search

cofe-ai / fast-gector

Apache License 2.0
54 stars 11 forks source link

What dataset? #9

Closed akesh1235 closed 1 year ago

akesh1235 commented 1 year ago

Hello sir, i was wondering how fast-gector is capturing context of long sentences when it is only trained on some english tokens/words.

Jason3900 commented 1 year ago

Hey, GECToR is a seq2edit model which mainly captures the edits from source to target instead of words/tokens. The long dependencies of error editing may require more iterations for a single instance. In the paper, the author indicates that there's a trade-off between speed and correction.

akesh1235 commented 1 year ago

@Jason3900 Thanks for replying, What do you mean by More iterations for single instance?

You mean running more number of epochs while training??

Jason3900 commented 1 year ago

What I mean is that GECToR uses iterative editing strategy in inference mode which modifies a sentence step by step. If an error requires long dependencies to find out, you can set more iteration number in order to capture it.