search

Vaidehi99 / OBPE

Apache License 2.0
6 stars 2 forks source link

Understanding the experiments #1

Closed zhang-jian closed 2 years ago

zhang-jian commented 2 years ago

Hi, thanks for the work. I have a question about the experiment section in the paper.

You mentioned BALANCED : all three HRLs get 160K documents each SKEWED : English gets one million, French half million, and Hindi 160K documents. In the result tables, you also mentioned several setups, e.g. Balanced-4, Balanced-12 and Skewed-12. What are the 4 and 12 mean here? Thanks

Vaidehi99 commented 2 years ago

Hi, thanks for reaching out. Balanced-4 setting involves training 3 separate models with 4 languages each (1 HRL and 3 LRLs) from the same family corresponding to each row in Table 2 of the paper. Balanced-12 setting involves training a single model with all 12 languages. I hope this makes it clear.