Upload dataset

README.md

@@ -0,0 +1,128 @@
+---
+languages:
+- ca
+---
+# TECA: Textual Entailment Catalan dataset
+
+## BibTeX  citation
+
+If you use any of these resources (datasets or models) in your work, please cite our latest paper:
+
+```bibtex
+@inproceedings{armengol-estape-etal-2021-multilingual,
+    title = "Are Multilingual Models the Best Choice for Moderately Under-resourced Languages? {A} Comprehensive Assessment for {C}atalan",
+    author = "Armengol-Estap{\'e}, Jordi  and
+      Carrino, Casimiro Pio  and
+      Rodriguez-Penagos, Carlos  and
+      de Gibert Bonet, Ona  and
+      Armentano-Oller, Carme  and
+      Gonzalez-Agirre, Aitor  and
+      Melero, Maite  and
+      Villegas, Marta",
+    booktitle = "Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021",
+    month = aug,
+    year = "2021",
+    address = "Online",
+    publisher = "Association for Computational Linguistics",
+    url = "https://aclanthology.org/2021.findings-acl.437",
+    doi = "10.18653/v1/2021.findings-acl.437",
+    pages = "4933--4946",
+}
+```
+
+
+## Digital Object Identifier (DOI) and access to dataset files
+
+[![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.4761458.svg)](https://doi.org/10.5281/zenodo.4761458)
+
+
+## Introduction
+
+TECA consists of two subsets of textual entailment in Catalan, *catalan_TE1* and *vilaweb_TE*, which contain 14997 and 6166 pairs of premises and hypotheses, annotated according to the inference relation they have (implication, contradiction or neutral).
+
+This dataset was developed by BSC TeMU as part of the AINA project and intended as part of the Catalan Language Understanding Benchmark (CLUB). It is part of the Catalan Language Understanding Benchmark (CLUB) as presented in: 
+
+Armengol-Estapé J., Carrino CP., Rodriguez-Penagos C., de Gibert Bonet O., Armentano-Oller C., Gonzalez-Agirre A., Melero M., and Villegas M., Are Multilingual Models the Best Choice for Moderately Under-resourced Languages? A Comprehensive Assessment for Catalan". Findings of ACL 2021 (ACL-IJCNLP 2021).
+
+
+### Supported Tasks and Leaderboards
+
+Text classification, Language Model
+
+###  Languages
+
+CA-  Catalan
+
+###  Directory structure
+
+* **.gitattributes**
+* **README.md**
+* **dev.json** - json-formatted file with the dev split of the dataset
+* **teca.py** - data loader script
+* **test.json** - json-formatted file with the test split of the dataset
+* **train.json** - json-formatted file with the train split of the dataset
+
+## Dataset Structure
+
+### Data Instances
+
+Two JSON files, one for each subset.
+
+### Example:
+
+<pre>
+{
+"id": 6940,
+"premise": "Podriem posar uns bons filtres a les xemeneies de les quimiques per tal de que poguin seguint havent"hypothesis": "Caldria eliminar tots els filtres de les xemeneies de les qu\u00edmiques.",
+"label": "2"
+}
+</pre>
+
+### Number of sentence pairs
+* catalan_TE1: 14,997
+* vilaweb_TE: 6,166 
+
+## Dataset Creation
+
+### Methodology
+
+catalan_TE1: 12000 sentences were chosen randomly from the BSC Catalan Textual Corpus, and filtered by different criteria, such as length and
+stand-alone intelligibility. From 6000 text sentences, we commissioned 3 hypotheses (one for each entailment category) to be written by a team of
+annotators. 
+
+vilaweb_TE: We randomly selected 6200 headers from the Catalan news site Vilaweb and filtered them to obtain 2100 text sentences. For each
+text, 3 hypotheses were likewise commissioned. 
+
+### Curation Rationale
+
+In both sub-datasets, some sentence pairs were excluded because of inconsistencies. 
+
+### Source Data
+
+#### Initial Data Collection and Normalization
+
+Source sentences are extracted from the [Catalan Textual Corpus](https://doi.org/10.5281/zenodo.4519349), and from Vilaweb newswire.
+
+## Annotations
+
+#### Inter-annotator agreement:
+
+From 600 randomly selected samples, the inter-annotator agreement was 83,57%.
+
+### Dataset Curators
+
+Casimiro Pio Carrino, Carlos Rodríguez and Carme Armentano, from BSC-CNS.
+
+### Personal and Sensitive Information
+
+No personal or sensitive information is included.
+
+## Contact
+
+- Carlos Rodríguez-Penagos (carlos.rodriguez1@bsc.es)
+- Carme Armentano-Oller (carme.armentano@bsc.es)
+
+## License
+
+<a rel="license" href="https://creativecommons.org/licenses/by-nc-nd/4.0/"><img alt="Attribution-NonCommercial-NoDerivatives 4.0 International License" style="border-width:0" src="http://d2klr1ixr44jla.cloudfront.net/306/125/0.5-0.5/assets/images/55132bfeb13b7b027c000041.png" width="100"/></a><br />This work is licensed under an <a rel="license" href="https://creativecommons.org/licenses/by-nc-nd/4.0/">Attribution-NonCommercial-NoDerivatives 4.0 International License</a>.
+

splitter.py

@@ -0,0 +1,41 @@
+import json
+import pandas as pd
+from sklearn.model_selection import train_test_split
+
+# both files downloaded from https://zenodo.org/record/4621378
+path_to_teca1 = 'dataset_te1.json'
+path_to_teca2 = 'dataset_te_vilaweb.json'
+
+# load data to pandas dataframes
+teca1 = pd.read_json(path_to_teca1) # Shape: (14997, 4)
+teca2 = pd.read_json(path_to_teca2) # Shape: (6166, 4)
+teca  = pd.concat([teca1, teca2])   # Shape: (21163, 4)
+
+# remove "id" column, now columns are: ['premise', 'hypothesis', 'label']
+teca.drop(['id'], axis=1, inplace=True)
+
+# shuffle rows
+teca = teca.sample(frac=1).reset_index(drop=True)
+
+# stratified split with harcoded percentages: 80% train, 10% dev, 10% test
+train, dev_test = train_test_split(teca, test_size=0.2, random_state=42, stratify=teca['label'])
+dev, test = train_test_split(dev_test, test_size=0.5, random_state=42, stratify=dev_test['label'])
+
+# report some stats
+print('### VALUE COUNTS TECA ###')
+print(teca['label'].value_counts())
+print('### VALUE COUNTS TRAIN ###')
+print(train['label'].value_counts())
+print('###  VALUE COUNTS DEV  ###')
+print(dev['label'].value_counts())
+print('### VALUE COUNTS TEST ###')
+print(test['label'].value_counts())
+print('train shape:', train.shape[0], ', dev shape:', dev.shape[0], ', test shape:', test.shape[0])
+
+# save train/dev/test sets as json files
+sets = {'train': train, 'dev': dev, 'test': test}
+for key in sets:
+    set_dict = sets[key].to_dict('records')
+    json_content = {"version": '1.0.1', "data": set_dict}
+    with open(key+'.json', 'w') as f:
+        json.dump(json_content, f)

splitter_with_ids.py

@@ -0,0 +1,42 @@
+import json
+import pandas as pd
+from sklearn.model_selection import train_test_split
+
+# both files downloaded from https://zenodo.org/record/4621378
+path_to_teca1 = 'dataset_te1.json'
+path_to_teca2 = 'dataset_te_vilaweb.json'
+
+teca1 = pd.read_json(path_to_teca1) # Shape: (14997, 4)
+teca2 = pd.read_json(path_to_teca2) # Shape: (6166, 4)
+
+teca1['id'] = 'te1_' + teca1['id'].astype(str)
+teca2['id'] = 'vila_' + teca2['id'].astype(str)
+
+teca = pd.concat([teca1, teca2])    # Shape: (21163, 4)
+#teca.drop(['id'], axis=1, inplace=True) # now columns are: ['premise', 'hypothesis', 'label']
+teca = teca.sample(frac=1).reset_index(drop=True) # shuffle rows
+
+print('### VALUE COUNTS TECA ###')
+print(teca['label'].value_counts())
+
+# stratified split with harcoded percentages: 80% train, 10% dev, 10% test
+train, dev_test = train_test_split(teca, test_size=0.2, random_state=42, stratify=teca['label'])
+dev, test = train_test_split(dev_test, test_size=0.5, random_state=42, stratify=dev_test['label'])
+
+print('### VALUE COUNTS TRAIN ###')
+print(train['label'].value_counts())
+print('###  VALUE COUNTS DEV  ###')
+print(dev['label'].value_counts())
+print('### VALUE COUNTS TEST ###')
+print(test['label'].value_counts())
+print('train shape:', train.shape[0], ', dev shape:', dev.shape[0], ', test shape:', test.shape[0])
+
+print(train.head())
+
+sets = {'train': train, 'dev': dev, 'test': test, 'full': teca}
+
+for key in sets:
+    set_dict = sets[key].to_dict('records')
+    json_content = {"version": '1.0.1', "data": set_dict}
+    with open(key+'.json', 'w') as f:
+        json.dump(json_content, f)

teca.py

@@ -0,0 +1,116 @@
+# Loading script for the TECA dataset.
+import json
+import datasets
+
+logger = datasets.logging.get_logger(__name__)
+
+_CITATION = """
+            @inproceedings{armengol-estape-etal-2021-multilingual,
+                title = "Are Multilingual Models the Best Choice for Moderately Under-resourced Languages? {A} Comprehensive Assessment for {C}atalan",
+                author = "Armengol-Estap{\'e}, Jordi  and
+                  Carrino, Casimiro Pio  and
+                  Rodriguez-Penagos, Carlos  and
+                  de Gibert Bonet, Ona  and
+                  Armentano-Oller, Carme  and
+                  Gonzalez-Agirre, Aitor  and
+                  Melero, Maite  and
+                  Villegas, Marta",
+                booktitle = "Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021",
+                month = aug,
+                year = "2021",
+                address = "Online",
+                publisher = "Association for Computational Linguistics",
+                url = "https://aclanthology.org/2021.findings-acl.437",
+                doi = "10.18653/v1/2021.findings-acl.437",
+                pages = "4933--4946",
+            }
+            """
+
+_DESCRIPTION = """
+               TECA consists of two subsets of textual entailment in Catalan, *catalan_TE1* and *vilaweb_TE*, which contain 14997 and 6166 pairs of premises and hypotheses, annotated according to the inference relation they have (implication, contradiction or neutral). This dataset was developed by BSC TeMU as part of the AINA project and intended as part of the Catalan Language Understanding Benchmark (CLUB).
+               """
+
+_HOMEPAGE = """https://zenodo.org/record/4621378"""
+
+# TODO: upload datasets to github
+_URL = "https://huggingface.co/datasets/BSC-TeMU/teca/resolve/main/"
+_TRAINING_FILE = "train.json"
+_DEV_FILE = "dev.json"
+_TEST_FILE = "test.json"
+
+
+class tecaConfig(datasets.BuilderConfig):
+    """ Builder config for the TECA dataset """
+
+    def __init__(self, **kwargs):
+        """BuilderConfig for TECA.
+        Args:
+          **kwargs: keyword arguments forwarded to super.
+        """
+        super(tecaConfig, self).__init__(**kwargs)
+
+
+class teca(datasets.GeneratorBasedBuilder):
+    """ TECA Dataset """
+
+    BUILDER_CONFIGS = [
+        tecaConfig(
+            name="teca",
+            version=datasets.Version("1.0.1"),
+            description="teca dataset",
+        ),
+    ]
+
+    def _info(self):
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION,
+            features=datasets.Features(
+                {
+                    "id": datasets.Value("string"),
+                    "premise": datasets.Value("string"),
+                    "hypothesis": datasets.Value("string"),
+                    "label": datasets.features.ClassLabel
+                        (names=
+                    [
+                        "entailment",
+                        "neutral",
+                        "contradiction"
+                    ]
+                    ),
+                }
+            ),
+            homepage=_HOMEPAGE,
+            citation=_CITATION,
+        )
+
+    def _split_generators(self, dl_manager):
+        """Returns SplitGenerators."""
+        urls_to_download = {
+            "train": f"{_URL}{_TRAINING_FILE}",
+            "dev": f"{_URL}{_DEV_FILE}",
+            "test": f"{_URL}{_TEST_FILE}",
+        }
+        downloaded_files = dl_manager.download_and_extract(urls_to_download)
+
+        return [
+            datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepath": downloaded_files["train"]}),
+            datasets.SplitGenerator(name=datasets.Split.VALIDATION, gen_kwargs={"filepath": downloaded_files["dev"]}),
+            datasets.SplitGenerator(name=datasets.Split.TEST, gen_kwargs={"filepath": downloaded_files["test"]}),
+        ]
+
+    def _generate_examples(self, filepath):
+        """This function returns the examples in the raw (text) form."""
+        logger.info("generating examples from = %s", filepath)
+        with open(filepath, encoding="utf-8") as f:
+            data_dict = json.load(f)
+            for id_, article in enumerate(data_dict["data"]):
+                original_id = article["id"]
+                premise     = article["premise"]
+                hypothesis  = article["hypothesis"]
+                label       = article["label"]
+                yield id_, {
+                    "id":           original_id,
+                    "premise":      premise,
+                    "hypothesis":   hypothesis,
+                    "label":        label,
+                }