README.md

---
annotations_creators: []
language:
- en
language_creators:
- found
license:
- cc-by-nc-sa-4.0
multilinguality:
- monolingual
pretty_name: AsyLex
size_categories:
- 1M<n<10M
source_datasets: []
tags:
- legal NLP
- Refugee Law
task_categories:
- text-classification
- token-classification
- text-retrieval
task_ids:
- multi-label-classification
- named-entity-recognition
- document-retrieval
- utterance-retrieval
configs:
  - config_name: raw_sentences
    data_files: all_sentences_anonymized.tar.xz
    default: true 
  - config_name: raw_documents
    data_files: cases_anonymized_txt_raw.tar.gz  
  - config_name: all_legal_entities
    data_files: main_and_case_cover_all_entities_inferred.csv
  - config_name: casecover_legal_entities
    data_files: case_cover/case_cover_anonymised_extracted_entities.csv 
  - config_name: casecover_entities_outcome
    data_files: case_cover/case_cover_entities_and_decision_outcome.csv
  - config_name: determination_sentences
    data_files: determination_label_extracted_sentences.csv
  - config_name: outcome_classification   
    data_files:
    - split: train
      path: "outcome_train_test/train_dataset_silver.csv"
    - split: test
      path: "outcome_train_test/test_dataset_gold.csv"
config_names:
  - raw_documents 
  - raw_sentences 
  - all_legal_entities
  - casecover_legal_entities
  - casecover_entities_outcome 
  - determination_sentencess 
  - outcome_classification 

---
# Dataset Card for AsyLex

The dataset introduces 59,112 documents of refugee status determination in Canada from 1996 to 2022, providing researchers and practitioners with essential material for training and evaluating NLP models for legal research and case review.

AsyLex contains labeled data suited for two NLP tasks: (1) Entity extraction and (2) Legal Judgment Prediction.

## Dataset Details

AsyLex includes gold-standard human-labeled annotations for 24 legally relevant entity types curated with the help of legal experts, and 1,682 gold-standard labeled documents for the outcome of the case.

The dataset can be split in two sets: 
- (1) a Case Covers set that consists of semi-structured data and displays meta-information (the first page of each case);
- (2) a Main Text set that contains the body of each case, in full text.

### Dataset Sources

The documents have been collected from the online services of the Canadian Legal Information Institute (CanLII). 

## Uses

- **License:** cc-by-nc-sa-4.0 

The dataset must be used for research purposes only. It must not be use for commercial purposes. 


## Dataset Structure

This dataset contains the following files: 

| Configuration | Files | Description |
| ------------- | ------------- | ------------- |
| raw_documents | cases_anonymized_txt_raw.tar.gz | contains the raw text from all documents, by case, with the corresponding case identifier |
| raw_sentences | all_sentences_anonymized.tar.xz | contains the raw text from all retrieved documents, split by sentences, with the corresponding case identifier |
|  all_legal_entities| main_and_case_cover_all_entities_inferred.csv | contains the structured dataset, all extracted entities (one column per entity type), with the corresponding case identifier |
| casecover_legal_entities| case_cover/case_cover_anonymised_extracted_entities.csv | contains the structured dataset derived from the case covers only (one column per entity type), with the corresponding case identifier|
|  casecover_entities_outcome | case_cover/case_cover_entities_and_decision_outcome.csv| same as above, with the addition of the decision outcome of the case |
| determination_sentences| determination_label_extracted_sentences.csv | contains all sentences that have been extracted with the Entity Type "determination". All sentences included here should therefore directly state the outcome of the decision, with the correspinding case identifier |
| outcome_classification | outcome_train_test| folder containing a train and test set for the task of outcome classificiation. Each set includes the case identifier and the decision outcome (0,1,2). The test set only contains gold-standard manually labeled data. |
|  | manual_annotations | contains jsonl files of the manually collected annotations for the case cover and the main text|

In all files containing the decision outcome, 0 refers to a "reject", 1 to a "granted", and 2 to "uncertain".

Each configuration can be load by passing its name as a second parameter:
 ```
  from datasets import load_dataset
  
  outcome_classification_data = load_dataset("clairebarale/AsyLex", "outcome_classification")
  raw_documents_data = load_dataset("clairebarale/AsyLex", "raw_documents")
```

#### Personal and Sensitive Information

All documents have been anonymized. 

## Citation [optional]

**Papers:**

- **NLLP @EMNLP Publication:** tba

- **ACL Publication:**
 ```
  @inproceedings{barale-etal-2023-automated,
    title = "Automated Refugee Case Analysis: A {NLP} Pipeline for Supporting Legal Practitioners",
    author = "Barale, Claire  and
      Rovatsos, Michael  and
      Bhuta, Nehal",
    booktitle = "Findings of the Association for Computational Linguistics: ACL 2023",
    month = jul,
    year = "2023",
    address = "Toronto, Canada",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2023.findings-acl.187",
    doi = "10.18653/v1/2023.findings-acl.187",
    pages = "2992--3005",
}
```

## Dataset Card Contact

Please contact the authors of the papers.