Update README.md

README.md

@@ -6,7 +6,7 @@ tags:
 ---
 
 # t5-small-spanish-nahuatl
-Nahuatl is the most widely spoken indigenous language in Mexico. However, training a neural network for the task of neural machine tranlation is hard due to the lack of structured data. The most popular datasets such as the Axolot dataset and the bible-corpus only consists of ~16,000 and ~7,000 samples respectivly. Moreover, there are multiple variants of Nahuatl, which makes this task even more difficult. For example, a single word from the Axolot dataset  can be found written in more than three different ways. In this work we leverage the T5 text-to-text training strategy to compensate for the lack of data. The resulting model successfully translates short sentences from Spanish to Nahuatl. We report Chrf and BLEU results.
+Nahuatl is the most widely spoken indigenous language in Mexico. However, training a neural network for the task of neural machine tranlation is hard due to the lack of structured data. The most popular datasets such as the Axolot dataset and the bible-corpus only consists of ~16,000 and ~7,000 samples respectivly. Moreover, there are multiple variants of Nahuatl, which makes this task even more difficult. For example, a single word from the Axolot dataset can be found written in more than three different ways. Therefore, in this work we leverage the T5 text-to-text sufix training strategy to compensate the lack of data. We first teach the  multilingual model Spanish unsing English, then we make the transition to Spanish-Nahuatl. The resulting model successfully translates short sentences from Spanish to Nahuatl. We report Chrf and BLEU results.
 
 
 ## Model description
@@ -30,15 +30,33 @@ outputs = tokenizer.batch_decode(outputs, skip_special_tokens=True)[0]
 ```
 
 ## Approach
-Since the Axolotl corpus contains misaligments, we just select the best samples (~10,000 samples). We use the [bible-sorpus](https://github.com/christos-c/bible-corpus) (7,821 samples) to compensate the lack of nahuatl data.
+### Dataset
+Since the Axolotl corpus contains misaligments, we just select the best samples (~10,000 samples). We also use the [bible-sorpus](https://github.com/christos-c/bible-corpus) (7,821 samples). 
+
+### Model and training
+We employ two training-stages using a multilingual T5-small. This model was chosen because it can handle different vocabularies and suffixes. The model is pretrained on different tasks and lenguages (French, Romanian, English, German).
+
+### Training-stage 1 (learning Spanish)
+In training stage 1 we first introduce Spanish to the model. The objective is to learn a new language rich in data (Spanish) and not lose the previous knowledge acquired. We use the English-Spanish [Anki](https://www.manythings.org/anki/) dataset, which consists of 118.964 text pairs. We train the model till convergence adding the suffix "Translate Spanish to English: ".
+
+
+### Training-stage 2 (learning Nahuatl)
+We use the pretrained Spanish-English model to learn Spanish-Nahuatl. Since the amount of Nahuatl pairs is limited, we also add to our dataset 20,000 samples from the English-Spanish Anki dataset. This two-task-trianing avoids overfitting end mekes the model more robust.
+
+### Training setup
+We train the models on the same datasets for 660k steps using batch size = 16 adn 2e-5 learning rate.
 
 
 ## Evaluation results
-The model is evaluated on 505 validation sentences. We report the results using chrf and sacrebleu hugging face metrics:
-- Validation loss: 1.31
-- BLEU: 6.18
-- Chrf: 28.21
+For a fair comparison, the models are evaluated on the same 505 validation  Nahuatl sentences. We report the results using chrf and sacrebleu hugging face metrics:
+
+| English-Spanish pretraining  | Validation loss | BLEU | Chrf   |
+|:----------------------------:|:---------------:|:-----|-------:|
+| False                        | 1.34            | 6.17 | 26.96  | 
+| True                         | 1.31            | 6.18 | 28.21  | 
 
+The English-Spanish pretrained model improves BLEU and Chrf.
+, and leads to faster convergence.
 
 ## References
 - Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu. 2019. Exploring the limits