Thirty English passages, 41 Japanese-speaking readers, and a finding that an individual’s departure from the group’s judgement carries information of its own.
Text difficulty is usually treated as a property of the text. Readability formulas count syllables, sentence lengths and word frequencies, and language teaching assigns levels accordingly — on the assumption that a passage is hard or easy more or less independently of who is reading it.
A study in Reading & Writing from the University of Tsukuba tests that assumption against two other sources of information: what readers say about a passage, and what their eyes do while reading it. The relationship between the three turns out to be close but not redundant.
The correspondence authors are Shingo Nahatame, assistant professor at the Institute of Human Sciences, University of Tsukuba, and Tomoko Ogiso, associate professor in the Faculty of Education at the University of Toyama.
Three measures of the same thirty passages
The study assessed 30 English reading passages spanning a broad range of difficulty, measured three ways.
Objective difficulty came from established readability measures together with an evaluation by a large language model — which is a notable methodological choice, since a language model’s judgement of difficulty is neither a formula nor a human rating but something in between, and using it alongside traditional metrics treats it as a third kind of instrument.
Subjective difficulty came from 41 Japanese first-language undergraduate and graduate students, who rated each passage on a nine-point scale.
And eye movements were recorded while they read — reading speed and word skipping being the measures reported.
The two kinds of difficulty agree
Objective and subjective difficulty were strongly related: passages judged harder by the objective measures were also rated harder by readers.
The eye-movement data agreed too. Harder texts were read more slowly, and readers skipped fewer words in them — skipping being the thing a reader does when a word is predictable enough that the eyes can pass over it, which stops happening when a text is demanding.
So far this is confirmation rather than discovery, and confirmation of a sort the field needed. Readability formulas were built on English-speaking readers and are applied routinely to second-language learners, so evidence that they track what second-language readers actually experience is worth having.
The finding that is not confirmation
The result the paper turns on is about disagreement. An individual reader’s deviation from the group average rating for a passage explained eye-movement behaviour beyond what the text-level measures could explain.
That sentence is doing something specific, and it is worth unpacking. For each passage there is an average rating across the 41 readers, and each reader’s own rating sits above or below it. That residual — this reader found this passage harder than most people did — predicted how their eyes moved, over and above everything the text’s own difficulty predicted.
Which means subjective difficulty is not simply a noisy estimate of objective difficulty. If it were, the deviations would be measurement error and would predict nothing. They predict reading behaviour, so they are carrying real information about the encounter between a particular reader and a particular text.
Why that should be is not mysterious. A reader’s difficulty depends on vocabulary overlap with the passage, on background knowledge of its topic, on familiarity with its genre and on how much working memory the syntax demands of them specifically. None of those is a property of the text, and all of them affect how long the eyes rest on a word.
What it implies for teaching
The applied reading is that levelling a text is necessary and insufficient.
A readability score tells a teacher roughly where a passage sits, and it cannot tell them that a particular student will find this passage harder than its level suggests because its subject is unfamiliar. The study’s finding is that students have usable information about this, since their own judgements predict their reading behaviour beyond the level — which is an argument for asking learners how hard something felt rather than only testing whether they understood it.
It also suggests those self-reports are not merely about confidence. A rating that predicts eye movements is tracking processing load, not just self-belief.
Thirty passages and forty-one readers
The limits are mostly about scale, and the release does not raise them.
Forty-one readers reading 30 passages is a reasonable eye-tracking study — the method is labour-intensive and samples this size are normal — and it is small for a claim about individual differences. The deviation effect is estimated across readers whose own reading ability, vocabulary size and topic knowledge varied, and 41 people cannot support much analysis of which of those explains the deviations.
The participants were also a homogeneous group: Japanese first-language university students learning English. Whether the same holds for readers of other first languages, at other proficiency levels, or outside higher education is untested here, and the relationship between first language and second-language reading difficulty is substantial — a Japanese speaker and a German speaker do not find the same English sentences hard.
The release reports no statistical values at all: no correlation coefficients, no effect sizes, no indication of how much additional variance the deviation measure explained. For a finding whose entire claim is that one predictor adds to another, the amount it adds is the result.
There are also no quotations from either researcher on the page, so the interpretation here follows the release’s summary rather than anything the authors said in their own words.
What the study contributes is a clean demonstration that a reader’s sense of difficulty is a measurement rather than an opinion — and that the part of it which disagrees with the text’s measured difficulty is the informative part.
The paper was published on 12 August 2026. The research was supported by JSPS KAKENHI grants 23K25335 and 26K00161; the release does not state any conflicts of interest or give a full author list.
Notes
- “Subjective and Objective Text Difficulty in L2 English Reading: Exploring Their Interplay in Relation to Eye-Movement Behavior.” Reading & Writing, 2026. DOI: 10.1007/s11145-026-10888-0.
- University of Tsukuba. “Subjective and objective difficulty of English texts are related—but not the same.” EurekAlert!, 5 October 2026. https://www.eurekalert.org/news-releases/1146426.
Photo by George Milton on Pexels.com