TOEIC Link Reading — Subvocalization Reduction and Reading-Rate Calibration Under the Speed-Accuracy Frontier

The single most common self-diagnosis among candidates who miss questions at the end of the reading module is "I read too slowly." It is almost always wrong. Reading speed measured in words per minute is not the constraint; reading rate calibrated to the wrong target is. A candidate who reads every clause of every passage at the deliberate, fully-comprehending rate that feels safe will run out of time on a module that never asked for full comprehension of every clause. The test rewards correct answers, and correct answers depend on comprehending only the clauses that carry answer-relevant information at the depth the question demands. Reading rate, seen correctly, is not a fixed personal trait to be trained upward like a muscle. It is a resource you allocate — spending seconds where the answer lives and withdrawing them where it does not.

This reframing matters because the two obvious interventions both fail. Reading faster uniformly — pushing your eyes across every line at higher speed — trades accuracy for time at a bad exchange rate: you skim the clause that held the answer and lose the point you were trying to save time to earn. Reading everything carefully guarantees you leave points unearned at the end because the clock stops before you reach them. Neither uniform speed nor uniform care wins. What wins is variable rate: fast where the text is scaffolding, slow where the text is load-bearing, and the judgment to tell which is which in real time. This guide is about building that judgment and about the mechanical habit — subvocalization control — that makes a variable rate physically possible.

What subvocalization actually costs

Subvocalization is the inner voice that "pronounces" words as you read. It is not a defect; for most readers it is inseparable from comprehension, and the popular advice to eliminate it entirely is both impossible and counterproductive. The useful question is narrower: subvocalization sets a ceiling on reading rate roughly equal to your speaking rate, because you cannot inner-voice words faster than you could say them. On dense, load-bearing text that ceiling is fine — you want to be at speaking rate when a clause carries the answer, because the inner voice is doing real comprehension work. The cost appears on scaffolding text: transitions, restatements, throat-clearing openers, and boilerplate whose only function is to connect load-bearing clauses. Subvocalizing that material spends full speaking-rate seconds on words that carry no answer-relevant information.

So the goal is not to suppress the inner voice. It is to let it run at full strength on load-bearing clauses and to let the eyes move ahead of it — skimming, not pronouncing — on scaffolding. High-band readers do this without naming it: their inner voice goes quiet across a transition sentence and comes back on for the clause that follows. Mid-band readers subvocalize uniformly, which locks them to speaking rate across the whole passage and makes a variable rate impossible. The trainable habit, then, is selective subvocalization — knowing which clauses deserve the inner voice — and it depends entirely on the skill of recognizing scaffolding versus load, which we develop below.

The speed-accuracy frontier

Think of every reading decision as a point on a frontier. The horizontal axis is time spent on a passage; the vertical axis is probability of answering its questions correctly. At very low time, accuracy is poor — you have not read enough to answer. As time rises, accuracy climbs steeply, then flattens: past a certain point, additional seconds buy almost no additional accuracy because you have already extracted what the questions need. The optimal operating point is not the top of the curve (maximum accuracy at any time cost) and not the far left (minimum time at any accuracy cost). It is the point where the marginal accuracy per second stops being worth more than that same second spent on the next passage.

The practical rule that falls out of this: leave a passage the moment additional reading stops raising your answer probability, and carry the saved seconds to a passage still on the steep part of its curve. Candidates lose points not by reading any single passage badly but by mis-allocating across passages — over-reading an easy single-passage item into the flat zone while a hard double-passage set never gets its steep-zone seconds. This is the same allocation logic developed for whole-section timing in the companion guide on time allocation and question triage; here we apply it inside a single passage, at the level of individual clauses.

Calibrating rate by passage type

The frontier sits in a different place for different passage types, so a single reading rate is wrong by construction. Three families cover most of the module.

Single-passage informational (notices, emails, ads)

These are short and the answer usually sits in one or two identifiable clauses. The correct rate is fast scaffolding, slow target: skim to locate the clause the question points at, then drop to full speaking rate on that clause and its immediate neighbors. Reading the whole notice at comprehension rate is over-reading into the flat zone. The saccade-efficiency mechanics that make the fast-scaffolding phase possible are covered in eye-movement and saccade control for skimming.

Single-passage reasoning (inference and implicature items)

Here the answer is not stated; it is licensed by the interaction of several clauses. Fast skimming fails because the load is distributed — you cannot locate a single target clause because there is no single target clause. The correct rate is slower and more even, closer to full comprehension, because the inference depends on holding several clauses together. This is precisely the passage type analyzed in inference and implicature resolution under indirect-statement passages: spending here is not over-reading, it is the reading the question requires.

Double-passage integration

The two texts are mostly scaffolding relative to each other; the answer lives at the junction where a fact in one passage constrains a fact in the other. The correct rate is fast within each passage to build a light model of what each contains, then slow and deliberate at the cross-reference where the two models meet. Candidates who read both passages fully at comprehension rate exhaust their time before reaching the junction where the points actually are.

Why accuracy usually rises when rate is calibrated

The counterintuitive finding, borne out repeatedly in practice-corpus timing data, is that calibrating rate upward on scaffolding often raises accuracy rather than trading it away. The mechanism is attentional: a reader who subvocalizes uniformly arrives at the load-bearing clause already fatigued, having spent full attention on boilerplate, and reads the clause that holds the answer with a depleted budget. A reader who skims scaffolding arrives at the load-bearing clause fresh and spends concentrated attention exactly where it pays. Reading rate and reading accuracy are not simply traded against each other; mis-allocated slow reading degrades both. The frontier is not a straight line you slide along — it is a curve you can move outward by spending your attention in the right places.

This also explains why the "read everything carefully" strategy feels safe but scores poorly. It maximizes effort uniformly, which is not the same as maximizing correct answers. Effort spent on a transition sentence is effort not available for the clause that decides the item.

A four-week calibration protocol

Rate calibration is a habit, not a fact, so it trains through repetition under observation.

• Week 1 — measure your baseline. Read ten single passages at your natural rate and record two numbers per passage: seconds spent and questions correct. You are not trying to change anything yet; you are learning where your personal frontier currently sits and whether your errors cluster in over-read or under-read passages.
• Week 2 — mark scaffolding aloud. On each practice passage, before answering, underline the one or two clauses you believe carry the answer and bracket everything you judge to be scaffolding. Check against the answer key: were the load-bearing clauses the ones you underlined? This trains the recognition that selective subvocalization depends on. The skill is the same coreference-and-entity tracking that lets you find where the answer lives — see coreference chain resolution and entity tracking.
• Week 3 — enforce the variable rate. Now read with the explicit instruction: skim bracketed scaffolding, full attention on underlined load. Re-measure seconds and accuracy. The target is fewer seconds with equal or higher accuracy — the signature of a frontier moved outward.
• Week 4 — allocate across passages. Run full timed sections and, when a passage reaches its flat zone, force yourself to leave and bank the seconds. Track whether the banked seconds raised your accuracy on the hard passages that used to go unread.

The end state is not a faster reader in the word-per-minute sense. It is a reader who spends the inner voice deliberately — quiet on scaffolding, full on load — and who allocates seconds to the passages still climbing their curve. That is what "reading faster" should have meant all along: not moving your eyes faster, but spending your attention where the answers are.

Summary

Reading rate on the TOEIC Link reading module is a resource-allocation problem, not a fixed speed. Subvocalization sets a rate ceiling that helps on load-bearing clauses and wastes seconds on scaffolding, so the trainable habit is selective subvocalization: inner voice full on the clauses that carry answers, eyes ahead of the voice on the clauses that do not. The speed-accuracy frontier sits in different places for informational, reasoning, and double-passage items, so a single rate is wrong by construction. Calibrating rate upward on scaffolding frees fresh attention for load-bearing clauses and usually raises accuracy rather than trading it away. Train the judgment with a four-week protocol that measures your baseline, teaches you to recognize load versus scaffolding, enforces a variable rate, and allocates banked seconds to the passages still climbing their curve.