Datasets:
metadata
dataset_info:
- config_name: simplified_ekman
features:
- name: lt_text
dtype: string
- name: text
dtype: string
- name: labels
dtype:
list:
class_label:
names:
- admiration
- amusement
- anger
- annoyance
- approval
- caring
- confusion
- curiosity
- desire
- disappointment
- disapproval
- disgust
- embarrassment
- excitement
- fear
- gratitude
- grief
- joy
- love
- nervousness
- optimism
- pride
- realization
- relief
- remorse
- sadness
- surprise
- neutral
- name: labels_ekman
dtype:
list:
class_label:
names:
- anger
- disgust
- fear
- joy
- sadness
- surprise
- neutral
- name: id
dtype: string
splits:
- name: train
num_bytes: 7095238
num_examples: 43410
- name: validation
num_bytes: 885284
num_examples: 5426
- name: test
num_bytes: 882333
num_examples: 5427
download_size: 6057071
dataset_size: 8862855
configs:
- config_name: simplified_ekman
data_files:
- split: train
path: simplified_ekman/train-*
- split: validation
path: simplified_ekman/validation-*
- split: test
path: simplified_ekman/test-*
license: apache-2.0
task_categories:
- text-classification
language:
- lt
- en
Lithuanian GoEmotions dataset
The original dataset: GoEmotions (paper).
The derived dataset was machine translated from English into Lithuanian using the free Google Translate API (with deep-translator). The translation script:
from datasets import load_dataset
from deep_translator import GoogleTranslator
from deep_translator.exceptions import TranslationNotFound
original_dataset = load_dataset("go_emotions", name="simplified")
translator = GoogleTranslator(source="en", target="lt")
def translate_batch(batch):
original_text = batch["text"]
while True:
try:
translated_batch = translator.translate_batch(original_text)
break
except TranslationNotFound:
print(f"Translation failed. Retrying...")
# We fix untranslated entries (None values) by replacing them with the original text
for i in range(len(translated_batch)):
if not translated_batch[i]:
translated_batch[i] = original_text[i]
print(f"Replaced {original_text[i]} vs {translated_batch[i]}")
batch["lt_text"] = translated_batch
return batch
translated_dataset = original_dataset.map(
translate_batch, batched=True, batch_size=500
)
The derived dataset uses two aligned tagsets:
- The original 27 +
neutralemotion labels (may contain more than one label per sample):
0: admiration
1: amusement
2: anger
3: annoyance
4: approval
5: caring
6: confusion
7: curiosity
8: desire
9: disappointment
10: disapproval
11: disgust
12: embarrassment
13: excitement
14: fear
15: gratitude
16: grief
17: joy
18: love
19: nervousness
20: optimism
21: pride
22: realization
23: relief
24: remorse
25: sadness
26: surprise
27: neutral
- The basic 6 +
neutralemotion labels as per Paul Ekman's theory (may contain more than one label per sample):
0: anger
1: disgust
2: fear
3: joy
4: sadness
5: surprise
6: neutral
Mapping from the 27 fine-grained emotions to the 6 basic emotions:
| GoEmotions | Ekman |
|---|---|
| admiration | joy |
| amusement | joy |
| anger | anger |
| annoyance | anger |
| approval | joy |
| caring | joy |
| confusion | surprise |
| curiosity | surprise |
| desire | joy |
| disappointment | sadness |
| disapproval | anger |
| disgust | disgust |
| embarrassment | sadness |
| excitement | joy |
| fear | fear |
| gratitude | joy |
| grief | sadness |
| joy | joy |
| love | joy |
| nervousness | fear |
| optimism | joy |
| pride | joy |
| realization | surprise |
| relief | joy |
| remorse | sadness |
| sadness | sadness |
| surprise | surprise |
Acknowledgements
This work was supported by the EU Recovery and Resilience Facility project Language Technology Initiative (2.3.1.1.i.0/1/22/I/CFLA/002).