ul-nl-parallel / README.md
genie-unlikely's picture
Update README.md
201b667 verified
metadata
dataset_info:
  features:
    - name: nl
      dtype: string
    - name: ul
      dtype: string
    - name: status
      dtype: string
    - name: level
      dtype: string
  splits:
    - name: train
      num_bytes: 1189650.9591673338
      num_examples: 5620
    - name: test
      num_bytes: 66256.3612489992
      num_examples: 313
    - name: valid
      num_bytes: 66044.67958366693
      num_examples: 312
  download_size: 536146
  dataset_size: 1321951.9999999998
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*
      - split: valid
        path: data/valid-*

Dataset Description

The dataset contains the parallel NL-UL examples which can be used to train models for translation of natural language claims to UL. For the purposes of tractability, in this dataset we focus only on claims, i.e., short texts translated into UL.

Features

Each example in the dataset contains 4 features:

  • "nl": Natural language utterances/short paragraphs
  • "ul": UL representation of the "nl"
  • "status": the status of UL translation, e.g., if it was approved (APPROVED) or if there were changes requested (CHANGE_REQUESTED) [Options: 'TRANSLATION_LEVEL_ASPIRATIONAL', 'TRANSLATION_LEVEL_INGESTIBLE_BUT_APPROXIMATE', 'TRANSLATION_LEVEL_EXACT', 'TRANSLATION_LEVEL_UNKNOWN']
  • "level": level of translation quality, e.g., exact or there are potential improvements [Options: 'REVIEW_REQUIRED', 'APPROVED', 'WONT_DO', 'DRAFT', 'CHANGES_REQUESTED', 'NOT_STARTED', 'BLOCKED']

Dataset Statistics

The dataset was randomly split into train-validation-test splits into 90-5-5%.

Split Percentage N Examples
Train 90% 5620
Valid 5% 313
Test 5% 313
------------- ------------- -------------
Total: 6246