README.md

metadata

license: cc-by-sa-4.0
task_categories:
  - conversational
  - translation
pretty_name: OLDI Seed eng-lij translation dataset (instruction-style)
dataset_info:
  features:
    - name: inputs
      dtype: string
    - name: targets
      dtype: string
    - name: template_id
      dtype: int64
    - name: template_lang
      sequence: string
  splits:
    - name: train
      num_bytes: 2347477
      num_examples: 5802
    - name: dev
      num_bytes: 79012
      num_examples: 189
    - name: test
      num_bytes: 86660
      num_examples: 202
  download_size: 1299002
  dataset_size: 2513149
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: dev
        path: data/dev-*
      - split: test
        path: data/test-*

This is an English→Ligurian sentence-level translation dataset. The original data comes from the OLDI Seed dataset, and it has been converted to the instruction format.

The prompts, written in English, ask the model to translate the text to Ligurian. There are several variants of the prompt which were randomly sampled for each sentence: The prompts variously refer to the language as Ligurian and Genoese (the specific dialect of Ligurian used in this datset):

Translate to Ligurian: \<sentence>
Translate to Ligurian (Genoese): \<sentence>
Translate to Genoese: \<sentence>
Translate from English to Ligurian: \<sentence>
Translate from English to Genoese: \<sentence>
Translate from English to Ligurian (Genoese dialect): \<sentence>
Translate this sentence to Ligurian: \<sentence>
Translate this sentence to Genoese: \<sentence>
What’s the Ligurian translation of this sentence? \<sentence>
What’s the Genoese translation of this sentence? \<sentence>
Can you translate this text to Ligurian? \<sentence>

The template used for each dataset entry is referenced in the column template_id, with ids ranging from 1 to 11 according to the order given above. The targets are always prefixed with the string "The Ligurian (Genoese) translation is: <sentence>".

The correspondence between template_id, prompt template and target template is therefore:

[
    (1, "Translate to Ligurian:\n", "The Ligurian (Genoese) translation is:\n"),
    (2, "Translate to Ligurian (Genoese):\n", "The Ligurian (Genoese) translation is:\n"),
    (3, "Translate to Genoese:\n", "The Ligurian (Genoese) translation is:\n"),
    (4, "Translate from English to Ligurian:\n", "The Ligurian (Genoese) translation is:\n"),
    (5, "Translate from English to Genoese:\n", "The Ligurian (Genoese) translation is:\n"),
    (6, "Translate from English to Ligurian (Genoese dialect):\n", "The Ligurian (Genoese) translation is:\n"),
    (7, "Translate this sentence to Ligurian:\n", "The Ligurian (Genoese) translation is:\n"),
    (8, "Translate this sentence to Genoese:\n", "The Ligurian (Genoese) translation is:\n"),
    (9, "What’s the Ligurian translation of this sentence?\n", "The Ligurian (Genoese) translation is:\n"),
    (10, "What’s the Genoese translation of this sentence?\n", "The Ligurian (Genoese) translation is:\n"),
    (11, "Can you translate this text to Ligurian?\n", "The Ligurian (Genoese) translation is:\n"),
]

The dataset contains 5802 train samples, 190 validation samples and 201 test samples.