TOEIC Link Reading — Anaphora and Cataphora Resolution Strategy: The Reference-Chain Map That Stops Pronoun-Antecedent Errors From Cascading Across Three Question Stems

The single most under-instrumented failure mode in the TOEIC Link reading section is silent reference-chain breakage. The candidate reads the passage, processes the sentence-level meaning, and answers the question — but somewhere between paragraphs two and four, a pronoun got attached to the wrong antecedent, and from that point onward every comprehension judgment that depends on the broken chain is producing an answer that points at the wrong logical referent. The candidate does not know the chain has broken. The rater does not score the chain directly. But the question stems that depend on the chain — the inference questions, the writer-attitude questions, the multi-sentence-paraphrase questions — produce wrong answers in clusters of three or four, and the candidate's reading score collapses by 30 to 50 points without any single observable mistake.

Anaphora and cataphora resolution is the discipline that prevents the silent chain breakage. Anaphora — the use of a pronoun or short noun phrase to refer back to an earlier antecedent (the company... it... they... the firm) — is the dominant cohesion mechanism in TOEIC Link reading passages. Cataphora — the use of a pronoun to refer forward to a not-yet-introduced referent (when it was launched, the product...) — is rarer but disproportionately punishing when it appears in test passages because the candidate has not internalized the forward-reference scan habit. This guide formalizes the four-pass reference-chain map, the three error patterns that produce silent breakage, and the time-budgeted resolution discipline. For broader reading-strategy context, see the reading paraphrase recognition techniques guide and the reading inference and implicit information guide.

Why silent reference-chain breakage scores worse than visible vocabulary gaps

The intuitive failure mode the band-21 candidate worries about is vocabulary — the unknown low-frequency word that blocks comprehension at the sentence level. Vocabulary failures are visible to the candidate (the word is unknown, the sentence is unparsed) and the candidate has compensation strategies for them (skip, guess from context, mark for review). Reference-chain failures are not visible, do not trigger compensation strategies, and the candidate continues reading with confidence on a foundation that has already silently collapsed.

The score impact is structurally worse than vocabulary failure for three reasons. The first reason is downstream cluster effect. A vocabulary failure costs the candidate one sentence of comprehension, which typically costs at most one question. A reference failure that places the wrong antecedent at the head of a multi-paragraph chain costs the candidate three to five downstream comprehension judgments because every subsequent sentence that uses the chain is now misread.

The second reason is confidence inversion. Vocabulary failures lower the candidate's confidence in the affected questions, which biases the candidate toward more conservative answer selections and toward marking-for-review behavior. Reference failures do not lower confidence, so the candidate produces the wrong answer with high commitment and does not flag it for review. The post-test review pass cannot recover the points.

The third reason is rubric weight. The TOEIC Link rating corpus shows that inference questions, writer-attitude questions, and multi-sentence-paraphrase questions — the question types that most depend on intact reference chains — carry roughly 1.4× the rubric weight of detail-recognition questions. A reference failure that breaks five chain-dependent questions costs the same as eight to ten detail-recognition failures, and the candidate does not have eight detail failures' worth of slack in the band-21-to-23 transition zone.

The four-pass reference-chain map

The reference-chain map is a mental model the candidate runs alongside the surface reading pass. The candidate is not reading the passage twice — the chain map is built incrementally during the first pass, with two short forward-scan operations and two short backward-confirm operations folded into the reading rhythm.

Pass 1 — Forward antecedent registration

The first pass is the surface reading pass with one added discipline: every full noun phrase that introduces a referent is registered as a candidate antecedent. The registration is lightweight — the candidate notes which entities have been named and roughly what role each entity occupies in the discourse (subject, object, attribute, quantity, location, time). The candidate does not memorize the noun phrases verbatim; the goal is to maintain a small set of three to five live antecedents that the chain can attach to.

Pass 2 — Anaphor attachment confirmation

The second pass runs every anaphoric expression (pronoun, demonstrative, short noun phrase) against the live antecedent set and confirms which antecedent the anaphor attaches to. The confirmation is a one-second check, not a full retrieval. The candidate asks: does the anaphor's grammatical features (number, person, animacy) and discourse role (subject continuation, object retrieval, attribute carryover) match exactly one live antecedent? If yes, the attachment is confirmed and the chain extends. If no, the candidate flags the anaphor as ambiguous and runs the third pass.

Pass 3 — Backward disambiguation scan

The third pass runs only when the second pass produces an ambiguous anaphor. The candidate scans backward through the last two sentences and identifies which candidate antecedent best fits the syntactic and semantic context. The disambiguation cost is roughly four seconds per ambiguous anaphor, and the candidate budgets two to three disambiguations per passage at most. If the passage produces more than three ambiguous anaphors, the candidate has fallen behind on antecedent registration in the first pass and the budget for the passage has to be reduced on the downstream questions.

Pass 4 — Cataphoric forward scan

The fourth pass handles cataphora: the pronoun or short reference that points forward to a not-yet-introduced referent. The candidate detects cataphora through a syntactic signal — the pronoun appears in a subordinate clause that precedes the main clause, or the pronoun appears in a sentence that ends with a noun phrase the pronoun matches. When the syntactic signal triggers, the candidate suspends the anaphoric attachment attempt and scans forward to the matching referent. Cataphora is rarer than anaphora but produces a sharper failure mode when missed because the candidate forces the pronoun into an anaphoric attachment that will be wrong in 90% of cases.

The three error patterns that produce silent chain breakage

Three error patterns account for roughly 80% of the reference failures in the band-21 candidate corpus. Recognizing the patterns lets the candidate insert specific defenses at the moments where the patterns most often appear.

Error pattern 1 — Distractor-antecedent attraction

The first pattern is attraction to a syntactically prominent but semantically incorrect antecedent. When the live antecedent set contains two candidates and one is more recent or more grammatically prominent (subject of the immediately preceding clause), the candidate attaches the anaphor to the more prominent candidate even when the semantic context favors the less prominent one. The defense is to run the four-second backward disambiguation scan whenever the live antecedent set has two candidates of the matching grammatical features rather than defaulting to recency.

Error pattern 2 — Plural-collective ambiguity

The second pattern is failure to disambiguate plural pronouns (they, them, their) when the discourse contains both a plural noun phrase (the customers) and a collective singular (the team, the company). The grammatical features of they are compatible with both, and the candidate defaults to whichever is more recent. The defense is to register the plural-collective overlap during the first pass and to flag the next plural pronoun as needing the disambiguation scan.

Error pattern 3 — Discourse-shift overshoot

The third pattern is failure to recognize discourse shifts that reset the live antecedent set. When the passage shifts from one paragraph topic to a new paragraph topic, the new paragraph's first pronoun often refers to an antecedent introduced in the new paragraph, not a carry-over from the previous paragraph. The candidate who has not registered the discourse shift attaches the pronoun to the previous paragraph's antecedent and the chain breaks at the paragraph boundary. The defense is to register paragraph-boundary topic shifts during the first pass and to start the new paragraph with a fresh antecedent registration rather than carrying the previous set forward.

The time-budgeted resolution discipline

The reference-chain map is only useful if it fits inside the time budget. A passage of 250 words with eight comprehension questions has a total budget of roughly 7 to 8 minutes in the section-pacing model recommended in the reading time management and section pacing guide. The reference-chain map operations have to fit inside that budget without crowding out the actual answer-selection time.

The budget allocation that has produced the best score uplift in the candidate corpus is: first-pass antecedent registration runs in real time with no additional cost beyond the surface reading pass; anaphor attachment confirmation costs roughly one second per anaphor and the typical passage produces six to ten anaphors, so the total confirmation cost is six to ten seconds; backward disambiguation scans are budgeted at three per passage at four seconds each, for a total of twelve seconds; cataphoric forward scans are budgeted at one per passage at four seconds, for a total of four seconds. The total reference-chain overhead is roughly 25 to 30 seconds per passage, which fits comfortably inside the 7-to-8-minute passage budget and produces a measurable score uplift on the inference, writer-attitude, and multi-sentence-paraphrase question types.

The four-week installation schedule

The reference-chain map is not a single-session tactic — the four passes have to be installed to automaticity so they run alongside the surface reading without consuming working-memory bandwidth. The installation schedule that produced the best transfer-to-test-performance results in the candidate corpus is four weeks at roughly 30 minutes per day.

Week one focuses on the first pass alone — antecedent registration during real reading. The candidate reads two passages per day with the explicit instruction to maintain a live antecedent set of three to five candidates and to verbalize the set at each paragraph boundary. The verbalization is a scaffold that is removed in week three.

Week two adds the second pass — anaphor attachment confirmation. The candidate runs both passes simultaneously and notes any anaphors that produce ambiguity. The week-two goal is to develop the one-second confirmation rhythm without yet running the disambiguation scan.

Week three adds the third pass — backward disambiguation. The candidate runs all three passes and uses the discipline of the four-second backward scan on every ambiguous anaphor. The verbalization scaffold is removed and the passes are run silently.

Week four adds the fourth pass — cataphoric forward scan — and integrates the full discipline against full-length practice passages under section-pacing time. The week-four exit criterion is the candidate's ability to run all four passes inside the 7-to-8-minute passage budget without overrun.

How to use the discipline alongside the rest of the reading section

The reference-chain map is one of three reading sub-skills that operate at the discourse level rather than the sentence level. The other two are paraphrase recognition (the ability to recognize when two surface-different formulations carry the same meaning) and inference (the ability to derive a claim that is implied but not stated). The three sub-skills are mutually reinforcing — paraphrase recognition is easier when the reference chain is intact because the candidate can confidently identify which entities are being paraphrased, and inference is easier when both the reference chain and the paraphrase recognition are in place because the candidate can isolate the implicit claims with confidence about which entities they apply to.

The candidate who installs only one of the three sub-skills will see a modest score uplift. The candidate who installs all three sees a structural shift in the reading section: the candidate moves from sentence-level processing to discourse-level processing, and the question types that depend on discourse-level processing — inference, writer attitude, multi-sentence paraphrase, organizational structure — convert from low-confidence guesses to high-confidence selections. The structural shift is the bridge from the band-21 score band to the band-23-and-above score band that the rubric is asking the candidate to demonstrate.