TOEIC Link Reading — Referent Tracking and Pronoun-Antecedent Resolution Across Paragraph Boundaries

Referent-tracking failure is the quietest score leak on the TOEIC Link reading module. It rarely shows up as an item the candidate finds hard; it shows up as an item the candidate finds easy and answers wrong, because somewhere a few sentences back a pronoun was attached to the wrong entity and every conclusion built on top of it inherited the error. The candidate decodes every word correctly, reads at full speed, and still selects a distractor — not because the vocabulary was hard but because "it," "they," "the latter," or "this figure" quietly pointed at the wrong thing. The skill under test is not word recognition. It is the maintenance of a stable mental model of who and what the passage is talking about as the entities accumulate and the pronouns pile up.

Internal practice-corpus data indicates that referent-linked errors account for roughly one in five wrong answers among 22-to-25 band candidates on multi-paragraph passages, and that the error rate on items requiring a cross-paragraph referent link runs nearly double the rate on items whose referent is in the same sentence. High-band candidates show almost no distance penalty: their accuracy on cross-paragraph referent items is within a few points of their same-sentence accuracy. That flatness is the marker of the skill. It means high-band candidates carry an explicit, updated model of the passage's entities, so distance does not degrade their resolution. Mid-band candidates carry an implicit model that decays with distance, which is why their errors cluster exactly where the referent and its antecedent are far apart. For the reasoning skill that referent tracking feeds — drawing only the conclusions the passage licenses — see the companion guide on inference and implicature resolution under indirect-statement passages, and for the foundational mechanics see coreference chain resolution and entity tracking.

The entity-stack model

The most reliable way to hold referents stable is to read as if maintaining an explicit stack of the passage's active entities. Every time the passage introduces a new referring expression — a proper noun, a definite noun phrase, a number, a named document — it pushes an entity onto the stack. Every pronoun or reduced reference pops the correct entity by matching it against grammatical and semantic constraints. The candidate who reads this way never asks "what does 'it' refer to?" as an isolated puzzle; they ask "which entity on my current stack satisfies the constraints on 'it'?" and the answer is usually forced.

The stack has three properties worth naming explicitly:

• Recency ordering. The most recently mentioned compatible entity is the default antecedent for a pronoun. English strongly prefers recent referents, and TOEIC Link passages follow the default unless a marked construction overrides it.
• Grammatical gating. Number and, where marked, gender filter the stack. "They" cannot pop a singular entity; "it" cannot pop a plural set. This gate alone resolves a large fraction of items mechanically.
• Semantic gating. Selectional restrictions filter further. In "the board reviewed the proposal and rejected it," "it" cannot be "the board" because boards are not the kind of thing one rejects in that clause; "the proposal" is forced.

The discipline is to keep the stack current — to update it at every new mention rather than reconstructing it under pressure when a question demands it. Reconstruction under pressure is where distance decay does its damage.

The four distance-decay failure zones

Referent errors are not uniformly distributed. They concentrate in four zones, and knowing the zones lets a candidate raise vigilance exactly where the risk is.

Zone 1 — The paragraph-boundary carryover

A paragraph ends with an entity in focus; the next paragraph opens with a pronoun that refers back across the boundary. "...the committee deferred the decision. They would reconvene the following week." The pronoun's antecedent is in the previous paragraph, and candidates who mentally "reset" at paragraph breaks lose it. Treat a paragraph-opening pronoun as a signal to hold the prior paragraph's focus entity active, not to flush it.

Zone 2 — The competing-antecedent cluster

Two entities of matching number sit close together, and a later pronoun could grammatically pop either. "The auditor sent the report to the director before she left for the conference." Both "the auditor" and "the director" are female-compatible; resolution requires semantic and syntactic cues (subject preference, the logic of who attends conferences in the passage's frame), not grammar alone.

Zone 3 — The reduced-reference chain

The passage refers to an entity through a shifting series of reduced expressions — "the acquisition," then "the deal," then "the transaction," then "it." Candidates who fail to recognize these as coreferential split one entity into several and lose the thread. Recognizing lexical variation as coreference is a distinct sub-skill; the surface form changes while the referent does not.

Zone 4 — The quantitative-referent hop

A number, percentage, or figure is introduced, discussed, and then referred to as "this figure," "that increase," or "the former." Numeric referents are especially error-prone because candidates track the value but not the role — they remember "forty percent" but lose whether "this figure" points at the growth rate or the market share. Anchor each number to its role, not just its magnitude.

Ambiguity arbitration: the cue hierarchy

When two entities compete for a pronoun, resolve in this fixed order and stop at the first cue that decides it.

1. Number and gender agreement. Eliminate any candidate the pronoun's grammatical features exclude. Often decisive on its own.
2. Selectional compatibility. Eliminate any candidate the predicate cannot semantically take. "Rejected it" cannot take an animate agent as the object.
3. Syntactic parallelism. In coordinate structures, a pronoun tends to pick the antecedent occupying the parallel grammatical slot. "The manager praised the analyst, and then she was promoted" leans toward "the analyst" by parallel subject-of-consequence logic in many business frames — but this cue is weaker than 1 and 2 and yields to them.
4. Subject-of-topic preference. All else equal, the pronoun favors the entity that is the paragraph's topical subject — the entity the passage is "about" — over an incidental mention.
5. Global coherence. If cues 1 through 4 leave a genuine tie, choose the reading that makes the surrounding discourse coherent. TOEIC Link items are engineered so that exactly one reading produces a coherent passage; the incoherent reading is the trap.

The value of a fixed hierarchy is that it converts an open-ended "which one feels right" judgment into a short deterministic checklist. High-band candidates run this checklist so fast it feels like intuition; mid-band candidates who adopt it explicitly close most of the gap.

Worked example

Passage excerpt: "The procurement team submitted two bids to the finance office. The first came from a domestic manufacturer; the second from an overseas supplier with a lower unit price but longer lead times. After reviewing both, the office recommended the latter — though it warned that its delivery schedule would need close monitoring."

• "the latter" pops "the overseas supplier" by the first/latter ordering — a reduced ordinal reference, Zone 3 style. Candidates who lose the "first/second" ordering across the sentence boundary invert it.
• "its" is the arbitration point. Grammatically "its" is singular and could pop "the office," "the overseas supplier," or "the delivery schedule." Selectional compatibility (cue 2) plus coherence (cue 5) force "the overseas supplier": it is that supplier's delivery schedule that needs monitoring, given the passage flagged its longer lead times. A distractor answer would attach the warning to the finance office's own schedule, which the passage never establishes.

The four-week drill protocol

Referent stability is built by making the stack explicit until maintaining it becomes automatic.

• Week 1 — Stack annotation. On every multi-paragraph passage, underline each pronoun and reduced reference and write its antecedent in the margin before answering any question. Build the habit of resolving referents as you read, not when a question forces it.
• Week 2 — Zone-targeted vigilance. Practice passages selected for dense Zone 1 and Zone 2 features — paragraph-boundary carryovers and competing-antecedent clusters. Raise annotation effort specifically at these high-risk sites.
• Week 3 — Cue-hierarchy application. On every ambiguous referent, name which cue in the hierarchy resolved it. This trains you to reach the deciding cue fast and to trust it over gut feeling.
• Week 4 — Timed compression. Drop the margin annotations; maintain the stack mentally under a per-passage time cap. The target is to preserve same-sentence accuracy on cross-paragraph referents — to flatten the distance-decay curve that marks the mid-band candidate.

Candidates who complete this protocol typically eliminate the distance penalty on referent items, which removes an entire category of "easy item answered wrong" losses and lifts multi-paragraph passage accuracy by several points without any gain in vocabulary or reading speed.

Summary

Referent tracking is the maintenance of a stable entity model across distance. Read as if pushing every new mention onto an entity stack and popping the correct antecedent by recency, grammatical gating, and semantic gating. Raise vigilance in the four distance-decay zones — paragraph-boundary carryover, competing-antecedent clusters, reduced-reference chains, and quantitative-referent hops — where mid-band errors concentrate. Arbitrate ambiguity with the fixed cue hierarchy: agreement, then selectional compatibility, then parallelism, then topic preference, then global coherence. The four-week protocol builds the discipline by externalizing the stack, then compressing it to test pace. For the reasoning layer that a stable referent model enables, continue with inference and implicature resolution under indirect-statement passages.