panagoa commited on
Commit
d69fe55
·
verified ·
1 Parent(s): 9d3bda6

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +13 -13
README.md CHANGED
@@ -11,7 +11,7 @@ language:
11
  license: cc-by-2.0
12
  task_categories:
13
  - translation
14
- pretty_name: Tatoeba Translations Dataset
15
  size_categories:
16
  - 1M<n<10M
17
  tags:
@@ -122,7 +122,7 @@ Each instance in the dataset contains:
122
  {
123
  'source_tatoeba_id': 7627627, # Tatoeba sentence ID
124
  'source_lang': 'eng', # Source language code
125
- 'source_text': 'He\'s always happy.', # Original text
126
  'target_text': 'Ар сыт щыгъуи насыпыфIэщ.', # Translated text
127
  'similarity_score': 0.9283, # Translation similarity score
128
  'nllb_model_version': 'rus-kbd-v1.3.4' # Model used for verification
@@ -134,7 +134,7 @@ Each instance in the dataset contains:
134
  - `source_tatoeba_id`: Unique identifier from Tatoeba
135
  - `source_lang`: Language code of the source text
136
  - `source_text`: Original text in source language
137
- - `target_text`: Translated text (predominantly in Russian)
138
  - `similarity_score`: Similarity score between source and target texts
139
  - `nllb_model_version`: Version of the translation model used for verification
140
 
@@ -160,12 +160,12 @@ The source data comes from [Tatoeba](https://tatoeba.org/), a free collaborative
160
  1. Sentences were extracted from Tatoeba's database
161
  2. Filtered by language pairs
162
  3. Translated using various versions of NLLB-200-kbd model
163
- 4. Translation quality was assessed using panagoa/LaBSE-kbd-v0.1 model to compute similarity scores
164
  5. For quality assurance, translations were filtered using the 75th percentile of similarity scores for each language pair as a threshold
165
 
166
  ### Annotations
167
 
168
- The translations were generated using different versions of NLLB-200-kbd model. The similarity scores were computed using panagoa/LaBSE-kbd-v0.1 model to ensure translation quality and consistency.
169
 
170
  ## Additional Information
171
 
@@ -180,12 +180,12 @@ This dataset is licensed under [Creative Commons Attribution 2.0](https://creati
180
  ### Citation Information
181
 
182
  ```bibtex
183
- @misc{tatoeba_translations_2024,
184
- title={Multilingual Tatoeba Translations Dataset},
185
- author={Your Name},
186
- year={2024},
187
  publisher={Hugging Face},
188
- howpublished={\url{https://huggingface.co/datasets/username/dataset-name}}
189
  }
190
  ```
191
 
@@ -199,11 +199,11 @@ Thanks to [@Tatoeba](https://tatoeba.org/) for providing the source data.
199
  from datasets import load_dataset
200
 
201
  # Load the entire dataset
202
- dataset = load_dataset("username/dataset-name")
203
 
204
  # Load specific language pairs
205
- eng_dataset = load_dataset("username/dataset-name", "eng")
206
- rus_dataset = load_dataset("username/dataset-name", "rus")
207
 
208
  # Example of accessing data
209
  for example in eng_dataset['train']:
 
11
  license: cc-by-2.0
12
  task_categories:
13
  - translation
14
+ pretty_name: Multilingual-to-Kabardian Tatoeba Translations Dataset
15
  size_categories:
16
  - 1M<n<10M
17
  tags:
 
122
  {
123
  'source_tatoeba_id': 7627627, # Tatoeba sentence ID
124
  'source_lang': 'eng', # Source language code
125
+ 'source_text': 'He\'s always happy.', # Original text
126
  'target_text': 'Ар сыт щыгъуи насыпыфIэщ.', # Translated text
127
  'similarity_score': 0.9283, # Translation similarity score
128
  'nllb_model_version': 'rus-kbd-v1.3.4' # Model used for verification
 
134
  - `source_tatoeba_id`: Unique identifier from Tatoeba
135
  - `source_lang`: Language code of the source text
136
  - `source_text`: Original text in source language
137
+ - `target_text`: Translated text
138
  - `similarity_score`: Similarity score between source and target texts
139
  - `nllb_model_version`: Version of the translation model used for verification
140
 
 
160
  1. Sentences were extracted from Tatoeba's database
161
  2. Filtered by language pairs
162
  3. Translated using various versions of NLLB-200-kbd model
163
+ 4. Translation quality was assessed using [panagoa/LaBSE-kbd-v0.1](https://huggingface.co/panagoa/LaBSE-kbd-v0.1) model to compute similarity scores
164
  5. For quality assurance, translations were filtered using the 75th percentile of similarity scores for each language pair as a threshold
165
 
166
  ### Annotations
167
 
168
+ The translations were generated using different versions of NLLB-200-kbd model. The similarity scores were computed using [panagoa/LaBSE-kbd-v0.1](https://huggingface.co/panagoa/LaBSE-kbd-v0.1) model to ensure translation quality and consistency.
169
 
170
  ## Additional Information
171
 
 
180
  ### Citation Information
181
 
182
  ```bibtex
183
+ @misc{tatoeba_translations_2025,
184
+ title={Multilingual-to-Kabardian Tatoeba Translations Dataset},
185
+ author={Adam Panagov},
186
+ year={2025},
187
  publisher={Hugging Face},
188
+ howpublished={\url{https://huggingface.co/datasets/panagoa/tatoeba_kbd}}
189
  }
190
  ```
191
 
 
199
  from datasets import load_dataset
200
 
201
  # Load the entire dataset
202
+ dataset = load_dataset("panagoa/tatoeba_kbd")
203
 
204
  # Load specific language pairs
205
+ eng_dataset = load_dataset("panagoa/tatoeba_kbd", "eng")
206
+ rus_dataset = load_dataset("panagoa/tatoeba_kbd", "rus")
207
 
208
  # Example of accessing data
209
  for example in eng_dataset['train']: