Welcome to our sixth team profile blog post! In this article, we interview Yuta Hayashibe, a Research Scientist at Megagon Labs Tokyo. Yuta is leading research on Japanese natural language processing (NLP) technology. His work has been featured at quite a number of domestic and international conferences. In the following interview, you’ll learn what led Yuta to Megagon, new developments in his work, and the secret to his success as a top researcher.
How did you arrive at your current role at Megagon Labs?
While working on my Master of Engineering (MEng) degree and Ph.D in Engineering at Nara Institute of Science and Technology, I began working on research in NLP. After completing my doctorate, I continued work on this as a researcher at Kyoto University. During this time, I became fascinated by spoken language processing and industry research. So I shifted my focus towards these endeavors and conducted research and development in speech recognition.
I met a few team members of Megagon Labs while working on speech recognition. Not only did I enjoy their personalities and their perspective on this research, but I also became captivated by the projects they were working on at the time. When it comes to research, a team’s vision is crucial. Data and learning resources are obviously also important. But the opportunity to work with amazing people on such ambitious goals were really the main factors that compelled me to join the Megagon Labs* team as a research scientist.
Please tell us about your current work. Are there any exciting events or developments you want to mention?
When people search for something on the internet (e.g., vacations, restaurants, lodging), computers can elicit requests from users through words. But in many scenarios, users have trouble looking for something specific because they are not able to verbalize what they want. The main objective of my research efforts these days is to help those “unspoken” or “unfulfilled” requests be brought out and realized.
In order to achieve this goal, we are currently working on data development with a focus on word-of-mouth as real linguistic information that lets people share their experiences and knowledge. To give an example of dataset development, we released the “Implication Relations Dataset” in October 2020. My team and I hope to contribute more to the development of Japanese NLP in the near future. Along with developing datasets, I am also writing research papers.
I had the amazing opportunity to present “Japanese Realistic Textual Entailment Corpus” at LREC 2020 and “Sentence Boundary Detection on Line Breaks in Japanese” at W-NUT 2020 — those are definitely career highlights! These studies are part of our pre-processing efforts to utilize actual word-of-mouth data. The paper I presented at W-NUT 2020 is about sentence breaks. Japanese NLP generally uses punctuation marks as sentence separators for sentence-by-sentence processing. But web text, including word-of-mouth text, is often separated by line breaks, which can cause sentence segmentation to fail. We are conducting research to solve these problems and to further utilize the data.
What is your favorite part about working at Megagon labs?
Three main factors make me excited to come work at Megagon Labs! The first is to be able to actually work on research. At many corporate research centers, it’s not unusual to get involved in technology development that has little to do with research. At Megagon labs, our mission is not to provide technical consulting — it is to create new value in both academia and business. This focus on purpose means that we prioritize research. As a research scientist, that’s where the excitement is for me!
Secondly, all Megagon Labs team members have a clear direction of where they want to go because the applications of their research work are clearly defined. We are constantly asking ourselves, “How will this research affect business and society?” Doing so always helps to elucidate the path we should take.
Lastly, I would like to mention the diversity of our team members. They come from a variety of backgrounds. This vast experience across numerous companies and research organizations makes it incredibly fun to have discussions. I’m always learning something new and regularly get to see different perspectives on important matters.
All of these factors help me make the learning data I have created and maintained through research useful for others involved in language processing. I believe this is key to opening up unprecedented value in several emerging fields of technology. Eventually, I would like to make it possible for Japanese internet users to clarify their unspoken requests and access all the information they need in a more intuitive manner.
What piece of advice would you offer to an aspiring research scientist?
For this question, I’ll describe some of the things I try to do every day. As a research scientist, it’s imperative to observe your data carefully. Don’t hesitate to take a close look at the actual information in front of you. All of my professors and research mentors have emphasized this, and I believe I owe a lot of my success to this practice.
Simply following the experimental data and figures in papers will not reveal what is important. In my case, the subject of research is language, which is subject to frequent exception handling. So it’s important to examine specific cases in detail. By doing so, we can hone in on the real issues that are present. This, in turn, shows us where the real value of research lies.
I also believe it’s vital that you ask yourself how you can best leverage your strengths. By understanding your own nature, you’ll be prepared to optimize your outcomes when opportunities arise.
Lastly, I’ll give some advice that I would have liked to receive as a student: Be proactive and get involved! There are many opportunities available. But you won’t encounter them if you don’t put yourself out there. Also, value the connections you have with people. Because at the end of the day, there are always many people who help you achieve your goals.
What are your favorite spots in Japan?
The Megagon Labs Tokyo office is located in Ginza, Tokyo. When you hear the word “Tokyo,” you may assume that there is not much nature in this area. But there are actually many places here that you can visit to experience the tranquility of nature! For example, the city has several gardens and large parks. Whenever I have time, I like to go for a walk through one of these to get refreshed.
My personal favorites are Odaiba Seaside Park and Todoroki Valley. I highly recommend visiting Todoroki Ravine for your first time at the latter; it’s so rich in nature that you’ll find it hard to believe that it’s located in the heart of Tokyo.
We hope you’ve enjoyed this special interview with Yuta! Check out our blog to see other team member profiles and learn more about Megagon’s recent work!
(Interviewee: Yuta Hayashibe / Interviewer: Megagon Labs Tokyo)
* Megagon Labs Tokyo is a research team within the Recruit, Co.