THE MASTERY CLUB · VOCABULARY SERIES · SELECTIONAL RESTRICTIONS · SELECTIONAL PREFERENCES · SEMANTIC COMPATIBILITY · WORD COMBINATION

Vocabulary | Selectional Restrictions and Semantic Compatibility — Why Some Word Combinations Make Sense and Others Do Not

Selectional restrictions and selectional preferences describe the semantic expectations words place on the other words they combine with. A verb such as drink normally takes something drinkable as its object; diagonalise in mathematics expects a matrix or related mathematical object; persuade usually expects an animate or person-like participant capable of changing a decision. Search questions such as “What are selectional restrictions?”, “What is semantic compatibility?”, “Why does a sentence sound semantically wrong even when the grammar is correct?”, “What are selectional preferences?”, and “Why do verbs prefer certain nouns?” all point toward the same lexical-semantic problem: grammar can permit a structure while word meaning makes some combinations natural, strange, metaphorical or impossible under the ordinary interpretation.

Selectional preferences are usually better treated as soft expectations rather than absolute rules. English can say eat lunch, eat glass, eat your words and even the spreadsheet ate my afternoon in figurative language. The first is ordinary, the second unusual but physically possible, the third idiomatic and the fourth metaphorical. Modern lexical-semantics and computational accounts therefore often model predicate–argument compatibility probabilistically or as graded semantic preference rather than as a simple grammatical ban. A combination can violate the usual preference and still be interpretable because context, coercion, metaphor, domain knowledge or pragmatic imagination repairs the mismatch.

This article is the canonical Selectional Restrictions and Semantic Compatibility specialist inside Vocabulary — The Mastery Club apex. It is a completely new article and does not rewrite existing eduKate pages. Existing specialists on Semantic Frames, Lexical Coercion, Lexical Entailment, Troponymy and Syntactic Bootstrapping remain protected. This page owns the broad system: semantic compatibility, predicate preferences, argument types, anomaly, coercion, metaphorical repair and how learners use compatibility to interpret and choose vocabulary.

The 60-Second Answer

A selectional restriction is a traditional way of stating that a predicate expects arguments with certain semantic properties. Drink prefers a liquid or drinkable substance; murder normally requires a living victim; pregnant applies only to organisms of an appropriate biological type under literal interpretation. Modern accounts often replace strict restrictions with selectional preferences because language regularly stretches or violates ordinary expectations.

Semantic compatibility is therefore graded. She drank water is highly compatible. She drank medicine is compatible but less prototypical. She drank sunlight is anomalous literally but interpretable poetically. Strong vocabulary knowledge includes knowing these normal compatibility patterns and recognising when a writer is deliberately breaking them.

Part I — Why Grammar Is Not Enough

1. Grammatical but semantically strange

The rock persuaded the teacher. The sentence has an ordinary subject–verb–object structure, yet literal interpretation is odd because rocks do not normally possess communicative agency.

2. Ungrammatical and semantically odd are different problems

Teacher the persuaded rock breaks English syntax. The rock persuaded the teacher is syntactically well formed but semantically unusual. Vocabulary instruction should help learners distinguish these failure types.

3. Lexical meaning helps compose sentence meaning

Words contribute more than dictionary definitions. Predicates bring expectations about participants, event types, instruments, results and states. Those expectations guide real-time comprehension.

Part II — From Restrictions to Preferences

4. Early strict-feature models

Classic accounts represented selectional restrictions as semantic features: eat might require a [+EDIBLE] object or sleep a [+ANIMATE] subject under literal use.

5. Why strict rules fail

People eat glass in extraordinary performances, computers can sleep, markets can digest news, and writers constantly produce metaphor. Real language uses categories flexibly.

6. Modern preference view

Selectional preferences describe typical compatibility rather than absolute possibility. Corpus and computational approaches estimate which semantic classes commonly fill a predicate’s roles.

Part III — Participant Roles

7. Agent

The participant that intentionally or causally initiates an action in many event descriptions: The chef cut the bread.

8. Experiencer

The participant that perceives, feels or undergoes a mental state: Maya feared the storm.

9. Theme or patient

The entity moved, affected or described by the predicate: She lifted the box.

10. Instrument

The means used in an action: He cut the rope with a knife. A knife is more compatible as a cutting instrument than milk under ordinary interpretation.

11. Location, source, goal and path

Motion and placement verbs constrain spatial roles differently: put strongly expects a location complement, while arrive encodes a goal-like endpoint.

Part IV — Compatibility Is Sense-Specific

Polysemous words change their selectional behaviour by sense. Run in run a race, run a company, run software and run water selects different participant types. Selectional knowledge therefore belongs to word-sense knowledge, not merely the spelling run.

12. Literal sense

The athlete ran the race. Human/animate agent plus event/path object.

13. Management sense

She runs the company. Human/institutional agent plus organisation object.

14. Computing sense

The server runs the program. Machine/system subject plus software object.

15. Flow sense

Water runs through the pipe. Liquid theme plus path.

Part V — Verb Selectional Preferences

eat

Typical preference: edible food-like theme. Ordinary combinations: eat bread, eat rice, eat fruit. Stretch/extension: eat glass is unusual but possible; eat your words is idiomatic.

drink

Typical preference: drinkable liquid. Ordinary combinations: drink water, drink tea. Stretch/extension: drink sunlight is poetic; drink medicine is possible despite lower prototypicality.

murder

Typical preference: living sentient victim. Ordinary combinations: murder a person. Stretch/extension: murder a spreadsheet is anomalous unless metaphorical.

kill

Typical preference: living target or metaphorical process. Ordinary combinations: kill an insect, kill a process. Stretch/extension: computing extends the verb to programs and tasks.

persuade

Typical preference: entity capable of belief/decision. Ordinary combinations: persuade a voter, persuade the committee. Stretch/extension: persuade the wall requires personification.

convince

Typical preference: cognitive experiencer. Ordinary combinations: convince a student, convince the jury. Stretch/extension: institutional groups can behave as collective experiencers.

apologise

Typical preference: typically animate/social agent. Ordinary combinations: she apologised to him. Stretch/extension: a machine can apologise only through personified interface language.

sleep

Typical preference: animate subject or machine metaphor. Ordinary combinations: the baby sleeps, the computer sleeps. Stretch/extension: device sense is conventional metaphor.

bark

Typical preference: dog-like animal subject in literal sense. Ordinary combinations: the dog barked. Stretch/extension: the command barked at him shifts to human speech manner.

meow

Typical preference: cat-like subject. Ordinary combinations: the cat meowed. Stretch/extension: humans can imitate the sound, context overrides preference.

bleed

Typical preference: living organism or fluid-containing system. Ordinary combinations: the patient bled, the brake line bled. Stretch/extension: technical senses extend source type.

evaporate

Typical preference: liquid/volatile substance. Ordinary combinations: water evaporates. Stretch/extension: ideas evaporate is metaphorical.

melt

Typical preference: solid that can become liquid or metaphorical state. Ordinary combinations: ice melts. Stretch/extension: resistance melted is figurative.

rust

Typical preference: metal/oxidisable material. Ordinary combinations: iron rusts. Stretch/extension: skills rust is metaphorical.

sprout

Typical preference: plant/plant-like source. Ordinary combinations: seeds sprout. Stretch/extension: new cafes sprouted across town is metaphorical extension.

germinate

Typical preference: seed/spore or idea metaphor. Ordinary combinations: seeds germinate. Stretch/extension: an idea germinated extends biological frame.

calculate

Typical preference: quantity/expression/value. Ordinary combinations: calculate area, calculate probability. Stretch/extension: calculate a sandwich is anomalous without special context.

diagonalise

Typical preference: matrix/operator. Ordinary combinations: diagonalise a matrix. Stretch/extension: highly technical selectional preference.

differentiate

Typical preference: function or conceptual distinction. Ordinary combinations: differentiate a function, differentiate the terms. Stretch/extension: sense determines compatible object.

integrate

Typical preference: function/system/group. Ordinary combinations: integrate a function, integrate systems. Stretch/extension: polysemy creates several compatibility classes.

vaccinate

Typical preference: animate patient/population. Ordinary combinations: vaccinate children. Stretch/extension: vaccinate software is metaphorical cyber language at best.

diagnose

Typical preference: person/condition/system problem depending sense. Ordinary combinations: diagnose a patient, diagnose a fault. Stretch/extension: technical extension broadens object type.

prescribe

Typical preference: medicine/treatment/rule. Ordinary combinations: prescribe antibiotics. Stretch/extension: prescribe a colour is odd unless design rules are meant.

encrypt

Typical preference: data/message/file. Ordinary combinations: encrypt data. Stretch/extension: encrypt a chair is anomalous literally.

compile

Typical preference: code/document/list. Ordinary combinations: compile software, compile a report. Stretch/extension: sense-specific object classes.

reboot

Typical preference: computer/device/system. Ordinary combinations: reboot the server. Stretch/extension: reboot a career is metaphorical.

upload

Typical preference: digital file/data. Ordinary combinations: upload a photo. Stretch/extension: upload a physical chair requires digitisation/metaphor.

download

Typical preference: digital data. Ordinary combinations: download a file. Stretch/extension: download knowledge into a brain is science-fiction/metaphor.

legislate

Typical preference: rules/laws or on an issue. Ordinary combinations: legislate standards. Stretch/extension: legislate a sandwich is semantically mismatched.

ratify

Typical preference: treaty/agreement/formal decision. Ordinary combinations: ratify a treaty. Stretch/extension: ratify a mood is anomalous.

acquit

Typical preference: defendant/person accused. Ordinary combinations: acquit the defendant. Stretch/extension: acquit a stone is impossible under literal legal sense.

sentence

Typical preference: convicted person. Ordinary combinations: sentence the offender. Stretch/extension: sentence the building is anomalous.

borrow

Typical preference: transferable possession/resource. Ordinary combinations: borrow a book, borrow money. Stretch/extension: borrow an idea is conventional metaphorical extension.

lend

Typical preference: transferable possession/resource. Ordinary combinations: lend money, lend support. Stretch/extension: lend a mountain is odd unless ownership/control is contextual.

owe

Typical preference: debt/obligation to person/institution. Ordinary combinations: owe money, owe an apology. Stretch/extension: semantic roles matter more than surface nouns.

purchase

Typical preference: goods/services/assets. Ordinary combinations: purchase equipment. Stretch/extension: purchase a sunset is anomalous unless rights/art are involved.

harvest

Typical preference: crop/resource/data metaphor. Ordinary combinations: harvest wheat, harvest energy, harvest data. Stretch/extension: technical metaphors broaden selection.

mine

Typical preference: mineral/resource/data. Ordinary combinations: mine coal, mine data. Stretch/extension: data mining is conventional extension.

brew

Typical preference: beverage/mixture; trouble metaphor. Ordinary combinations: brew tea, a storm is brewing. Stretch/extension: event sense is lexicalised metaphor.

forge

Typical preference: metal object/document/relationship. Ordinary combinations: forge a sword, forge a signature, forge an alliance. Stretch/extension: polysemy yields distinct compatibility profiles.

Part VI — Adjective Selectional Preferences

pregnant

Typical compatibility: biologically capable organism. Ordinary: pregnant woman / pregnant animal. Extended: pregnant pause is lexicalised metaphor.

bald

Typical compatibility: head/person/animal with hair or surface metaphor. Ordinary: bald man / bald patch. Extended: bald statement = plain/unadorned.

rancid

Typical compatibility: fats/oils/food. Ordinary: rancid butter. Extended: rancid idea is figurative and marked.

stale

Typical compatibility: bread/air/news/ideas. Ordinary: stale bread, stale joke. Extended: meaning extends from freshness to novelty.

ripe

Typical compatibility: fruit/biological maturity/opportunity. Ordinary: ripe mango, ripe age, ripe opportunity. Extended: semantic metaphor broadens selection.

edible

Typical compatibility: food-like substance. Ordinary: edible mushroom. Extended: edible paper works if physically consumable.

audible

Typical compatibility: sound/event capable of being heard. Ordinary: audible whisper. Extended: audible colour is cross-sensory metaphor.

legible

Typical compatibility: writing/print. Ordinary: legible handwriting. Extended: legible landscape is metaphorical visual organisation.

solvent

Typical compatibility: financial entity under finance sense. Ordinary: solvent company. Extended: chemical sense has different noun behaviour.

flammable

Typical compatibility: material/substance. Ordinary: flammable liquid. Extended: flammable argument is metaphorical.

bilingual

Typical compatibility: person/community/text in language sense. Ordinary: bilingual child, bilingual sign. Extended: bilingual rock is odd without art/metaphor.

sentient

Typical compatibility: organism/entity with experience. Ordinary: sentient being. Extended: sentient algorithm is debated/speculative.

fragile

Typical compatibility: physical object/system/state. Ordinary: fragile glass, fragile peace. Extended: metaphorical extension is common.

torrential

Typical compatibility: rain/flow-like event. Ordinary: torrential rain. Extended: torrential criticism is metaphorical intensity.

punctual

Typical compatibility: agent/event occurring on time. Ordinary: punctual employee, punctual train. Extended: punctual stone is anomalous.

eloquent

Typical compatibility: speaker/text/expression. Ordinary: eloquent speech, eloquent silence. Extended: inanimate nouns possible through expressive metaphor.

nutritious

Typical compatibility: food/diet. Ordinary: nutritious meal. Extended: nutritious argument is anomalous.

contagious

Typical compatibility: disease/emotion/behaviour. Ordinary: contagious virus, contagious laughter. Extended: metaphorical extension conventional.

transparent

Typical compatibility: material/process/communication. Ordinary: transparent glass, transparent policy. Extended: polysemy changes semantic class.

aromatic

Typical compatibility: substance with smell. Ordinary: aromatic herbs. Extended: aromatic equation is anomalous.

Part VII — Noun–Noun Semantic Compatibility

Noun compounds and attributive nouns also display preferences. English permits creative compounding, but interpretation depends on plausible relations between the nouns.

coffee cup

Relation: container-for-content. Interpretation: cup used for coffee.

stone wall

Relation: material-of-object. Interpretation: wall made of stone.

school bus

Relation: purpose/destination. Interpretation: bus serving a school.

chicken soup

Relation: ingredient/content. Interpretation: soup containing chicken.

data centre

Relation: function/domain. Interpretation: centre for data/computing infrastructure.

language model

Relation: domain/object of modelling. Interpretation: model involving language.

policy failure

Relation: event/result relation. Interpretation: failure of a policy.

rain jacket

Relation: protection-purpose. Interpretation: jacket for rain.

summer school

Relation: time relation. Interpretation: school/program in summer.

student loan

Relation: recipient relation. Interpretation: loan for/to a student.

banana phone

Relation: normally semantically odd. Interpretation: interpretable as novelty object, toy or metaphor through context.

stone idea

Relation: semantically underdetermined/odd. Interpretation: requires poetic interpretation such as rigid/heavy idea.

emotion engine

Relation: creative metaphor. Interpretation: engine that produces/drives emotion.

memory palace

Relation: conventional metaphor. Interpretation: imagined spatial mnemonic system.

Part VIII — Coercion: Repairing a Mismatch

Lexical coercion occurs when the semantic requirements of one expression force another expression to be interpreted in a richer or shifted way. Begin the book normally means begin reading, writing or otherwise engaging in an event involving the book, even though book is an object noun rather than an event.

begin the book

Interpretation: supply an event such as reading/writing. Repair mechanism: event-selecting verb coerces object noun.

finish the sandwich

Interpretation: eat/prepare depending context. Repair mechanism: object linked to conventional event.

enjoy the book

Interpretation: reading/experiencing content. Repair mechanism: experiencer predicate accesses event/content.

the newspaper fired the editor

Interpretation: institution stands for management. Repair mechanism: metonymic coercion.

the ham sandwich wants the bill

Interpretation: customer identified by order. Repair mechanism: restaurant metonymy.

drink the entire bottle

Interpretation: contents rather than container. Repair mechanism: container-for-content coercion.

the city voted

Interpretation: residents/electorate. Repair mechanism: place-for-people metonymy.

the White House announced

Interpretation: administration/institution. Repair mechanism: place-for-institution.

the kettle is boiling

Interpretation: water in kettle. Repair mechanism: container-for-content.

she read Shakespeare

Interpretation: works by Shakespeare. Repair mechanism: author-for-work metonymy.

Part IX — Literal Anomaly, Metaphor and Personification

16. Literal anomaly

The number seven drank the bicycle. Without special context, the semantic types conflict too strongly for an ordinary literal interpretation.

17. Metaphor

She drank in the view. The drinking frame is mapped onto perception and appreciation. The selectional violation is the mechanism that signals figurative interpretation.

18. Personification

The city refuses to sleep. An inanimate place receives an animate volitional predicate, creating personification.

19. Science-fiction/domain override

In a fictional world with sentient rocks, the rock persuaded the teacher may become literal. Selectional preferences depend partly on world knowledge.

20. Humour

Comedians deliberately violate normal compatibility to surprise listeners. Anomaly can be a creative resource, not merely an error.

Part X — Semantic Compatibility Casebook: 70 Combinations

1. drink water

Compatibility: high compatibility. Semantic structure: drink + liquid. Interpretation: ordinary literal combination.

2. drink medicine

Compatibility: compatible but less prototypical. Semantic structure: drink + ingestible liquid. Interpretation: context makes medicinal liquid normal.

3. drink sunlight

Compatibility: literal mismatch. Semantic structure: drink + nonliquid abstract/physical phenomenon. Interpretation: poetic metaphor can repair it.

4. eat rice

Compatibility: high compatibility. Semantic structure: eat + food. Interpretation: ordinary literal.

5. eat glass

Compatibility: low preference but physically possible. Semantic structure: eat + inedible solid. Interpretation: world knowledge marks unusual event, not grammatical error.

6. eat your words

Compatibility: idiomatic. Semantic structure: eat + linguistic object. Interpretation: idiom overrides literal food preference.

7. persuade the voter

Compatibility: high compatibility. Semantic structure: persuade + cognitive agent. Interpretation: ordinary.

8. persuade the committee

Compatibility: compatible collective. Semantic structure: institution/group interpreted as decision-makers. Interpretation: metonymic collective reading.

9. persuade the wall

Compatibility: literal mismatch. Semantic structure: persuade + noncognitive object. Interpretation: personification needed.

10. convince the jury

Compatibility: high compatibility. Semantic structure: convince + reasoning participants. Interpretation: ordinary.

11. convince the spreadsheet

Compatibility: mismatch. Semantic structure: convince + nonmental object. Interpretation: metaphor/personification required.

12. murder the victim

Compatibility: high literal compatibility. Semantic structure: murder + living person. Interpretation: ordinary criminal sense.

13. murder the song

Compatibility: figurative. Semantic structure: murder + performance/work. Interpretation: means perform very badly.

14. kill the mosquito

Compatibility: high literal compatibility. Semantic structure: kill + living organism. Interpretation: ordinary.

15. kill the process

Compatibility: technical extension. Semantic structure: kill + software process. Interpretation: computing lexicalisation.

16. kill the mood

Compatibility: metaphorical conventional extension. Semantic structure: kill + abstract state. Interpretation: means destroy atmosphere.

17. vaccinate children

Compatibility: high compatibility. Semantic structure: vaccinate + animate patient. Interpretation: ordinary.

18. vaccinate the server

Compatibility: mismatch/metaphor. Semantic structure: vaccinate + machine. Interpretation: cybersecurity metaphor only.

19. diagnose the patient

Compatibility: high medical compatibility. Semantic structure: diagnose + patient. Interpretation: ordinary clinical use.

20. diagnose the fault

Compatibility: technical extension. Semantic structure: diagnose + system problem. Interpretation: engineering/computing sense.

21. diagnose the sandwich

Compatibility: anomalous. Semantic structure: diagnose + ordinary food. Interpretation: requires comic/metaphorical frame.

22. prescribe antibiotics

Compatibility: high compatibility. Semantic structure: prescribe + treatment. Interpretation: ordinary medical.

23. prescribe silence

Compatibility: extended institutional use. Semantic structure: prescribe + behaviour/state. Interpretation: formal recommendation/order.

24. prescribe a mountain

Compatibility: mismatch. Semantic structure: prescribe + physical object not treatment. Interpretation: requires special context.

25. encrypt data

Compatibility: high technical compatibility. Semantic structure: encrypt + information. Interpretation: ordinary computing.

26. encrypt a chair

Compatibility: mismatch. Semantic structure: encrypt + physical furniture. Interpretation: only possible if chair data/model meant.

27. compile code

Compatibility: high technical compatibility. Semantic structure: compile + source code. Interpretation: ordinary computing.

28. compile a bibliography

Compatibility: different sense. Semantic structure: compile + collected information. Interpretation: lexical polysemy changes preference.

29. compile a cloud

Compatibility: mismatch. Semantic structure: compile + natural object. Interpretation: figurative context needed.

30. reboot the computer

Compatibility: high compatibility. Semantic structure: reboot + computing device. Interpretation: ordinary.

31. reboot the economy

Compatibility: metaphorical conventional extension. Semantic structure: reboot + system. Interpretation: restart metaphor.

32. upload a photo

Compatibility: high compatibility. Semantic structure: upload + digital content. Interpretation: ordinary.

33. upload a bicycle

Compatibility: literal mismatch. Semantic structure: upload + physical object. Interpretation: requires digital model/photo meaning.

34. ratify the treaty

Compatibility: high legal compatibility. Semantic structure: ratify + formal agreement. Interpretation: ordinary.

35. ratify a sandwich

Compatibility: anomalous. Semantic structure: ratify + food. Interpretation: semantic type mismatch.

36. acquit the defendant

Compatibility: high legal compatibility. Semantic structure: acquit + accused person. Interpretation: ordinary.

37. acquit the building

Compatibility: mismatch. Semantic structure: acquit + nonaccused inanimate object. Interpretation: requires personification/legal fiction.

38. borrow a book

Compatibility: high compatibility. Semantic structure: borrow + transferable possession. Interpretation: ordinary.

39. borrow money

Compatibility: high compatibility. Semantic structure: borrow + fungible resource. Interpretation: ordinary.

40. borrow an idea

Compatibility: metaphorical/conventional. Semantic structure: borrow + abstract concept. Interpretation: temporary/adoptive metaphor.

41. borrow a sunset

Compatibility: mismatch/poetic. Semantic structure: borrow + nonpossessable event. Interpretation: metaphor only.

42. lend money

Compatibility: high compatibility. Semantic structure: lend + transferable resource. Interpretation: ordinary.

43. lend support

Compatibility: metaphorical conventional. Semantic structure: lend + abstract assistance. Interpretation: lexicalised extension.

44. lend a mountain

Compatibility: low compatibility. Semantic structure: lend + large property/entity. Interpretation: possible only under ownership/legal scenario.

45. owe money

Compatibility: high compatibility. Semantic structure: owe + debt. Interpretation: ordinary.

46. owe an apology

Compatibility: abstract obligation. Semantic structure: owe + social act. Interpretation: conventional.

47. owe a cloud

Compatibility: mismatch. Semantic structure: owe + nonpossessable object. Interpretation: context required.

48. harvest wheat

Compatibility: high compatibility. Semantic structure: harvest + crop. Interpretation: literal.

49. harvest energy

Compatibility: technical metaphor. Semantic structure: harvest + resource. Interpretation: conventional engineering extension.

50. harvest data

Compatibility: technical metaphor. Semantic structure: harvest + information. Interpretation: conventional computing/analytics.

51. mine coal

Compatibility: high literal compatibility. Semantic structure: mine + mineral/resource. Interpretation: ordinary.

52. mine data

Compatibility: technical metaphor. Semantic structure: mine + information. Interpretation: conventional computing.

53. mine happiness

Compatibility: marked metaphor. Semantic structure: mine + emotion. Interpretation: requires figurative frame.

54. brew tea

Compatibility: high compatibility. Semantic structure: brew + beverage. Interpretation: ordinary.

55. brew beer

Compatibility: high compatibility. Semantic structure: brew + fermented beverage. Interpretation: ordinary.

56. trouble is brewing

Compatibility: metaphorical conventional. Semantic structure: brew + event/state. Interpretation: develop gradually.

57. rusted iron

Compatibility: high compatibility. Semantic structure: rust + metal. Interpretation: ordinary.

58. rusted skills

Compatibility: metaphorical. Semantic structure: rust + ability. Interpretation: decline through disuse.

59. ripe mango

Compatibility: high compatibility. Semantic structure: ripe + fruit. Interpretation: ordinary.

60. ripe opportunity

Compatibility: metaphorical conventional. Semantic structure: ripe + abstract opportunity. Interpretation: ready for action.

61. stale bread

Compatibility: high compatibility. Semantic structure: stale + food. Interpretation: ordinary.

62. stale joke

Compatibility: metaphorical conventional. Semantic structure: stale + idea/content. Interpretation: old/unoriginal.

63. fragile glass

Compatibility: high compatibility. Semantic structure: fragile + breakable object. Interpretation: ordinary.

64. fragile peace

Compatibility: metaphorical conventional. Semantic structure: fragile + abstract state. Interpretation: easily disrupted.

65. audible whisper

Compatibility: high compatibility. Semantic structure: audible + sound event. Interpretation: ordinary.

66. audible colour

Compatibility: cross-modal mismatch. Semantic structure: audible + visual property. Interpretation: synesthetic metaphor only.

67. legible handwriting

Compatibility: high compatibility. Semantic structure: legible + written marks. Interpretation: ordinary.

68. legible landscape

Compatibility: marked metaphor. Semantic structure: legible + visual/spatial system. Interpretation: interpretable as readable/understandable.

69. punctual train

Compatibility: compatible event/system. Semantic structure: punctual + scheduled entity. Interpretation: ordinary metonymic use.

70. punctual stone

Compatibility: mismatch. Semantic structure: punctual + nonscheduled object. Interpretation: requires personification.

71. nutritious meal

Compatibility: high compatibility. Semantic structure: nutritious + food. Interpretation: ordinary.

72. nutritious argument

Compatibility: mismatch/metaphor. Semantic structure: nutritious + discourse. Interpretation: figurative ‘intellectually nourishing’ possible.

73. transparent glass

Compatibility: literal sense. Semantic structure: transparent + material. Interpretation: ordinary.

74. transparent policy

Compatibility: abstract sense. Semantic structure: transparent + process/governance. Interpretation: lexicalised metaphor meaning open/clear.

75. sentient organism

Compatibility: high compatibility. Semantic structure: sentient + living/experiencing entity. Interpretation: ordinary philosophical/biological.

76. sentient algorithm

Compatibility: speculative/debated. Semantic structure: sentient + artificial system. Interpretation: compatibility depends on theory/world assumptions.

Part XI — Selectional Preferences and Metaphor Detection

One of the most useful consequences of selectional knowledge is metaphor detection. When a predicate combines with an argument outside its usual semantic class, the reader often searches for a figurative interpretation. The market swallowed the bad news violates literal eating preferences but activates a metaphor of ingestion and absorption.

time flies

Violated preference: fly prefers animate/airborne movers. Metaphorical repair: time is conceptualised as moving quickly.

ideas bloom

Violated preference: bloom prefers plants. Metaphorical repair: ideas are conceptualised as living growth.

anger boiled

Violated preference: boil prefers liquids. Metaphorical repair: emotion is conceptualised as heated fluid.

the news hit him

Violated preference: hit prefers physical contact. Metaphorical repair: information is conceptualised as impact.

the argument collapsed

Violated preference: collapse prefers structures. Metaphorical repair: reasoning is conceptualised as a building.

she digested the report

Violated preference: digest prefers food. Metaphorical repair: understanding is conceptualised as digestion.

the policy strangled growth

Violated preference: strangle prefers living victim. Metaphorical repair: constraint is conceptualised as bodily restriction.

the rumour infected the town

Violated preference: infect prefers organisms/populations. Metaphorical repair: information spread is conceptualised as disease.

the idea took root

Violated preference: root prefers plants. Metaphorical repair: concept establishment is conceptualised as plant growth.

the company weathered the storm

Violated preference: weather normally relates exposure to weather. Metaphorical repair: difficulty is conceptualised as a storm.

Part XII — Selectional Preferences and Word Learning

21. New verb meaning from arguments

If a child hears She daxed the ball into the box, the direct object, goal phrase and event scene constrain what dax could mean. Grammar and semantic compatibility cooperate.

22. New adjective meaning from noun class

If glorpous repeatedly modifies foods that cause illness, learners can infer an evaluative/edibility-related meaning even before a precise definition is available.

23. Technical vocabulary

Domain-specific predicates narrow possible meanings sharply. In mathematics, diagonalise selects matrices/operators; in computing, encrypt selects data; in law, ratify selects formal agreements.

24. Semantic bootstrapping

Known lexical classes can help infer unknown words because compatible argument types reveal event structure. This complements syntactic bootstrapping rather than replacing it.

Part XIII — Selectional Preferences and Collocation

Selectional preference and collocation overlap but are not identical. Drink water is both semantically compatible and frequent. Drink seawater is semantically compatible but less common. Drink sunlight is collocationally rare and semantically anomalous literally. Collocation concerns conventional co-occurrence; selectional preference concerns semantic type compatibility.

25. Semantic class first

A verb may accept a broad class, such as liquids for drink.

26. Collocational preference within the class

Within liquids, drink water/tea/coffee are far more conventional than drink varnish, even though varnish is physically liquid.

27. World knowledge

Some combinations are semantically type-compatible but pragmatically implausible because the object is poisonous, socially inappropriate or physically inaccessible.

Part XIV — Semantic Compatibility vs World Knowledge

Selectional preferences cannot be reduced entirely to dictionary semantics. The child lifted the piano is semantically compatible—children and pianos fit the participant types—but world knowledge makes the event improbable. The piano lifted the child is more anomalous because agency and physical causation expectations are reversed, yet a crane-like robotic piano in a fictional world could make it true.

Strong comprehension therefore integrates lexical semantics with encyclopedic knowledge. The reader asks both “Can this kind of word combine with that kind of participant?” and “Could this event plausibly occur in the current world?”

Part XV — Selectional Restrictions and Semantic Compatibility Glossary

Selectional restriction

A traditional semantic constraint stating what kinds of arguments a predicate can take.

Selectional preference

A softer, graded expectation about the semantic types that typically fill a predicate’s roles.

Semantic compatibility

Degree to which the meanings of combined expressions fit together naturally under a given interpretation.

Semantic anomaly

A combination whose literal interpretation clashes strongly with normal lexical-semantic expectations.

Semantic deviance

Marked mismatch in meaning composition; often used for unusual but syntactically possible combinations.

Predicate

Expression that attributes a property or relation, often a verb or adjective.

Argument

Participant or value required/selected by a predicate.

Agent

Participant that intentionally or causally initiates an action in many event descriptions.

Patient

Participant affected by an event.

Theme

Participant moved, located, described or otherwise centrally involved without necessarily being changed.

Experiencer

Participant that perceives, feels or undergoes a mental state.

Instrument

Means used to carry out an action.

Goal

Endpoint of movement or transfer.

Source

Starting point of movement or transfer.

Location

Place where a state/event holds.

Beneficiary

Participant that benefits from an action.

Recipient

Participant that receives something.

Semantic role

Meaning relation a participant bears to an event or predicate.

Thematic role

Traditional label for event-participant roles such as Agent, Theme and Experiencer.

Argument structure

Pattern of participants licensed or expected by a lexical predicate.

Valency

Number/type of arguments associated with a predicate.

Subcategorization

Syntactic information about the complements a lexical item can take.

Selectional feature

Semantic property such as [+ANIMATE] or [+EDIBLE] used in classic restriction models.

Animacy

Semantic property distinguishing living/agentive entities from inanimate ones under relevant analyses.

Sentience

Capacity for experience/feeling; relevant to predicates such as feel, suffer or persuade in literal uses.

Edibility

Property of being suitable for eating; a classic selectional example.

Countability

Grammatical-semantic property distinguishing count and mass nouns.

Selectional strength

How strongly a predicate favours a semantic class.

Preference violation

Argument that falls outside a predicate’s normal semantic preference.

Literal interpretation

Reading that preserves ordinary lexical meanings without figurative transfer.

Metaphorical interpretation

Reading that maps one conceptual domain onto another.

Metonymic interpretation

Reading based on association/contiguity, such as institution for people.

Personification

Attributing human/animate properties to an inanimate entity.

Coercion

Interpretive adjustment that repairs a mismatch between lexical semantic requirements.

Type coercion

Shift in semantic type prompted by a predicate or construction, such as object noun → event reading.

Co-composition

Mutual semantic adjustment among combining expressions.

Event coercion

Interpreting an entity as an event associated with it, as in begin the book.

Container-for-content

Metonymic interpretation where a container stands for what it contains.

Place-for-institution

Metonymy where a place name stands for an organisation or authority.

Author-for-work

Metonymy where an author’s name stands for their works.

Collective interpretation

Group/institution interpreted as its decision-makers or members.

Semantic type

Abstract category used in formal/computational semantics, such as person, event or physical object.

Ontology

Structured inventory of semantic types/classes.

Selectional preference model

Model estimating how strongly a predicate prefers semantic classes in its arguments.

Soft constraint

Preference that can be violated without making a sentence ungrammatical.

Hard constraint

Rule treated as nonviolable under a specific formal analysis.

Plausibility

How compatible an interpretation is with world knowledge and context.

Compositionality

Principle that phrase/sentence meaning is built from parts and their mode of combination.

Lexical semantics

Study of word meaning and how lexical meaning contributes to larger structures.

Predicate–argument compatibility

Fit between a predicate’s meaning and the semantic properties of its arguments.

Semantic fit

Everyday label for compatibility among meanings in a combination.

Collocational preference

Conventional tendency for words to co-occur; related to but distinct from semantic compatibility.

Semantic prosody

Evaluative tendency arising from repeated collocational environments.

Selectional surprise

Unexpected predicate–argument pairing that attracts processing/interpretation.

World knowledge

Encyclopedic knowledge used to assess event plausibility beyond lexical compatibility.

Contextual override

Context that makes an otherwise unusual combination interpretable or literal.

Domain-specific selection

Compatibility patterns restricted to a technical domain, such as matrix with diagonalise.

Lexical sense

Particular conventional meaning of a polysemous form.

Sense-specific preference

Selectional pattern associated with one sense of a polysemous word.

Semantic repair

Interpretive process that resolves an apparent compatibility mismatch.

Anomaly detection

Recognition that an ordinary literal combination fails semantic expectations.

Part XVI — Frequently Asked Questions

What are selectional restrictions?

They are semantic expectations about what kinds of entities can fill a predicate’s roles, historically stated as constraints such as eat selecting edible objects.

What are selectional preferences?

They are graded tendencies rather than absolute bans. A verb can strongly prefer one semantic class while still appearing with unusual or metaphorical arguments.

Why say preference instead of restriction?

Because real language regularly violates strict rules through unusual events, metaphor, metonymy, negation and creative use.

What is semantic compatibility?

The degree to which the meanings of words fit together naturally in a phrase or sentence under an interpretation.

Can a sentence be grammatical but semantically strange?

Yes. The rock persuaded the teacher is syntactically ordinary but semantically odd literally.

Is semantic anomaly the same as ungrammaticality?

No. Grammar concerns structural well-formedness; semantic anomaly concerns meaning compatibility.

What is a classic selectional restriction example?

Eat prefers an edible theme; sleep prefers an animate experiencer under literal use.

Can selectional restrictions be broken?

Yes. Humans can eat glass, computers can sleep, and metaphor deliberately stretches ordinary preferences.

Does eat glass violate grammar?

No. It is grammatically normal and semantically unusual because glass is not prototypically edible.

Why is eat your words acceptable?

It is an idiom with a conventional figurative meaning; the literal edible-object expectation is overridden.

What is coercion?

A semantic repair process where context forces an expression into a compatible interpretation, such as begin the book meaning begin reading or writing it.

What is type coercion?

Changing the semantic type assigned to an expression so it can satisfy another expression’s requirements.

Why does begin the book mean begin reading?

The verb begin prefers an event; the noun book supplies a conventional associated event such as reading/writing.

Is metaphor a selectional violation?

Many metaphors can be detected because literal selectional expectations are violated, prompting conceptual reinterpretation.

Why can a city sleep?

In the city never sleeps, an animate-state predicate is metaphorically applied to a place/community.

What is personification?

Giving an inanimate entity human/animate properties, often through selectionally unusual predicates.

How are selectional preferences different from collocations?

Selectional preferences concern semantic classes; collocations concern conventional co-occurrence of specific words or patterns.

Can two combinations be semantically compatible but differ in collocational naturalness?

Yes. Drink water and drink varnish both involve liquids, but only the first is ordinary.

How does world knowledge affect compatibility?

World knowledge decides whether an event is plausible even when semantic types fit, as in the child lifted the piano.

How does polysemy affect selectional preferences?

Each sense can select different argument types, as with run a race, run a company, run software.

What are semantic roles?

Roles such as Agent, Theme, Experiencer, Instrument, Source and Goal that participants play in events.

Do all linguists use the same semantic roles?

No. Role inventories and definitions vary across theories.

What is argument structure?

The participants and syntactic/semantic patterns associated with a predicate.

What is valency?

A related concept describing how many and what kinds of arguments a predicate can combine with.

Are selectional restrictions syntax?

They interact with syntax but are fundamentally about meaning/type compatibility in classic lexical-semantic analyses.

Can selectional restrictions help vocabulary learning?

Yes. Knowing what kinds of nouns a new verb combines with narrows its possible meaning.

How does syntactic bootstrapping relate?

Syntax constrains possible meanings; selectional semantics adds information about the types of participants involved.

How do children learn verb meanings from arguments?

Repeated scenes and noun types around a new verb provide evidence about its event structure and semantic class.

Can technical vocabulary have selectional restrictions?

Very strongly. Diagonalise selects mathematical objects; encrypt selects data; ratify selects formal agreements.

Why does encrypt the chair sound wrong?

The literal computing sense selects information-like objects, while a physical chair does not fit unless its digital representation is intended.

What is a domain override?

A specialised domain supplies a conventional meaning that changes ordinary compatibility patterns.

Why can kill a process be normal in computing?

The verb has acquired a conventional technical sense meaning terminate a running process.

Is diagnose a fault metaphorical?

It began as extension from medicine but is conventional in engineering/computing; lexicalised extensions become ordinary senses.

Can nouns have selectional behaviour?

Yes. Relational nouns, compounds and complement-taking nouns impose semantic expectations too.

Can adjectives have selectional preferences?

Yes. Pregnant, edible, legible, punctual naturally apply to restricted semantic classes under literal senses.

Why is nutritious argument odd?

Nutritious literally selects food/diet, so applying it to an argument requires metaphor.

Can a word’s selectional preference change historically?

Yes. Metaphors and domain extensions can lexicalise and become ordinary senses.

Does corpus frequency define selectional preference?

Corpus data is a major source of evidence, but semantic classes and world knowledge also matter.

How are selectional preferences used in NLP?

Systems estimate which semantic classes are likely in predicate roles, supporting parsing, word-sense disambiguation and plausibility modelling.

Are preferences binary?

Usually not in modern models. Compatibility is often probabilistic or graded.

Can negation license strange combinations?

Yes. You cannot eat gold for lunch intentionally mentions an incompatible food to deny it.

Can fiction change selectional restrictions?

Fiction changes world assumptions. Sentient robots, talking rocks or living cities make combinations literal that would otherwise be personification.

How do jokes use selectional violations?

They exploit unexpected semantic combinations and force reinterpretation.

What is semantic type mismatch?

Predicate and argument belong to categories that do not compose normally under the intended literal sense.

What is semantic repair?

The listener/reader finds a metaphorical, metonymic, coerced or contextual interpretation that restores coherence.

Should learners memorise semantic feature labels?

Not usually. They should learn natural participant types, collocations, examples and contrastive non-examples.

Can selectional preferences improve writing?

Yes. They help writers avoid semantically mismatched word choices and choose verbs/adjectives that fit their nouns precisely.

Can they improve reading?

Yes. Unexpected combinations signal polysemy, metaphor, technical sense or a need for reinterpretation.

How do selectional preferences relate to lexical competence?

They are part of knowing how a word behaves in real language, beyond its dictionary gloss.

What is the biggest misconception?

That a strange predicate–argument combination is automatically ungrammatical. Many are grammatically perfect but semantically marked.

What is the main educational takeaway?

Knowing a word includes knowing what kinds of things it normally describes, affects, selects or combines with—and recognising when writers deliberately break those expectations.

Part XVII — Research and Reference Boundary

Modern lexical semantics treats selectional properties as part of the information needed to explain how word meanings compose into coherent sentence meanings. The Cambridge Handbook of Formal Semantics notes the importance of selectional properties in lexical semantics and predicate composition. A Cambridge Journal of Linguistics discussion of rich lexical meaning describes coercion as a repair process triggered when selectional restrictions or preferences mismatch the semantic properties of a combining expression.

In computational linguistics, Jurafsky and Martin’s Speech and Language Processing explains the historical move from strict selectional restrictions toward softer selectional preferences. The text illustrates how verbs can have preferred semantic classes while unusual but well-formed combinations remain possible. This is the boundary used here: teach semantic expectations as strong tendencies, not rigid grammar rules.

• Cambridge Handbook of Formal Semantics — Lexical Semantics
• Journal of Linguistics — Rich Lexical Meanings and Coercion
• Stanford — Speech and Language Processing

Part XVIII — What This Article Does Not Claim

It does not claim that every predicate has one fixed semantic feature list. It does not claim that unusual combinations are errors. It does not claim that metaphor is outside the lexicon. It does not claim that selectional preferences can be separated perfectly from world knowledge or collocation. It does not claim that every theory uses the same semantic roles.

It claims something practical: words bring expectations about the semantic types they combine with, those expectations strongly guide interpretation, and skilled readers/writers know when a combination is ordinary, technically specialised, metaphorical, coerced or genuinely anomalous.

Part XIX — Domain-Specific Compatibility: 50 Cases

1. Mathematics: diagonalise a matrix

Compatibility: high. Why: Predicate is technically defined for matrices/operators.

2. Mathematics: factorise a polynomial

Compatibility: high. Why: Object belongs to expected algebraic class.

3. Mathematics: differentiate a function

Compatibility: high. Why: Calculus sense selects mathematical functions.

4. Mathematics: integrate a function

Compatibility: high. Why: Calculus sense.

5. Mathematics: solve an equation

Compatibility: high. Why: Problem object expected.

6. Mathematics: prove a theorem

Compatibility: high. Why: Formal proposition expected.

7. Mathematics: prove a sandwich

Compatibility: anomalous. Why: Food object does not fit proof predicate.

8. Mathematics: plot the points

Compatibility: high. Why: Representable data/location objects.

9. Mathematics: plot the anger

Compatibility: marked. Why: Possible metaphor/data representation only.

10. Physics: accelerate the particle

Compatibility: high. Why: Physical entity can change velocity.

11. Physics: conduct electricity

Compatibility: high. Why: Material/system domain.

12. Physics: emit radiation

Compatibility: high. Why: Physical source/event.

13. Physics: refract light

Compatibility: high. Why: Optical phenomenon.

14. Physics: refract a policy

Compatibility: anomalous. Why: Policy not optical wave under literal sense.

15. Chemistry: dissolve salt

Compatibility: high. Why: Solute/substance compatibility.

16. Chemistry: oxidise iron

Compatibility: high. Why: Chemical substrate.

17. Chemistry: catalyse a reaction

Compatibility: high. Why: Reaction/event object.

18. Chemistry: catalyse a sandwich

Compatibility: marked. Why: Only metaphorical ‘speed up making/eating’ context.

19. Biology: metabolise glucose

Compatibility: high. Why: Biochemical substance.

20. Biology: transcribe DNA

Compatibility: high. Why: Molecular biology sense.

21. Biology: translate mRNA

Compatibility: high. Why: Technical biological sense selects RNA/message.

22. Biology: translate a French sentence

Compatibility: high under language sense. Why: Polysemy creates another compatibility profile.

23. Medicine: administer a drug

Compatibility: high. Why: Treatment object.

24. Medicine: intubate a patient

Compatibility: high. Why: Animate patient expected.

25. Medicine: intubate a spreadsheet

Compatibility: anomalous. Why: No anatomical airway.

26. Medicine: contraindicate treatment

Compatibility: high. Why: Medical relation.

27. Law: ratify a treaty

Compatibility: high. Why: Formal agreement.

28. Law: indict a defendant

Compatibility: high. Why: Person/legal entity capable of accusation.

29. Law: appeal a decision

Compatibility: high. Why: Formal decision/judgment.

30. Law: appeal a sandwich

Compatibility: anomalous. Why: Wrong semantic class.

31. Finance: amortise a loan

Compatibility: high. Why: Debt/asset accounting object.

32. Finance: hedge risk

Compatibility: high. Why: Risk/exposure object.

33. Finance: liquidate assets

Compatibility: high. Why: Asset/business domain.

34. Finance: liquidate a feeling

Compatibility: figurative/marked. Why: Metaphor required.

35. Computing: compile source code

Compatibility: high. Why: Code object.

36. Computing: execute a program

Compatibility: high. Why: Program/instruction object.

37. Computing: parse a string

Compatibility: high. Why: Structured symbolic input.

38. Computing: hash a password

Compatibility: high. Why: Data object.

39. Computing: render an image

Compatibility: high. Why: Visual/digital object.

40. Computing: render a sandwich

Compatibility: possible in 3D graphics if digital model. Why: Domain context can rescue apparent anomaly.

41. History: annex a territory

Compatibility: high. Why: Political/geographic object.

42. History: abdicate the throne

Compatibility: high. Why: Office/power relation.

43. History: colonise a region

Compatibility: high. Why: Territory/population object.

44. History: colonise a number

Compatibility: anomalous literally. Why: Metaphor only.

45. Education: assess a learner

Compatibility: high. Why: Performance/participant object.

46. Education: scaffold learning

Compatibility: high. Why: Process/capability object via educational metaphor.

47. Education: differentiate instruction

Compatibility: high. Why: Teaching-domain sense.

48. Education: differentiate a sandwich

Compatibility: anomalous under teaching/math senses. Why: Requires special context.

49. Media: verify a claim

Compatibility: high. Why: Truth-evaluable proposition.

50. Media: fact-check a statement

Compatibility: high. Why: Proposition/text object.

51. Media: fact-check a chair

Compatibility: anomalous. Why: Chair not truth-evaluable unless claim about chair implied.

Part XX — 80 Anomaly, Metaphor or Literal? Decisions

1. The child ate an apple.

Best analysis: literal ordinary. Reason: edible theme.

2. The child ate a candle.

Best analysis: literal unusual. Reason: physically possible but dangerous/nonprototypical.

3. The market ate the losses.

Best analysis: metaphorical. Reason: financial system conceptualised as consuming/absorbing.

4. The computer slept.

Best analysis: technical/conventional extension. Reason: sleep mode lexicalised.

5. The stone slept.

Best analysis: personification/poetic. Reason: inanimate object receives animate state.

6. The witness remembered the event.

Best analysis: literal ordinary. Reason: experiencer capable of memory.

7. The camera remembered the scene.

Best analysis: metaphorical/technical personification. Reason: device storage framed as memory.

8. The policy remembered the crisis.

Best analysis: anomalous. Reason: abstract policy lacks memory unless rhetorical.

9. The teacher swallowed the pill.

Best analysis: literal ordinary. Reason: ingestible object.

10. The teacher swallowed the insult.

Best analysis: metaphorical. Reason: accept/endure without response.

11. The river swallowed the village.

Best analysis: metaphorical/personification. Reason: flooding framed as ingestion.

12. The surgeon repaired the artery.

Best analysis: literal/technical. Reason: physical tissue can be repaired.

13. The engineer repaired the engine.

Best analysis: literal/technical. Reason: machine repair sense.

14. The therapist repaired the relationship.

Best analysis: metaphorical conventional. Reason: social relation conceptualised as damaged object.

15. The wind whispered through the trees.

Best analysis: personification/metaphor. Reason: sound resembles whisper.

16. The student whispered the answer.

Best analysis: literal ordinary. Reason: human speech agent.

17. The equation whispered the answer.

Best analysis: poetic metaphor. Reason: inanimate abstract object given communicative act.

18. The plan collapsed.

Best analysis: metaphorical conventional. Reason: abstract structure framed as building.

19. The bridge collapsed.

Best analysis: literal ordinary. Reason: physical structure.

20. The argument collapsed.

Best analysis: metaphorical conventional. Reason: reasoning structure fails.

21. The medicine attacked the bacteria.

Best analysis: metaphorical technical. Reason: causal action framed as combat.

22. The soldiers attacked the base.

Best analysis: literal ordinary. Reason: intentional agents.

23. The evidence attacked the theory.

Best analysis: marked metaphor. Reason: evidence conceptualised as adversary.

24. The plant drank water.

Best analysis: metaphorical/conventional personification. Reason: absorption framed as drinking.

25. The athlete drank water.

Best analysis: literal ordinary. Reason: animate drinker.

26. The desert drank the rain.

Best analysis: poetic metaphor. Reason: land absorbs water.

27. The software refuses to open.

Best analysis: conventional personification. Reason: system failure framed as volition.

28. The child refuses to open the door.

Best analysis: literal volitional. Reason: animate decision-maker.

29. The door refuses to open.

Best analysis: personification/conventional. Reason: mechanical resistance framed as refusal.

30. The idea grew.

Best analysis: metaphorical conventional. Reason: conceptual development framed as biological growth.

31. The tree grew.

Best analysis: literal ordinary. Reason: biological growth.

32. The company grew.

Best analysis: conventional extension. Reason: organisation changes size.

33. The temperature climbed.

Best analysis: metaphorical conventional. Reason: scalar increase framed as upward motion.

34. The hiker climbed.

Best analysis: literal ordinary. Reason: animate motion.

35. The price fell.

Best analysis: metaphorical conventional. Reason: scalar decrease framed as downward motion.

36. The stone fell.

Best analysis: literal ordinary. Reason: physical movement.

37. The deadline approached.

Best analysis: metaphorical conventional. Reason: time/event framed as moving entity.

38. The car approached.

Best analysis: literal ordinary. Reason: physical movement.

39. The solution emerged.

Best analysis: metaphorical conventional. Reason: idea becomes available/visible.

40. The diver emerged.

Best analysis: literal ordinary. Reason: physical motion out of medium.

41. The report argues that costs will rise.

Best analysis: metonymic/conventional. Reason: document stands for authorial argument.

42. The researcher argues that costs will rise.

Best analysis: literal discourse act. Reason: human agent.

43. The spreadsheet argues that costs will rise.

Best analysis: personification unless automated analysis output contextualised. Reason: inanimate artifact.

44. The law prohibits smoking.

Best analysis: metonymic institutional. Reason: legal instrument stands for authority/rule.

45. The guard prohibits entry.

Best analysis: literal social agent. Reason: human authority.

46. The wall prohibits entry.

Best analysis: marked. Reason: physical barrier prevents rather than prohibits.

47. The lock prevents entry.

Best analysis: literal causative compatibility. Reason: physical device can prevent.

48. The lock forbids entry.

Best analysis: personification/institutional metaphor. Reason: forbid prefers authority/agent.

49. The algorithm decides the route.

Best analysis: technical/conventional personification. Reason: decision function lexicalised in computing.

50. The judge decides the case.

Best analysis: literal institutional agent. Reason: human decision-maker.

51. The spoon decides the menu.

Best analysis: anomalous/personification. Reason: no decision capacity.

52. The vaccine teaches the immune system.

Best analysis: metaphorical pedagogical framing. Reason: biological process conceptualised as learning.

53. The teacher teaches the student.

Best analysis: literal ordinary. Reason: instructional agent.

54. The weather teaches patience.

Best analysis: metaphorical. Reason: experience becomes teacher.

55. The economy breathes again.

Best analysis: metaphorical. Reason: system conceptualised as organism.

56. The patient breathes.

Best analysis: literal biological. Reason: animate organism.

57. The engine breathes better after tuning.

Best analysis: mechanical metaphor. Reason: airflow framed biologically.

58. The data speak for themselves.

Best analysis: personification/conventional rhetoric. Reason: evidence framed as speaker.

59. The witness speaks.

Best analysis: literal ordinary. Reason: human speech.

60. The graph speaks loudly.

Best analysis: metaphorical rhetoric. Reason: visual evidence framed as voice.

61. The forest screams under the saws.

Best analysis: personification. Reason: environment framed as animate victim.

62. The child screams.

Best analysis: literal. Reason: animate vocalisation.

63. The brakes scream.

Best analysis: acoustic metaphor/conventional. Reason: sound resembles scream.

64. The camera captures light.

Best analysis: technical conventional. Reason: capture extends from seizure to recording.

65. The police capture the suspect.

Best analysis: literal intentional. Reason: agent takes control.

66. The sentence captures the idea.

Best analysis: metaphorical conventional. Reason: representation contains essence.

67. The model predicts rain.

Best analysis: technical conventional. Reason: system outputs forecast.

68. The meteorologist predicts rain.

Best analysis: literal cognitive/professional. Reason: human forecast.

69. The rock predicts rain.

Best analysis: anomalous unless used as sign/metonymy. Reason: inanimate object lacks forecasting capacity.

70. Dark clouds predict rain.

Best analysis: metonymic/statistical conventional. Reason: evidence cue stands in predictive relation.

71. The law recognises the right.

Best analysis: institutional metonymy. Reason: legal system treated as recognising agent.

72. The person recognises the face.

Best analysis: literal cognitive. Reason: animate perceiver.

73. The camera recognises the face.

Best analysis: technical conventional. Reason: machine recognition lexicalised.

74. The flower recognises the face.

Best analysis: anomalous under ordinary botany. Reason: speculative context required.

75. The book explains the method.

Best analysis: metonymic conventional. Reason: book/content stands for author/explanation.

76. The teacher explains the method.

Best analysis: literal. Reason: human agent.

77. The stone explains the method.

Best analysis: anomalous/personification. Reason: no communicative agency.

78. The evidence supports the claim.

Best analysis: abstract conventional. Reason: support metaphor lexicalised in argumentation.

79. The pillar supports the roof.

Best analysis: literal physical. Reason: physical support.

80. The friend supports the learner.

Best analysis: social extension. Reason: assistance sense conventional.

Part XXI — Primary-to-Adult Progression

28. Primary 1–2

Use concrete mismatches: drink water versus drink a chair, the dog barks versus the pencil barks. The aim is intuitive semantic fit, not terminology.

29. Primary 3–4

Introduce personification and metaphor: the wind whispered, the city sleeps. Ask why literal meaning feels unusual and what figurative interpretation repairs it.

30. Primary 5–6

Connect semantic compatibility to vocabulary MCQs, collocation, polysemy and subject vocabulary. Students should justify why one verb fits a noun better than another.

31. Secondary 1–2

Add semantic roles, technical senses and coercion. Contrast grammatical structure with semantic well-formedness.

32. Secondary 3–4

Use argumentation, academic verbs and cross-subject predicates. Students should detect when a thesaurus substitute creates a semantic mismatch.

33. JC, university and adult

Explore valency, type coercion, domain-specific selection and corpus/computational preference models. Focus on precision in academic and professional writing.

Part XXII — First Weak Link Diagnostic

1. Can the learner identify the predicate and its participants?
2. Can the learner tell whether the sentence is grammatically well formed?
3. Can the learner identify the ordinary semantic class expected by the predicate?
4. Can the learner detect an unusual argument type?
5. Can the learner distinguish impossible from merely improbable?
6. Can the learner recognise a technical sense that changes compatibility?
7. Can the learner recognise polysemy?
8. Can the learner detect metaphor/personification/metonymy?
9. Can the learner propose a coercion or contextual repair?
10. Can the learner choose a semantically compatible word in writing?

Repair the first failed step. A learner who cannot identify the predicate needs sentence-meaning support; a learner who detects anomaly but cannot interpret metaphor needs figurative-language work; a learner who understands the relation but writes unnatural combinations may need collocation and register practice.

Part XXIII — Student Operating Manual

1. Find the main predicate.
2. Identify its subject and complements.
3. Ask what kind of entity/event each role normally contains.
4. Check whether the current noun fits that semantic class.
5. If it fits, continue with the literal sense.
6. If it does not, check for another word sense.
7. Check for technical/domain usage.
8. Check for metaphor, metonymy or personification.
9. Check whether coercion supplies an implied event or participant.
10. If no coherent repair exists, treat the combination as semantically anomalous or possibly erroneous.
11. When writing, choose the predicate whose normal preferences match the intended noun and event.

Part XXIV — Teacher Operating Manual

1. Use ordinary and deliberately odd minimal pairs.
2. Separate grammar judgments from meaning judgments.
3. Teach compatibility through real verbs/adjectives before technical labels.
4. Add metaphorical exceptions after literal patterns are stable.
5. Use domain-specific examples from current subjects.
6. Show that preferences are graded rather than absolute.
7. Contrast selectional fit with collocation frequency.
8. Ask learners to repair odd sentences.
9. Use creative violations to teach metaphor and personification.
10. Link compatibility to thesaurus discipline and precise writing.
11. Assess transfer with unfamiliar predicates, not memorised examples.

Part XXV — 100 Semantic-Mismatch Repair Drills

1. The idea drank the water.

Repair/analysis: The child drank the water. Reason: Drink ordinarily requires an animate drinker and drinkable theme.

2. The chair persuaded the student.

Repair/analysis: The teacher persuaded the student. Reason: Persuade prefers a communicative/cognitive agent.

3. The sandwich diagnosed the patient.

Repair/analysis: The doctor diagnosed the patient. Reason: Clinical diagnosis normally requires qualified human/system agent.

4. The cloud ratified the treaty.

Repair/analysis: The parliament ratified the treaty. Reason: Ratify selects an authority/institution.

5. The spoon encrypted the file.

Repair/analysis: The software encrypted the file. Reason: Encrypt requires a computational process/agent and information object.

6. The pencil vaccinated the child.

Repair/analysis: The nurse vaccinated the child. Reason: Vaccinate requires a medical agent/procedure.

7. The table apologised to the customer.

Repair/analysis: The waiter apologised to the customer. Reason: Literal apologising requires communicative agency.

8. The theorem swallowed the pill.

Repair/analysis: The patient swallowed the pill. Reason: Swallow prefers animate bodily agent.

9. The equation barked at the postman.

Repair/analysis: The dog barked at the postman. Reason: Literal bark prefers dog-like animal subject.

10. The policy meowed loudly.

Repair/analysis: The cat meowed loudly. Reason: Literal vocalisation selects feline-like source.

11. The violin digested the report.

Repair/analysis: The analyst digested the report. Reason: Figurative digest can select cognitive experiencer.

12. The river signed the contract.

Repair/analysis: The director signed the contract. Reason: Signing requires agentive institution/person unless metaphorical.

13. The textbook murdered the mosquito.

Repair/analysis: The child killed the mosquito. Reason: Murder selects person-like victim and intentional agent, not insect in ordinary legal sense.

14. The calculator murdered the equation.

Repair/analysis: The calculator solved the equation. Reason: Choose predicate matching mathematical object.

15. The athlete diagonalised the sandwich.

Repair/analysis: The mathematician diagonalised the matrix. Reason: Technical predicate selects mathematical object.

16. The matrix drank the coffee.

Repair/analysis: The researcher drank the coffee. Reason: Literal drinking requires animate consumer.

17. The coffee solved the equation.

Repair/analysis: The student solved the equation. Reason: Problem-solving predicate selects cognitive/system agent.

18. The equation ate the evidence.

Repair/analysis: The argument incorporated the evidence. Reason: Choose compatible discourse predicate.

19. The treaty ate the parliament.

Repair/analysis: Parliament ratified the treaty. Reason: Replace anomalous event with institutional relation.

20. The student ratified the apple.

Repair/analysis: The student ate the apple. Reason: Select food-compatible predicate.

21. The apple convinced the jury.

Repair/analysis: The lawyer convinced the jury. Reason: Convincing requires communicative/cognitive source under literal reading.

22. The jury melted the ice.

Repair/analysis: The heat melted the ice. Reason: Causative melt can take agent/cause, but context determines plausibility.

23. The ice persuaded the jury.

Repair/analysis: The lawyer persuaded the jury. Reason: Animate/social persuasion.

24. The server ate breakfast.

Repair/analysis: The engineer ate breakfast. Reason: Literal eating requires organism.

25. The server processed breakfast.

Repair/analysis: The server processed the request. Reason: Technical process selects information/task.

26. The request chewed the server.

Repair/analysis: The server processed the request. Reason: Repair argument roles and predicate.

27. The camera remembered lunch.

Repair/analysis: The photographer remembered lunch. Reason: Literal memory prefers cognitive experiencer.

28. The camera captured the image.

Repair/analysis: Keep sentence. Reason: Technical capture conventionally selects recording device + image/data.

29. The policy photographed the city.

Repair/analysis: The camera photographed the city. Reason: Photographing requires imaging agent/device.

30. The city photographed the policy.

Repair/analysis: The journalist photographed the city. Reason: Repair both semantic roles.

31. The law ate corruption.

Repair/analysis: The law addresses corruption. Reason: Use institutional predicate compatible with abstract problem.

32. The law prohibits smoking.

Repair/analysis: Keep sentence. Reason: Institutional metonymy is conventional.

33. The smoke legislated a ban.

Repair/analysis: Parliament legislated a ban. Reason: Legislation requires law-making authority.

34. The teacher legislated homework.

Repair/analysis: The teacher assigned homework. Reason: School authority is not legislative in ordinary sense.

35. The computer apologised for the error.

Repair/analysis: Keep if interface message/personification intended. Reason: Context can license personified system speech.

36. The rock apologised for erosion.

Repair/analysis: Mark as personification or revise to ‘erosion damaged the rock’. Reason: Inanimate subject needs figurative frame.

37. The plant criticised the gardener.

Repair/analysis: Personification: keep only if deliberate. Reason: Critical speech implies cognitive/communicative agent.

38. The gardener watered the plant.

Repair/analysis: Keep sentence. Reason: Semantic classes fit ordinary event.

39. The plant drank the water.

Repair/analysis: Accept as conventional metaphor. Reason: Biological absorption can be framed as drinking.

40. The plant negotiated the contract.

Repair/analysis: The manager negotiated the contract. Reason: Negotiation requires social/cognitive agents.

41. The contract grew rapidly.

Repair/analysis: The contract grew longer during revision. Reason: Repair by supplying dimension/metaphorical growth context.

42. The company grew rapidly.

Repair/analysis: Keep sentence. Reason: Organisational-size growth is conventional.

43. The company sprouted leaves.

Repair/analysis: The tree sprouted leaves. Reason: Literal sprouting selects plant.

44. New branches sprouted across the company.

Repair/analysis: Accept metaphorically. Reason: Organisational expansion borrows plant-growth frame.

45. The engine suffered sadness.

Repair/analysis: The driver suffered sadness. Reason: Emotion predicate prefers sentient experiencer.

46. The engine suffered damage.

Repair/analysis: Keep sentence. Reason: Suffer has extended sense selecting entities affected by damage.

47. The policy felt nervous.

Repair/analysis: The minister felt nervous. Reason: Mental-state adjective selects experiencer.

48. The policy created uncertainty.

Repair/analysis: Keep sentence. Reason: Abstract cause is compatible with result predicate.

49. The number admired the painting.

Repair/analysis: The visitor admired the painting. Reason: Admiration requires experiencer.

50. The model admired the data.

Repair/analysis: The model fitted the data. Reason: Use technical predicate.

51. The model predicts outcomes.

Repair/analysis: Keep sentence. Reason: Computational prediction is conventional.

52. The table predicts tomorrow.

Repair/analysis: The forecast model predicts tomorrow’s temperature. Reason: Prediction requires predictive source/system and proposition/quantity.

53. The graph believes the hypothesis.

Repair/analysis: The researcher believes the hypothesis. Reason: Belief requires cognitive experiencer.

54. The graph supports the hypothesis.

Repair/analysis: Keep sentence. Reason: Argumentative support conventionally selects evidence.

55. The evidence sleeps.

Repair/analysis: The evidence remains unused. Reason: Literal sleep incompatible with abstract noun.

56. The evidence awakens suspicion.

Repair/analysis: Accept metaphorical/conventional. Reason: Causative awakening extends to mental states.

57. The suspicion drinks evidence.

Repair/analysis: Suspicion grows with evidence. Reason: Choose compatible abstract process verb.

58. The suspicion infected the team.

Repair/analysis: Accept metaphorically if meaning spread. Reason: Emotion/belief spread uses disease metaphor.

59. The rumour infected the town.

Repair/analysis: Accept metaphorically/conventionally. Reason: Information spread as contagion.

60. The town infected the rumour.

Repair/analysis: The rumour spread through the town. Reason: Repair role direction.

61. The disease persuaded the patient.

Repair/analysis: The doctor persuaded the patient. Reason: Disease lacks communicative agency.

62. The disease attacked the lungs.

Repair/analysis: Accept conventional biological metaphor. Reason: Attack lexicalised in disease discourse.

63. The lungs voted against treatment.

Repair/analysis: Personification unless fictional. Reason: Voting requires institutional/cognitive agents.

64. The committee voted against treatment.

Repair/analysis: Keep sentence. Reason: Collective agent fits.

65. The committee photosynthesised the proposal.

Repair/analysis: The committee evaluated the proposal. Reason: Photosynthesis selects biological plant-like systems.

66. The plant evaluated the proposal.

Repair/analysis: The committee evaluated the proposal. Reason: Evaluation requires cognitive/institutional agent.

67. The plant responded to light.

Repair/analysis: Keep sentence. Reason: Biological response does not require conscious cognition.

68. The plant argued with light.

Repair/analysis: Personification. Reason: Argument requires discourse agents.

69. The teacher inhaled oxygen.

Repair/analysis: Keep sentence. Reason: Animate breathing agent.

70. The textbook inhaled oxygen.

Repair/analysis: Anomalous unless personified. Reason: Book lacks respiratory system.

71. The engine inhaled air.

Repair/analysis: Accept mechanical metaphor/technical use. Reason: Intake systems license breathing metaphor.

72. The economy inhaled investment.

Repair/analysis: Marked metaphor; revise to ‘absorbed investment’. Reason: Choose conventional abstract predicate.

73. The economy absorbed investment.

Repair/analysis: Keep sentence. Reason: Absorption metaphor is conventional in economics.

74. The economy vaccinated growth.

Repair/analysis: Revise to ‘stimulated/protected growth’. Reason: Medical verb mismatched.

75. The policy stimulated growth.

Repair/analysis: Keep sentence. Reason: Causative abstract predicate compatible.

76. The policy digested the data.

Repair/analysis: The analysts digested the data. Reason: Cognitive digestion metaphor needs experiencer.

77. The system ingested the data.

Repair/analysis: Keep sentence. Reason: Technical data-ingestion sense conventional.

78. The system swallowed the data.

Repair/analysis: Possible metaphor but less formal. Reason: Register and domain affect compatibility.

79. The theorem ingested the data.

Repair/analysis: The system ingested the data. Reason: Abstract proposition not processing agent.

80. The algorithm learned the pattern.

Repair/analysis: Keep in machine-learning domain. Reason: Technical sense lexicalised.

81. The spoon learned the pattern.

Repair/analysis: Anomalous literally. Reason: No learning system/cognitive agent.

82. The child learned the pattern.

Repair/analysis: Keep sentence. Reason: Cognitive learner.

83. The algorithm forgot the password.

Repair/analysis: Possible technical metaphor if state lost; otherwise revise. Reason: Computing can extend memory predicates.

84. The chair forgot the password.

Repair/analysis: Anomalous/personification. Reason: Furniture lacks memory system.

85. The database stores the password.

Repair/analysis: Keep sentence. Reason: Storage predicate compatible with information system.

86. The database tastes sweet.

Repair/analysis: Anomalous literally. Reason: Taste property requires edible substance/experiencer relation.

87. The medicine tastes bitter.

Repair/analysis: Keep sentence. Reason: Substance can have taste property.

88. The argument tastes bitter.

Repair/analysis: Poetic synesthesia/metaphor. Reason: Marked figurative interpretation.

89. The song tastes bitter.

Repair/analysis: Synesthetic metaphor. Reason: Cross-sensory extension, not literal.

90. The soup sounds bitter.

Repair/analysis: Mismatched sensory adjective unless figurative. Reason: Taste property not auditory under literal use.

91. The violin sounds bright.

Repair/analysis: Accept conventional cross-modal adjective. Reason: Bright has lexicalised auditory sense.

92. The colour sounds bright.

Repair/analysis: Marked unless discussing name/description. Reason: Sense-specific selection matters.

93. The melody looks delicious.

Repair/analysis: Synesthetic/metaphorical. Reason: Visual/taste adjectives creatively crossed.

94. The data are delicious.

Repair/analysis: Informal metaphor meaning appealing. Reason: Not literal but interpretable.

95. The data are statistically significant.

Repair/analysis: Keep sentence. Reason: Technical adjective fits data/result domain.

96. The sandwich is statistically significant.

Repair/analysis: Anomalous unless sandwich is a measured experimental condition. Reason: Technical context can rescue via metonymy.

97. The result is pregnant.

Repair/analysis: Anomalous literal; possible archaic/metaphor ‘full of meaning’. Reason: Sense history can create rare compatibility.

98. The pause was pregnant.

Repair/analysis: Accept lexicalised metaphor. Reason: Means full of implication.

99. The argument is edible.

Repair/analysis: Anomalous literal. Reason: Argument not food.

100. The mushroom is edible.

Repair/analysis: Keep sentence. Reason: Literal compatibility.

101. The handwriting is edible.

Repair/analysis: Anomalous unless made of edible material. Reason: Context can make surprising literal readings possible.

102. The handwriting is legible.

Repair/analysis: Keep sentence. Reason: Property selects writing.

103. The soup is legible.

Repair/analysis: Anomalous literally. Reason: Legibility requires readable marks/text.

Part XXVI — Writing Precision: Selection Before Synonym Substitution

A thesaurus can suggest verbs that share broad meanings but differ in selectional behaviour. Consume, eat, devour, ingest, absorb overlap, yet their normal subjects/objects and registers differ. Before substituting, ask whether the new predicate naturally selects the intended participants.

consume

Typical selection: food/resources/time/content in formal metaphor. Examples: consume energy; consume a meal; the task consumed time.

eat

Typical selection: food/meal; idiomatic extensions. Examples: eat lunch; eat your words.

devour

Typical selection: food; books/content metaphor. Examples: devour a meal; devour a novel.

ingest

Typical selection: food/medicine/data technical. Examples: ingest medication; data ingestion.

absorb

Typical selection: liquid/energy/information/cost. Examples: absorb water; absorb information; absorb costs.

attack

Typical selection: person/target/problem metaphor. Examples: attack enemy; attack a problem.

address

Typical selection: person/audience/problem. Examples: address the audience; address the issue.

tackle

Typical selection: physical opponent/problem metaphor. Examples: tackle a player; tackle a challenge.

solve

Typical selection: problem/equation/mystery. Examples: solve the equation.

resolve

Typical selection: dispute/issue/ambiguity. Examples: resolve conflict; resolve an issue.

Part XXVII — Parent Guide

Parents can teach semantic compatibility through playful nonsense. Ask a child which sentence sounds ordinary: The dog drank water or The pencil drank water. Then ask whether the strange sentence could make sense in a cartoon. This builds an important distinction: “odd literally” is not the same as “bad language.”

For older children, use school vocabulary: Can you calculate a sandwich? Can you measure a sandwich? Can you weigh a sandwich? The answers reveal that words select different semantic properties even when they belong to the same general topic.

Part XXVIII — Predicate Profiles: 80 High-Utility Words

consume

Typical argument profile: agent/system + resource/food/time. Examples: consume food, consume energy, consume time. Note: formal broad consumption predicate.

devour

Typical argument profile: animate eater + food; reader + content metaphor. Examples: devour dinner, devour a novel. Note: intensity stronger than eat/read.

ingest

Typical argument profile: organism/system + substance/data. Examples: ingest medicine, ingest data. Note: technical/formal.

absorb

Typical argument profile: material/person/system + liquid/energy/information/cost. Examples: absorb water, absorb heat, absorb information. Note: cross-domain polysemy.

emit

Typical argument profile: source + radiation/sound/gas. Examples: emit light, emit sound, emit carbon. Note: technical source relation.

exhale

Typical argument profile: animate respiratory agent + air/gas. Examples: exhale slowly. Note: bodily process.

evaporate

Typical argument profile: liquid/substance + change-of-state. Examples: water evaporates. Note: physical process; metaphorical disappearance possible.

condense

Typical argument profile: vapour/information + reduction/concentration. Examples: vapour condenses; condense a report. Note: physical and discourse senses.

dissolve

Typical argument profile: solute/material/organisation + disintegration. Examples: dissolve salt; dissolve parliament. Note: sense-specific classes.

erode

Typical argument profile: surface/material/trust/rights. Examples: erosion erodes soil; scandal erodes trust. Note: literal + abstract metaphor.

corrode

Typical argument profile: metal/material + chemical damage; trust metaphor. Examples: salt corrodes metal. Note: strong material preference.

fracture

Typical argument profile: bone/material/alliance. Examples: fracture a bone; alliance fractured. Note: physical-to-social extension.

inflate

Typical argument profile: gas-filled object/price/number. Examples: inflate tyre; inflate estimate. Note: literal + quantitative metaphor.

deflate

Typical argument profile: inflated object/expectation. Examples: deflate balloon; deflate hopes. Note: reversive extension.

measure

Typical argument profile: measurable quantity/object. Examples: measure length, measure performance. Note: requires defined dimension.

weigh

Typical argument profile: physical mass; evidence metaphor. Examples: weigh package; weigh evidence. Note: literal + deliberative extension.

quantify

Typical argument profile: measurable property. Examples: quantify risk, quantify emissions. Note: abstract numerical domain.

estimate

Typical argument profile: quantity/cost/probability. Examples: estimate cost, estimate population. Note: epistemic numerical predicate.

calculate

Typical argument profile: mathematical quantity/result. Examples: calculate area, calculate interest. Note: strong formal preference.

compute

Typical argument profile: algorithmically derivable value. Examples: compute checksum, compute mean. Note: technical/computational.

prove

Typical argument profile: proposition/theorem/claim under standard. Examples: prove theorem, prove identity. Note: truth-evaluable object.

demonstrate

Typical argument profile: claim/effect/method. Examples: demonstrate effect, demonstrate technique. Note: evidence or showing sense.

verify

Typical argument profile: claim/identity/measurement. Examples: verify result, verify identity. Note: checking predicate.

validate

Typical argument profile: method/model/ticket/claim under criteria. Examples: validate model, validate ticket. Note: criterion-based acceptance.

corroborate

Typical argument profile: claim/account/evidence. Examples: corroborate testimony. Note: support through independent evidence.

substantiate

Typical argument profile: claim/allegation. Examples: substantiate allegation. Note: evidence-strengthening.

infer

Typical argument profile: proposition/conclusion from evidence. Examples: infer cause, infer meaning. Note: cognitive reasoning predicate.

assume

Typical argument profile: proposition/condition. Examples: assume independence. Note: epistemic stance.

conclude

Typical argument profile: proposition from reasoning. Examples: conclude that policy failed. Note: reasoning endpoint.

hypothesise

Typical argument profile: proposition/explanation. Examples: hypothesise a mechanism. Note: scientific reasoning.

predict

Typical argument profile: future outcome/value. Examples: predict rain, predict score. Note: forecast relation.

forecast

Typical argument profile: future economic/weather quantity. Examples: forecast inflation, forecast rain. Note: domain-specific prediction.

remember

Typical argument profile: event/fact/person. Examples: remember meeting, remember name. Note: cognitive experiencer.

recognise

Typical argument profile: person/pattern/status. Examples: recognise face, recognise right. Note: perception/institutional senses.

notice

Typical argument profile: perceptible event/property. Examples: notice error, notice change. Note: attention predicate.

perceive

Typical argument profile: sensory/conceptual object. Examples: perceive sound, perceive risk. Note: experiencer-based.

admire

Typical argument profile: person/object/quality. Examples: admire painting, admire courage. Note: evaluative experiencer.

resent

Typical argument profile: event/person/action. Examples: resent criticism. Note: negative social emotion.

fear

Typical argument profile: threat/event/entity. Examples: fear storm, fear failure. Note: experiencer + perceived threat.

prefer

Typical argument profile: alternative/options. Examples: prefer tea to coffee. Note: comparative choice relation.

own

Typical argument profile: possessable entity/right. Examples: own house, own shares. Note: legal/social possession.

possess

Typical argument profile: entity/property/quality. Examples: possess skill, possess weapon. Note: broad possession.

contain

Typical argument profile: container/set/text + contents. Examples: box contains tools; report contains data. Note: part-whole/inclusion.

include

Typical argument profile: set/text/event + member. Examples: include example. Note: membership relation.

exclude

Typical argument profile: set/system + member. Examples: exclude outlier. Note: membership negation.

comprise

Typical argument profile: whole + parts. Examples: team comprises five members. Note: part-whole lexical pattern.

consist

Typical argument profile: whole + of-parts. Examples: team consists of five members. Note: fixed complement pattern.

resemble

Typical argument profile: entity + comparable entity. Examples: resemble parent. Note: similarity predicate.

differ

Typical argument profile: entity + from entity. Examples: differ from baseline. Note: contrast relation.

match

Typical argument profile: entity + counterpart/criterion. Examples: match colour, match record. Note: similarity/fit.

fit

Typical argument profile: object/person + space/criterion. Examples: fit box, fit profile. Note: compatibility predicate.

suit

Typical argument profile: person/purpose. Examples: colour suits her; plan suits purpose. Note: appropriateness.

qualify

Typical argument profile: person/claim + criterion. Examples: qualify for role. Note: eligibility relation.

entitle

Typical argument profile: person + benefit/right. Examples: entitle worker to leave. Note: institutional relation.

authorize

Typical argument profile: agent/institution + action/person. Examples: authorise payment. Note: permission relation.

prohibit

Typical argument profile: authority/rule + action. Examples: prohibit smoking. Note: institutional restriction.

permit

Typical argument profile: authority/rule + action. Examples: permit entry. Note: permission relation.

require

Typical argument profile: rule/task + condition/object. Examples: require ID. Note: necessity.

necessitate

Typical argument profile: event/condition + response. Examples: damage necessitates repair. Note: causal necessity.

cause

Typical argument profile: cause/event + effect. Examples: heat causes expansion. Note: causal relation.

trigger

Typical argument profile: event/stimulus + response. Examples: trigger alarm, trigger reaction. Note: initiating causation.

stimulate

Typical argument profile: agent/input + growth/activity. Examples: stimulate demand, stimulate nerve. Note: increase activity.

inhibit

Typical argument profile: agent/factor + process. Examples: inhibit enzyme, inhibit growth. Note: reduce activity.

prevent

Typical argument profile: cause/agent + event. Examples: prevent infection. Note: block occurrence.

enable

Typical argument profile: resource/design + agent/event. Examples: enable access. Note: make possible.

facilitate

Typical argument profile: resource/action + process. Examples: facilitate learning. Note: make easier.

support

Typical argument profile: structure/person/evidence + target. Examples: support roof, support claim. Note: polysemous but role-sensitive.

undermine

Typical argument profile: action/evidence + confidence/system. Examples: undermine trust. Note: metaphorical structural weakening.

strengthen

Typical argument profile: agent/change + structure/argument. Examples: strengthen bridge, strengthen case. Note: physical/abstract extension.

weaken

Typical argument profile: agent/change + structure/claim. Examples: weaken immune response. Note: broad scalar reduction.

allocate

Typical argument profile: authority/system + resource + recipient. Examples: allocate funds to schools. Note: transfer/distribution frame.

distribute

Typical argument profile: agent/system + items across recipients. Examples: distribute food. Note: movement/allocation.

assign

Typical argument profile: authority + task/value + recipient. Examples: assign homework. Note: institutional mapping.

appoint

Typical argument profile: authority + person + role. Examples: appoint chairperson. Note: role assignment.

elect

Typical argument profile: group + person + office. Examples: elect president. Note: collective-choice frame.

hire

Typical argument profile: employer + person. Examples: hire engineer. Note: employment frame.

employ

Typical argument profile: organisation + worker/resource/method. Examples: employ staff; employ technique. Note: work/use senses.

dismiss

Typical argument profile: authority + person/idea. Examples: dismiss employee; dismiss claim. Note: different sense-specific objects.

fire

Typical argument profile: employer + employee; weapon + projectile. Examples: fire worker; fire missile. Note: polysemous selection.

launch

Typical argument profile: agent + product/campaign/vehicle. Examples: launch satellite, launch campaign. Note: initiation frame.

deploy

Typical argument profile: organisation + resources/personnel/system. Examples: deploy troops, deploy software. Note: strategic placement.

install

Typical argument profile: agent + equipment/software. Examples: install app, install equipment. Note: technical placement.

configure

Typical argument profile: agent + system/settings. Examples: configure router. Note: technical setup.

calibrate

Typical argument profile: agent + instrument/model. Examples: calibrate sensor. Note: measurement domain.

Part XXIX — Selectional Compatibility in Cloze and Vocabulary Questions

Many vocabulary questions can be solved partly through selectional fit. If the noun is evidence, verbs such as support, corroborate, weaken, contradict are semantically compatible; drink is not. If the object is a treaty, ratify, sign, negotiate, violate fit institutional frames; photosynthesise does not.

34. Use noun class

Identify whether the noun is a person, event, physical object, proposition, quantity, institution or process.

35. Use event frame

Ask what kind of event the sentence describes: transfer, communication, movement, evaluation, measurement, causation or perception.

36. Use domain

Technical contexts sharply narrow compatibility: differentiate in mathematics, differentiate instruction in education.

37. Use collocation after semantic fit

First eliminate semantically impossible options, then choose the most conventional collocation among the survivors.

Part XXX — Assessment Ladder

Part XXXI — A 30-Day Selectional Compatibility Programme

Part XXXII — Canonical Map Inside the Vocabulary Ecosystem

This page owns the broad selectional-compatibility question. Existing Semantic Frames, Coercion, Entailment, Troponymy and Syntactic Bootstrapping pages keep their narrower specialist roles and remain unchanged.

Part XXXIII — Final Operating Standard

• Know the ordinary semantic class selected by a predicate.
• Treat compatibility as graded rather than blindly binary.
• Distinguish semantic anomaly from syntax error.
• Check the active sense of polysemous words.
• Use domain knowledge for technical predicates.
• Separate semantic fit from collocational frequency.
• Use world knowledge to evaluate plausibility.
• Recognise metaphor, personification and metonymy as repairs/extensions.
• Recognise coercion when an implied event or participant is supplied.
• Use selectional fit to choose precise vocabulary in writing.
• Expect creative language to violate ordinary preferences deliberately.

Conclusion — Words Bring Expectations About Other Words

Vocabulary knowledge includes more than definitions. Verbs, adjectives and relational nouns carry expectations about the kinds of participants, objects, events and properties they normally combine with. Those expectations help readers predict meaning, reject implausible interpretations, detect figurative language and choose natural vocabulary.

The most important modern shift is from rigid “restrictions” to graded “preferences.” Eat glass is unusual but possible; eat your words is idiomatic; the market ate the losses is metaphorical. Language stretches because people stretch semantic frames.

The practical sequence is predicate → participant types → ordinary fit → sense/domain check → contextual repair → interpretation. For writers, reverse it: decide the event and participant types first, then choose a predicate that naturally fits them.

When this becomes automatic, semantically strange sentences start to reveal exactly why they are strange—and creative sentences reveal exactly how they make themselves understood.

Appendix A — 100 Predicate–Argument Judgments

1. boil water

Judgment: ordinary literal. Why: boil selects liquid/substance capable of heating phase change.

2. boil anger

Judgment: figurative. Why: emotion conceptualised as heated liquid.

3. freeze water

Judgment: ordinary literal. Why: freezable substance.

4. freeze a bank account

Judgment: institutional metaphor. Why: financial access state lexicalised as frozen.

5. freeze a smile

Judgment: figurative. Why: expression becomes fixed.

6. digest food

Judgment: ordinary literal. Why: biological processing.

7. digest information

Judgment: conventional metaphor. Why: understanding framed as bodily processing.

8. digest a chair

Judgment: anomalous. Why: physical furniture not edible or informational under default sense.

9. absorb water

Judgment: ordinary literal. Why: material + liquid.

10. absorb a lesson

Judgment: metaphorical/conventional. Why: learner internalises information.

11. absorb costs

Judgment: economic metaphor. Why: organisation takes financial burden.

12. taste soup

Judgment: ordinary literal. Why: edible substance.

13. taste victory

Judgment: metaphorical. Why: experience conceptualised through taste.

14. taste mathematics

Judgment: anomalous unless poetic. Why: abstract discipline lacks literal flavour.

15. smell smoke

Judgment: ordinary literal. Why: odor source.

16. smell danger

Judgment: metaphorical/conventional. Why: detect threat without literal smell.

17. hear music

Judgment: ordinary literal. Why: auditory stimulus.

18. hear an argument

Judgment: ordinary literal/discourse. Why: speech/event.

19. hear a colour

Judgment: cross-modal anomaly. Why: synesthetic metaphor needed.

20. see a mountain

Judgment: ordinary literal. Why: visible object.

21. see the point

Judgment: conventional abstract sense. Why: understand.

22. see justice

Judgment: abstract/metaphorical. Why: may mean witness/understand justice.

23. touch fabric

Judgment: ordinary literal. Why: physical contact.

24. touch a heart

Judgment: metaphorical. Why: emotionally affect.

25. touch an equation

Judgment: marked. Why: physical sheet possible; abstract equation not literally touched.

26. carry a box

Judgment: ordinary literal. Why: physical transport.

27. carry responsibility

Judgment: metaphorical/conventional. Why: abstract burden.

28. carry a melody

Judgment: conventional extension. Why: voice/instrument sustains tune.

29. lift a suitcase

Judgment: ordinary literal. Why: physical object.

30. lift a ban

Judgment: institutional metaphor. Why: remove restriction.

31. lift a mood

Judgment: metaphorical. Why: improve emotional state.

32. break a glass

Judgment: ordinary literal. Why: physical object.

33. break a promise

Judgment: conventional metaphor. Why: violate commitment.

34. break the news

Judgment: idiomatic. Why: communicate information.

35. repair a machine

Judgment: ordinary literal. Why: physical system.

36. repair trust

Judgment: metaphorical. Why: restore relationship/state.

37. repair a hypothesis

Judgment: marked. Why: better revise/refine hypothesis.

38. build a house

Judgment: ordinary literal. Why: construction.

39. build confidence

Judgment: metaphorical conventional. Why: increase abstract capacity.

40. build an argument

Judgment: metaphorical conventional. Why: construct reasoning.

41. construct a bridge

Judgment: ordinary literal. Why: physical object.

42. construct an explanation

Judgment: academic metaphor. Why: assemble conceptual structure.

43. construct a sandwich

Judgment: possible literal. Why: assemble food; less prototypical verb but compatible.

44. destroy a building

Judgment: ordinary literal. Why: physical object.

45. destroy evidence

Judgment: ordinary/legal. Why: physical/digital information object.

46. destroy happiness

Judgment: metaphorical. Why: eliminate emotional state.

47. open a door

Judgment: ordinary literal. Why: physical closure.

48. open a meeting

Judgment: institutional extension. Why: begin event.

49. open an argument

Judgment: marked but interpretable. Why: begin line of discussion.

50. close a window

Judgment: ordinary literal. Why: physical/digital aperture.

51. close a deal

Judgment: institutional metaphor. Why: finalise transaction.

52. close an idea

Judgment: anomalous. Why: use reject/conclude discussion instead.

53. launch a rocket

Judgment: ordinary literal/technical. Why: vehicle.

54. launch a product

Judgment: conventional business extension. Why: initiate release.

55. launch a complaint

Judgment: marked. Why: better lodge/file complaint.

56. file a document

Judgment: ordinary institutional. Why: document object.

57. file a complaint

Judgment: institutional lexicalisation. Why: submit formally.

58. file a sandwich

Judgment: anomalous unless physically using a file tool. Why: sense mismatch.

59. table a proposal

Judgment: institutional/political. Why: meaning varies by regional parliamentary practice.

60. table a meal

Judgment: ordinary arrangement. Why: different sense of table as verb.

61. table an emotion

Judgment: anomalous. Why: no conventional sense.

62. draw a picture

Judgment: ordinary literal. Why: graphic production.

63. draw a conclusion

Judgment: conventional reasoning metaphor. Why: derive conclusion.

64. draw water

Judgment: ordinary/older literal. Why: extract liquid.

65. draw anger

Judgment: possible causative. Why: elicit reaction; context needed.

66. frame a picture

Judgment: ordinary literal. Why: place in frame.

67. frame a question

Judgment: conventional abstract. Why: structure/formulate.

68. frame a suspect

Judgment: legal colloquial. Why: falsely incriminate.

69. frame a cloud

Judgment: odd unless artistic. Why: requires special context.

70. plant a seed

Judgment: ordinary literal. Why: agricultural.

71. plant evidence

Judgment: legal/criminal metaphor lexicalised. Why: place secretly.

72. plant an idea

Judgment: metaphorical conventional. Why: introduce idea subtly.

73. plant a theorem

Judgment: anomalous. Why: no conventional relation.

74. seed a field

Judgment: ordinary literal. Why: agricultural.

75. seed a tournament

Judgment: institutional sports. Why: rank/place competitors.

76. seed a database

Judgment: technical computing. Why: populate initial data.

77. seed a spoon

Judgment: anomalous. Why: wrong semantic class.

78. feed a child

Judgment: ordinary literal. Why: animate recipient.

79. feed a machine

Judgment: technical metaphor. Why: supply material/data.

80. feed data into model

Judgment: technical conventional. Why: information-as-input.

81. feed a theorem

Judgment: anomalous without system context. Why: theorem not recipient/input by default.

82. train a dog

Judgment: ordinary literal. Why: learning agent.

83. train a model

Judgment: technical machine-learning sense. Why: system learning metaphor lexicalised.

84. train a mountain

Judgment: anomalous. Why: no learning capacity.

85. coach a team

Judgment: ordinary institutional. Why: human group.

86. coach a model

Judgment: possible if model = person, not computational model usually. Why: sense-specific ambiguity.

87. educate a child

Judgment: ordinary literal/institutional. Why: learner agent.

88. educate the market

Judgment: business metaphor. Why: market participants conceptualised as learners.

89. educate a chair

Judgment: anomalous. Why: no learner capacity.

90. teach a lesson

Judgment: ordinary. Why: content object.

91. teach a machine

Judgment: technical metaphor/conventional AI context. Why: learning system.

92. teach a sandwich

Judgment: anomalous. Why: no learning capacity.

93. ask a person

Judgment: ordinary. Why: addressee agent.

94. ask a question

Judgment: ordinary. Why: content object.

95. ask a chair

Judgment: personification. Why: inanimate addressee.

96. answer a question

Judgment: ordinary. Why: question content.

97. answer a person

Judgment: ordinary communicative. Why: respond to addressee.

98. answer a mountain

Judgment: poetic/personification. Why: no ordinary conversational partner.

99. promise help

Judgment: ordinary. Why: future commitment.

100. promise a person help

Judgment: ordinary ditransitive. Why: recipient + content.

101. promise a stone

Judgment: anomalous unless personified. Why: stone cannot socially receive commitment.

102. accuse a person

Judgment: ordinary legal/social. Why: agentive target.

103. accuse a theory

Judgment: marked. Why: better criticise/challenge theory.

104. challenge a theory

Judgment: ordinary academic. Why: proposition/model.

105. challenge a person

Judgment: ordinary social. Why: animate target.

106. challenge a sandwich

Judgment: anomalous unless eating contest context. Why: context can rescue.

107. evaluate a proposal

Judgment: ordinary academic. Why: decision object.

108. evaluate a student

Judgment: ordinary educational. Why: participant performance.

109. evaluate a cloud

Judgment: possible meteorological analysis, but default vague. Why: domain needed.

110. grade an essay

Judgment: ordinary educational. Why: assessable work.

111. grade a student

Judgment: institutional shorthand. Why: assign grade to performance/person.

112. grade a mountain

Judgment: geography verb sense possible. Why: slope/classification context changes compatibility.

113. classify a species

Judgment: ordinary scientific. Why: taxonomic entity.

114. classify a document

Judgment: ordinary information management. Why: categorisable object.

115. classify a feeling

Judgment: possible psychology. Why: category system required.

116. categorise a chair

Judgment: possible. Why: physical object can be assigned category.

117. categorise sunlight

Judgment: possible under scientific scheme. Why: category context needed.

118. calculate risk

Judgment: ordinary quantitative. Why: measurable/estimated variable.

119. calculate happiness

Judgment: possible only with operational metric. Why: abstract construct needs measurement definition.

120. calculate a joke

Judgment: anomalous. Why: not normally quantitative.

121. measure temperature

Judgment: ordinary. Why: measurable quantity.

122. measure intelligence

Judgment: possible psychometric. Why: construct requires operationalisation.

123. measure beauty

Judgment: possible under study but conceptually contested. Why: metric context needed.

124. quantify emissions

Judgment: ordinary. Why: measurable output.

125. quantify uncertainty

Judgment: ordinary technical. Why: probabilistic/statistical representation.

126. quantify kindness

Judgment: possible research operationalisation. Why: not direct physical quantity.

127. prove identity

Judgment: ordinary math/legal depending sense. Why: proposition/identity.

128. prove innocence

Judgment: legal/evidential. Why: truth-evaluable proposition/status.

129. prove flavour

Judgment: odd. Why: flavour not proposition unless claim implied.

130. refute a claim

Judgment: ordinary argument. Why: proposition.

131. refute a person

Judgment: marked. Why: better refute person’s claim.

132. refute a sandwich

Judgment: anomalous. Why: not truth-evaluable.

133. contradict a statement

Judgment: ordinary discourse. Why: proposition.

134. contradict a person

Judgment: ordinary shorthand. Why: person’s statement/position.

135. contradict a chair

Judgment: anomalous. Why: no proposition.

136. translate a sentence

Judgment: ordinary language. Why: linguistic content.

137. translate a protein

Judgment: marked under language sense. Why: biology translates mRNA into protein; protein itself is not normally translated.

138. translate an mRNA sequence

Judgment: ordinary molecular biology. Why: domain sense.

139. translate a mountain

Judgment: anomalous literal. Why: unless translate = move geometrically and mountain is an object in a model.

140. rotate a shape

Judgment: ordinary math/physical. Why: geometric object.

141. rotate a policy

Judgment: anomalous. Why: no conventional transformation.

142. rotate staff

Judgment: ordinary organisational. Why: workers cycled through roles.

143. compress a file

Judgment: technical computing. Why: digital data.

144. compress a gas

Judgment: ordinary physics. Why: physical substance.

145. compress an argument

Judgment: metaphorical/marked. Why: better condense argument.

146. condense a report

Judgment: conventional abstract. Why: shorten information.

147. condense steam

Judgment: ordinary physical. Why: gas to liquid.

148. condense a person

Judgment: anomalous. Why: wrong type.

149. filter water

Judgment: ordinary physical. Why: fluid mixture.

150. filter data

Judgment: technical computing. Why: information set.

151. filter a theorem

Judgment: marked. Why: requires selection criteria context.

152. screen patients

Judgment: medical institutional. Why: people assessed.

153. screen a film

Judgment: media sense. Why: show film.

154. screen light

Judgment: physical filtering. Why: block/reduce light.

155. screen a sandwich

Judgment: anomalous. Why: unless inspect food under specific procedure.

156. sample water

Judgment: scientific. Why: substance/population.

157. sample voters

Judgment: survey sense. Why: population.

158. sample music

Judgment: audio production. Why: recorded sound source.

159. sample a theorem

Judgment: anomalous. Why: not population/material/audio by default.

160. model climate

Judgment: scientific. Why: system/process.

161. model behaviour

Judgment: scientific/social. Why: process.

162. model clothing

Judgment: fashion sense. Why: person displays garment.

163. model a sandwich

Judgment: possible 3D modelling. Why: domain changes compatibility.

164. simulate weather

Judgment: computational. Why: process/system.

165. simulate a patient

Judgment: training context. Why: represented person/system.

166. simulate hunger

Judgment: possible experimental/computational. Why: state.

167. simulate a theorem

Judgment: odd. Why: proposition not process unless proof process modelled.

Appendix B — Five-Step Compatibility Test

1. Identify the active sense of the predicate.
2. Identify the semantic type of each participant: person, physical object, substance, event, proposition, quantity, institution, information, etc.
3. Ask whether that type commonly fills the predicate’s role.
4. If not, search for a technical sense, metaphor, metonymy, personification or coercion that repairs the mismatch.
5. If no coherent repair exists, treat the combination as anomalous or revise the word choice.

Appendix C — Final Writer’s Check

Before accepting a thesaurus replacement, ask whether the new word selects the same kind of subject, object and complement as the original. Similar dictionary meanings do not guarantee identical semantic compatibility. Eat, consume, ingest, absorb overlap, but each prefers different semantic environments and registers.

Selectional knowledge becomes practical when it prevents silent lexical errors: a technically grammatical sentence whose verbs and nouns simply do not belong together under the intended meaning.

Appendix D — Forty Myths About Selectional Restrictions

Myth 1: A semantically odd sentence is ungrammatical.

Reality: syntax can be perfectly well formed while lexical meanings clash.

Myth 2: Selectional restrictions are absolute rules.

Reality: modern approaches often model graded preferences because language tolerates unusual, metaphorical and domain-specific combinations.

Myth 3: If a verb usually takes people, it can never take machines.

Reality: technical senses often extend human predicates to systems, as in the model learns.

Myth 4: Metaphor ignores lexical meaning.

Reality: metaphor often exploits ordinary selectional expectations precisely by violating them.

Myth 5: Any violation is metaphorical.

Reality: some are simply incoherent or errors, while others are literal but unusual.

Myth 6: World knowledge and lexical semantics are the same.

Reality: a combination can be semantically type-compatible yet pragmatically implausible.

Myth 7: The child lifted the piano is semantically impossible.

Reality: participant types fit; the event is merely difficult/improbable.

Myth 8: The piano lifted the child is automatically impossible.

Reality: machinery, fiction or context could make it literal; default world knowledge simply resists it.

Myth 9: Collocation and selectional preference are identical.

Reality: collocation tracks conventional co-occurrence; selectional preference concerns semantic type fit.

Myth 10: A rare collocation is semantically wrong.

Reality: many low-frequency combinations are perfectly coherent.

Myth 11: A frequent collocation is always literal.

Reality: highly frequent combinations may be conventional metaphors.

Myth 12: Technical language obeys ordinary selectional preferences.

Reality: technical senses create new compatibility profiles.

Myth 13: Kill a process is figurative every time.

Reality: in computing it is conventional technical vocabulary.

Myth 14: Run a program is merely poetic.

Reality: it is a lexicalised computing sense.

Myth 15: Polysemy is unrelated to selectional behaviour.

Reality: different senses select different semantic classes.

Myth 16: One verb has one fixed argument structure.

Reality: polysemous verbs often have several frames and valencies.

Myth 17: Semantic roles are universally agreed categories.

Reality: linguistic theories use overlapping but different role inventories.

Myth 18: Selectional preferences belong only to verbs.

Reality: adjectives, nouns and constructions also constrain semantic types.

Myth 19: Pregnant pause is an error.

Reality: it is a conventional metaphorical extension.

Myth 20: Fragile peace is literally impossible and therefore wrong.

Reality: physical-breakability meaning has extended conventionally to unstable abstract states.

Myth 21: Coercion means forcing bad English to work.

Reality: coercion is a normal interpretive mechanism in ordinary language.

Myth 22: Begin the book is incomplete English.

Reality: context supplies a conventional event such as reading or writing.

Myth 23: Metonymy is just metaphor.

Reality: metonymy relies on association/contiguity, such as institution for decision-makers.

Myth 24: The White House announced is literal place speech.

Reality: place stands for administration/institution.

Myth 25: Machines cannot be subjects of cognitive verbs.

Reality: technical discourse increasingly lexicalises machine-learning and recognition predicates.

Myth 26: If an AI model learns, the verb must mean exactly human learning.

Reality: technical senses can be conventional while retaining metaphorical ancestry.

Myth 27: Personification is only literary.

Reality: everyday language personifies systems, markets, devices and institutions constantly.

Myth 28: A selectional mismatch always needs repair.

Reality: poetry and humour may preserve the mismatch deliberately.

Myth 29: Creative language has no semantic rules.

Reality: creativity is interpretable because readers know the ordinary rules being stretched.

Myth 30: Selectional preference can be learned from a definition alone.

Reality: examples, collocations, argument patterns and domains reveal the actual behaviour.

Myth 31: Synonyms share identical selectional preferences.

Reality: near-synonyms often differ in argument type and register.

Myth 32: Eat, ingest, consume, absorb are interchangeable.

Reality: each has distinct semantic classes and discourse environments.

Myth 33: Semantic compatibility is binary.

Reality: compatibility often exists on a gradient from prototypical to unusual to coerced/anomalous.

Myth 34: Unusual participant roles are always errors.

Reality: context can change semantic type through metonymy or coercion.

Myth 35: Selectional preferences are useless for children.

Reality: even young children use participant types to infer new verb meanings.

Myth 36: Selectional preference is too theoretical for exams.

Reality: vocabulary/cloze items frequently depend on semantic fit.

Myth 37: Thesaurus replacements preserve compatibility automatically.

Reality: broad semantic similarity does not guarantee the same argument profile.

Myth 38: Selectional preference only matters for comprehension.

Reality: it is equally important for precise writing and translation.

Myth 39: Translation can ignore selectional preferences if definitions match.

Reality: target-language predicates may select different argument classes or constructions.

Myth 40: The goal is to memorise feature labels.

Reality: the practical goal is knowing what kinds of things a word normally combines with and why.

Appendix E — Fifty Transfer Cases

1. consume

Transfer: Move from consume food to consume energy. What stays stable: The common idea is using up a resource, not merely eating.

2. absorb

Transfer: Move from absorb water to absorb information. What stays stable: Physical intake supports a conventional cognitive metaphor.

3. digest

Transfer: Move from digest food to digest a report. What stays stable: Understanding is framed as processing food.

4. attack

Transfer: Move from attack an enemy to attack a problem. What stays stable: Aggressive physical action becomes energetic problem-solving.

5. defend

Transfer: Move from defend a city to defend a claim. What stays stable: Protection frame extends to argumentation.

6. support

Transfer: Move from support a roof to support a claim. What stays stable: Structural support becomes evidential support.

7. undermine

Transfer: Move from undermine a wall to undermine confidence. What stays stable: Physical weakening becomes abstract weakening.

8. strengthen

Transfer: Move from strengthen a bridge to strengthen an argument. What stays stable: Structural robustness becomes evidential robustness.

9. weaken

Transfer: Move from weaken a material to weaken a case. What stays stable: Scalar reduction transfers domains.

10. launch

Transfer: Move from launch a rocket to launch a campaign. What stays stable: Initiation becomes the shared semantic core.

11. run

Transfer: Move from run a race to run a company. What stays stable: Different lexical senses require different participant profiles.

12. drive

Transfer: Move from drive a car to drive change. What stays stable: Physical propulsion becomes causal force.

13. navigate

Transfer: Move from navigate a river to navigate a crisis. What stays stable: Path management extends to problem management.

14. steer

Transfer: Move from steer a boat to steer policy. What stays stable: Directional control extends to institutional guidance.

15. fuel

Transfer: Move from fuel an engine to fuel anger. What stays stable: Providing energy extends to increasing emotion.

16. ignite

Transfer: Move from ignite fuel to ignite debate. What stays stable: Beginning combustion extends to starting intense social activity.

17. spark

Transfer: Move from spark a fire to spark interest. What stays stable: Small initiating cause transfers abstractly.

18. trigger

Transfer: Move from trigger an alarm to trigger anxiety. What stays stable: Initiating mechanism generalises across domains.

19. seed

Transfer: Move from seed a field to seed a database. What stays stable: Initial population/growth metaphor becomes technical.

20. harvest

Transfer: Move from harvest crops to harvest energy. What stays stable: Collecting usable output extends technically.

21. mine

Transfer: Move from mine coal to mine data. What stays stable: Extracting valuable material extends to information.

22. filter

Transfer: Move from filter water to filter results. What stays stable: Physical separation becomes selection.

23. screen

Transfer: Move from screen patients to screen calls. What stays stable: Evaluation/filtering sense generalises.

24. process

Transfer: Move from process food to process data. What stays stable: Transformation through stages becomes abstract/technical.

25. store

Transfer: Move from store grain to store data. What stays stable: Retention/container relation generalises.

26. retrieve

Transfer: Move from retrieve a ball to retrieve a memory. What stays stable: Recovering a physical item becomes mental access.

27. encode

Transfer: Move from encode a message to encode a memory. What stays stable: Representation into a system extends cognitive metaphor.

28. consolidate

Transfer: Move from consolidate companies to consolidate memory. What stays stable: Making stable/combined extends to learning.

29. activate

Transfer: Move from activate a device to activate knowledge. What stays stable: Turning on becomes making available.

30. inhibit

Transfer: Move from inhibit enzyme to inhibit speech. What stays stable: Reducing activity generalises.

31. facilitate

Transfer: Move from facilitate access to facilitate learning. What stays stable: Making easier fits processes across domains.

32. mediate

Transfer: Move from mediate dispute to mediate an effect. What stays stable: Intermediate causal relation becomes technical research vocabulary.

33. moderate

Transfer: Move from moderate discussion to moderate an effect. What stays stable: Control intensity becomes statistical interaction sense.

34. predict

Transfer: Move from predict weather to predict exam performance. What stays stable: Future/unknown outcome remains central.

35. model

Transfer: Move from model a building to model a process. What stays stable: Representation sense generalises.

36. simulate

Transfer: Move from simulate flight to simulate a market. What stays stable: Represent dynamic process computationally.

37. map

Transfer: Move from map a coastline to map relationships. What stays stable: Spatial representation extends to abstract structure.

38. trace

Transfer: Move from trace a line to trace a cause. What stays stable: Following a path extends to causal investigation.

39. frame

Transfer: Move from frame a picture to frame a problem. What stays stable: Providing boundary/structure extends cognitively.

40. construct

Transfer: Move from construct a bridge to construct an argument. What stays stable: Building metaphor structures reasoning.

41. collapse

Transfer: Move from collapse a building to collapse a distinction. What stays stable: Loss of structure extends conceptually.

42. bridge

Transfer: Move from bridge a river to bridge a knowledge gap. What stays stable: Connecting separated regions extends abstractly.

43. anchor

Transfer: Move from anchor a boat to anchor an explanation. What stays stable: Stabilising location becomes conceptual grounding.

44. ground

Transfer: Move from ground a wire to ground a theory in evidence. What stays stable: Providing stable basis extends abstractly.

45. channel

Transfer: Move from channel water to channel resources. What stays stable: Directed flow generalises.

46. stream

Transfer: Move from stream water to stream video. What stays stable: Continuous flow becomes digital transmission.

47. broadcast

Transfer: Move from broadcast seed historically to broadcast news. What stays stable: Wide distribution becomes communication term.

48. circulate

Transfer: Move from blood circulates to rumours circulate. What stays stable: Movement through network extends information spread.

49. spread

Transfer: Move from spread butter to spread information. What stays stable: Expansion/distribution core generalises.

50. infect

Transfer: Move from virus infects cells to laughter infects the room. What stays stable: Contagion frame supports emotional/social metaphor.

Appendix F — Selectional Preference Notebook

For difficult academic verbs, use a five-column notebook: predicate sense, preferred subject class, preferred object/complement class, natural examples, marked/metaphorical examples. This format captures how the word behaves rather than storing a definition alone.

Example: corroborate — sense: independently support evidence/claim; subject: witness/data/source; object: account/claim/evidence; ordinary: records corroborate the testimony; marked: the sandwich corroborates the weather. The final contrast makes the semantic boundaries visible.

Appendix G — Final Independent Transfer Test

1. Choose ten unfamiliar verbs from a current textbook.
2. For each, collect three ordinary subject–verb–object examples.
3. Classify the semantic types of subjects and objects.
4. Write one deliberately incompatible combination.
5. Decide whether metaphor or technical context can repair it.
6. Check a corpus/dictionary for real usage.
7. Use the verb in a new sentence with a compatible argument.
8. Repeat one week later without notes.

Appendix H — 60 Minimal-Pair Compatibility Tests

1. Contrast

A: The chef sliced the bread. B: The chef sliced the thunder. Compatibility lesson: Slice prefers a divisible physical object; thunder needs metaphor.

2. Contrast

A: The child hugged the dog. B: The child hugged the equation. Compatibility lesson: Hug prefers a physical/animate target; abstract object is figurative.

3. Contrast

A: The scientist tested the sample. B: The scientist tested the sunrise. Compatibility lesson: Test prefers an assessable object/hypothesis; sunrise needs operational context.

4. Contrast

A: The lawyer challenged the claim. B: The lawyer challenged the sandwich. Compatibility lesson: Argumentative challenge selects proposition/person, not food by default.

5. Contrast

A: The gardener pruned the tree. B: The gardener pruned the hypothesis. Compatibility lesson: Prune literally selects plant branches; abstract editing use possible metaphorically.

6. Contrast

A: The editor pruned the paragraph. B: The editor pruned the spoon. Compatibility lesson: Editing sense selects text/content.

7. Contrast

A: The nurse monitored the patient. B: The nurse monitored the sandwich. Compatibility lesson: Monitoring selects changing state/system unless food safety context is supplied.

8. Contrast

A: The sensor monitored temperature. B: The sensor monitored justice. Compatibility lesson: Instrument can monitor measurable variable; justice requires operationalisation.

9. Contrast

A: The researcher sampled water. B: The researcher sampled honesty. Compatibility lesson: Abstract trait requires defined population/measurement method.

10. Contrast

A: The survey sampled voters. B: The survey sampled mountains. Compatibility lesson: Mountains could be sampled in geography only if units/population defined.

11. Contrast

A: The teacher graded essays. B: The teacher graded sunlight. Compatibility lesson: Assessment requires assessable work/performance or measurable scale.

12. Contrast

A: The geologist graded the slope. B: The geologist graded kindness. Compatibility lesson: Different sense of grade can classify physical slope; kindness needs rating scheme.

13. Contrast

A: The doctor prescribed medicine. B: The doctor prescribed a cloud. Compatibility lesson: Prescription requires treatment/action unless poetic.

14. Contrast

A: The coach prescribed rest. B: The coach prescribed geometry. Compatibility lesson: Rest can be recommended; geometry needs learning-plan context.

15. Contrast

A: The judge sentenced the offender. B: The judge sentenced the chair. Compatibility lesson: Legal sentencing selects convicted person/entity.

16. Contrast

A: The editor deleted the sentence. B: The editor deleted the mountain. Compatibility lesson: Digital/text deletion selects information; mountain only via representation.

17. Contrast

A: The programmer deleted the file. B: The programmer deleted the sandwich. Compatibility lesson: Food can only be deleted as an order/data record.

18. Contrast

A: The analyst interpreted the data. B: The analyst interpreted the table. Compatibility lesson: Possible if table means data display; polysemy matters.

19. Contrast

A: The tourist interpreted the mountain. B: The tourist described the mountain. Compatibility lesson: Interpretation requires sign/text/evidence unless symbolic reading intended.

20. Contrast

A: The translator interpreted the speech. B: The translator interpreted the bicycle. Compatibility lesson: Interpreting sense selects language/communication.

21. Contrast

A: The doctor examined the patient. B: The doctor examined the theorem. Compatibility lesson: Possible generally, but medical sense selects patient while academic sense changes predicate domain.

22. Contrast

A: The mathematician examined the theorem. B: The mathematician examined the patient. Compatibility lesson: Possible general sense, but domain expectations shift.

23. Contrast

A: The artist sketched the building. B: The artist sketched the smell. Compatibility lesson: Visual representation selects representable form; synesthetic art could repair.

24. Contrast

A: The engineer sketched the design. B: The engineer sketched the anxiety. Compatibility lesson: Abstract state needs metaphor/diagram context.

25. Contrast

A: The poet captured the mood. B: The camera captured the mood. Compatibility lesson: Camera can capture visual signs of mood metaphorically; agent/sense differs.

26. Contrast

A: The camera captured the image. B: The camera captured the argument. Compatibility lesson: Argument only via recording/document context.

27. Contrast

A: The historian documented the event. B: The historian documented the flavour. Compatibility lesson: Flavour can be documented descriptively, so compatibility is possible with context.

28. Contrast

A: The sensor detected smoke. B: The sensor detected democracy. Compatibility lesson: Abstract political system not directly sensor-detectable.

29. Contrast

A: The survey detected a trend. B: The survey detected a sandwich. Compatibility lesson: Data instrument can reveal trend; sandwich only if survey asks about it.

30. Contrast

A: The microscope revealed cells. B: The microscope revealed injustice. Compatibility lesson: Instrument reveals visible structure; injustice needs metaphor/evidence chain.

31. Contrast

A: The report revealed fraud. B: The report revealed bacteria. Compatibility lesson: Possible only if report contains laboratory findings; metonymy matters.

32. Contrast

A: The doctor treated the patient. B: The doctor treated the equation. Compatibility lesson: Medical sense selects patient; general ‘deal with’ could extend.

33. Contrast

A: The mathematician treated the equation as a special case. B: The mathematician treated the patient. Compatibility lesson: General conceptual sense vs medical sense.

34. Contrast

A: The company hired an engineer. B: The company hired a thunderstorm. Compatibility lesson: Employment frame selects person/contractor.

35. Contrast

A: The company acquired a startup. B: The company acquired a headache. Compatibility lesson: Both possible under different senses: purchase vs come to have.

36. Contrast

A: The child acquired vocabulary. B: The child acquired a skyscraper. Compatibility lesson: Possession/learning senses; skyscraper requires ownership context.

37. Contrast

A: The library acquired books. B: The library acquired jealousy. Compatibility lesson: Institution can acquire collections, not emotions literally.

38. Contrast

A: The patient developed a fever. B: The city developed a transport network. Compatibility lesson: Different senses of develop select biological state vs system.

39. Contrast

A: The sandwich developed a policy. B: Anomalous. Compatibility lesson: No institutional agency.

40. Contrast

A: The committee adopted the policy. B: The committee adopted the sandwich. Compatibility lesson: Adopt can take plan/policy; sandwich only humorous ‘choose as mascot/menu’ context.

41. Contrast

A: The family adopted a child. B: The family adopted an equation. Compatibility lesson: Different adopt sense selects person vs method/practice.

42. Contrast

A: The class adopted a new method. B: The class adopted a mountain. Compatibility lesson: Method compatible; mountain only symbolic/contextual.

43. Contrast

A: The scientist rejected the hypothesis. B: The scientist rejected the molecule. Compatibility lesson: A molecule can be rejected from sample under technical context; sense matters.

44. Contrast

A: The editor rejected the article. B: The editor rejected the colour. Compatibility lesson: Possible if colour proposal; object must be decision candidate.

45. Contrast

A: The machine rejected the coin. B: The machine rejected the theory. Compatibility lesson: Technical accept/reject can select input objects; theory only if evaluation system.

46. Contrast

A: The voter rejected the proposal. B: The voter rejected the sunrise. Compatibility lesson: Proposal is choice object; sunrise not ordinarily decidable.

47. Contrast

A: The engine consumed fuel. B: The engine consumed a theorem. Compatibility lesson: Fuel/resource compatible; theorem only metaphor/data system context.

48. Contrast

A: The model consumed training data. B: The model consumed soup. Compatibility lesson: Data ingestion technical; soup requires physical agent.

49. Contrast

A: The plant absorbed water. B: The plant absorbed a lecture. Compatibility lesson: Information absorption requires cognitive agent.

50. Contrast

A: The student absorbed the lecture. B: The student absorbed the river. Compatibility lesson: Lecture metaphorical information; river literal physically impossible beyond context.

51. Contrast

A: The sponge absorbed the spill. B: The sponge absorbed the theorem. Compatibility lesson: Material absorption vs conceptual object.

52. Contrast

A: The market absorbed the shock. B: The market absorbed the sandwich. Compatibility lesson: System shock metaphor conventional; sandwich odd.

53. Contrast

A: The policy addressed inequality. B: The policy addressed the chair. Compatibility lesson: Address as deal-with selects issue; as speak-to selects addressee.

54. Contrast

A: The speaker addressed the chair. B: The speaker addressed inequality. Compatibility lesson: Both possible under different senses: speak to vs deal with topic.

55. Contrast

A: The chair addressed the audience. B: The chair addressed the weather. Compatibility lesson: Both possible if weather is topic; person sense of chair via role metonymy.

56. Contrast

A: The weather addressed the audience. B: Personification. Compatibility lesson: Natural phenomenon lacks speech agency.

57. Contrast

A: The software handled the request. B: The software handled the sandwich. Compatibility lesson: Request/data compatible; sandwich only if order-management system.

58. Contrast

A: The waiter handled the sandwich. B: The waiter handled the request. Compatibility lesson: Both possible literally/socially under different objects.

59. Contrast

A: The theory explains the data. B: The theory explains the sandwich. Compatibility lesson: Sandwich needs phenomenon/recipe question to be explainable.

60. Contrast

A: The recipe explains the sandwich. B: Metonymic/conventional. Compatibility lesson: Recipe explains how sandwich is made.

61. Contrast

A: The sandwich explains the recipe. B: Marked/anomalous. Compatibility lesson: Reverse explanatory relation unless visual example context.

62. Contrast

A: The evidence justifies the conclusion. B: The evidence justifies the bicycle. Compatibility lesson: Bicycle requires decision/action context; object alone is not proposition.

63. Contrast

A: The cost justifies the purchase. B: The cost justifies the cloud. Compatibility lesson: Cloud needs purchasable service/context.

64. Contrast

A: The rule governs behaviour. B: The rule governs the sandwich. Compatibility lesson: Food needs process/category context.

65. Contrast

A: The thermostat regulates temperature. B: The thermostat regulates friendship. Compatibility lesson: Temperature measurable control variable; friendship not ordinary system variable.

66. Contrast

A: The law regulates commerce. B: The law regulates sadness. Compatibility lesson: Institutional rules can regulate behaviour, not inner emotion directly.

67. Contrast

A: The medicine regulates blood pressure. B: The medicine regulates democracy. Compatibility lesson: Biological system variable vs political abstraction.

68. Contrast

A: The algorithm ranks pages. B: The algorithm ranks sandwiches. Compatibility lesson: Both possible if sandwiches are items in a ranking dataset.

69. Contrast

A: The critic ranks films. B: The critic ranks oxygen. Compatibility lesson: Oxygen can be ranked only among alternatives under defined criterion.

70. Contrast

A: The app recommends restaurants. B: The app recommends gravity. Compatibility lesson: Gravity is not a choice/action unless educational content is meant.

71. Contrast

A: The doctor recommends rest. B: The doctor recommends a cloud. Compatibility lesson: Cloud only if service/product context.

72. Contrast

A: The sensor records temperature. B: The sensor records justice. Compatibility lesson: Abstract construct needs operational indicator.

73. Contrast

A: The historian records events. B: The historian records sunlight. Compatibility lesson: Could record data about sunlight; context turns abstract into measurable phenomenon.

74. Contrast

A: The experiment controls temperature. B: The experiment controls happiness. Compatibility lesson: Happiness can be experimentally influenced/controlled only through operational design.

75. Contrast

A: The pilot controls the aircraft. B: The pilot controls the theorem. Compatibility lesson: Theorem not controllable entity.

76. Contrast

A: The moderator controls the discussion. B: The moderator controls the mountain. Compatibility lesson: Discussion is social process; mountain only metaphor/infrastructure context.

Appendix I — Final Assessment Rubric

Appendix J — Final Mastery Questions

• Can I distinguish a grammar error from a semantic mismatch?
• Can I name the ordinary semantic class selected by a verb or adjective?
• Can I explain why an unusual combination is still literally possible?
• Can I detect when a technical sense changes the selectional pattern?
• Can I recognise coercion, metaphor, metonymy and personification?
• Can I separate collocational rarity from semantic incompatibility?
• Can I use world knowledge without mistaking implausibility for impossibility?
• Can I choose a better predicate when a thesaurus substitute creates a mismatch?
• Can I analyse unfamiliar combinations rather than memorising examples?
• Can I deliberately violate selectional preferences to create controlled figurative language?

Final Canonical Statement

This page is the eduKateSG broad owner for selectional restrictions, selectional preferences, predicate–argument semantic compatibility and semantic anomaly. Narrower pages on Semantic Frames, Lexical Coercion, Lexical Entailment, Troponymy, Deictic Verbs, argument alternations and Syntactic Bootstrapping remain protected specialists.

The practical mastery test is simple: when a sentence is grammatically sound but something about the word combination feels wrong, the learner can identify the semantic expectation being violated—and can tell whether the result is an error, an unusual literal event, a technical sense, or creative language doing deliberate work.

Appendix K — Semantic Compatibility in Definitions and Dictionaries

Dictionary definitions often imply selectional behaviour even when they do not label it explicitly. A definition of persuade that says “cause someone to do or believe something through reasoning or urging” already signals that the object is normally a person or person-like cognitive entity. A definition of diagonalise refers to matrices or operators. A definition of vaccinate refers to people or animals receiving a vaccine. Strong dictionary use therefore reads examples and grammatical patterns as part of meaning.

1. Definition words reveal semantic classes

Look for labels such as person, animal, substance, statement, event, quantity, organisation, document or process. These are selectional clues.

2. Example sentences reveal ordinary participants

If every example of corroborate contains evidence, testimony, records or accounts, the learner sees a truth-evaluable/evidential semantic field around the verb.

3. Grammar labels and selection interact

Transitive/intransitive patterns, prepositions and complement types show how semantic roles are realised syntactically.

4. Domain labels change expectations

A computing label beside kill or run tells the reader not to apply only the biological/movement senses.

Appendix L — Semantic Compatibility in Translation

Translation exposes selectional differences because two languages can carve up predicates differently. A source verb may have one broad compatibility range while the target language uses several verbs depending on object class, animacy, instrument or manner. Literal word-for-word substitution can therefore produce a target sentence that is grammatical yet semantically unnatural.

Appendix M — Selectional Preferences in AI and NLP

Computational language processing has long used selectional preferences to model plausibility. Early systems tried strict semantic constraints; later models learned graded preferences from corpora. Modern neural language models absorb vast numbers of predicate–argument patterns statistically, but they can still generate semantically odd combinations when local word probability does not align with world knowledge.

5. Word-sense disambiguation

If bank appears as the object of deposit money at, selectional and collocational evidence favours the financial sense. Semantic role patterns can narrow ambiguity.

6. Semantic role labelling

Computational systems identify who did what to whom, where and how. Selectional preferences can help distinguish plausible role assignments.

7. Plausibility scoring

The chef sliced bread receives higher ordinary plausibility than the thunder sliced bread. Models can estimate such gradients from data.

8. Metaphor and anomaly

A system must not reject every low-preference combination, because creative and metaphorical language is pervasive. The challenge is interpreting, not merely filtering.

9. Generative AI writing

When AI produces a sentence that is grammatically smooth but semantically strange, selectional compatibility is one useful diagnostic. Ask whether the predicate’s participants make sense under the intended domain and word sense.

Appendix N — Final 25 Compatibility Questions

• What kind of subject does this verb normally take?
• What kind of object does it normally take?
• Does the active sense change those preferences?
• Is the noun animate, inanimate, abstract, informational, institutional or propositional?
• Is the event literally possible?
• Is it merely improbable rather than semantically mismatched?
• Is there a technical/domain sense that licenses the combination?
• Is a collective institution standing for people?
• Is a place standing for an institution?
• Is a container standing for its contents?
• Is an author standing for a work?
• Is an object noun coerced into an event reading?
• Is the sentence metaphorical?
• Is it personification?
• Is it metonymy?
• Is it an idiom?
• Is the unusual combination humorous or poetic on purpose?
• Does world knowledge rescue the literal interpretation?
• Does fiction change the ontology of the world?
• Would a near-synonym have the same selectional profile?
• Is the issue really collocation rather than semantic compatibility?
• Could a dictionary example reveal the normal participant class?
• Could corpus examples show whether the pattern is conventional?
• In translation, does the target verb select the same semantic class?
• If writing for clarity, should the predicate be replaced with a more compatible one?

Final Synthesis

Selectional knowledge sits at the junction of vocabulary, grammar, world knowledge and figurative language. It explains why learners can produce sentences that are structurally correct yet feel wrong, why metaphor becomes noticeable, why technical vocabulary narrows meaning so sharply, and why thesaurus substitution can quietly damage a sentence.

A learner does not need to memorise formal semantic features for every verb. They need a practical habit: notice what kinds of subjects and objects repeatedly accompany a word, learn the event frame those participants create, and treat unusual combinations as signals to check sense, domain or figurative interpretation.

That habit makes vocabulary more predictive. Instead of knowing ratify = formally approve in isolation, the learner knows that treaties, agreements and formal decisions are natural objects. Instead of knowing corroborate = support, the learner knows that evidence, testimony and accounts participate in an evidential frame. Instead of knowing ingest = take in, the learner knows the difference between ingesting medicine, ingesting data and the oddity of ingesting a chair.

Vocabulary has become relational rather than definitional. That is the level at which selectional restrictions and preferences become genuinely useful.

Appendix O — When Is an Unusual Combination Actually Good Writing?

Not every selectional violation should be repaired. Creative language often becomes memorable precisely because it places a predicate beside an argument it would not normally select. The test is whether the mismatch creates an interpretable effect rather than accidental confusion.

The city swallowed the rain.

Analysis: personification/metaphor. Why it works: Urban drains and surfaces absorb huge rainfall; ingestion metaphor compresses the event.

Her silence screamed.

Analysis: paradox/personification. Why it works: A non-sound is given a loud communicative force, highlighting significance.

The deadline chased us.

Analysis: personification. Why it works: Time pressure is framed as a pursuing agent.

The argument limped toward its conclusion.

Analysis: manner metaphor. Why it works: Weak reasoning is framed as impaired motion.

The evidence refused to disappear.

Analysis: personification. Why it works: Persistence of evidence is framed as volitional resistance.

The market digested the news.

Analysis: conventional economic metaphor. Why it works: System response is framed as processing food/information.

The idea infected the team.

Analysis: contagion metaphor. Why it works: Rapid interpersonal spread of a concept.

The policy strangled innovation.

Analysis: bodily-force metaphor. Why it works: Constraint is conceptualised as blocked breathing/growth.

Hope took root.

Analysis: plant-growth metaphor. Why it works: An abstract state becomes established like a plant.

The numbers whisper a warning.

Analysis: personification. Why it works: Weak but detectable statistical signal is framed as quiet speech.

A Three-Part Creative-Language Test

1. The ordinary preference is clear. The reader knows what kind of participant the predicate usually selects. 2. The violation supports a recoverable mapping. The reader can explain the intended conceptual transfer. 3. The effect serves the text. It adds precision, imagery, tone or compression rather than merely sounding strange.

If all three conditions hold, an unusual combination may be strong figurative language. If the writer cannot explain the mapping and the reader cannot recover one, the mismatch is more likely lexical error than creativity.

Final Writer’s Rule

Before deleting a strange combination, ask whether it is accidentally incompatible or deliberately figurative. Before keeping it, ask whether the intended metaphor is recoverable. Selectional knowledge gives writers the control to break ordinary expectations on purpose rather than by accident.

Appendix P — Compatibility Is a Gradient, Not a Switch

Selectional fit is often easiest to understand as a gradient. Drink water is highly typical. Drink medicine is ordinary but less prototypical. Drink seawater is semantically compatible but pragmatically unusual. Drink varnish is physically possible but dangerous and unexpected. Drink sunlight fails the ordinary literal liquid requirement but can become poetic. A useful vocabulary system therefore separates semantic type, real-world plausibility, conventionality and figurative interpretation instead of collapsing them into one yes/no judgment.

This gradient view also improves writing feedback. A phrase can be technically interpretable yet unnatural, semantically compatible yet collocationally weak, or metaphorical but insufficiently motivated. The right repair depends on the failure. A learner who writes drink medicine may need no correction at all; a learner who writes drink a chair probably needs a different verb or a deliberate fictional context; a writer who writes drink the silence may be creating imagery intentionally.

The mature question is therefore not simply “Can these words go together?” It is “How naturally do they go together under this sense, domain and world model—and what interpretive work is required if they do not?” That question captures the real value of selectional preferences for vocabulary, comprehension and precise word choice.

Final synthesis: Selectional compatibility becomes durable vocabulary knowledge when learners can predict the kinds of subjects, objects and complements a word normally accepts, notice when a sentence violates those expectations, and decide whether the result is literal but unusual, technically specialised, metaphorical, metonymic, coerced or genuinely anomalous. That ability turns definitions into usable lexical behaviour.

For readers, the benefit is faster interpretation. For writers, it is more precise word choice. For multilingual learners and translators, it is protection against target-language combinations that are grammatically possible but semantically unnatural. For advanced learners, it becomes a way to analyse how creative language deliberately bends ordinary semantic expectations without becoming meaningless.

Selectional knowledge is therefore one of the hidden systems behind natural vocabulary use. Learners rarely name it in ordinary conversation, yet they use it every time they decide that a verb, adjective or noun “fits” one context better than another. Making that system visible gives them a practical way to diagnose awkward phrasing, interpret metaphor, learn technical senses and choose words with greater semantic control.

That control is the practical goal: recognise ordinary semantic fit quickly, notice meaningful violations, and use those judgments to read, write and learn vocabulary with greater precision.

Semantic fit becomes usable knowledge.