Trends in NLP Research: Highlights from the ACL 2019

Summary
A review of major trends and the most popular papers presented at the Association for Computational Linguistics Conference.  The community is tackling the issue of bias in NLP, both from expanding the geographical representation of practitioners, to addressing gender bias in models.  Use this as a source for your reading list.

Commentary
Another trend highlighted is the advent of pre-training and then fine-tuning models with BERT, ELMO and OpenAI GPT to achieve newfound performance on datasets.  This permits models to be trained with smaller domain-specific datasets and expand the application space. However, as more powerful models crush current benchmark datasets to near human-level performance, we need to have more complex datasets that test the models’ ability to learn the task, not just the dataset.  Researchers are moving the goalpost with crafted datasets that are specifically challenging for pre trained-based models; this is undoubtedly a good thing that will continue to push progress in the NLP field.

What Do You Think?
What is the most exciting development from the ACL you read about in this review and how will you implement it in your application?