Quick Notes - Day 10
All previous posts in this series here
Expertise Style Transfer: A New Task Towards Better Communication between Experts and Laymen
Authors: Yixin Cao, Ruihao Shui, Liangming Pan, Min-Yen Kan, Zhiyuan Liu, Tat-Seng Chua
Published at: ACL 2020 url
This paper proposes a task of style transfer from experts’ language to layman language in medical domain. They constructed a dataset for this task and established some benchmarks using SOTA models for similar tasks. They performed an analysis of different analysis metrics, and concluded that there is a huge gap between machine and human performance.
While the authors repeatedly say this is a different task from style transfer and text simplification, I did not quite understand the difference. text simplification can be seen as a form of style transfer where we simplify the content. Similarly, what this paper proposes is also a form of simplification specific to medical domain. So, I wasn’t sure I understood the repeated insistence on the differences. That aside, it seems like an interesting work and definitely is a much needed topic.
Beneath the Tip of the Iceberg: Current Challenges and New Directions in Sentiment Analysis Research
Authors: Soujanya Poria, Devamanyu Hazarika, Navonil Majumder, Rada Mihalcea
Published at: ArXiv (may not be peer-reviewed) url
This is a sort of perspective paper on sentiment analysis research pointing to under explored topics in the area and discussing challenges in current research. They started with summarizing major milestones in this strand of research, identified shortcomings in current research and suggested potential ways to address them. They also list a broad range of under-explored topics related to sentiment analysis (e.g., transfer learning, multilingual and code mixed sentiment analysis, multimodal analysis etc).
Like the authors mentioned in the beginning of the article, I also felt sentiment analysis is a saturated topic in terms of research, as I came across the same kind of problems and same datasets whenever I browsed through a paper on that topic. This paper told me that is not the case, and in fact, there are several interesting challenges related to sentiment analysis still waiting to be solved.