ACL 2020 highlights – Canwen

The original Twitter thread here.

The selecting criterion here is being interesting. Not an exhaustive list.

Let Me Choose: From Verbal Context to Font Selection

Bridging text with its font! Very interesting application paper from Adobe. They even have emojis play a role in it! Even fonts have their semantics and sentiments.

Contextualized Weak Supervision for Text Classification

This paper cleverly introduces word disambiguation into weakly supervised text classification and the method for data augmentation is also great!

Human Attention Maps for Text Classification: Do Humans and Neural Networks Focus on the Same Words?

This paper answered the interesting question that if machines read text just like us humans? Though the conclusion may not be surprising, it opens a new path to understand attention.

Pre-train and Plug-in: Flexible Conditional Text Generation with Variational Auto-Encoders

Sorry for self-promoting but this paper is actually very interesting. The flexible framework can be extended to many more fields including text style transfer, image generation, voice conversion, etc.

5 Likes

I like the human attention maps one. It’s interesting that humans have much more peaked distributions, focusing in a few key words where as the ML system attends to a larger sweet of words with varying weights.

2 Likes