Millions of people open Duolingo every day to learn a new language, but few of these learners likely understand the science behind language processing.
Prof Roland, a psycholinguistics professor at the University of Tokyo, researches how people process language and why second-language comprehension is often less automatic than comprehension in a first language. In an interview with the Todai Shimbun, he discusses his unconventional academic path, the challenges of language processing research, and the differences between humans and artificial intelligence. (Interview and Photos by Kamil Zeribi)
Growing up in Delaware, United States—home to a significant chemical industry presence—Prof Roland initially pursued chemistry during his undergraduate years. By chance, he ended up in a linguistics class in order to complete university credits, where he was drawn to its similar analytical nature, commenting that linguistics was “chemistry but with words”. Since then, Prof Roland’s interest in the subject increasingly grew until his graduation.
It wasn’t just linguistics that caught his attention during his undergraduate years. Prof Roland’s interest in Japan also grew throughout university, where he came across an opportunity to travel across the United States for $500 to help visiting international students. During the trip, he interacted with many Japanese students, which sparked his interest in visiting Japan. Prof Roland’s initial plan – to only go to Japan for a short time – was extended to two years. Ultimately, teaching in Japan convinced Prof Roland to decide to pursue psycholinguistics at graduate school, moving back to the US to study first language comprehension and production at the University of Colorado Boulder. His return ended up being temporary, as Prof Roland soon found himself longing for Tokyo — the only place he had “ever been homesick for”.
After returning to Japan, his research focus shifted from first to second language acquisition. Unlike the US, where there are many first-language English speakers, this was not the case for Japan. While conducting his research, Prof Roland faced logistical challenges where he had enough subjects for “one experiment, but if [he] needed 40 more people, [he’d] run out.” The natural pivot was to study second-language acquisition — in a nation where English learning is a multi-billion dollar market, data was in ready supply.
Prof Roland specifically studies second-language acquisition through “how people make decisions as they read.” As an individual reads, their eyes move across the text, with various factors determining how long they look at a word. He examines whether something is preventing second-language learners from making decisions during reading at the “right point in time”, and whether their eye movements become more automatic as they become more fluent.
This is difficult to measure exactly. Prof Roland approaches it through the use of eye-tracker technology, which encounters multiple obstacles. The eye tracker’s limited resolution makes it difficult when a target word is very short, such as “I” or “a,” and precise accuracy is needed. Beyond technical issues, acquiring a truly random sample is incredibly difficult. Due to being the most accessible for researchers, most research is conducted on university students. Prof Roland wonders if they are “representative of human beings in general”, or if they are “just a subset of highly educated people” that may distort data if not carefully considered.
The important thing is to minimize the impact of these challenges on the final data, in order to ensure the clearest results possible. A critical realization from Prof Roland’s research is the difference in how first-language and second-language learners read. For first-language speakers, the process is automatic as they engage in “low-level automatic processing”, where their eyes are influenced by specific linguistic factors almost immediately. In contrast, his research reveals that “the factors that are influencing the native speaker’s decisions are not influencing the non-native speakers”. Even when a second-language learner is highly proficient, their brain is either processing the information more slowly or physically unable to make the decisions at the same speed.
In language acquisition, this discrepancy is referred to as a “boundary point”. Psycholinguists are unsure why it exists, yet, it is clear that a specific threshold where certain linguistic structures become automatic exists. Prof Roland references his own journey of learning Japanese, where certain grammatical structures and other difficulties that he struggled with for years have become automatic – now his brain simply processes them as “the right way” without conscious effort. To this, he responds, “the more exposure you have, the more automatic” the process becomes.
Generative artificial intelligence (AI) learns in a similar way, improving accuracy with increased input. However, modern-day generative AI provides the common misconception of talking to a cognitive machine that responds thoughtfully, when the reality is a complex algorithm that generates a series of probable next words. Prof Roland argues that this distinction separates humans from generative AI, referring to the term “embodied cognition” which is the concept that humans gain experience through physical and emotional interaction with the world around them. Prof Roland uses dogs as an example. A large language model (LLM) may be able to find pictures of dogs and understand it through text statistics, but “doesn’t know a dog by interacting with one as a kid”, which is a fundamental part of how humans generate linguistic predictions.
Rather, LLMs are “closer to having a dictionary and a grammar book” than a true understanding of language. Humans acquire negative and positive semantic associations through unique interpersonal experiences, formulating a nuanced, layered understanding of language that generative AI lacks.
These issues translate into the classroom. Prof Roland does not “necessarily mind” generative AI use, in fact, he encourages responsible use, but sees an issue when generative AI use results in “superficial and fluffy” content. For example, Prof Roland noted that in one of his classes, students’ excessive reliance on generative AI was apparent from their limited understanding of the material they were presenting, which in turn made it difficult for the audience to understand the content. Excessive reliance on generative AI means students do not engage in the encoding process, which is essential for learning and memory, and fail to store information effectively in their brains.
For Prof Roland, language learning is a human process. While technologies such as generative AI can and should be useful tools, they cannot replicate the experience and interactions that shape how people understand language. Learning a language extends beyond word and grammar memorization, extending into contextual understanding and the cultural nuances that make communication so difficult, but worthwhile. Prof Roland intends to keep pursuing his research with the same question at heart: “I need to understand why we understand.”
The post UTokyo’s Prof Roland and the Human Process Behind Language Learning first appeared on 東大新聞オンライン.
UTokyo’s Prof Roland and the Human Process Behind Language Learning 2026
IUSTITIA.BG – Investigations 2009-2026 2026-10-04 02:36:30
最新ニュース、世界のニュース、国内ニュースと本日の重要情報。ペタル・ニザモフ裁判官とフェザーズ・ペタル・ニザモフがブルガスおよびブルガリア全土で分析と調査を提供します。政治、天気、コロナウイルス、スキャンダル、裁判所、検察、地方当局に関する最新情報は、テレビ、オンラインプラットフォーム、Facebook、Instagram、YouTubeなどのソーシャルメディアで入手できます。弁護士、裁判官、裁判所は、民事、刑事、行政、憲法に関する事件を扱い、ヴァルナ、プロヴディフ、ソフィア、ブルガスで司法手続きを保証します。Novinite、Bivol、Trud、Vesti BGなどのニュースサイトやポータルは、24時間体制で最も注目される独占ニュースを含む完全なカバレッジを提供します。
