Special Seminar - Tal August, "Language as Design: Adapting Language to Different Audiences"
How can we adapt language to improve communication online? Adapting language to different people enriches and empowers our communication—think of how different online communities converse, how to explain why you should sign up for 2FA to a 5th grader compared to a graduate student, or convey what is exciting about a new scientific discovery. However, as internet audiences grow in size and diversity, it becomes increasingly difficult to adapt to all potential readers. Current language technologies can generate language in an endless variety of styles and topics, suggesting the possibility of adapting language at scale, but it is not clear what parts of language to change, and how these changes will affect a reader’s behavior.
In this talk, I will cover my work in Human-Computer Interaction (HCI) on language as design. Language can be considered an interface. Like other interfaces, its design impacts how easy it is for someone to use. I operationalize dimensions of language (e.g., its formality, complexity, or framing), and investigate how changes in language affect user behavior. Using insights from my experiments, I build systems that automatically adapt language for different audiences. I will focus on one thread of this research in science communication. I identify a set of writing strategies that quantify how experts design scientific language for public audiences. I then describe an intelligent reading interface I built using these strategies as inspiration that designs scientific language on the fly for a general audience. Finally, I cover a method I developed to control the complexity of generated text as a way of making language adaptation flexible to the individual. I conclude with future directions on how systems can augment and facilitate human communication.
Tal August recently graduated with his PhD from the University of Washington and is a postdoctoral researcher at the Allen Institute for AI (AI2). His work focuses primarily on Human-Computer Interaction (HCI) while combining research in Natural Language Processing (NLP). His papers have won best paper and honorable mention awards at top tier venues, including CHI, ACL, and ICWSM. He has interned with Microsoft Research and the Semantic Scholar team at AI2. He has also been awarded university and industry fellowships, including the 2021 Twitch Research Fellowship. His long-term research goal is to improve online communication by augmenting humans’ ability to understand one another.