
1007 |
|
1008 |
## Tasks:
|
1009 |
|