Halfway through a short story, the main character leaves a passport on a café table and boards a bus to the airport. Stop there. Ask the learner to predict one consequence and point to the sentence that supports it. That thirty-second interruption changes the reading task from collecting facts to building a possible next event.
Prediction is often treated as a warm-up based on a title or picture. A more useful version happens inside the text, after enough evidence has accumulated. The learner commits to one outcome, reads on, and checks which clue deserved attention. For a B1 or B2 reader in Taiwan who can understand most sentences yet loses the thread across paragraphs, this is practice in joining events together.
Prediction belongs inside comprehension
Reading a narrative requires more than decoding the words on the page. Readers connect causes, goals, locations, and likely consequences. The text may say that Mei has an interview at 9:10, the train stalls at Banqiao, and her phone battery shows 3 percent. A prediction about calling the employer or arriving late comes from those details. A prediction about winning a lottery does not.
This distinction matters because free guessing can distract from reading. The task needs a textual warrant. Use the frame “I predict ___ because the text says ___.” The second blank controls the first. It also gives the teacher an observable answer: perhaps the learner understood every noun yet missed that by the time placed one event before another.
Vocabulary coverage still sets a boundary. A learner who meets too many unknown words has little working attention left for inference. Choc's discussion of one unknown word in fifty explains why apparently small differences in known-word coverage alter the reading experience. Prediction works on a text the learner can mostly read, with a few consequential gaps rather than a thicket of dictionary searches.
A study that caught prediction happening
Nahatame's 2014 experiment examined Japanese university students reading short English narratives. One group received ordinary comprehension instructions. Another was told to anticipate the outcome of events. A lexical-decision measure tested whether words related to likely consequences had become active during reading. Predictive inferences appeared under the prediction instruction, with a stronger effect among readers with higher second-language proficiency. Recall of explicit information did not suffer.
The result is useful and bounded. It shows that instructions can change whether readers generate predictive inferences during carefully designed narratives. It does not establish that stopping after every paragraph improves a semester's reading scores. The proficiency pattern also warns teachers against treating a failed prediction as laziness. A learner may be spending available attention on word recognition and syntax.
A 2026 cross-linguistic study of pre-reading prediction and metacomprehension adds another caution. It compared prediction and monitoring across first- and second-language reading and examined proficiency as a moderator. The paper describes how relevant knowledge can exist in the first language while slower lexical and syntactic processing makes that knowledge harder to apply during English reading. Prediction, confidence, and actual comprehension can separate. A confident forecast is still a hypothesis.
Research on contextual predictability also shows that proficiency changes the picture. An eye-tracking study of 80 Chinese learners of English tested how sentence predictability interacted with word frequency and orthographic neighbourhood size. Context did not operate as a universal shortcut around weak lexical access. This is another reason to choose readable material and keep the inference demand modest.
Where to place the pause
A random stop produces a random guess. Pause where the story contains an unresolved cause, goal, or constraint. Four good locations recur:
- A character makes a choice with a visible cost.
- An object or fact has been introduced and left unresolved.
- Two plans conflict over time, place, or responsibility.
- A speaker says something that changes another person's options.
Suppose a workplace email story says that Priya attached the old price sheet, noticed the error after sending it, and sees the client typing a reply. The prediction point is ready. “The client will ask about the lower price” is defensible. “Priya will quit her job” reaches far beyond the evidence. Both may be possible in life. only one is useful for close reading.
One prediction is enough. Pause once. When a reader has to select one outcome, quote the words that support it, weigh that clue against another plausible signal, and later account for the gap between forecast and story, the exercise joins lexical access, event structure, causal reasoning, and monitoring without turning a short narrative into a worksheet whose interruptions matter more than its plot. Asking for five outcomes rewards imagination and verbal speed. The reading goal is to select the consequence best supported by the current text. After the reveal, compare the forecast with the actual event and locate the clue that was overweighted, underweighted, or missed.
Use wrong predictions as evidence
A wrong prediction can show sound comprehension. Stories are written to surprise, and authors plant plausible alternatives. If the learner cites a real clue and proposes a coherent result, the inference process worked. The next question is how the author redirected it.
A correct prediction can also be empty. Someone familiar with airport stories may guess that a passport will be forgotten without reading the sentence carefully. Ask for the supporting phrase before awarding the point. This keeps the activity tied to English rather than genre luck.
For lower-proficiency readers, allow the prediction in華語 and require the evidence phrase in English. Later, reverse the balance: a short English prediction with a brief explanation. The cognitive work is connecting evidence to consequence. A long spoken answer adds production demands that can obscure whether the story was understood.
Keep the interruption rare
Too many prediction stops destroy the rhythm that makes a story worth reading. On a 900-word graded reader, one pause near the middle and one optional check near the end are plenty. News reports and expository texts may need a different prompt, such as predicting the next claim or example, because their structure is less event-driven.
This task also fits some learners better than others. A2 readers working hard on basic clause boundaries need shorter passages and obvious causal links. Advanced readers may benefit from predicting a character's decision under conflicting motives, then explaining why two outcomes remain possible. A test passage should usually be read under test conditions. the pauses belong in teaching and independent practice.
Tomorrow, take a story the learner can read with roughly 98 percent known words. Mark one stopping point after a decision or problem. Before continuing, write one predicted consequence and copy six to twelve words that support it. Read the next section, then mark the forecast as supported, contradicted, or still open. Keep the copied clue. It is more informative than the tick beside the prediction.