Update parquet files

.gitattributes

@@ -14,3 +14,18 @@
 *.pb filter=lfs diff=lfs merge=lfs -text
 *.pt filter=lfs diff=lfs merge=lfs -text
 *.pth filter=lfs diff=lfs merge=lfs -text
+pairs/german_legal_sentences-train-00001-of-00002.parquet filter=lfs diff=lfs merge=lfs -text
+pairs/german_legal_sentences-train-00000-of-00002.parquet filter=lfs diff=lfs merge=lfs -text
+pairs/german_legal_sentences-validation.parquet filter=lfs diff=lfs merge=lfs -text
+pairs/german_legal_sentences-test.parquet filter=lfs diff=lfs merge=lfs -text
+pairs+es/german_legal_sentences-train-00003-of-00006.parquet filter=lfs diff=lfs merge=lfs -text
+pairs+es/german_legal_sentences-train-00005-of-00006.parquet filter=lfs diff=lfs merge=lfs -text
+pairs+es/german_legal_sentences-train-00002-of-00006.parquet filter=lfs diff=lfs merge=lfs -text
+pairs+es/german_legal_sentences-train-00000-of-00006.parquet filter=lfs diff=lfs merge=lfs -text
+pairs+es/german_legal_sentences-train-00004-of-00006.parquet filter=lfs diff=lfs merge=lfs -text
+pairs+es/german_legal_sentences-train-00001-of-00006.parquet filter=lfs diff=lfs merge=lfs -text
+pairs+es/german_legal_sentences-validation.parquet filter=lfs diff=lfs merge=lfs -text
+pairs+es/german_legal_sentences-test.parquet filter=lfs diff=lfs merge=lfs -text
+sentences/german_legal_sentences-train.parquet filter=lfs diff=lfs merge=lfs -text
+sentences/german_legal_sentences-validation.parquet filter=lfs diff=lfs merge=lfs -text
+sentences/german_legal_sentences-test.parquet filter=lfs diff=lfs merge=lfs -text

README.md

@@ -1,186 +0,0 @@
----
-annotations_creators:
-- machine-generated
-language_creators:
-- found
-language:
-- de
-license:
-- unknown
-multilinguality:
-- monolingual
-size_categories:
-- n>1M
-source_datasets:
-- original
-task_categories:
-- text-retrieval
-- text-scoring
-task_ids:
-- semantic-similarity-scoring
-- text-retrieval-other-example-based-retrieval
----
-
-# Dataset Card for German Legal Sentences
-
-## Table of Contents
-- [Dataset Card for [Dataset Name]](#dataset-card-for-dataset-name)
-  - [Table of Contents](#table-of-contents)
-  - [Dataset Description](#dataset-description)
-    - [Dataset Summary](#dataset-summary)
-    - [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)
-    - [Languages](#languages)
-  - [Dataset Structure](#dataset-structure)
-    - [Data Instances](#data-instances)
-    - [Data Fields](#data-fields)
-    - [Data Splits](#data-splits)
-  - [Dataset Creation](#dataset-creation)
-    - [Curation Rationale](#curation-rationale)
-    - [Source Data](#source-data)
-      - [Initial Data Collection and Normalization](#initial-data-collection-and-normalization)
-      - [Who are the source language producers?](#who-are-the-source-language-producers)
-    - [Annotations](#annotations)
-      - [Annotation process](#annotation-process)
-      - [Who are the annotators?](#who-are-the-annotators)
-    - [Personal and Sensitive Information](#personal-and-sensitive-information)
-  - [Considerations for Using the Data](#considerations-for-using-the-data)
-    - [Social Impact of Dataset](#social-impact-of-dataset)
-    - [Discussion of Biases](#discussion-of-biases)
-    - [Other Known Limitations](#other-known-limitations)
-  - [Additional Information](#additional-information)
-    - [Dataset Curators](#dataset-curators)
-    - [Licensing Information](#licensing-information)
-    - [Citation Information](#citation-information)
-    - [Contributions](#contributions)
-
-## Dataset Description
-
-- **Homepage:** https://lavis-nlp.github.io/german_legal_sentences/
-- **Repository:** https://github.com/lavis-nlp/german_legal_sentences
-- **Paper:** coming soon
-- **Leaderboard:**
-- **Point of Contact:** [Marco Wrzalik](mailto:marco.wrzalik@hs-rm.de)
-
-### Dataset Summary
-
-German Legal Sentences (GLS) is an automatically generated training dataset for semantic sentence matching and citation recommendation in the domain in german legal documents. It follows the concept of weak supervision, where imperfect labels are generated using multiple heuristics. For this purpose we use a combination of legal citation matching and BM25 similarity. The contained sentences and their citations are parsed from real judicial decisions provided by [Open Legal Data](http://openlegaldata.io/) (https://arxiv.org/abs/2005.13342).
-
-### Supported Tasks and Leaderboards
-
-The main associated task is *Semantic Similarity Ranking*. We propose to use the *Mean Reciprocal Rank* (MRR) cut at the tenth position as well as MAP and Recall on Rankings of size 200. As baselines we provide the follows: 
-
-| Method                            | MRR@10   | MAP@200    | Recall@200  |
-|-----------------------------------|---------:|-----------:|------------:|
-| BM25 - default `(k1=1.2; b=0.75)` |     25.7 |       17.6 |        42.9 |
-| BM25 - tuned `(k1=0.47; b=0.97)`  |     26.2 |       18.1 |        43.3 |
-| [CoRT](https://arxiv.org/abs/2010.10252)             |     31.2 |       21.4 |        56.2 |
-| [CoRT + BM25](https://arxiv.org/abs/2010.10252)      |     32.1 |       22.1 |        67.1 |
-
-In addition, we want to support a *Citation Recommendation* task in the future.
-
-If you wish to contribute evaluation measures or give any suggestion or critique, please write an [e-mail](mailto:marco.wrzalik@hs-rm.de).
-
-### Languages
-
-This dataset contains texts from the specific domain of German court decisions.
-
-## Dataset Structure
-
-### Data Instances
-
-```
-{'query.doc_id': 28860,
- 'query.ref_ids': [6215, 248, 248],
- 'query.sent_id': 304863,
- 'query.text': 'Zudem ist zu berücksichtigen , dass die Vollverzinsung nach '
-               '[REF] i. V. m. [REF] gleichermaßen zugunsten wie zulasten des '
-               'Steuerpflichtigen wirkt , sodass bei einer Überzahlung durch '
-               'den Steuerpflichtigen der Staat dem Steuerpflichtigen neben '
-               'der Erstattung ebenfalls den entstandenen potentiellen Zins- '
-               'und Liquiditätsnachteil in der pauschalierten Höhe des [REF] '
-               'zu ersetzen hat , unabhängig davon , in welcher Höhe dem '
-               'Berechtigten tatsächlich Zinsen entgangen sind .',
- 'related.doc_id': 56348,
- 'related.ref_ids': [248, 6215, 62375],
- 'related.sent_id': 558646,
- 'related.text': 'Ferner ist zu berücksichtigen , dass der Zinssatz des [REF] '
-                 'im Rahmen des [REF] sowohl für Steuernachforderung wie auch '
-                 'für Steuererstattungen und damit gleichermaßen zugunsten wie '
-                 'zulasten des Steuerpflichtigen wirkt , Vgl. BVerfG , '
-                 'Nichtannahmebeschluss vom [DATE] [REF] , juris , mit der '
-                 'Folge , dass auch Erstattungsansprüche unabhängig davon , ob '
-                 'und in welcher Höhe dem Berechtigten tatsächlich Zinsen '
-                 'entgangen sind , mit monatlich 0,0 % verzinst werden .'}
-```
-
-### Data Fields
-
-[More Information Needed]
-
-### Data Splits
-
-[More Information Needed]
-
-## Dataset Creation
-
-### Curation Rationale
-
-[More Information Needed]
-
-### Source Data
-
-#### Initial Data Collection and Normalization
-
-The documents we take from [Open Legal Data](http://openlegaldata.io/) (https://arxiv.org/abs/2005.13342) are first preprocessed by removing line breaks, enumeration characters and headings. Afterwards we parse legal citations using hand-crafted regular expressions. Each citation is split into it components and normalized, thus different variants of the same citation are matched together. For instance, "§211 Absatz 1 des Strafgesetzbuches" is normalized to "§ 211 Abs. 1 StGB". Every time we discover an unknown citation, we assign an unique id to it. We use these ids to replace parsed citations in the document text with a simple reference tag containing this id (e.g `[REF321]`). At the same time we parse dates and replace them with the date tag `[DATE]`. Both remove dots which can may be confused with the end of a sentence, which makes the next stage easier.
-
-We use [SoMaJo](https://github.com/tsproisl/SoMaJo) to perform sentence tokenizing on the pre-processed documents. Each sentence that does not contain at least one legal citation is discarded. For the rest we assign sentence ids, remove all reference ids from them as well as any contents in braces (braces often contain large enumerations of citations and their sources). At the same time we keep track of the corresponding document from which a sentence originates and which references occur in it. 
-
-#### Who are the source language producers?
-
-The source language originates in the context of German court proceedings.
-
-### Annotations
-
-#### Annotation process
-
-[More Information Needed]
-
-#### Who are the annotators?
-
-The annotations are machine-generated.
-
-### Personal and Sensitive Information
-
-The source documents are already public and anonymized.
-
-## Considerations for Using the Data
-
-### Social Impact of Dataset
-
-With this dataset, we strive towards better accessibility of court decisions to the general public by accelerating research on semantic search technologies. We hope that emerging search technologies will enable the layperson to find relevant information without knowing the specific terms used by lawyers.
-
-### Discussion of Biases
-
-[More Information Needed]
-
-### Other Known Limitations
-
-[More Information Needed]
-
-## Additional Information
-
-### Dataset Curators
-
-[More Information Needed]
-
-### Licensing Information
-
-[More Information Needed]
-
-### Citation Information
-
-Coming soon!
-
-### Contributions
-
-Thanks to [@mwrzalik](https://github.com/mwrzalik) for adding this dataset.

dataset_infos.json

@@ -1 +0,0 @@
-{"sentences": {"description": "German Legal Sentences (GLS) is an automatically generated training dataset for semantic sentence \nmatching in the domain in german legal documents. It follows the concept of weak supervision, where \nimperfect labels are generated using multiple heuristics. For this purpose we use a combination of \nlegal citation matching and BM25 similarity. The contained sentences and their citations are parsed \nfrom real judicial decisions provided by [Open Legal Data](http://openlegaldata.io/)\n", "citation": "coming soon\n", "homepage": "", "license": "", "features": {"sent_id": {"dtype": "uint32", "id": null, "_type": "Value"}, "doc_id": {"dtype": "uint32", "id": null, "_type": "Value"}, "text": {"dtype": "string", "id": null, "_type": "Value"}, "references": {"feature": {"ref_id": {"dtype": "uint32", "id": null, "_type": "Value"}, "name": {"dtype": "string", "id": null, "_type": "Value"}, "type": {"num_classes": 2, "names": ["AZ", "LAW"], "names_file": null, "id": null, "_type": "ClassLabel"}}, "length": -1, "id": null, "_type": "Sequence"}}, "post_processed": null, "supervised_keys": null, "builder_name": "german_legal_sentences", "config_name": "sentences", "version": {"version_str": "0.0.2", "description": "", "major": 0, "minor": 0, "patch": 2}, "splits": {"train": {"name": "train", "num_bytes": 470336071, "num_examples": 1542499, "dataset_name": "german_legal_sentences"}, "validation": {"name": "validation", "num_bytes": 26119884, "num_examples": 85375, "dataset_name": "german_legal_sentences"}, "test": {"name": "test", "num_bytes": 26082080, "num_examples": 85405, "dataset_name": "german_legal_sentences"}}, "download_checksums": {"http://lavis.cs.hs-rm.de/storage/german-legal-sentences/GermanLegalSentences_v0.0.2.zip": {"num_bytes": 289263658, "checksum": "57ec7c5ba6c800383bee938cd979305d064163585a5b2fc4f46ae385e0973a1f"}}, "download_size": 289263658, "post_processing_size": null, "dataset_size": 522538035, "size_in_bytes": 811801693}, "pairs": {"description": "German Legal Sentences (GLS) is an automatically generated training dataset for semantic sentence \nmatching in the domain in german legal documents. It follows the concept of weak supervision, where \nimperfect labels are generated using multiple heuristics. For this purpose we use a combination of \nlegal citation matching and BM25 similarity. The contained sentences and their citations are parsed \nfrom real judicial decisions provided by [Open Legal Data](http://openlegaldata.io/)\n", "citation": "coming soon\n", "homepage": "", "license": "", "features": {"query.sent_id": {"dtype": "uint32", "id": null, "_type": "Value"}, "query.doc_id": {"dtype": "uint32", "id": null, "_type": "Value"}, "query.text": {"dtype": "string", "id": null, "_type": "Value"}, "query.ref_ids": {"feature": {"dtype": "uint32", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "related.sent_id": {"dtype": "uint32", "id": null, "_type": "Value"}, "related.doc_id": {"dtype": "uint32", "id": null, "_type": "Value"}, "related.text": {"dtype": "string", "id": null, "_type": "Value"}, "related.ref_ids": {"feature": {"dtype": "uint32", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}}, "post_processed": null, "supervised_keys": null, "builder_name": "german_legal_sentences", "config_name": "pairs", "version": {"version_str": "0.0.2", "description": "", "major": 0, "minor": 0, "patch": 2}, "splits": {"train": {"name": "train", "num_bytes": 754039911, "num_examples": 1404271, "dataset_name": "german_legal_sentences"}, "validation": {"name": "validation", "num_bytes": 42311363, "num_examples": 78472, "dataset_name": "german_legal_sentences"}, "test": {"name": "test", "num_bytes": 41120928, "num_examples": 76626, "dataset_name": "german_legal_sentences"}}, "download_checksums": {"http://lavis.cs.hs-rm.de/storage/german-legal-sentences/GermanLegalSentences_v0.0.2.zip": {"num_bytes": 289263658, "checksum": "57ec7c5ba6c800383bee938cd979305d064163585a5b2fc4f46ae385e0973a1f"}}, "download_size": 289263658, "post_processing_size": null, "dataset_size": 837472202, "size_in_bytes": 1126735860}, "pairs+es": {"description": "German Legal Sentences (GLS) is an automatically generated training dataset for semantic sentence \nmatching in the domain in german legal documents. It follows the concept of weak supervision, where \nimperfect labels are generated using multiple heuristics. For this purpose we use a combination of \nlegal citation matching and BM25 similarity. The contained sentences and their citations are parsed \nfrom real judicial decisions provided by [Open Legal Data](http://openlegaldata.io/)\n", "citation": "coming soon\n", "homepage": "", "license": "", "features": {"query.sent_id": {"dtype": "uint32", "id": null, "_type": "Value"}, "query.doc_id": {"dtype": "uint32", "id": null, "_type": "Value"}, "query.text": {"dtype": "string", "id": null, "_type": "Value"}, "query.ref_ids": {"feature": {"dtype": "uint32", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "related.sent_id": {"dtype": "uint32", "id": null, "_type": "Value"}, "related.doc_id": {"dtype": "uint32", "id": null, "_type": "Value"}, "related.text": {"dtype": "string", "id": null, "_type": "Value"}, "related.ref_ids": {"feature": {"dtype": "uint32", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "es_neighbors.text": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "es_neighbors.sent_id": {"feature": {"dtype": "uint32", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "es_neighbors.doc_id": {"feature": {"dtype": "uint32", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "es_neighbors.ref_ids": {"feature": {"feature": {"dtype": "uint32", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "length": -1, "id": null, "_type": "Sequence"}}, "post_processed": null, "supervised_keys": null, "builder_name": "german_legal_sentences", "config_name": "pairs+es", "version": {"version_str": "0.0.2", "description": "", "major": 0, "minor": 0, "patch": 2}, "splits": {"train": {"name": "train", "num_bytes": 2543172549, "num_examples": 1396670, "dataset_name": "german_legal_sentences"}, "validation": {"name": "validation", "num_bytes": 128326675, "num_examples": 69765, "dataset_name": "german_legal_sentences"}, "test": {"name": "test", "num_bytes": 123911313, "num_examples": 67569, "dataset_name": "german_legal_sentences"}}, "download_checksums": {"http://lavis.cs.hs-rm.de/storage/german-legal-sentences/GermanLegalSentences_v0.0.2.zip": {"num_bytes": 289263658, "checksum": "57ec7c5ba6c800383bee938cd979305d064163585a5b2fc4f46ae385e0973a1f"}}, "download_size": 289263658, "post_processing_size": null, "dataset_size": 2795410537, "size_in_bytes": 3084674195}}

german_legal_sentences.py

@@ -1,285 +0,0 @@
-import random
-
-from pathlib import Path
-import datasets
-from datasets import Value, Sequence, ClassLabel, Features
-
-_CITATION = """\
-coming soon
-"""
-
-_DESCRIPTION = """\
-German Legal Sentences (GLS) is an automatically generated training dataset for semantic sentence 
-matching in the domain in german legal documents. It follows the concept of weak supervision, where 
-imperfect labels are generated using multiple heuristics. For this purpose we use a combination of 
-legal citation matching and BM25 similarity. The contained sentences and their citations are parsed 
-from real judicial decisions provided by [Open Legal Data](http://openlegaldata.io/)
-"""
-
-_VERSION = "0.0.2"
-_DATA_URL = f"http://lavis.cs.hs-rm.de/storage/german-legal-sentences/GermanLegalSentences_v{_VERSION}.zip"
-
-
-class GLSConfig(datasets.BuilderConfig):
-    """BuilderConfig."""
-
-    def __init__(
-        self,
-        load_collection,
-        load_es_neighbors=None,
-        n_es_neighbors=None,
-        **kwargs,
-    ):
-        """BuilderConfig.
-        Args:
-          **kwargs: keyword arguments forwarded to super.
-        """
-        super(GLSConfig, self).__init__(**kwargs)
-        self.load_collection = load_collection
-        self.load_es_neighbors = load_es_neighbors
-        self.n_es_neighbors = n_es_neighbors
-
-
-class GermanLegalSentences(datasets.GeneratorBasedBuilder):
-    BUILDER_CONFIGS = [
-        GLSConfig(
-            name="sentences",
-            load_es_neighbors=False,
-            load_collection=False,
-            version=datasets.Version(_VERSION, ""),
-            description="Just the sentences and their masked references",
-        ),
-        GLSConfig(
-            name="pairs",
-            load_es_neighbors=False,
-            load_collection=True,
-            version=datasets.Version(_VERSION, ""),
-            description="Sentence pairs sharing references",
-        ),
-        GLSConfig(
-            name="pairs+es",
-            load_es_neighbors=True,
-            load_collection=True,
-            n_es_neighbors=5,
-            version=datasets.Version(_VERSION, ""),
-            description="Sentence pairs sharing references plus ES neighbors",
-        ),
-    ]
-
-    def _features(self):
-        if self.config.name == "sentences":
-            return datasets.Features(
-                {
-                    "sent_id": Value("uint32"),
-                    "doc_id": Value("uint32"),
-                    "text": Value("string"),
-                    "references": Sequence(
-                        {
-                            "ref_id": Value("uint32"),
-                            "name": Value("string"),
-                            "type": ClassLabel(names=["AZ", "LAW"]),
-                        }
-                    ),
-                }
-            )
-        elif self.config.name == "pairs":
-            return Features(
-                {
-                    "query.sent_id": Value("uint32"),
-                    "query.doc_id": Value("uint32"),
-                    "query.text": Value("string"),
-                    "query.ref_ids": Sequence(Value("uint32")),
-                    "related.sent_id": Value("uint32"),
-                    "related.doc_id": Value("uint32"),
-                    "related.text": Value("string"),
-                    "related.ref_ids": Sequence(Value("uint32")),
-                }
-            )
-        elif self.config.name == "pairs+es":
-            return Features(
-                {
-                    "query.sent_id": Value("uint32"),
-                    "query.doc_id": Value("uint32"),
-                    "query.text": Value("string"),
-                    "query.ref_ids": Sequence(Value("uint32")),
-                    "related.sent_id": Value("uint32"),
-                    "related.doc_id": Value("uint32"),
-                    "related.text": Value("string"),
-                    "related.ref_ids": Sequence(Value("uint32")),
-                    "es_neighbors.text": Sequence(Value("string")),
-                    "es_neighbors.sent_id": Sequence(Value("uint32")),
-                    "es_neighbors.doc_id": Sequence(Value("uint32")),
-                    "es_neighbors.ref_ids": Sequence(
-                        Sequence(datasets.Value("uint32"))
-                    ),
-                }
-            )
-        assert True
-
-    def _info(self):
-        return datasets.DatasetInfo(
-            description=_DESCRIPTION,
-            features=self._features(),
-            supervised_keys=None,
-            homepage="",
-            citation=_CITATION,
-        )
-
-    def _split_generators(self, dl_manager):
-        if dl_manager.manual_dir:
-            data_dir = Path(dl_manager.manual_dir)
-        else:
-            data_dir = Path(dl_manager.download_and_extract(_DATA_URL))
-        collection = _load_collection(data_dir) if self.config.load_collection else None
-        sent_ref_map = _load_sent_references(data_dir)
-        references = (
-            _load_reference_info(data_dir) if self.config.name == "sentences" else None
-        )
-        es_neighbors = (
-            _load_es_neighbors(data_dir) if self.config.load_es_neighbors else None
-        )
-
-        gen_kwargs = dict()
-        for split in ("train", "valid", "test"):
-            gen_kwargs[split] = {
-                "collection": collection,
-                "pair_id_file": data_dir / f"{split}.pairs.tsv",
-                "sentence_file": data_dir / f"{split}.sentences.tsv",
-                "references": references,
-                "sent_ref_map": sent_ref_map,
-                "es_neighbors": es_neighbors,
-            }
-        return [
-            datasets.SplitGenerator(
-                name=datasets.Split.TRAIN, gen_kwargs=gen_kwargs["train"]
-            ),
-            datasets.SplitGenerator(
-                name=datasets.Split.VALIDATION, gen_kwargs=gen_kwargs["valid"]
-            ),
-            datasets.SplitGenerator(
-                name=datasets.Split.TEST, gen_kwargs=gen_kwargs["test"]
-            ),
-        ]
-
-    def _generate_examples(self, **kwargs):
-        if self.config.name.startswith("pairs"):
-            yield from self._generate_pairs(**kwargs)
-        elif self.config.name == "sentences":
-            yield from self._generate_sentences(**kwargs)
-        else:
-            assert True
-
-    def _generate_pairs(
-        self, pair_id_file, collection, sent_ref_map, es_neighbors, **kwargs
-    ):
-        random.seed(17)
-        with open(pair_id_file, encoding="utf-8") as r:
-            idx = 0
-            for line in r:
-                stripped = line.rstrip()
-                if stripped:
-                    a, b = stripped.split("\t")
-                    features = {
-                        "query.sent_id": int(a),
-                        "query.doc_id": int(collection[a]["doc_id"]),
-                        "query.text": collection[a]["text"],
-                        "query.ref_ids": sent_ref_map[a],
-                        "related.sent_id": int(b),
-                        "related.doc_id": int(collection[b]["doc_id"]),
-                        "related.text": collection[b]["text"],
-                        "related.ref_ids": sent_ref_map[b],
-                    }
-                    if self.config.name == "pairs+es":
-                        curr_es_neighbors = es_neighbors.get(a) or []
-                        if len(curr_es_neighbors) < self.config.n_es_neighbors:
-                            continue
-
-                        es_sent_ids = random.sample(
-                            curr_es_neighbors, k=self.config.n_es_neighbors
-                        )
-                        additional_features = {
-                            "es_neighbors.sent_id": [int(i) for i in es_sent_ids],
-                            "es_neighbors.doc_id": [
-                                int(collection[i]["doc_id"]) for i in es_sent_ids
-                            ],
-                            "es_neighbors.text": [
-                                collection[i]["text"] for i in es_sent_ids
-                            ],
-                            "es_neighbors.ref_ids": [
-                                sent_ref_map[i] for i in es_sent_ids
-                            ],
-                        }
-                        features.update(additional_features)
-                    yield idx, features
-                    idx += 1
-
-    def _generate_sentences(
-        self,
-        sentence_file,
-        references,
-        sent_ref_map,
-        **kwargs,
-    ):
-        with open(sentence_file, encoding="utf-8") as r:
-            for idx, line in enumerate(r):
-                stripped = line.rstrip()
-                if stripped == "":
-                    continue
-                s_id, doc_id, text = stripped.split("\t", maxsplit=2)
-                yield idx, {
-                    "sent_id": int(s_id),
-                    "doc_id": int(doc_id),
-                    "text": text,
-                    "references": [
-                        {
-                            "ref_id": int(r_id),
-                            "name": references[r_id][1],
-                            "type": references[r_id][0],
-                        }
-                        for r_id in sent_ref_map[s_id]
-                    ],
-                }
-
-
-def _load_collection(data_dir):
-    collection = dict()
-    for split in ("train", "valid", "test"):
-        with open(data_dir / f"{split}.sentences.tsv", encoding="utf-8") as r:
-            for line in r:
-                s_id, d_id, sent = line.strip().split("\t", maxsplit=2)
-                collection[s_id] = {"doc_id": d_id, "text": sent}
-    return collection
-
-
-def _load_reference_info(data_dir):
-    with open(data_dir / "refs.tsv", encoding="utf-8") as r:
-        references = {
-            r_id: (r_type, r_name.rstrip())
-            for r_id, r_type, r_name in (
-                line.split("\t", maxsplit=2) for line in r if len(line) > 2
-            )
-        }
-
-    return references
-
-
-def _load_sent_references(data_dir):
-    with open(data_dir / "sent_ref_map.tsv", encoding="utf-8") as r:
-        sent_ref_map = {
-            s_id: r_ids.rstrip().split()
-            for s_id, r_ids in (
-                line.split("\t", maxsplit=1) for line in r if len(line) > 2
-            )
-        }
-    return sent_ref_map
-
-
-def _load_es_neighbors(data_dir):
-    with open(data_dir / "es_neighbors.tsv", encoding="utf-8") as r:
-        es_neighbors = {
-            s_id: other_s_ids.rstrip().split()
-            for s_id, other_s_ids in (
-                line.split("\t", maxsplit=1) for line in r if len(line) > 2
-            )
-        }
-    return es_neighbors

pairs+es/german_legal_sentences-test.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:83c8c25358770c544b342e8d54ac81927baec3ae828c21c0585d933e20a4d1fc
+size 47844504

pairs+es/german_legal_sentences-train-00000-of-00006.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:53b3f1d89e13be279f4ab9f4cd8c68b8f304ccca3df28c09cf35584bf3f5f1f0
+size 145586262

pairs+es/german_legal_sentences-train-00001-of-00006.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e7879dc5984c05e6c53a367eed214ca19fe6b5415c62b6d3a74494d18dda7768
+size 142565537

pairs+es/german_legal_sentences-train-00002-of-00006.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:84c9686be6469018da4df3b862f1bd858cebf8e4fd2b4ac1c4dcf14ea4cfb081
+size 147929304

pairs+es/german_legal_sentences-train-00003-of-00006.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:57ededabde8ae4f99908d2191c6d4759118d7c98de3a775f26650103527c96b5
+size 155381619

pairs+es/german_legal_sentences-train-00004-of-00006.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:65f4bd2d620b1adf35078c24d265a77ef01a494f6ec17cec0c3f3411a9b0f365
+size 165843896

pairs+es/german_legal_sentences-train-00005-of-00006.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:505b048d68f9069032c2033661fee0d712005e9ab055ebb6dced540c72789215
+size 16172022

pairs+es/german_legal_sentences-validation.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d51f8433fb525afd76d4dda2aff60fa07d4a7acb55a804a6b7c20a245b5033c2
+size 49213428

pairs/german_legal_sentences-test.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c474545a9a08553e021646d8e8738fbada1531643f54cdfe94f88121c107d802
+size 11233367

pairs/german_legal_sentences-train-00000-of-00002.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eeafc81e0e81fd3f73461c93fa5c09e57e43dfdaacbe41afcba98ea68cff7081
+size 80990173

pairs/german_legal_sentences-train-00001-of-00002.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:af84883aba011408b32c7b65cb88c950efdf42b6412155eb0fcf9e3d8cc24e4f
+size 45307302

pairs/german_legal_sentences-validation.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c8223f1b831736ca8e9001bee43d6269acd2b41ce2fe7226a571421c3b4d71dd
+size 11440806

sentences/german_legal_sentences-test.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e6307a54ced9b9b7f1ed03417adcb45ade149b308c08befe768ea7236314a7f3
+size 12943167

sentences/german_legal_sentences-train.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:91bf7b312ea9843a4dd2c1651332d12194bd2ff064ee785f340e1d0898046e02
+size 233268726

sentences/german_legal_sentences-validation.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ac592778e57cf68aae9ccc6a88cde5e3df81f1790cdc9f6de9972409bb648d76
+size 12945125