News & Events
ACL publication success
The paper Gender-preserving Debiasing for Pre-trained Word Embeddings by Masahiro Kaneko, Danushka Bollegala has been accepted to the Annual Conference of the Association for Computational Linguistics (ACL-19) to be held in Florence, Italy in August.
ACL is the premiere venue for Natural Language Processing (NLP) related research.
It has been found that gender biases are learnt and amplified by machine learning models, trained on BigData collections such as from social media in applications such as machine translation and dialogue systems. The paper proposes a novel method to remove such gender-related stereotypical biases, while retaining the gender-related information needed for an NLP application.
The work was done at Liverpool, while Masahiro was a visiting research student at the Department of Computer Science.