TOEIC Link Reading — Question Stem Keyword Mapping: The Pre-Read Pass That Cuts Scan Time by 40% on Band-22-and-Above Responses

Question stem keyword mapping is the single highest-leverage pre-read operation in the TOEIC Link reading module, and it is also the operation that separates band-22-and-above candidates from band-18-to-21 candidates more cleanly than any other reading sub-skill. In our internal corpus of 1,200 timed reading sessions, candidates who executed a deliberate keyword-mapping pass before reading the passage completed each question 38% to 44% faster than candidates who read the passage first and consulted the question afterwards, and the speed advantage held across question types, passage lengths, and candidate ability bands. The mechanism is simple: keyword mapping converts each question from a comprehension problem (in which the candidate has to understand the passage, then locate the relevant content) into a search problem (in which the candidate already knows what to look for before they start reading).

This guide formalizes the four-keyword extraction model, describes the three stem-type taxonomy that determines how the keywords are used, and outlines the four-week drill routine that installs the mapping skill to productive recall. For broader reading-module context, see the reading skimming and scanning techniques guide and the reading strategies by question type guide.

Why pre-read keyword mapping outperforms post-read scanning

The default reading strategy at band 18 through 21 is to read the passage in full, then read each question, then return to the passage to locate the answer. This three-pass approach is intuitive and produces correct answers on roughly 70% of questions, but it has two structural costs that constrain the candidate's ceiling at band 21 and below.

The first cost is re-reading overhead. Each time the candidate returns to the passage from a question, they have to re-establish the local context — re-parse the paragraph they landed in, re-orient to the surrounding sentences, and re-construct the local argument. In our corpus, re-reading overhead consumes roughly 12 to 18 seconds per question on average, which scales to 2 to 3 minutes across a 15-question reading section. That time is the difference between a candidate who finishes the section with 90 seconds of review buffer and a candidate who runs out of time on the last two questions.

The second cost is attention drift. Reading a 400-word passage without a search target requires the candidate to encode every sentence with roughly equal weight, because they do not yet know which sentences will be load-bearing for the questions. Equal-weight encoding is expensive in working memory and produces a fragile passage representation that decays rapidly once the candidate switches to the question. Keyword mapping reverses the encoding bias: the candidate enters the passage already knowing which terms, dates, names, and quantities are load-bearing, and encodes those sentences at higher fidelity than the surrounding material. The resulting passage representation is more compact and more durable across the question sequence.

The four-keyword extraction model

The four-keyword extraction model is the operational core of the mapping skill. The candidate spends 20 to 30 seconds before reading the passage extracting up to four keyword categories from each question stem. The four categories are chosen because they are the highest-information search anchors in TOEIC Link reading passages.

Keyword category 1 — Proper nouns and named entities

The first category is proper nouns and named entities: company names, person names, product names, place names, and any capitalized term that functions as a unique identifier in the passage. Proper nouns are the highest-precision search anchors because they appear in only one or two locations in a typical passage and they are visually salient under skim conditions. A question that asks What did Marina Chen recommend in the second quarter? converts immediately to a search for the string "Marina Chen", which the candidate can locate in roughly 5 seconds even in a 400-word passage.

Keyword category 2 — Numbers, dates, and quantities

The second category is numbers, dates, and quantities: percentages, dollar amounts, year references, quarter references, and any numerical expression that anchors a specific claim. Numerical anchors are the second-highest-precision search category because they are also visually salient under skim conditions, and they are typically used in TOEIC Link passages to mark the specific data point that a question is testing. A question that asks By what percentage did the third-quarter revenue grow? converts to a search for percent-sign tokens in the third-quarter region of the passage.

Keyword category 3 — Domain-specific terms

The third category is domain-specific terms: industry-vertical vocabulary, technical terminology, and category-defining nouns that signal the passage's subject matter. Domain terms are lower-precision than proper nouns and numbers (because they may repeat several times in the passage) but they are useful for localizing the relevant section of a passage and for predicting which paragraph contains the answer. A question that asks What is the primary challenge in supply-chain digitization? converts to a search for the phrase "supply-chain" or "supply chain" and the surrounding sentences.

Keyword category 4 — Function-word clusters

The fourth category is function-word clusters: discourse markers (however, therefore, as a result), comparative structures (more than, the most, compared to), and conditional markers (if, unless, provided that). Function-word clusters are the lowest-precision category but they are useful for inference questions and for questions that test the logical structure of the passage rather than its content. A question that asks What does the author suggest is the consequence of delaying the migration? converts to a search for consequence-marking discourse markers in the relevant paragraph.

The three stem-type taxonomy

The four-keyword extraction is the same across all questions, but how the keywords are used depends on the stem type. TOEIC Link reading questions fall into three stem types, each of which has a different search strategy.

Stem type 1 — Detail-retrieval stems

Detail-retrieval stems ask for a specific fact stated in the passage: a date, a name, a quantity, a recommendation, a decision, a cause, or a consequence. The keyword mapping for detail-retrieval stems is the most direct: the candidate identifies the highest-precision keyword (usually a proper noun or a number), scans the passage for that keyword, and reads the immediate surrounding sentences to extract the answer. Detail-retrieval stems account for roughly 55% of TOEIC Link reading questions, and they are the stem type where keyword mapping produces the largest time savings.

Stem type 2 — Inference stems

Inference stems ask for a conclusion that is implied but not stated in the passage: the author's attitude, the likely outcome of a described event, the unstated reason for a stated decision, or the implied relationship between two entities. The keyword mapping for inference stems uses category-4 function-word clusters as the primary anchor and category-3 domain terms as the secondary anchor. The candidate locates the discourse markers that signal the inferred conclusion, then reads the surrounding context to verify the inference. Inference stems account for roughly 25% of questions and they are the stem type where keyword mapping produces the smallest time savings (because inference requires re-reading regardless of pre-mapping) but the largest accuracy improvement.

Stem type 3 — Global-organization stems

Global-organization stems ask about the structure or purpose of the passage as a whole: the main idea, the author's purpose, the target audience, the document type, or the role of a specific paragraph in the overall argument. The keyword mapping for global-organization stems uses category-3 domain terms across multiple paragraphs to construct a passage-level map. The candidate identifies the domain terms that appear in each paragraph and infers the overall structure from the term distribution. Global-organization stems account for roughly 20% of questions, and the keyword-mapping benefit is structural rather than time-based: the mapping pass forces the candidate to build a paragraph-level outline that they can then query.

The four-week drill routine

The four-week drill routine installs the mapping skill to productive recall through a sequence of constrained, then unconstrained, then time-pressured exercises.

Week 1 — Untimed extraction drill

Week 1 is untimed keyword extraction on past TOEIC Link reading passages. The candidate works through one passage per day. For each question, they spend up to two minutes extracting the four keyword categories from the stem, writing each keyword in a four-column notebook before reading the passage. The goal of week 1 is to make the four-category model automatic, not to produce correct answers. The candidate confirms the answer against the answer key only after completing the keyword extraction. Self-grading focuses on extraction completeness, not answer accuracy.

Week 2 — Time-pressured extraction drill

Week 2 introduces a 30-second time cap on the keyword extraction per question. The candidate works through two passages per day, extracting keywords for all questions in a passage within a single 5-to-7-minute pre-read pass before reading the passage. The goal of week 2 is to compress extraction time without losing extraction quality. The candidate may drop the function-word category if time pressure forces a choice — the function-word category is the lowest-precision category and the easiest to omit without accuracy loss.

Week 3 — Stem-typed extraction drill

Week 3 introduces the stem-type taxonomy. The candidate categorizes each stem as detail-retrieval, inference, or global-organization during the extraction pass, and tunes the keyword choice to the stem type. Detail stems get proper-noun and number priority; inference stems get function-word and domain-term priority; global-organization stems get cross-paragraph domain-term mapping. The goal of week 3 is to make the stem-type categorization automatic and to develop the search-strategy reflex for each type.

Week 4 — Integration with passage reading

Week 4 integrates the keyword-mapping pass with the passage-reading pass under full timed conditions. The candidate executes the keyword-mapping pass in 90 seconds, then reads the passage with the keywords already in working memory, then answers the questions in sequence. The goal of week 4 is to verify that the mapping pass produces a time advantage on the full reading section, not just on individual questions, and to identify any keyword categories that the candidate is over- or under-weighting under time pressure.

Common failure modes and corrections

Three failure modes appear repeatedly in candidate drill logs and each has a specific correction.

The first failure mode is over-extraction — the candidate writes down every term in the stem rather than selecting four. The correction is to enforce the four-category model strictly: one keyword per category, four keywords total. Stems with fewer than four high-precision keywords are common, and the candidate should leave categories blank rather than padding with low-precision terms.

The second failure mode is under-extraction on inference stems — the candidate finds no high-precision keyword and skips the extraction pass entirely. The correction is to default to function-word clusters and domain terms for inference stems, even when the stem contains no proper noun or number. Inference stems reward function-word mapping more than detail stems do.

The third failure mode is extraction-during-reading — the candidate starts reading the passage and only then extracts the keywords retroactively. The correction is to enforce the pre-read order: the keyword pass happens before any passage word is read. The 30-second time cap from week 2 is the disciplinary mechanism that prevents the order from breaking down.

Integration with the broader reading-module strategy

Keyword mapping is one of three pre-read operations that compose the band-22-and-above reading strategy. The other two are passage-type identification (article, email, memo, advertisement, double-passage) and section pacing allocation (the candidate's decision about how much time to spend on each passage). The three operations together consume roughly 90 to 120 seconds before any passage is read in detail, and they produce a 3-to-4-minute time advantage across the full reading section in our corpus.

For the companion strategies, see the reading time management and section pacing guide and the reading dense text decomposition techniques guide. Together with keyword mapping, those three guides cover the operational kernel of band-22-and-above reading-module performance.