What Is Vocabulary | Lexical Entropy — How Context Narrows Uncertainty Before the Next Word Arrives

LEXICAL ENTROPY · CONTEXTUAL UNCERTAINTY · INFORMATION THEORY · WORD PREDICTION · SURPRISAL · CONTEXTUAL CONSTRAINT · CANDIDATE WORDS

Lexical entropy is the uncertainty over possible upcoming words or tokens given a context. When one continuation dominates, entropy is low. When many alternatives remain plausible with similar probabilities, entropy is high. Entropy therefore describes the shape of the entire prediction distribution before the next item appears.

This is different from surprisal. Surprisal measures how unexpected the word that actually arrives is; entropy measures how uncertain the system was beforehand. A low-entropy context can still produce a highly surprising word if the writer deliberately chooses an unlikely continuation.

This guide explains probability distributions, contextual constraint, entropy, surprisal, uncertainty reduction, candidate sets, syntax, collocation, discourse, world knowledge and language-model prediction. Existing eduKateSG Contextual Predictability and Lexical Prediction articles remain untouched.

Surprisal asks, “How unexpected was this word?” Entropy asks, “How many possibilities were still alive before it arrived?”

2. Lexical entropy

Lexical entropy is the uncertainty over possible upcoming lexical items given a context. If many continuations remain plausible with similar probabilities, entropy is high; if only one or two words dominate, entropy is low.

The key is to think about a distribution, not a single guess. Uncertainty remains high when probability is spread across many candidates even if one of them later turns out to be correct.

3. Entropy versus surprisal

Entropy describes uncertainty before the next word appears. Surprisal describes how unexpected the word that actually appears is. Low-entropy contexts can still produce high surprisal if an unlikely continuation occurs.

The key is to think about a distribution, not a single guess. Uncertainty remains high when probability is spread across many candidates even if one of them later turns out to be correct.

4. Probability distribution

Entropy depends on the whole distribution of possible continuations, not just the probability of one target. A context with ten equally likely words is more uncertain than one dominated by a single continuation.

The key is to think about a distribution, not a single guess. Uncertainty remains high when probability is spread across many candidates even if one of them later turns out to be correct.

5. Contextual constraint

Strong contextual constraint lowers entropy by concentrating probability on a smaller set of candidates. Weak contexts leave probability spread widely.

The key is to think about a distribution, not a single guess. Uncertainty remains high when probability is spread across many candidates even if one of them later turns out to be correct.

6. Vocabulary and candidate sets

A learner’s effective lexical entropy depends partly on vocabulary knowledge. If likely candidates are unknown, human expectations may be poorer or differently distributed.

The key is to think about a distribution, not a single guess. Uncertainty remains high when probability is spread across many candidates even if one of them later turns out to be correct.

7. Syntax reduces uncertainty

Grammar narrows word class, morphology and construction. After an article, a noun-like continuation becomes more probable than a finite verb.

The key is to think about a distribution, not a single guess. Uncertainty remains high when probability is spread across many candidates even if one of them later turns out to be correct.

8. Collocation reduces uncertainty

Fixed and semi-fixed expressions strongly narrow possible continuations. After by the way or make a decision, phrase knowledge compresses the candidate space.

The key is to think about a distribution, not a single guess. Uncertainty remains high when probability is spread across many candidates even if one of them later turns out to be correct.

9. World knowledge reduces uncertainty

Event scripts and causal knowledge narrow what is plausible in a scene. A restaurant context activates different candidate words from a courtroom.

The key is to think about a distribution, not a single guess. Uncertainty remains high when probability is spread across many candidates even if one of them later turns out to be correct.

10. Discourse reduces uncertainty

Earlier sentences establish topic, referents and goals, making some lexical continuations more likely.

The key is to think about a distribution, not a single guess. Uncertainty remains high when probability is spread across many candidates even if one of them later turns out to be correct.

11. Register reduces uncertainty

Formal academic prose and casual chat have different lexical distributions. Knowing the register changes expectations before the next word arrives.

The key is to think about a distribution, not a single guess. Uncertainty remains high when probability is spread across many candidates even if one of them later turns out to be correct.

12. Entropy and reading

High uncertainty can increase processing demands because comprehenders maintain or evaluate more alternatives.

The key is to think about a distribution, not a single guess. Uncertainty remains high when probability is spread across many candidates even if one of them later turns out to be correct.

13. Entropy and listening

Listening is time-sensitive, so narrowing candidate sets quickly helps word recognition under rapidly unfolding input.

Entropy should always be tied to a defined prediction unit and model. Human words, subword tokens and experimental response options can yield different uncertainty values.

14. Entropy and learning

Unexpected but interpretable words can create prediction error that updates lexical expectations. Too much uncertainty, however, can make inference weak.

Entropy should always be tied to a defined prediction unit and model. Human words, subword tokens and experimental response options can yield different uncertainty values.

15. Entropy and ambiguity

Lexical ambiguity concerns multiple meanings of one form; lexical entropy concerns uncertainty over possible future forms. The two can interact but are not the same construct.

Entropy should always be tied to a defined prediction unit and model. Human words, subword tokens and experimental response options can yield different uncertainty values.

16. Entropy and semantic diversity

A word can appear across many contexts yet occur in locally low-entropy frames. Distribution across contexts and uncertainty within one context answer different questions.

Entropy should always be tied to a defined prediction unit and model. Human words, subword tokens and experimental response options can yield different uncertainty values.

17. Entropy in language models

Language models assign probabilities to possible next tokens. Entropy summarises how spread those probabilities are at a position.

Entropy should always be tied to a defined prediction unit and model. Human words, subword tokens and experimental response options can yield different uncertainty values.

18. Token versus word entropy

Modern models often predict tokens rather than human words. Token-level entropy should not be interpreted automatically as whole-word lexical uncertainty.

Entropy should always be tied to a defined prediction unit and model. Human words, subword tokens and experimental response options can yield different uncertainty values.

19. Entropy reduction

Each new word can reduce uncertainty about what comes next by constraining syntax, semantics and discourse.

Entropy should always be tied to a defined prediction unit and model. Human words, subword tokens and experimental response options can yield different uncertainty values.

20. Information gain

When context eliminates many alternatives, it provides information. Entropy reduction is one formal way to describe that narrowing.

Entropy should always be tied to a defined prediction unit and model. Human words, subword tokens and experimental response options can yield different uncertainty values.

21. Teaching

Teachers can manipulate entropy: high-constraint contexts support first learning, while moderate-entropy contexts later test flexible lexical knowledge.

Entropy should always be tied to a defined prediction unit and model. Human words, subword tokens and experimental response options can yield different uncertainty values.

22. AI-era vocabulary

Understanding entropy helps learners see that prediction systems do not simply ‘know the next word’; they maintain probability distributions whose uncertainty changes with context.

Entropy should always be tied to a defined prediction unit and model. Human words, subword tokens and experimental response options can yield different uncertainty values.

23. Entropy laboratory — Cases 1–20

1. The chef sharpened the ___

Plausible candidates: knife,blade,tool. Approximate entropy: low. Main constraint source: restaurant/kitchen script.

Distribution test: assign rough probabilities to the candidates. A context dominated by one continuation has lower entropy than a context where probability is spread evenly.

Surprisal contrast: choose a low-probability but grammatical continuation. The context may have low entropy while that actual word has high surprisal.

Constraint removal: delete one informative phrase and observe how the candidate set expands.

Vocabulary implication: decide whether the context supports learning because it narrows meaning, or merely makes a familiar continuation easy to predict.

2. She spread butter on the ___

Plausible candidates: bread,toast,roll. Approximate entropy: low. Main constraint source: strong food collocation.

Distribution test: assign rough probabilities to the candidates. A context dominated by one continuation has lower entropy than a context where probability is spread evenly.

Surprisal contrast: choose a low-probability but grammatical continuation. The context may have low entropy while that actual word has high surprisal.

Constraint removal: delete one informative phrase and observe how the candidate set expands.

Vocabulary implication: decide whether the context supports learning because it narrows meaning, or merely makes a familiar continuation easy to predict.

3. The baby was tired, so he put her to ___

Plausible candidates: sleep,bed,rest. Approximate entropy: low. Main constraint source: formulaic event.

Distribution test: assign rough probabilities to the candidates. A context dominated by one continuation has lower entropy than a context where probability is spread evenly.

Surprisal contrast: choose a low-probability but grammatical continuation. The context may have low entropy while that actual word has high surprisal.

Constraint removal: delete one informative phrase and observe how the candidate set expands.

Vocabulary implication: decide whether the context supports learning because it narrows meaning, or merely makes a familiar continuation easy to predict.

4. He unlocked the door with a ___

Plausible candidates: key,code,card. Approximate entropy: low. Main constraint source: instrument relation.

Distribution test: assign rough probabilities to the candidates. A context dominated by one continuation has lower entropy than a context where probability is spread evenly.

Surprisal contrast: choose a low-probability but grammatical continuation. The context may have low entropy while that actual word has high surprisal.

Constraint removal: delete one informative phrase and observe how the candidate set expands.

Vocabulary implication: decide whether the context supports learning because it narrows meaning, or merely makes a familiar continuation easy to predict.

5. The referee blew the ___

Plausible candidates: whistle. Approximate entropy: very low. Main constraint source: fixed sports phrase.

Distribution test: assign rough probabilities to the candidates. A context dominated by one continuation has lower entropy than a context where probability is spread evenly.

Surprisal contrast: choose a low-probability but grammatical continuation. The context may have low entropy while that actual word has high surprisal.

Constraint removal: delete one informative phrase and observe how the candidate set expands.

Vocabulary implication: decide whether the context supports learning because it narrows meaning, or merely makes a familiar continuation easy to predict.

6. The nurse measured the patient’s blood ___

Plausible candidates: pressure. Approximate entropy: very low. Main constraint source: medical collocation.

Distribution test: assign rough probabilities to the candidates. A context dominated by one continuation has lower entropy than a context where probability is spread evenly.

Surprisal contrast: choose a low-probability but grammatical continuation. The context may have low entropy while that actual word has high surprisal.

Constraint removal: delete one informative phrase and observe how the candidate set expands.

Vocabulary implication: decide whether the context supports learning because it narrows meaning, or merely makes a familiar continuation easy to predict.

7. The committee reached a final ___

Plausible candidates: decision,agreement,conclusion. Approximate entropy: low. Main constraint source: institutional collocation.

Distribution test: assign rough probabilities to the candidates. A context dominated by one continuation has lower entropy than a context where probability is spread evenly.

Surprisal contrast: choose a low-probability but grammatical continuation. The context may have low entropy while that actual word has high surprisal.

Constraint removal: delete one informative phrase and observe how the candidate set expands.

Vocabulary implication: decide whether the context supports learning because it narrows meaning, or merely makes a familiar continuation easy to predict.

8. The scientist analysed the ___

Plausible candidates: data,results,samples,evidence. Approximate entropy: medium. Main constraint source: research domain.

Distribution test: assign rough probabilities to the candidates. A context dominated by one continuation has lower entropy than a context where probability is spread evenly.

Surprisal contrast: choose a low-probability but grammatical continuation. The context may have low entropy while that actual word has high surprisal.

Constraint removal: delete one informative phrase and observe how the candidate set expands.

Vocabulary implication: decide whether the context supports learning because it narrows meaning, or merely makes a familiar continuation easy to predict.

9. She walked into the ___

Plausible candidates: room,house,building,office,shop. Approximate entropy: high. Main constraint source: weak spatial context.

Distribution test: assign rough probabilities to the candidates. A context dominated by one continuation has lower entropy than a context where probability is spread evenly.

Surprisal contrast: choose a low-probability but grammatical continuation. The context may have low entropy while that actual word has high surprisal.

Constraint removal: delete one informative phrase and observe how the candidate set expands.

Vocabulary implication: decide whether the context supports learning because it narrows meaning, or merely makes a familiar continuation easy to predict.

10. He looked at the ___

Plausible candidates: screen,clock,door,table,sky. Approximate entropy: high. Main constraint source: very broad object slot.

Distribution test: assign rough probabilities to the candidates. A context dominated by one continuation has lower entropy than a context where probability is spread evenly.

Surprisal contrast: choose a low-probability but grammatical continuation. The context may have low entropy while that actual word has high surprisal.

Constraint removal: delete one informative phrase and observe how the candidate set expands.

Vocabulary implication: decide whether the context supports learning because it narrows meaning, or merely makes a familiar continuation easy to predict.

11. The child picked up a ___

Plausible candidates: toy,book,stone,pencil,ball. Approximate entropy: high. Main constraint source: broad object context.

Distribution test: assign rough probabilities to the candidates. A context dominated by one continuation has lower entropy than a context where probability is spread evenly.

Surprisal contrast: choose a low-probability but grammatical continuation. The context may have low entropy while that actual word has high surprisal.

Constraint removal: delete one informative phrase and observe how the candidate set expands.

Vocabulary implication: decide whether the context supports learning because it narrows meaning, or merely makes a familiar continuation easy to predict.

12. After three days of rain, the river ___

Plausible candidates: rose,overflowed,swelled,flooded. Approximate entropy: medium. Main constraint source: weather script.

Distribution test: assign rough probabilities to the candidates. A context dominated by one continuation has lower entropy than a context where probability is spread evenly.

Surprisal contrast: choose a low-probability but grammatical continuation. The context may have low entropy while that actual word has high surprisal.

Constraint removal: delete one informative phrase and observe how the candidate set expands.

Vocabulary implication: decide whether the context supports learning because it narrows meaning, or merely makes a familiar continuation easy to predict.

13. The judge announced the ___

Plausible candidates: verdict,sentence,decision,ruling. Approximate entropy: low/medium. Main constraint source: courtroom script.

Distribution test: assign rough probabilities to the candidates. A context dominated by one continuation has lower entropy than a context where probability is spread evenly.

Surprisal contrast: choose a low-probability but grammatical continuation. The context may have low entropy while that actual word has high surprisal.

Constraint removal: delete one informative phrase and observe how the candidate set expands.

Vocabulary implication: decide whether the context supports learning because it narrows meaning, or merely makes a familiar continuation easy to predict.

14. The lawyer presented new ___

Plausible candidates: evidence,arguments,documents,facts. Approximate entropy: medium. Main constraint source: legal discourse.

Distribution test: assign rough probabilities to the candidates. A context dominated by one continuation has lower entropy than a context where probability is spread evenly.

Surprisal contrast: choose a low-probability but grammatical continuation. The context may have low entropy while that actual word has high surprisal.

Constraint removal: delete one informative phrase and observe how the candidate set expands.

Vocabulary implication: decide whether the context supports learning because it narrows meaning, or merely makes a familiar continuation easy to predict.

15. The doctor prescribed ___

Plausible candidates: antibiotics,medicine,treatment,rest. Approximate entropy: medium. Main constraint source: medical script.

Distribution test: assign rough probabilities to the candidates. A context dominated by one continuation has lower entropy than a context where probability is spread evenly.

Surprisal contrast: choose a low-probability but grammatical continuation. The context may have low entropy while that actual word has high surprisal.

Constraint removal: delete one informative phrase and observe how the candidate set expands.

Vocabulary implication: decide whether the context supports learning because it narrows meaning, or merely makes a familiar continuation easy to predict.

16. The cyclist wore a helmet for ___

Plausible candidates: safety,protection. Approximate entropy: low. Main constraint source: purpose relation.

Distribution test: assign rough probabilities to the candidates. A context dominated by one continuation has lower entropy than a context where probability is spread evenly.

Surprisal contrast: choose a low-probability but grammatical continuation. The context may have low entropy while that actual word has high surprisal.

Constraint removal: delete one informative phrase and observe how the candidate set expands.

Vocabulary implication: decide whether the context supports learning because it narrows meaning, or merely makes a familiar continuation easy to predict.

17. The phone battery was almost ___

Plausible candidates: dead,empty,flat. Approximate entropy: low/medium. Main constraint source: device state.

Distribution test: assign rough probabilities to the candidates. A context dominated by one continuation has lower entropy than a context where probability is spread evenly.

Surprisal contrast: choose a low-probability but grammatical continuation. The context may have low entropy while that actual word has high surprisal.

Constraint removal: delete one informative phrase and observe how the candidate set expands.

Vocabulary implication: decide whether the context supports learning because it narrows meaning, or merely makes a familiar continuation easy to predict.

18. The librarian asked everyone to speak ___

Plausible candidates: quietly,softly. Approximate entropy: low. Main constraint source: institutional expectation.

Distribution test: assign rough probabilities to the candidates. A context dominated by one continuation has lower entropy than a context where probability is spread evenly.

Surprisal contrast: choose a low-probability but grammatical continuation. The context may have low entropy while that actual word has high surprisal.

Constraint removal: delete one informative phrase and observe how the candidate set expands.

Vocabulary implication: decide whether the context supports learning because it narrows meaning, or merely makes a familiar continuation easy to predict.

19. The musician tuned the guitar before the ___

Plausible candidates: performance,concert,show,rehearsal. Approximate entropy: medium. Main constraint source: music event.

Distribution test: assign rough probabilities to the candidates. A context dominated by one continuation has lower entropy than a context where probability is spread evenly.

Surprisal contrast: choose a low-probability but grammatical continuation. The context may have low entropy while that actual word has high surprisal.

Constraint removal: delete one informative phrase and observe how the candidate set expands.

Vocabulary implication: decide whether the context supports learning because it narrows meaning, or merely makes a familiar continuation easy to predict.

20. The students opened their books to page ___

Plausible candidates: ten,twenty,five,thirty. Approximate entropy: high. Main constraint source: many numeric continuations.

Distribution test: assign rough probabilities to the candidates. A context dominated by one continuation has lower entropy than a context where probability is spread evenly.

Surprisal contrast: choose a low-probability but grammatical continuation. The context may have low entropy while that actual word has high surprisal.

Constraint removal: delete one informative phrase and observe how the candidate set expands.

Vocabulary implication: decide whether the context supports learning because it narrows meaning, or merely makes a familiar continuation easy to predict.

24. Entropy laboratory — Cases 21–40

21. The road was closed because of ___

Plausible candidates: an accident,flooding,construction,snow. Approximate entropy: high. Main source: multiple causes.

Syntax contribution: identify how grammar narrows the candidate class before semantics operates. Removing syntactic constraints typically raises uncertainty.

World-knowledge contribution: identify the event script that concentrates probability on particular lexical fields.

Register contribution: rewrite the context as casual chat or formal academic prose and predict how the distribution changes.

Processing prediction: explain whether readers should maintain a small focused candidate set or a broad diffuse set at the blank.

22. The report identified three major ___

Plausible candidates: problems,issues,risks,trends. Approximate entropy: medium. Main source: reporting frame.

Syntax contribution: identify how grammar narrows the candidate class before semantics operates. Removing syntactic constraints typically raises uncertainty.

World-knowledge contribution: identify the event script that concentrates probability on particular lexical fields.

Register contribution: rewrite the context as casual chat or formal academic prose and predict how the distribution changes.

Processing prediction: explain whether readers should maintain a small focused candidate set or a broad diffuse set at the blank.

23. The hikers followed the marked ___

Plausible candidates: trail,path,route. Approximate entropy: low. Main source: outdoor collocation.

Syntax contribution: identify how grammar narrows the candidate class before semantics operates. Removing syntactic constraints typically raises uncertainty.

World-knowledge contribution: identify the event script that concentrates probability on particular lexical fields.

Register contribution: rewrite the context as casual chat or formal academic prose and predict how the distribution changes.

Processing prediction: explain whether readers should maintain a small focused candidate set or a broad diffuse set at the blank.

24. The gardener watered the plants every ___

Plausible candidates: day,morning,evening,week. Approximate entropy: medium/high. Main source: temporal options.

Syntax contribution: identify how grammar narrows the candidate class before semantics operates. Removing syntactic constraints typically raises uncertainty.

World-knowledge contribution: identify the event script that concentrates probability on particular lexical fields.

Register contribution: rewrite the context as casual chat or formal academic prose and predict how the distribution changes.

Processing prediction: explain whether readers should maintain a small focused candidate set or a broad diffuse set at the blank.

25. The detective searched the room for ___

Plausible candidates: clues,evidence,fingerprints,weapons. Approximate entropy: medium. Main source: genre script.

Syntax contribution: identify how grammar narrows the candidate class before semantics operates. Removing syntactic constraints typically raises uncertainty.

World-knowledge contribution: identify the event script that concentrates probability on particular lexical fields.

Register contribution: rewrite the context as casual chat or formal academic prose and predict how the distribution changes.

Processing prediction: explain whether readers should maintain a small focused candidate set or a broad diffuse set at the blank.

26. The company reduced costs without lowering ___

Plausible candidates: quality,staffing,output,service. Approximate entropy: medium. Main source: business discourse.

Syntax contribution: identify how grammar narrows the candidate class before semantics operates. Removing syntactic constraints typically raises uncertainty.

World-knowledge contribution: identify the event script that concentrates probability on particular lexical fields.

Register contribution: rewrite the context as casual chat or formal academic prose and predict how the distribution changes.

Processing prediction: explain whether readers should maintain a small focused candidate set or a broad diffuse set at the blank.

27. The patient felt better after taking the ___

Plausible candidates: medicine,drug,tablets,treatment. Approximate entropy: medium. Main source: health script.

Syntax contribution: identify how grammar narrows the candidate class before semantics operates. Removing syntactic constraints typically raises uncertainty.

World-knowledge contribution: identify the event script that concentrates probability on particular lexical fields.

Register contribution: rewrite the context as casual chat or formal academic prose and predict how the distribution changes.

Processing prediction: explain whether readers should maintain a small focused candidate set or a broad diffuse set at the blank.

28. She reread the paragraph because it was ___

Plausible candidates: unclear,confusing,difficult,ambiguous. Approximate entropy: medium. Main source: reading evaluation.

Syntax contribution: identify how grammar narrows the candidate class before semantics operates. Removing syntactic constraints typically raises uncertainty.

World-knowledge contribution: identify the event script that concentrates probability on particular lexical fields.

Register contribution: rewrite the context as casual chat or formal academic prose and predict how the distribution changes.

Processing prediction: explain whether readers should maintain a small focused candidate set or a broad diffuse set at the blank.

29. The researcher calculated the average from the ___

Plausible candidates: data,scores,measurements,values. Approximate entropy: medium. Main source: research/math.

Syntax contribution: identify how grammar narrows the candidate class before semantics operates. Removing syntactic constraints typically raises uncertainty.

World-knowledge contribution: identify the event script that concentrates probability on particular lexical fields.

Register contribution: rewrite the context as casual chat or formal academic prose and predict how the distribution changes.

Processing prediction: explain whether readers should maintain a small focused candidate set or a broad diffuse set at the blank.

30. The manager scheduled another meeting to discuss the ___

Plausible candidates: issue,proposal,problem,budget,plan. Approximate entropy: high. Main source: business discourse.

Syntax contribution: identify how grammar narrows the candidate class before semantics operates. Removing syntactic constraints typically raises uncertainty.

World-knowledge contribution: identify the event script that concentrates probability on particular lexical fields.

Register contribution: rewrite the context as casual chat or formal academic prose and predict how the distribution changes.

Processing prediction: explain whether readers should maintain a small focused candidate set or a broad diffuse set at the blank.

31. The children built a sandcastle near the ___

Plausible candidates: sea,water,shore,beach. Approximate entropy: medium. Main source: beach script.

Syntax contribution: identify how grammar narrows the candidate class before semantics operates. Removing syntactic constraints typically raises uncertainty.

World-knowledge contribution: identify the event script that concentrates probability on particular lexical fields.

Register contribution: rewrite the context as casual chat or formal academic prose and predict how the distribution changes.

Processing prediction: explain whether readers should maintain a small focused candidate set or a broad diffuse set at the blank.

32. The farmer repaired the fence around the ___

Plausible candidates: field,farm,pasture,garden. Approximate entropy: medium. Main source: rural script.

Syntax contribution: identify how grammar narrows the candidate class before semantics operates. Removing syntactic constraints typically raises uncertainty.

World-knowledge contribution: identify the event script that concentrates probability on particular lexical fields.

Register contribution: rewrite the context as casual chat or formal academic prose and predict how the distribution changes.

Processing prediction: explain whether readers should maintain a small focused candidate set or a broad diffuse set at the blank.

33. The speaker paused to allow the audience to ___

Plausible candidates: reflect,think,respond,applaud. Approximate entropy: high. Main source: discourse action.

Syntax contribution: identify how grammar narrows the candidate class before semantics operates. Removing syntactic constraints typically raises uncertainty.

World-knowledge contribution: identify the event script that concentrates probability on particular lexical fields.

Register contribution: rewrite the context as casual chat or formal academic prose and predict how the distribution changes.

Processing prediction: explain whether readers should maintain a small focused candidate set or a broad diffuse set at the blank.

34. The engineer tested the bridge under heavy ___

Plausible candidates: load,traffic,stress,weight. Approximate entropy: low/medium. Main source: engineering semantics.

Syntax contribution: identify how grammar narrows the candidate class before semantics operates. Removing syntactic constraints typically raises uncertainty.

World-knowledge contribution: identify the event script that concentrates probability on particular lexical fields.

Register contribution: rewrite the context as casual chat or formal academic prose and predict how the distribution changes.

Processing prediction: explain whether readers should maintain a small focused candidate set or a broad diffuse set at the blank.

35. The journalist verified the information before publishing the ___

Plausible candidates: story,article,report,claim. Approximate entropy: medium. Main source: media script.

Syntax contribution: identify how grammar narrows the candidate class before semantics operates. Removing syntactic constraints typically raises uncertainty.

World-knowledge contribution: identify the event script that concentrates probability on particular lexical fields.

Register contribution: rewrite the context as casual chat or formal academic prose and predict how the distribution changes.

Processing prediction: explain whether readers should maintain a small focused candidate set or a broad diffuse set at the blank.

36. The customer asked for a full ___

Plausible candidates: refund,explanation,receipt,replacement. Approximate entropy: medium. Main source: commercial script.

Syntax contribution: identify how grammar narrows the candidate class before semantics operates. Removing syntactic constraints typically raises uncertainty.

World-knowledge contribution: identify the event script that concentrates probability on particular lexical fields.

Register contribution: rewrite the context as casual chat or formal academic prose and predict how the distribution changes.

Processing prediction: explain whether readers should maintain a small focused candidate set or a broad diffuse set at the blank.

37. The experiment was repeated to confirm the ___

Plausible candidates: result,finding,effect,hypothesis. Approximate entropy: medium. Main source: research script.

Syntax contribution: identify how grammar narrows the candidate class before semantics operates. Removing syntactic constraints typically raises uncertainty.

World-knowledge contribution: identify the event script that concentrates probability on particular lexical fields.

Register contribution: rewrite the context as casual chat or formal academic prose and predict how the distribution changes.

Processing prediction: explain whether readers should maintain a small focused candidate set or a broad diffuse set at the blank.

38. The witness gave a detailed ___

Plausible candidates: account,description,statement,report. Approximate entropy: medium. Main source: legal/reporting.

Syntax contribution: identify how grammar narrows the candidate class before semantics operates. Removing syntactic constraints typically raises uncertainty.

World-knowledge contribution: identify the event script that concentrates probability on particular lexical fields.

Register contribution: rewrite the context as casual chat or formal academic prose and predict how the distribution changes.

Processing prediction: explain whether readers should maintain a small focused candidate set or a broad diffuse set at the blank.

39. The crowd cheered when the striker scored a ___

Plausible candidates: goal. Approximate entropy: very low. Main source: football script.

Syntax contribution: identify how grammar narrows the candidate class before semantics operates. Removing syntactic constraints typically raises uncertainty.

World-knowledge contribution: identify the event script that concentrates probability on particular lexical fields.

Register contribution: rewrite the context as casual chat or formal academic prose and predict how the distribution changes.

Processing prediction: explain whether readers should maintain a small focused candidate set or a broad diffuse set at the blank.

40. She placed the flowers in a glass ___

Plausible candidates: vase,jar,bottle. Approximate entropy: low. Main source: object-function relation.

Syntax contribution: identify how grammar narrows the candidate class before semantics operates. Removing syntactic constraints typically raises uncertainty.

World-knowledge contribution: identify the event script that concentrates probability on particular lexical fields.

Register contribution: rewrite the context as casual chat or formal academic prose and predict how the distribution changes.

Processing prediction: explain whether readers should maintain a small focused candidate set or a broad diffuse set at the blank.

25. Entropy laboratory — Cases 41–60

41. The fire alarm sounded and everyone left the ___

Plausible candidates: building,room,office,school. Approximate entropy: medium. Main source: evacuation script.

Fixedness test: decide whether phrase knowledge collapses the distribution toward one continuation. Formulaic language is a major source of low lexical entropy.

Discourse-extension test: add a preceding sentence that makes one candidate much more likely. Entropy can fall because of information outside the immediate clause.

Unexpected-word test: insert a surprising but interpretable word and explain the update required by the reader.

Learning use: high-constraint, low-entropy contexts can scaffold first encounters; moderate-entropy contexts later test whether vocabulary is independently accessible.

42. He raised his hand because he wanted to ask a ___

Plausible candidates: question. Approximate entropy: very low. Main source: classroom formula.

Fixedness test: decide whether phrase knowledge collapses the distribution toward one continuation. Formulaic language is a major source of low lexical entropy.

Discourse-extension test: add a preceding sentence that makes one candidate much more likely. Entropy can fall because of information outside the immediate clause.

Unexpected-word test: insert a surprising but interpretable word and explain the update required by the reader.

Learning use: high-constraint, low-entropy contexts can scaffold first encounters; moderate-entropy contexts later test whether vocabulary is independently accessible.

43. The team practised every day before the ___

Plausible candidates: match,game,tournament,competition. Approximate entropy: medium. Main source: sport script.

Fixedness test: decide whether phrase knowledge collapses the distribution toward one continuation. Formulaic language is a major source of low lexical entropy.

Discourse-extension test: add a preceding sentence that makes one candidate much more likely. Entropy can fall because of information outside the immediate clause.

Unexpected-word test: insert a surprising but interpretable word and explain the update required by the reader.

Learning use: high-constraint, low-entropy contexts can scaffold first encounters; moderate-entropy contexts later test whether vocabulary is independently accessible.

44. She turned down the music because it was too ___

Plausible candidates: loud,noisy. Approximate entropy: low. Main source: causal adjective.

Fixedness test: decide whether phrase knowledge collapses the distribution toward one continuation. Formulaic language is a major source of low lexical entropy.

Discourse-extension test: add a preceding sentence that makes one candidate much more likely. Entropy can fall because of information outside the immediate clause.

Unexpected-word test: insert a surprising but interpretable word and explain the update required by the reader.

Learning use: high-constraint, low-entropy contexts can scaffold first encounters; moderate-entropy contexts later test whether vocabulary is independently accessible.

45. He saved the document before closing the ___

Plausible candidates: program,file,computer,window. Approximate entropy: medium. Main source: computing script.

Fixedness test: decide whether phrase knowledge collapses the distribution toward one continuation. Formulaic language is a major source of low lexical entropy.

Discourse-extension test: add a preceding sentence that makes one candidate much more likely. Entropy can fall because of information outside the immediate clause.

Unexpected-word test: insert a surprising but interpretable word and explain the update required by the reader.

Learning use: high-constraint, low-entropy contexts can scaffold first encounters; moderate-entropy contexts later test whether vocabulary is independently accessible.

46. They discussed the proposal during the ___

Plausible candidates: meeting,session,conference,debate. Approximate entropy: medium. Main source: institutional event.

Fixedness test: decide whether phrase knowledge collapses the distribution toward one continuation. Formulaic language is a major source of low lexical entropy.

Discourse-extension test: add a preceding sentence that makes one candidate much more likely. Entropy can fall because of information outside the immediate clause.

Unexpected-word test: insert a surprising but interpretable word and explain the update required by the reader.

Learning use: high-constraint, low-entropy contexts can scaffold first encounters; moderate-entropy contexts later test whether vocabulary is independently accessible.

47. The bridge was designed to withstand ___

Plausible candidates: pressure,wind,earthquakes,loads. Approximate entropy: medium. Main source: engineering.

Fixedness test: decide whether phrase knowledge collapses the distribution toward one continuation. Formulaic language is a major source of low lexical entropy.

Discourse-extension test: add a preceding sentence that makes one candidate much more likely. Entropy can fall because of information outside the immediate clause.

Unexpected-word test: insert a surprising but interpretable word and explain the update required by the reader.

Learning use: high-constraint, low-entropy contexts can scaffold first encounters; moderate-entropy contexts later test whether vocabulary is independently accessible.

48. The new evidence could change the ___

Plausible candidates: outcome,decision,case,interpretation. Approximate entropy: high. Main source: abstract discourse.

Fixedness test: decide whether phrase knowledge collapses the distribution toward one continuation. Formulaic language is a major source of low lexical entropy.

Discourse-extension test: add a preceding sentence that makes one candidate much more likely. Entropy can fall because of information outside the immediate clause.

Unexpected-word test: insert a surprising but interpretable word and explain the update required by the reader.

Learning use: high-constraint, low-entropy contexts can scaffold first encounters; moderate-entropy contexts later test whether vocabulary is independently accessible.

49. The policy aims to reduce ___

Plausible candidates: costs,emissions,risk,crime,inequality. Approximate entropy: high. Main source: policy frame.

Fixedness test: decide whether phrase knowledge collapses the distribution toward one continuation. Formulaic language is a major source of low lexical entropy.

Discourse-extension test: add a preceding sentence that makes one candidate much more likely. Entropy can fall because of information outside the immediate clause.

Unexpected-word test: insert a surprising but interpretable word and explain the update required by the reader.

Learning use: high-constraint, low-entropy contexts can scaffold first encounters; moderate-entropy contexts later test whether vocabulary is independently accessible.

50. The medicine may cause ___

Plausible candidates: nausea,drowsiness,headache,side effects. Approximate entropy: medium. Main source: medical.

Fixedness test: decide whether phrase knowledge collapses the distribution toward one continuation. Formulaic language is a major source of low lexical entropy.

Discourse-extension test: add a preceding sentence that makes one candidate much more likely. Entropy can fall because of information outside the immediate clause.

Unexpected-word test: insert a surprising but interpretable word and explain the update required by the reader.

Learning use: high-constraint, low-entropy contexts can scaffold first encounters; moderate-entropy contexts later test whether vocabulary is independently accessible.

51. The article argues that ___

Plausible candidates: policy X is wrong,the evidence is weak,change is needed. Approximate entropy: very high. Main source: propositional slot.

Fixedness test: decide whether phrase knowledge collapses the distribution toward one continuation. Formulaic language is a major source of low lexical entropy.

Discourse-extension test: add a preceding sentence that makes one candidate much more likely. Entropy can fall because of information outside the immediate clause.

Unexpected-word test: insert a surprising but interpretable word and explain the update required by the reader.

Learning use: high-constraint, low-entropy contexts can scaffold first encounters; moderate-entropy contexts later test whether vocabulary is independently accessible.

52. She said that ___

Plausible candidates: she was tired,he had left,the plan worked. Approximate entropy: very high. Main source: open clause.

Fixedness test: decide whether phrase knowledge collapses the distribution toward one continuation. Formulaic language is a major source of low lexical entropy.

Discourse-extension test: add a preceding sentence that makes one candidate much more likely. Entropy can fall because of information outside the immediate clause.

Unexpected-word test: insert a surprising but interpretable word and explain the update required by the reader.

Learning use: high-constraint, low-entropy contexts can scaffold first encounters; moderate-entropy contexts later test whether vocabulary is independently accessible.

53. The book is about ___

Plausible candidates: history,science,a family,a journey,war. Approximate entropy: very high. Main source: topic slot.

Fixedness test: decide whether phrase knowledge collapses the distribution toward one continuation. Formulaic language is a major source of low lexical entropy.

Discourse-extension test: add a preceding sentence that makes one candidate much more likely. Entropy can fall because of information outside the immediate clause.

Unexpected-word test: insert a surprising but interpretable word and explain the update required by the reader.

Learning use: high-constraint, low-entropy contexts can scaffold first encounters; moderate-entropy contexts later test whether vocabulary is independently accessible.

54. They decided to ___

Plausible candidates: leave,wait,continue,apply,cancel. Approximate entropy: very high. Main source: verb complement.

Fixedness test: decide whether phrase knowledge collapses the distribution toward one continuation. Formulaic language is a major source of low lexical entropy.

Discourse-extension test: add a preceding sentence that makes one candidate much more likely. Entropy can fall because of information outside the immediate clause.

Unexpected-word test: insert a surprising but interpretable word and explain the update required by the reader.

Learning use: high-constraint, low-entropy contexts can scaffold first encounters; moderate-entropy contexts later test whether vocabulary is independently accessible.

55. He is interested in ___

Plausible candidates: music,science,history,learning,art. Approximate entropy: very high. Main source: broad noun/-ing slot.

Fixedness test: decide whether phrase knowledge collapses the distribution toward one continuation. Formulaic language is a major source of low lexical entropy.

Discourse-extension test: add a preceding sentence that makes one candidate much more likely. Entropy can fall because of information outside the immediate clause.

Unexpected-word test: insert a surprising but interpretable word and explain the update required by the reader.

Learning use: high-constraint, low-entropy contexts can scaffold first encounters; moderate-entropy contexts later test whether vocabulary is independently accessible.

56. The result was ___

Plausible candidates: good,bad,unexpected,significant,clear. Approximate entropy: very high. Main source: copular adjective slot.

Fixedness test: decide whether phrase knowledge collapses the distribution toward one continuation. Formulaic language is a major source of low lexical entropy.

Discourse-extension test: add a preceding sentence that makes one candidate much more likely. Entropy can fall because of information outside the immediate clause.

Unexpected-word test: insert a surprising but interpretable word and explain the update required by the reader.

Learning use: high-constraint, low-entropy contexts can scaffold first encounters; moderate-entropy contexts later test whether vocabulary is independently accessible.

57. The problem can be solved by ___

Plausible candidates: changing the design,adding data,reducing cost. Approximate entropy: very high. Main source: open method phrase.

Fixedness test: decide whether phrase knowledge collapses the distribution toward one continuation. Formulaic language is a major source of low lexical entropy.

Discourse-extension test: add a preceding sentence that makes one candidate much more likely. Entropy can fall because of information outside the immediate clause.

Unexpected-word test: insert a surprising but interpretable word and explain the update required by the reader.

Learning use: high-constraint, low-entropy contexts can scaffold first encounters; moderate-entropy contexts later test whether vocabulary is independently accessible.

58. In conclusion, the study ___

Plausible candidates: supports the claim,shows a trend,raises questions. Approximate entropy: high. Main source: academic discourse.

Fixedness test: decide whether phrase knowledge collapses the distribution toward one continuation. Formulaic language is a major source of low lexical entropy.

Discourse-extension test: add a preceding sentence that makes one candidate much more likely. Entropy can fall because of information outside the immediate clause.

Unexpected-word test: insert a surprising but interpretable word and explain the update required by the reader.

Learning use: high-constraint, low-entropy contexts can scaffold first encounters; moderate-entropy contexts later test whether vocabulary is independently accessible.

59. Once upon a ___

Plausible candidates: time. Approximate entropy: very low. Main source: fixed phrase.

Fixedness test: decide whether phrase knowledge collapses the distribution toward one continuation. Formulaic language is a major source of low lexical entropy.

Discourse-extension test: add a preceding sentence that makes one candidate much more likely. Entropy can fall because of information outside the immediate clause.

Unexpected-word test: insert a surprising but interpretable word and explain the update required by the reader.

Learning use: high-constraint, low-entropy contexts can scaffold first encounters; moderate-entropy contexts later test whether vocabulary is independently accessible.

60. By the ___

Plausible candidates: way. Approximate entropy: very low. Main source: fixed expression.

Fixedness test: decide whether phrase knowledge collapses the distribution toward one continuation. Formulaic language is a major source of low lexical entropy.

Discourse-extension test: add a preceding sentence that makes one candidate much more likely. Entropy can fall because of information outside the immediate clause.

Unexpected-word test: insert a surprising but interpretable word and explain the update required by the reader.

Learning use: high-constraint, low-entropy contexts can scaffold first encounters; moderate-entropy contexts later test whether vocabulary is independently accessible.

26. Teaching vocabulary through controlled uncertainty

1. The chef sharpened the

Stage 1 — low entropy: strengthen restaurant/kitchen script so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as knife,blade,tool are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

2. She spread butter on the

Stage 1 — low entropy: strengthen strong food collocation so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as bread,toast,roll are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

3. The baby was tired, so he put her to

Stage 1 — low entropy: strengthen formulaic event so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as sleep,bed,rest are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

4. He unlocked the door with a

Stage 1 — low entropy: strengthen instrument relation so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as key,code,card are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

5. The referee blew the

Stage 1 — low entropy: strengthen fixed sports phrase so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as whistle are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

6. The nurse measured the patient’s blood

Stage 1 — low entropy: strengthen medical collocation so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as pressure are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

7. The committee reached a final

Stage 1 — low entropy: strengthen institutional collocation so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as decision,agreement,conclusion are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

8. The scientist analysed the

Stage 1 — low entropy: strengthen research domain so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as data,results,samples,evidence are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

9. She walked into the

Stage 1 — low entropy: strengthen weak spatial context so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as room,house,building,office,shop are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

10. He looked at the

Stage 1 — low entropy: strengthen very broad object slot so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as screen,clock,door,table,sky are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

11. The child picked up a

Stage 1 — low entropy: strengthen broad object context so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as toy,book,stone,pencil,ball are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

12. After three days of rain, the river

Stage 1 — low entropy: strengthen weather script so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as rose,overflowed,swelled,flooded are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

13. The judge announced the

Stage 1 — low entropy: strengthen courtroom script so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as verdict,sentence,decision,ruling are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

14. The lawyer presented new

Stage 1 — low entropy: strengthen legal discourse so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as evidence,arguments,documents,facts are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

15. The doctor prescribed

Stage 1 — low entropy: strengthen medical script so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as antibiotics,medicine,treatment,rest are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

16. The cyclist wore a helmet for

Stage 1 — low entropy: strengthen purpose relation so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as safety,protection are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

17. The phone battery was almost

Stage 1 — low entropy: strengthen device state so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as dead,empty,flat are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

18. The librarian asked everyone to speak

Stage 1 — low entropy: strengthen institutional expectation so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as quietly,softly are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

19. The musician tuned the guitar before the

Stage 1 — low entropy: strengthen music event so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as performance,concert,show,rehearsal are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

20. The students opened their books to page

Stage 1 — low entropy: strengthen many numeric continuations so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as ten,twenty,five,thirty are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

21. The road was closed because of

Stage 1 — low entropy: strengthen multiple causes so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as an accident,flooding,construction,snow are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

22. The report identified three major

Stage 1 — low entropy: strengthen reporting frame so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as problems,issues,risks,trends are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

23. The hikers followed the marked

Stage 1 — low entropy: strengthen outdoor collocation so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as trail,path,route are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

24. The gardener watered the plants every

Stage 1 — low entropy: strengthen temporal options so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as day,morning,evening,week are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

25. The detective searched the room for

Stage 1 — low entropy: strengthen genre script so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as clues,evidence,fingerprints,weapons are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

26. The company reduced costs without lowering

Stage 1 — low entropy: strengthen business discourse so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as quality,staffing,output,service are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

27. The patient felt better after taking the

Stage 1 — low entropy: strengthen health script so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as medicine,drug,tablets,treatment are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

28. She reread the paragraph because it was

Stage 1 — low entropy: strengthen reading evaluation so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as unclear,confusing,difficult,ambiguous are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

29. The researcher calculated the average from the

Stage 1 — low entropy: strengthen research/math so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as data,scores,measurements,values are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

30. The manager scheduled another meeting to discuss the

Stage 1 — low entropy: strengthen business discourse so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as issue,proposal,problem,budget,plan are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

31. The children built a sandcastle near the

Stage 1 — low entropy: strengthen beach script so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as sea,water,shore,beach are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

32. The farmer repaired the fence around the

Stage 1 — low entropy: strengthen rural script so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as field,farm,pasture,garden are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

33. The speaker paused to allow the audience to

Stage 1 — low entropy: strengthen discourse action so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as reflect,think,respond,applaud are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

34. The engineer tested the bridge under heavy

Stage 1 — low entropy: strengthen engineering semantics so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as load,traffic,stress,weight are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

35. The journalist verified the information before publishing the

Stage 1 — low entropy: strengthen media script so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as story,article,report,claim are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

36. The customer asked for a full

Stage 1 — low entropy: strengthen commercial script so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as refund,explanation,receipt,replacement are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

37. The experiment was repeated to confirm the

Stage 1 — low entropy: strengthen research script so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as result,finding,effect,hypothesis are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

38. The witness gave a detailed

Stage 1 — low entropy: strengthen legal/reporting so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as account,description,statement,report are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

39. The crowd cheered when the striker scored a

Stage 1 — low entropy: strengthen football script so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as goal are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

40. She placed the flowers in a glass

Stage 1 — low entropy: strengthen object-function relation so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as vase,jar,bottle are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

41. The fire alarm sounded and everyone left the

Stage 1 — low entropy: strengthen evacuation script so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as building,room,office,school are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

42. He raised his hand because he wanted to ask a

Stage 1 — low entropy: strengthen classroom formula so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as question are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

43. The team practised every day before the

Stage 1 — low entropy: strengthen sport script so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as match,game,tournament,competition are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

44. She turned down the music because it was too

Stage 1 — low entropy: strengthen causal adjective so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as loud,noisy are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

45. He saved the document before closing the

Stage 1 — low entropy: strengthen computing script so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as program,file,computer,window are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

46. They discussed the proposal during the

Stage 1 — low entropy: strengthen institutional event so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as meeting,session,conference,debate are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

47. The bridge was designed to withstand

Stage 1 — low entropy: strengthen engineering so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as pressure,wind,earthquakes,loads are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

48. The new evidence could change the

Stage 1 — low entropy: strengthen abstract discourse so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as outcome,decision,case,interpretation are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

49. The policy aims to reduce

Stage 1 — low entropy: strengthen policy frame so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as costs,emissions,risk,crime,inequality are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

50. The medicine may cause

Stage 1 — low entropy: strengthen medical so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as nausea,drowsiness,headache,side effects are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

51. The article argues that

Stage 1 — low entropy: strengthen propositional slot so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as policy X is wrong,the evidence is weak,change is needed are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

52. She said that

Stage 1 — low entropy: strengthen open clause so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as she was tired,he had left,the plan worked are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

53. The book is about

Stage 1 — low entropy: strengthen topic slot so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as history,science,a family,a journey,war are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

54. They decided to

Stage 1 — low entropy: strengthen verb complement so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as leave,wait,continue,apply,cancel are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

55. He is interested in

Stage 1 — low entropy: strengthen broad noun/-ing slot so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as music,science,history,learning,art are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

56. The result was

Stage 1 — low entropy: strengthen copular adjective slot so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as good,bad,unexpected,significant,clear are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

57. The problem can be solved by

Stage 1 — low entropy: strengthen open method phrase so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as changing the design,adding data,reducing cost are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

58. In conclusion, the study

Stage 1 — low entropy: strengthen academic discourse so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as supports the claim,shows a trend,raises questions are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

59. Once upon a

Stage 1 — low entropy: strengthen fixed phrase so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as time are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

60. By the

Stage 1 — low entropy: strengthen fixed expression so one or two plausible lexical candidates dominate. This helps a learner infer or retrieve the target with manageable uncertainty.

Stage 2 — verification: reveal the actual word and clarify why competing candidates such as way are better or worse fits.

Stage 3 — raise entropy: remove one strong cue, change topic or weaken the collocation. The learner should now rely more on stored lexical knowledge.

Stage 4 — productive choice: present an open context with several legitimate options and require the learner to select the word that best fits meaning, register and collocation.

27. Frequently asked questions

What is lexical entropy?

Uncertainty over possible upcoming lexical items given a context.

How is entropy different from surprisal?

Entropy describes uncertainty before the next word; surprisal describes how unexpected the actual word is.

What does low entropy mean?

A small number of continuations dominate the probability distribution.

What does high entropy mean?

Probability is spread across many plausible continuations.

Is entropy the same as contextual predictability?

They are related, but entropy summarises the whole continuation distribution rather than the probability of one target.

How does syntax affect entropy?

Grammar narrows the classes and forms that can appear next.

How do collocations affect entropy?

Strong phrase patterns concentrate probability on a small set of continuations.

Does high entropy always mean difficult language?

Not necessarily. It means greater lexical uncertainty, though processing cost also depends on familiarity, syntax and task.

Do language models compute entropy over words?

Often over tokens; token entropy should not automatically be treated as whole-word uncertainty.

What is the simplest rule?

Ask how many plausible continuations remain and how evenly probability is spread among them.

28. Research grounding

The 2025 Cambridge Handbook of Language and Brain distinguishes entropy from surprisal in predictive language processing: entropy captures uncertainty over the set of possible continuations, whereas surprisal captures the unexpectedness of a specific observed word. Cambridge work on predictive processing similarly explains that entropy is low when only a small set of words is likely and high when the next-word distribution remains diffuse. Information-theoretic accounts treat comprehension as progressive uncertainty reduction as context accumulates.

• Cambridge — Prediction, entropy and surprisal in language processing
• Cambridge — Entropy versus surprisal
• Cambridge — Reducing surprisal and entropy

29. eduKateSG routes

• Contextual Predictability
• Lexical Prediction
• Contextual Informativeness
• Vocabulary Master Router

30. Final model

Lexical entropy measures the width of the road ahead. Low entropy means context has narrowed the route to a few likely words; high entropy means many lexical futures remain open.

Strong comprehenders use syntax, collocation, discourse and world knowledge to reduce uncertainty while remaining ready to update when the actual word violates expectation.

Context does not merely predict the next word. It compresses the space of possibilities.

31. Final entropy calibration lab

1. The chef sharpened the

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using restaurant/kitchen script; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

2. She spread butter on the

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using strong food collocation; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

3. The baby was tired, so he put her to

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using formulaic event; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

4. He unlocked the door with a

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using instrument relation; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

5. The referee blew the

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using fixed sports phrase; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

6. The nurse measured the patient’s blood

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using medical collocation; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

7. The committee reached a final

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using institutional collocation; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

8. The scientist analysed the

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using research domain; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

9. She walked into the

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using weak spatial context; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

10. He looked at the

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using very broad object slot; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

11. The child picked up a

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using broad object context; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

12. After three days of rain, the river

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using weather script; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

13. The judge announced the

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using courtroom script; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

14. The lawyer presented new

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using legal discourse; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

15. The doctor prescribed

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using medical script; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

16. The cyclist wore a helmet for

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using purpose relation; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

17. The phone battery was almost

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using device state; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

18. The librarian asked everyone to speak

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using institutional expectation; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

19. The musician tuned the guitar before the

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using music event; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

20. The students opened their books to page

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using many numeric continuations; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

21. The road was closed because of

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using multiple causes; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

22. The report identified three major

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using reporting frame; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

23. The hikers followed the marked

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using outdoor collocation; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

24. The gardener watered the plants every

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using temporal options; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

25. The detective searched the room for

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using genre script; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

26. The company reduced costs without lowering

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using business discourse; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

27. The patient felt better after taking the

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using health script; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

28. She reread the paragraph because it was

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using reading evaluation; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

29. The researcher calculated the average from the

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using research/math; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

30. The manager scheduled another meeting to discuss the

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using business discourse; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

31. The children built a sandcastle near the

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using beach script; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

32. The farmer repaired the fence around the

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using rural script; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

33. The speaker paused to allow the audience to

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using discourse action; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

34. The engineer tested the bridge under heavy

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using engineering semantics; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

35. The journalist verified the information before publishing the

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using media script; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

36. The customer asked for a full

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using commercial script; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

37. The experiment was repeated to confirm the

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using research script; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

38. The witness gave a detailed

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using legal/reporting; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

39. The crowd cheered when the striker scored a

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using football script; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

40. She placed the flowers in a glass

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using object-function relation; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

41. The fire alarm sounded and everyone left the

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using evacuation script; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

42. He raised his hand because he wanted to ask a

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using classroom formula; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

43. The team practised every day before the

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using sport script; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

44. She turned down the music because it was too

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using causal adjective; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

45. He saved the document before closing the

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using computing script; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

46. They discussed the proposal during the

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using institutional event; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

47. The bridge was designed to withstand

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using engineering; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

48. The new evidence could change the

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using abstract discourse; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

49. The policy aims to reduce

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using policy frame; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

50. The medicine may cause

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using medical; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

51. The article argues that

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using propositional slot; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

52. She said that

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using open clause; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

53. The book is about

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using topic slot; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

54. They decided to

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using verb complement; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

55. He is interested in

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using broad noun/-ing slot; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

56. The result was

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using copular adjective slot; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

57. The problem can be solved by

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using open method phrase; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

58. In conclusion, the study

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using academic discourse; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

59. Once upon a

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using fixed phrase; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

60. By the

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using fixed expression; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

61. The chef sharpened the

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using restaurant/kitchen script; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

62. She spread butter on the

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using strong food collocation; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

63. The baby was tired, so he put her to

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using formulaic event; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.

64. He unlocked the door with a

Probability-shape exercise: create one distribution where the top candidate has 80% probability and another where five candidates share probability almost evenly. Explain why the first has lower entropy even if both contain the same target.

Surprisal exercise: choose the least likely plausible candidate and insert it. The actual continuation now has high surprisal even though the pre-word entropy may have been low.

Pedagogical decision: if the learner is seeing a target for the first time, reduce uncertainty using instrument relation; if the learner already knows the word, allow more alternatives so lexical selection is genuinely tested.