The Lexical-Overlap Lure
A class of 14 upper-intermediate learners takes a weekly inference quiz. On last week's 90-second audio about coral bleaching, 11 of 14 correctly listed the main causes and the 2 °C threshold, so literal comprehension is solid. On the four inference items, 9 of 14 chose an option that reused the audio's exact wording ('the reef will not recover') over the correct paraphrase ('recovery depends on how long the heat lasts'). In reading the pattern repeats: across 12 items, errors were 3.2 times more likely when the correct answer was a paraphrase and a distractor shared two or more content words with the text. Two of the three strongest students said they 'read the question first, then hunt for the sentence that matches.' The class's vocabulary scores are unremarkable but not low.
1. Is the constraint vocabulary, memory, or strategy? Which observations decide the question, and which one is only suggestive?
Read the worked answer
Strategy and monitoring - specifically, matching on surface form instead of testing meaning.
- Vocabulary is ruled out. If unknown words were the issue, errors would spread across item types and the class would miss literal items too; literal accuracy is high and vocabulary scores are mid-range, not low.
- Memory is ruled out as the primary cause. The failure appears in reading as well as listening, where the text stays visible. A memory account predicts a listening-only deficit.
- The deciding observation is the error asymmetry: errors are 3.2× more likely precisely when a distractor shares two or more content words with the source. That is an item-level fingerprint of lexical matching, not of comprehension ability.
- Only suggestive: the two strong students' self-report ('hunt for the sentence that matches'). It is consistent with the diagnosis and worth acting on, but self-reports about process are unreliable; the 3.2× figure is what carries the argument.
The mechanism: overlap feels like correctness, so no verification step ever runs.
2. Give a three-step procedure students can apply to an inference item, and name the habit it must replace.
Read the worked answer
Procedure:
- Predict before looking. After hearing or reading the question, state in one clause what the passage implies, in your own words, with the options covered.
- Then read every option and paraphrase each one back into your own words. Any option that will not paraphrase is suspect; any option that copies the source's wording is a candidate distractor, not a candidate answer.
- Demand a warrant. Point to the two or three words in the passage that support the chosen option, and say what would have to be true in the passage for the strongest distractor to be right.
Habit replaced: 'question first, then find the matching sentence' - which converts an inference item into a word-search task and rewards overlap. Note that step 2 inverts the usual instinct: it makes exact wording a warning sign rather than a comfort.
3. What short check would show whether the fix transferred to new material?
Read the worked answer
A ten-item transfer set built to make overlap a trap on purpose: five items where the correct answer is a paraphrase and a distractor is verbatim, and five where the correct answer is verbatim and the distractor is a loose paraphrase. Mix audio and reading; give it cold, two weeks later, with no coaching language in the instructions.
- Success looks like near-parity between the two halves. If the paraphrase-correct items remain much weaker, the class has learned a rule about tests ('beware exact wording') rather than a reading habit - a classic over-correction, and the reverse of the original error.
- The reverse half is the part most teachers skip. Without it, improvement is unfalsifiable, because 'always avoid the verbatim option' would also produce a gain.
- Secondary check: ask two students to justify a choice aloud. A warrant pointing to the text is transfer; a warrant pointing to the option's wording is not.