search

jerbarnes / semeval22_structured_sentiment

SemEval-2022 Shared Task 10: Structured Sentiment Analysis
75 stars 42 forks source link

Conflicting data usage rules about crosslingual task #28

Closed luxinyu1 closed 2 years ago

luxinyu1 commented 2 years ago

In ./README.md the data usage rule about subtask 2 is:

For training, you can use any of the other datasets, as well as any other resource that does not contain sentiment annotations in the target language.

In ./data/README.md the data usage rule about subtask 2 is:

This track will instead train only on a high-resource language (English) and test on several languages.

From my perspective, these two descriptions are quite different. We have spent a lot of time optimizing our cross-lingual model under the rule in ./README.md. And now it is the evaluation phase, we hope that there's a solution that fairly merges the results under different rules.

jerbarnes commented 2 years ago

The rules in the full readme are the valid ones. I've changed the information in the data/README.md to reflect this.

luxinyu1 commented 2 years ago

The rules in the full readme are the valid ones. I've changed the information in the data/README.md to reflect this.

Happy to hear this, thanks for your reply.