TOEIC Link Reading — Eye Movement and Saccade Control for Skimming Efficiency: How Fixation Discipline, Saccade Length, and Regression Suppression Determine Your Reading Throughput

Most TOEIC Link Reading candidates who plateau around the CEFR B1-B2 boundary attribute their plateau to vocabulary gaps or to slow grammar processing. Vocabulary and grammar do constrain comprehension, but the bottleneck for the median adult Japanese-L1 candidate at this band is rarely lexical — it is eye-movement behaviour inherited from L1 reading habits. The eye that reads Japanese-language news at a comfortable pace lands on roughly two characters per fixation, executes short saccades of two-to-three characters between fixations, and regresses (backtracks) rarely because Japanese orthography is dense with semantic content per character. The same eye, reading English prose, lands on a single short function word per fixation, executes saccades of three-to-four letters that do not even span a content word, and regresses constantly because the reader does not yet trust the parafoveal preview to deliver the next word. The result is a reading throughput around 120-150 words per minute — below the threshold required for TOEIC Link Reading's longer dense-text passages, which presume a sustained throughput of 200-240 words per minute for skimming and 160-200 words per minute for careful reading.

This guide explains the saccade-fixation-regression cycle that governs reading speed, the three eye-movement failure modes that cap reading throughput around 150 words per minute, and the four drills that retrain the eye for adaptive-test-grade skimming without the comprehension loss that usually accompanies speed-reading attempts. For related reading guides, see reading time management and section pacing and reading dense-text decomposition techniques.

The saccade-fixation-regression cycle

Reading is not a smooth left-to-right scan. The eye executes a discrete cycle of three movements thousands of times per page, and the efficiency of each cycle component determines reading throughput.

Fixation is the brief pause during which the eye is stationary and the brain extracts visual-and-lexical information from the page. A fixation lasts roughly 200-250 milliseconds for a fluent adult reader of English and 250-350 milliseconds for an L2 reader at the B1 band. The fixation does not sample only the letter directly under the gaze — the foveal window covers roughly seven-to-nine characters around the gaze point, and the parafoveal window extends another nine-to-fifteen characters to the right, delivering enough preview information to plan the next saccade.

Saccade is the rapid ballistic jump from one fixation to the next. A saccade takes roughly 20-40 milliseconds, during which the eye is in motion and the brain receives no useful visual input. A skilled adult reader of English executes saccades of seven-to-nine letters on average — spanning approximately one short content word or one-and-a-half function words. An L2 reader at the B1 band typically executes saccades of three-to-five letters, which is too short to span most content words and forces the eye to land twice within a single word.

Regression is a backwards saccade that returns the gaze to material already read. Skilled adult readers regress on roughly ten-to-fifteen percent of saccades, usually to disambiguate a syntactic attachment or to confirm a referent. L2 readers at the B1 band regress on thirty-to-forty percent of saccades, usually because the parafoveal preview was insufficient to identify the next word and the reader needed to reread the previous word to maintain comprehension.

The cycle is multiplicative. A reader who fixates for 300 milliseconds, executes 30-millisecond saccades of four letters, and regresses on thirty-five percent of saccades will read at roughly 135 words per minute. A reader who fixates for 220 milliseconds, executes 30-millisecond saccades of eight letters, and regresses on twelve percent of saccades will read at roughly 240 words per minute — nearly twice the throughput from optimizing the same three variables.

The three eye-movement failure modes that cap throughput

The L2 reader who has plateaued around 150 words per minute is almost always exhibiting one or more of three failure modes. Identifying which failure mode dominates is the prerequisite for selecting the correct drill.

Failure mode one: short saccades from foveal over-reliance

The reader does not trust the parafoveal preview to deliver word-identity information and consequently relies on foveal vision for every word. The eye lands directly on each word's centre, extracts the word, and saccades to the centre of the next word. Saccades become three-to-five letters long because the reader is unwilling to land between words or on a word's edge. This produces a high fixation count per line — twelve to fifteen fixations on a line that a skilled reader would cover in seven to nine fixations.

The diagnostic signal is that the reader's eye visibly stops on every word, including short function words ("the", "of", "to", "and"). A skilled reader skips approximately sixty percent of short function words because the parafoveal preview is sufficient to identify them without a direct fixation. The L2 reader who fixates on every "the" is leaking roughly thirty percent of available throughput.

Failure mode two: long fixations from insufficient lexical-retrieval speed

The reader's saccade and regression behaviour may be reasonable, but each fixation lasts 350-450 milliseconds because the lexical retrieval — the cognitive step of recognizing the word and accessing its meaning — is slow. This failure mode is the dominant one for readers whose vocabulary is genuinely thin or whose word-recognition has not been automatized through volume reading. Even with optimal saccades and minimal regressions, the throughput is capped by fixation duration.

The diagnostic signal is that the reader's comprehension is good on a passage they have read before but their first-pass speed on unfamiliar text is slow. Re-reading is fast because the lexical retrievals are pre-warmed; first-pass reading is slow because each lexical retrieval is being executed from cold.

Failure mode three: high regression rate from syntactic-parsing fragility

The reader's saccade length and fixation duration may be reasonable, but the eye regresses on thirty-to-forty percent of saccades because the reader's syntactic-parsing pipeline is fragile and frequently fails on long modifier chains, embedded clauses, or reduced relatives. Each parse failure triggers a regression to the parse-failure point, where the reader reattempts the parse from scratch.

The diagnostic signal is that regressions are not uniformly distributed — they cluster around grammatical constructions the reader has not automatized. A reader who regresses on every relative clause but not on simple SVO sentences is exhibiting failure mode three.

The four drills that retrain eye movement

Eye-movement retraining is unusual among reading-skill drills because the gains are mechanical rather than knowledge-based. A vocabulary drill produces gains only on the words drilled; an eye-movement drill produces gains on all text the reader subsequently encounters, including text that contains words and structures the reader has not yet seen.

Drill one: column-width pacing. The reader reads a passage formatted in a narrow column (roughly forty characters wide) at a metronome-driven pace of three lines per second initially, scaling up to four-to-five lines per second over four weeks. The narrow column forces the eye to execute fewer fixations per line, which retrains the saccade-length distribution toward longer saccades. The metronome forces the reader to execute the saccade on schedule rather than waiting for full word recognition, which retrains the parafoveal-preview confidence. This drill targets failure mode one.

Drill two: timed re-reading. The reader reads a passage for full comprehension at a comfortable pace, then re-reads the same passage three additional times under decreasing time pressure (75 percent, 50 percent, 35 percent of the original time). The re-readings build lexical-retrieval speed for the passage's vocabulary and syntactic-parsing speed for the passage's structures, and the speed gains transfer partially to subsequent unfamiliar passages. This drill targets failure mode two.

Drill three: structural-pattern recognition. The reader works through a deck of grammatical constructions that produce parse failures (reduced relatives, garden-path sentences, heavy-NP-shifted constructions, extraposed clauses) and trains the eye to recognize the construction's shape before attempting to parse its content. The recognition step pre-loads the parser with the construction's expected shape, which reduces the parse-failure rate and the consequent regressions. This drill targets failure mode three.

Drill four: meta-fixation suppression. The reader practices a deliberate suppression of the urge to regress when comprehension feels uncertain. The instruction is: when uncertainty arises, continue forward for at least three additional words before considering a regression. The rule exploits the fact that downstream context frequently resolves the uncertainty without an explicit regression. This drill reduces unnecessary regressions across all three failure modes.

How to deploy eye-movement training on TOEIC Link Reading

Eye-movement gains transfer to TOEIC Link Reading at scale only if the training is sustained across four-to-eight weeks. A two-week sprint produces measurable gains on the drill passages but fades within ten days of stopping. The recommended deployment is fifteen minutes per day across four-to-eight weeks, with the drill mix weighted toward whichever failure mode the diagnostic indicates is dominant.

The throughput target for TOEIC Link Reading is sustained skimming at 200-240 words per minute on first-pass dense business prose, with careful re-reading reserved for the two-to-four sentences per passage that contain the question-relevant information. A reader who reaches this throughput will complete the Reading module within the typical 30-35 minute window with three-to-five minutes of buffer for question-stem review and final-answer confirmation, rather than running out of time on the final third of the passages — the failure mode that caps most B1-band candidates' Reading scores.

For complementary reading-strategy guides, see reading question-stem distractor pattern recognition and reading inference and implicit information.