Measuring Language Proficiency through Eye Movements

Measuring Language Proficiency through Eye Movements

Xác định Năng lực Ngôn ngữ thông qua Chuyển động Mắt!

A study by Massachusetts Institute of Technology (MIT) researchers has uncovered a new way of telling how well people are learning English by tracking their eyes. The findings of this study could lead to the development of a method to determine if people are really ready to understand a new language. It also helps measure people’s ability to learn language.


This study was conducted by MIT researchers Yevgeni Berzak, Boris Katz and Roger Levy. Berzak is a postdoctoral fellow in the Department of Brain and Cognitive Sciences at MIT. Katz is the leader of MIT’s InfoLab Group and one of the prominent principal research scientists in the academy. And Levy is the director of the Computational Psycholinguistics Laboratory in MIT’s Department of Brain and Cognitive Sciences.

This study is detailed in the report “Gauging language proficiency through eye movement: Study tracks eye movement to determine how well people understand English as a foreign language.” The study was sponsored in part by a grant from the National Science Foundation and by MIT’s Center for Brains, Minds and Machines. The report was published in the Proceedings of the 16th Annual Meeting of the North American Region of the Society for Computational Linguistics: Human Language Technology.

How was the study conducted?

For this study, records from previous work done by Berzak were examined. Berzak’s previous work involved 145 student volunteers who were learning English as a second language and 37 native English speakers. Those foreign learners were divided into 4 groups based on their native language: Japanese, Spanish, Portuguese and Chinese.

Berzak’s earlier research collected eye movements from volunteers while they were reading texts in English, including 156 sentences, half of which were the same to every volunteer. Their eye movements were recorded on video and analyzed.

Xác định năng lực ngôn ngữ thông qua chuyển động mắt
Measuring language proficiency through eye movements


The study (“Assessing Language Proficiency from Eye Movements in Reading”) assigned an “EyeScore” for each participant who was made to read the 156 sentences. This score was based on how the participants’ eyes reacted to the words they were reading, as they were observed to intensively focus their sight on certain words. There were specific focus durations or patterns in their eye movements noted as they went along reading the sentences they were asked to read.

The EyeScore is anchored on how the eye movement patterns of English as a second language learners compare to those of native English speakers. Higher scores are given to those whose eye movement patterns were closely matched to the eye movement patterns of those who speak fluent English (as their first language). As mentioned, the study used a group of people who were learning English as a second language as well as a group of native English speakers.

After evaluating the EyeScore of the participants in the study, the researchers found a direct correlation (of the EyeScore results) with the Michigan English Test as well as the TOEFL or the Test of English as a Foreign Language. This led the researchers to conclude that EyeScore can be used as a metric for gauging language proficiency as it appeared to generate results that were comparable to those of standardized English proficiency tests.

The authors of the study consider this as the initial proof of concept that can be used to develop a system based on eye tracking for the evaluation of linguistic ability. It can bring about the development of a formal system that can be used alongside other language proficiency tests or as a standalone test.

Berzak, for his part, says that eye movement tends to reveal language proficiency to a large extent based on how the EyeScore can be measured against standardized test benchmarks. He thinks that eye movement data obtained while a person is reading is “very rich and very informative.”

Breaking into the illusion of continuity

Studying eye movements while reading is like an attempt to dig deep into the illusion of continuity in reading. There’s this phenomenon in reading wherein eye movements actually fluctuate depending on what is being read, something that hardly anyone notices and nobody may have noticed if not for the study conducted by Berzak.

As Berzak’s dataset revealed, the eyes do not move continuously upon reading a line of texts. The eyes don’t smoothly and continuously pass over words. What happens is that the eyes focus on certain words for up to 250 milliseconds. Likewise, the eyes tend to jump from one word to another for a duration of about a twentieth of a second.

According to Roger Levy, one of the three researchers, the eyes are leaping around, usually forward and occasionally backward. The brain then stitches together the information the eyes obtain and creates the illusion of a smooth and continuous reading experience. Levy says this is a demonstration of the mind’s ability to trick itself, to produce illusions.

Things are a little different, though, when it comes to learning a new language. When reading texts in a new language in particular, the eyes may dwell longer on certain words as the reader takes slightly more time to understand them or to try recalling what they mean. The eye movements of those who use English as a second or as a new language can reveal patterns that reflect their level of comprehension of the language.

Phân tích ảo tưởng về tính liên tục
Breaking into the illusion of continuity

Applications of the study

To emphasize, the findings of the research present a novel approach in measuring a person’s second language proficiency. As described by the researchers, this approach can provide standalone English proficiency scores based on eye-tracking data. This may introduce a new way to evaluate the English proficiency of immigrants entering the United States, especially in light of recent immigration policy changes in the country.

The EyeScore can serve as an objective and reliable way to test English proficiency, which can be used alongside written and spoken tests. It can even be better than written tests since it’s not possible to use a test “leak” to prepare for it. Eye movements when reading tend to be instinctive and may not be faked to imitate the eye movements of native English speakers.

Using reading eye movement tests to evaluate English proficiency might be an overkill for immigration screening, though. It entails additional costs and may not produce significant benefits. As mentioned in the description of the research, the EyeScore only shows strong correlation with standardized tests for English proficiency. Hence, it only predicts the results of such proficiency tests or affirms these results. It’s farfetched for it to become a replacement for the written tests for English proficiency.

Perhaps, it would be more applicable in the scrutiny of those who are expected to perform special language services. Providers of translation services including those working in the government, for example, can employ the EyeScore approach to ascertain that the translators they are getting are truly proficient or fluent in the foreign languages they are specializing in.

It would be great if the study included information on the cost of doing the test. The cost factor is certainly something that will be considered by users in deciding whether or not to adopt this test. The EyeScore can be used to evaluate how students are learning a new language. Language learning schools may want to consider using it as it can help them scrutinize in-depth the learning progress of students and implement pedagogical changes to achieve better learning results.

Outlook of the study

The findings of this study opens new questions that are worth exploring. Boris Katz, one of the researchers, notes that the human reading ability exemplifies the “amazing plasticity” of the brain, pointing out that humans only started processing written texts in the last several millennia. This sounds like a long time ago but it actually does not compare to the time when humans started demonstrating learning skills. Katz thinks the bigger question in all of these is how language affects the human brain.

Roger Levy, on the other hand, thinks that their eye test study can be extended beyond 156 sentences to explore more potential new information about human intelligence. It can be made more specific so experts can come up with more definitive judgments concerning even smaller strings of texts.

Levy is optimistic of the possibility that in the future, a person’s understanding of written texts can be determined on a sentence-by-sentence basis. Through the eye movement test, it may be possible to gauge how well someone understands a sentence by monitoring the movements of the eyes while a person is made to read a sentence.

Incorporating eye movement tracking in English proficiency gauging is a good demonstration of how technology improves various things or processes. Hopefully, this is something that can be done cheaply, with just a smartphone and a special app for example, so it can be adopted by more users.