OpenNMT Forum

Multi Label for Synonyms

Hello everyone,

I have a question and I’m waiting for your suggestions. I have synonyms of the word in the sentence. I want to train this as input. I think I’m trying to talk about Multi Label.

For example:
train-features.de.txt > Ich sah alte frau .
train-labels.en.txt > I saw old|elderly|aged|senior woman .

How should I set my inputs? Thanks you for the answers.

Hi,

Multiple labels are not supported. What would be the expected behavior during inference?

You can split the multiple synonym labels into many (normal) sentences pairs. Try to change both parallel sentences to reduce overfitting. Therefore, have a further look at data augmentation:


Thanks for answers. Actually, it’s not my purpose to augment the data. But it was helpful to read. (link)

I thought it might be useful to identify alternative words. Maybe we can create a weight list.
I have no idea for during inference. If you think it is useless, I will give up this. @guillaumekln