An advanced vocabulary agency casebook: nominalisation, passive voice, event nouns, impersonal constructions, responsibility, causation and the hidden question “who did what?”

The sentence is polished:

A decision was made to change the schedule.

Maya reads it once and moves on.

Ethan does not.

“Who made the decision?”

The sentence does not say.

Jia Jun tries another:

The cancellation of the session caused considerable inconvenience.

Hana looks at the sentence.

“Who cancelled it?”

Again, the grammar does not tell us.

Neither sentence is automatically bad English. In fact, both structures are common and often useful in formal writing. Cambridge Grammar explains that English frequently forms nouns from verbs and adjectives, a process called nominalisation, and that formal styles often use noun phrases and passive constructions to create a more detached, impersonal tone. British Council likewise notes that passive voice can shift focus away from the doer when the agent is unknown, obvious, unimportant or deliberately not mentioned.

The advanced problem begins when that flexibility changes what the reader can see.

The committee changed the schedule.

We can see the actor.

The schedule was changed.

The actor is absent.

A change to the schedule was made.

The action itself has become a noun phrase, and the actor remains absent.

There was a change to the schedule.

Now the event can sound almost as if it simply appeared.

This article owns one narrow applied job in the eduKatePunggol vocabulary estate: how readers and writers track agency when English turns actions into nouns, moves affected objects into subject position, uses impersonal reporting, or leaves the doer unstated.

The broad specialist theory already exists elsewhere in the eduKate ecosystem. How English Works | Nominalisation and Event Nouns owns how English turns actions into things. Nominalisation in English owns the broader academic-writing mechanism. eduKateSengkang also has a Chinese-edition advanced-vocabulary lesson on nominalisation and word-class shifts. This Punggol article deliberately does not recreate those owners.

The Punggol job is the practical audit:

• who acted?
• who decided?
• who caused the change?
• who is affected?
• what event has been turned into a noun?
• what information disappeared when the grammar changed?
• is hiding the actor useful, harmless or misleading?
• would restoring the agent make the sentence clearer, fairer or more accountable?

This continues from Advanced Vocabulary Synthesis | The Contrast Audit, Advanced Vocabulary Argument | The Qualification Audit, Advanced Vocabulary Reading | The Assumption Audit and Advanced Vocabulary Editing | The Counterexample Test.

The teaching cases below are constructed examples rather than reports of actual eduKate students, families, institutions or measured interventions. The linguistic principles are grounded in standard descriptions of nominalisation and passive voice. The Hidden-Agent Audit itself is an original classroom application for making agency visible.

The fifty-second route: restore the event before judging the sentence

When a sentence feels abstract, formal or oddly detached, do this:

1. Find the event noun or passive verb. Examples: decision, closure, increase, reduction, failure, implementation, was changed, was reported, was completed.
2. Turn the noun back into a verb. Decision → someone decided. Closure → someone closed something. Reduction → something or someone reduced something.
3. Ask who or what performed the action. Is the agent stated, obvious, unknown, irrelevant or missing?
4. Ask who or what received the action. What changed, was closed, was reduced, was denied, was moved?
5. Restore the timeline. When did the action happen? Was it completed, attempted, proposed or merely reported?
6. Restore the source. Who knows this? Who reported it?
7. Decide whether the hidden agent matters for this task. Sometimes it does not. Sometimes it is the whole point.

The audit is not “active voice good, passive voice bad”.

British Council explicitly notes useful reasons for passive voice: the most important information may be the thing affected; the agent may be unknown, obvious or unimportant; and formal or scientific writing often focuses on processes. Cambridge likewise notes that passives without agents are very common.

The audit asks a narrower question:

What did the sentence gain by hiding the agent—and what did it lose?

Reader routes: Agency map · Nominalisation laboratory · Passive laboratory · Responsibility language · Scientific and technical writing · AI hidden-agent drift · Extended case files · Independent practice · Sources.

The agency map: five common ways English can move the doer out of sight

There is no single “hidden agent grammar”. Several structures can reduce the visibility of the doer.

1. Passive without an agent

The schedule was changed.

The affected thing becomes the grammatical subject. The agent is omitted.

Restore:

The organiser changed the schedule.

Whether that restoration is appropriate depends on whether the source actually tells us the organiser acted.

2. Nominalisation

The cancellation of the session caused inconvenience.

Cancellation packages the event as a noun.

Possible unpacking:

When the organiser cancelled the session, families were inconvenienced.

But that version adds an organiser unless the source identifies one. A safer diagnostic version is:

Someone cancelled the session, and that inconvenienced families.

The awkward someone is useful during analysis because it reveals the missing slot.

3. Existential or event-presenting structure

There was a reduction in practice time.

The sentence presents the reduction as an event or state.

Possible unpackings:

• Practice time fell.
• The new timetable reduced practice time.
• The student reduced practice time.
• The school reduced practice time.

Those alternatives describe different causal structures.

4. Impersonal reporting

It is believed that the new method is more effective.

Who believes it?

British Council describes this kind of impersonal passive with reporting verbs such as believe, claim, estimate, expect, know, report, say, think and understand.

The construction is useful when the source group is unimportant or widely understood. It is weak when source credibility is central.

5. Abstract evaluative nouns

The failure led to a review.

What failed?

Who defined it as failure?

The improvement was welcomed.

What improved?

By what measure?

Who welcomed it?

Not every abstract noun comes from a verb in the same simple way, but many perform similar discourse work: they compress an event or judgement into a noun phrase that can then behave like a settled thing.

The first principle: hiding the agent can be useful

Consider a laboratory instruction:

The solution was heated to 60°C.

The important information may be the procedure, not the identity of the student holding the beaker.

Or a notice:

All forms must be submitted by Friday.

The actor is the reader or applicant and may be obvious from context.

Or a damaged object:

The window was broken overnight.

If nobody knows who broke it, the passive accurately represents the information available.

Or formal focus:

The final draft was reviewed by Hana.

The passive keeps the paragraph topic on the draft while still naming the agent in a by-phrase.

The mature question is therefore not:

Is this passive?

It is:

Does the reader need the agent here?

The second principle: formal does not automatically mean clearer

Cambridge notes that nominalisation is common in formal writing because it allows writers to use noun phrases where a more conversational version might use verbs.

Compare:

Trade grew after 1900.

There was growth in trade after 1900.

The second sounds more formal. It is not automatically more precise.

Formal noun-heavy style becomes useful when the noun phrase helps organise ideas:

The rapid growth in trade after 1900 altered the region’s transport priorities.

Now growth lets the writer treat the whole earlier process as one topic and connect it to a later consequence.

That is real information compression.

The danger is confusing compression with precision.

The implementation of changes produced improvements.

Sounds formal.

But who implemented what?

What changed?

What improved?

By what measure?

The sentence has high abstraction and low inspectability.

Advanced vocabulary control aims for the opposite: compress only after the underlying event is clear.

Laboratory A: nominalisation — when actions become things

Cambridge defines nominalisation as forming a noun from another part of speech, especially a verb or adjective. In formal writing this can make whole events easier to package, modify, compare and refer back to.

Verb:

The team decided to delay the launch.

Nominalised:

The team’s decision to delay the launch…

The noun decision lets the writer treat the whole act as a thing that can become the subject or object of another sentence:

The team’s decision to delay the launch reduced the risk of releasing an incomplete product.

This is useful compression.

But remove the possessor:

The decision to delay the launch reduced risk.

Who decided?

If the previous paragraph makes the actor obvious, no problem.

If responsibility for the decision matters, the omission matters.

Case 1. “The decision” can hide the decider

Source:

Hana decided to remove the final example because it duplicated an earlier case.

Rewrite:

The removal of the final example improved the section.

The new sentence performs two compressions.

• Hana decided to remove becomes the removal.
• The reason for removal disappears.

It also adds an evaluation: the section improved.

If the task is a concise report of the final document, that may be acceptable.

If the task is to explain editorial reasoning, the compression is damaging.

A strong diagnostic rewrite is:

Hana removed the final example because it duplicated an earlier case, which made the section less repetitive.

Now actor, action and reason remain visible.

Case 2. “The increase” hides what increased and why

Sentence:

The increase led to difficulties.

This is almost empty without context.

Increase in what?

How much?

Over what period?

What caused it?

What difficulties?

Nominalisation often depends on earlier context. If the previous sentence says:

Class size rose from eight to sixteen students after two groups were merged.

then:

The increase created longer waiting times for individual feedback.

is cohesive and useful.

The earlier sentence supplies the event. The noun phrase compresses it for the next step.

The audit therefore checks the paragraph, not just the isolated sentence.

Case 3. “Failure” can turn a disputed judgement into a noun

Compare:

The programme did not reach its enrolment target.

The programme failed.

The failure of the programme led to a review.

The third sentence packages failure as an established thing.

But a programme can miss one target and meet others.

The noun therefore needs a criterion.

A safer version is:

Because the programme missed its enrolment target, the team reviewed its outreach strategy.

Now the event and criterion are explicit.

If the organisation itself officially defined missing the target as programme failure, then failure may be justified—but the source of that evaluation should still be recoverable.

Case 4. “Implementation” may hide who implemented what

Sentence:

The implementation of the new system led to improved efficiency.

The sentence sounds professional.

But it carries several hidden variables.

• Who implemented it?
• What exactly changed?
• What efficiency measure improved?
• Did implementation cause the improvement, or did the improvement merely follow it?

A more inspectable sentence is:

After the team automated duplicate data entry, average processing time fell from nine minutes to six.

Now action, actor, mechanism and outcome are visible.

The formal noun phrase was not wrong. It was under-specified.

Case 5. “The refusal” can hide whose refusal and what was refused

The refusal caused frustration.

Who refused?

Refused what?

Frustrated whom?

Now compare:

The organiser refused Maya’s request to move the session, which frustrated her because the new time clashed with school.

This version may be too detailed for a summary, but it reveals the event structure.

A concise but accountable version is:

The organiser’s refusal to move the session frustrated Maya.

Nominalisation does not require deleting the agent. Possessive phrases and of-phrases can keep agency visible.

Case 6. “Improvement” can conceal the measurement

The improvement was significant.

Improvement in what?

The noun already assumes a positive change.

Then significant adds either ordinary importance or, in some contexts, statistical meaning.

A more transparent sentence might be:

Average retrieval accuracy rose from 62% to 74% across the two practice sets.

Now the reader can see the change directly.

If the task later needs a noun phrase:

This twelve-point increase was concentrated among the words reviewed twice.

The noun phrase becomes useful because the measurement has already been established.

Case 7. “The rejection” can make responsibility visible or invisible

Compare:

• The committee rejected the proposal.
• The proposal was rejected by the committee.
• The committee’s rejection of the proposal…
• The rejection of the proposal…

All four can describe the same decision.

They distribute attention differently.

The first foregrounds the committee.

The second foregrounds the proposal but keeps the committee in a by-phrase.

The third turns the whole decision into a noun while retaining the actor.

The fourth removes the actor entirely.

No single form is always best.

The Hidden-Agent Audit asks what the paragraph needs to keep visible.

Case 8. “The discovery” can hide the discoverer and the path to discovery

The discovery of the error led to a correction.

This sentence packages two events:

• someone discovered an error;
• someone corrected something.

Possibly the same person.

Possibly not.

If accountability for checking matters:

Ethan found the copied value error during his final check and corrected it before submission.

If the report cares only about process:

The error was detected and corrected before submission.

Both are legitimate under different briefs.

The unpacking test: turn every abstract event noun back into a sentence

For a difficult paragraph, underline nouns ending in forms such as:

• -tion / -sion: decision, reduction, expansion, rejection;
• -ment: improvement, development, implementation;
• -al: refusal, approval, removal;
• -ance / -ence: resistance, acceptance, dependence;
• -ure: failure, closure, exposure;
• -ing forms functioning as nouns: monitoring, testing, checking.

These suffixes are clues, not a complete definition of nominalisation.

Then unpack:

The reduction in errors…

becomes:

Errors decreased…

or:

The checking routine reduced errors…

Those two unpackings are not equivalent.

The first reports a change.

The second adds a cause.

The audit therefore separates event recovery from causal attribution.

Laboratory B: passive voice — changing focus without automatically changing truth

Cambridge describes passive voice as a way of organising a clause so that the person or thing affected by the action becomes the subject or topic. The doer can be omitted or added in a by-phrase.

Active:

Maya checked the labels.

Passive with agent:

The labels were checked by Maya.

Passive without agent:

The labels were checked.

The first two can preserve the same core event while changing focus.

The third removes the actor from the sentence.

Case 9. Passive is useful when the affected object is the paragraph topic

Paragraph:

The final draft contained two duplicated examples. The draft was reviewed by Hana, and both duplicates were removed before publication.

The passive keeps the paragraph focused on the draft and examples.

An active version:

Hana reviewed the draft and removed both duplicates before publication.

is shorter and foregrounds Hana.

Neither is universally better.

The paragraph’s information flow determines the choice.

Case 10. Passive is useful when the agent is genuinely unknown

The classroom window was broken overnight.

If nobody knows who broke it, an active sentence such as Someone broke the window adds no useful information.

The passive is efficient.

But later evidence may identify the actor.

Then a report can update:

Maintenance staff later found that a loose exterior panel had struck the window.

The correct grammar changes as the information state changes.

Case 11. Passive can hide responsibility when the agent is known and relevant

The wrong file was uploaded.

Who uploaded it?

If the purpose is to fix the immediate problem, the actor may not matter.

If the purpose is to understand the workflow failure, the actor and process may matter:

Jia Jun uploaded the earlier draft because both files had the same visible filename.

Now the report can distinguish personal action from system design.

A blame-only rewrite would be poor analysis:

Jia Jun caused the upload failure.

The full event includes a confusing file system.

Restoring the agent should improve causal clarity, not merely assign blame.

Case 12. A by-phrase can restore agency while keeping passive focus

The decision was made by the organising team after reviewing the updated attendance list.

The sentence keeps the decision as the topic while naming the agent.

This can be useful when a paragraph is already about the decision rather than the team.

Active alternative:

The organising team made the decision after reviewing the updated attendance list.

Both preserve agency.

The choice is information structure, not moral virtue.

Case 13. Passive can conceal source in reporting verbs

It is believed that the new schedule will improve attendance.

Who believes this?

Teachers?

Parents?

Researchers?

The construction may be acceptable if a previously identified group is obvious:

The organising team reviewed last year’s attendance data. It is believed that moving the session earlier will reduce late arrivals.

Even here, the source could be clearer:

The team believes that moving the session earlier will reduce late arrivals.

When evidence quality depends on who holds the belief, explicit attribution is usually better.

Case 14. Passive can turn an instruction into a process focus

The samples should be labelled before they are stored.

This is natural procedural language.

The actors may be the technicians or students already addressed by the procedure.

Active:

Label the samples before storing them.

This is clearer as a direct instruction.

The genre decides.

Method section: passive may fit.

Student checklist: imperative may be better.

Case 15. “Mistakes were made” can hide who made them—but sometimes that is deliberate conflict management

Sentence:

Mistakes were made during the preparation.

The agent is absent.

In an accountability report, that may be inadequate.

In a first-stage team debrief designed to identify process failures without triggering immediate personal defensiveness, the temporary agentless form can be useful:

Several version-control mistakes occurred during preparation. We will trace where the workflow allowed them before assigning corrective actions.

The key word is temporary.

A mature report can first stabilise the event, then restore responsibility where evidence supports it.

Agency should not be hidden forever merely to keep the tone comfortable.

The passive audit: four questions before you rewrite

1. Is the agent known?
2. Is the agent relevant to the reader’s task?
3. What should be the paragraph topic—the actor or the affected thing?
4. Would adding the agent improve explanation, accountability or source evaluation?

If the agent is unknown or irrelevant, keep the passive.

If the agent is central to responsibility or causation, restore it.

Laboratory C: responsibility language — when grammar changes who appears answerable

Agency matters most when the reader is trying to understand responsibility.

Compare:

• I did not submit the form.
• The form was not submitted.
• There was a failure to submit the form.

The first sentence names the responsible actor.

The second foregrounds the missing outcome.

The third turns the non-submission into an abstract event.

All three can be useful under different conditions.

The Hidden-Agent Audit becomes important when the grammatical choice makes responsibility easier or harder to see.

Case 16. A family sentence can move from accountability to accusation

Parent:

Your homework was not completed.

The passive describes the state of the homework.

If the child was responsible for completing it, that responsibility is understood but unstated.

A more direct version is:

You did not complete the homework.

That sentence can be appropriate when the fact is clear.

But now compare:

You chose not to complete the homework.

This adds intention.

The child may have tried and failed, misunderstood the instructions, run out of time or forgotten.

Restoring agency does not give the writer permission to invent motive.

A strong parent sentence is:

The homework is incomplete. Tell me what happened before we decide what needs fixing.

The event is visible; responsibility is investigated rather than guessed.

Case 17. “The deadline was missed” can protect the wrong layer

The deadline was missed.

Who missed it?

Was the deadline communicated?

Did the submission system fail?

Was the task completed but uploaded late?

One passive sentence can conceal several distinct failure mechanisms.

For a diagnostic report:

Hana completed the work before the deadline but uploaded it twenty minutes late after the submission page failed to load.

Now the missed deadline is decomposed into completion, upload, timing and system failure.

This is much more useful than assigning the global label late submission and stopping there.

Case 18. “Concerns were raised” can hide who is concerned

Concerns were raised about the new schedule.

By whom?

One parent?

Several families?

Teachers?

An administrator?

The passive may be useful in a short meeting note where the participants are already known.

It is weak in a public summary if the scale and source of concern matter.

Better:

Three parents raised concerns about the earlier start time.

Now the reader can judge scope.

This connects to the Qualification Audit: agency and population boundaries often work together.

Case 19. “Approval was granted” can hide who has authority

Approval was granted for the change.

This may be enough if the approving authority is obvious.

If the question is whether the approval was valid, the agent matters.

The programme coordinator approved the change.

Now the reader can compare the actor with the formal decision rules.

In organisational writing, agentless approval language can make authority look automatic.

The Hidden-Agent Audit asks:

Who had the power to make this action valid?

Case 20. “Responsibility was shared” can blur unequal roles

Responsibility was shared across the team.

This can be accurate.

It can also hide different levels of control.

Perhaps:

• Maya prepared the data;
• Ethan checked the calculations;
• Hana approved the final version;
• Jia Jun uploaded it.

If an error appears in the uploaded file, “shared responsibility” may be too broad to diagnose what happened.

A good report maps role to action.

Responsibility can still be shared at the system level while particular actions remain attributable.

Those two statements are not contradictory.

Case 21. “An error occurred” can be better than premature blame

Imagine a calculation is wrong and nobody yet knows why.

An error occurred in the final total.

This is an appropriately neutral starting point.

It avoids accusing someone before the process is traced.

Later, the investigation finds:

The spreadsheet imported one column as text, and Ethan copied the displayed subtotal without noticing the omitted values.

Now the causal chain has both system and human components.

The correct final report should not remain at an error occurred once the mechanism is known.

Agentless language can be appropriate at the beginning of inquiry and inadequate at the end.

The accountability ladder: from event to actor to mechanism

A mature report can move through three levels.

Level 1: event

The wrong file was uploaded.

Level 2: actor

Jia Jun uploaded the wrong file.

Level 3: mechanism

Jia Jun uploaded the earlier draft because both files displayed the same truncated name and the folder contained no version marker.

The first level stabilises the event.

The second identifies the human action.

The third explains how the system allowed the action to produce the outcome.

Stopping at Level 2 can create blame without learning.

Stopping at Level 1 can create vagueness without accountability.

Strong agency writing reaches the level the task requires.

Laboratory D: scientific, mathematical and technical writing — when impersonal language is doing real work

Passive voice and nominalisation are common in scientific and technical prose for good reasons.

Processes may matter more than individual people.

Methods should often be reproducible regardless of who performs them.

Technical concepts need to be packaged so they can become objects of later reasoning.

The problem begins when impersonal style hides variables that the reader actually needs.

Case 22. Procedure focus can be appropriate

The solution was heated to 60°C for three minutes before the indicator was added.

This is clear procedural writing.

The person performing the action is irrelevant to reproducing the procedure.

An active alternative:

We heated the solution to 60°C for three minutes before adding the indicator.

is also common in many modern scientific styles.

The discipline or publication convention decides which form sounds more natural.

The Hidden-Agent Audit does not impose one house style on all science.

Case 23. “The increase in temperature caused…” can be useful when temperature is the causal variable

The increase in temperature caused the gas to expand.

Here the nominalised phrase increase in temperature names a physical change, not a hidden human decision.

Unpacking:

As the temperature increased, the gas expanded.

Both can be useful.

The noun phrase may help when the paragraph later compares the temperature increase with the pressure increase.

Agency is not always human.

The audit asks what causal participant is hidden or compressed, whether human or physical.

Case 24. Mathematics often prefers process nouns once definitions are stable

The substitution of x = 3 into the equation gives y = 11.

Nominalisation packages the procedure.

For a beginner, active instruction may be clearer:

Substitute x = 3 into the equation. Then y = 11.

The more experienced learner can handle the noun phrase because the operation is already conceptually stable.

This creates a teaching principle:

Use verbs to teach the action; use nouns to package the action after the learner understands it.

Case 25. Scientific nouns can hide measurement choices

The improvement in performance was recorded.

Performance of what?

Measured how?

Compared with what?

Technical nouns often sound objective because they are abstract.

Objectivity still requires operational definitions.

Better:

Average completion accuracy rose from 71% to 79% on the second test.

Now the measurement is visible.

Later:

This eight-point improvement was concentrated in inference questions.

Nominalisation becomes useful after definition.

Case 26. “It was found that…” can hide the researcher without harming the result

It was found that the second method produced fewer errors.

The sentence does not name the researchers.

That may be fine when the paragraph is already discussing one study.

But if several studies disagree, source identity matters:

Study A found fewer errors with Method 2, whereas Study B found no difference.

Impersonal reporting is strongest when provenance is already clear.

Case 27. Procedure nouns can hide order

After preparation and measurement, analysis was conducted.

What was prepared?

What was measured?

Who analysed what?

For an expert audience, the omitted detail may be supplied by a preceding method section.

For a learner, the sentence may be too compressed.

Unpack:

Students prepared the samples, measured their mass and then analysed the results.

The sequence becomes visible.

This is why noun-heavy academic style can increase reading load even when it is concise on the page.

The technical-writing rule: compress after the mechanism is established

Good technical writing often moves in two stages.

First, establish the event clearly:

When the temperature increased from 20°C to 40°C, the reaction rate doubled.

Then package it:

This temperature increase was associated with a twofold rise in reaction rate.

The second sentence can now refer compactly to the established event.

Readers are not forced to reconstruct an event they have never seen.

Laboratory E: AI hidden-agent drift — when a rewrite makes agency disappear or appear

AI-assisted rewriting creates two opposite risks.

• It can remove an agent and make a sentence sound more formal or neutral.
• It can invent an agent while trying to make an abstract sentence clearer.

Both change the evidential structure.

Case 28. AI removes the actor because passive sounds “professional”

Source:

The tutor changed the practice sequence after reviewing the marked paper.

Generated rewrite:

The practice sequence was adjusted following a review of the marked paper.

The rewrite sounds polished.

It removes the tutor.

It also changes changed to adjusted, which may sound smaller or more technical.

If the article is explaining how a tutor diagnoses and responds, removing the tutor damages the mechanism.

Better:

After reviewing the marked paper, the tutor adjusted the practice sequence.

Clarity improves without losing agency.

Case 29. AI invents an actor to unpack nominalisation

Source:

The cancellation caused families to rearrange their evening plans.

Generated rewrite:

The school cancelled the session, forcing families to rearrange their evening plans.

The rewrite may be easier to read.

It also invents the school as the cancelling agent.

If the source never identifies who cancelled the session, the AI has supplied a fact.

Safer diagnostic rewrite:

When the session was cancelled, families had to rearrange their evening plans.

The event is verbal again, but the unknown agent remains unknown.

This is a crucial distinction:

unpack the event without filling an empty evidence slot.

Case 30. AI turns a system failure into a person failure

Source:

The submission was delayed after the upload page repeatedly timed out.

Generated rewrite:

The student submitted the work late because they did not upload it on time.

This is a major causal distortion.

The source says the upload page timed out.

The rewrite recasts a system event as personal failure.

AI can therefore add agency in the wrong place, just as easily as it can remove it.

A faithful version is:

The student attempted to submit the work before the deadline, but repeated upload-page timeouts delayed the final submission.

Now both human action and system mechanism are visible.

Case 31. AI can erase source ownership in academic synthesis

Source:

One study reported that the method reduced errors.

Generated rewrite:

Error reduction was achieved through the method.

The rewrite removes the source and upgrades a reported finding into an apparently direct fact.

It also nominalises reduced errors into error reduction.

This is exactly the kind of polished drift the Assumption Audit and Hidden-Agent Audit are designed to catch together.

Better:

One study reported fewer errors after the method was introduced.

Even this wording should be checked for causality. If the source says associated with, preserve that relationship.

Case 32. AI can overcorrect passive into awkward actor-heavy prose

Source method sentence:

The samples were stored at 4°C overnight.

Generated “active voice” rewrite:

The researchers stored the samples at 4°C overnight.

This is grammatical and faithful if researchers performed the storage.

But it may be unnecessary.

The method cares about storage conditions, not researcher identity.

A rule telling AI to “remove all passive voice” would therefore make technical prose noisier.

The correct instruction is not:

Use active voice.

It is:

Restore agents when they matter to causation, accountability, source or clarity; keep agentless process focus where the agent is irrelevant or already obvious.

A practical AI prompt for agency-preserving revision

Use:

Rewrite for clarity without changing agency. Preserve every stated actor, source and cause. Do not invent an actor when the source leaves the agent unknown. Do not remove an actor when responsibility, causation or provenance depends on that person or group. Keep passive voice when the agent is irrelevant, unknown or already obvious and the process or affected object is the paragraph focus. Flag any change from active to passive, passive to active, or verb to nominalisation before applying it.

Then inspect the output manually.

Cross-subject agency: what gets hidden in English, Mathematics, Science and Humanities

English: interpretation can hide the interpreter

The character’s behaviour is seen as selfish.

Seen by whom?

The narrator?

The student?

A critic?

Better analytical writing usually names the source of interpretation:

The repeated refusal to share information presents the character as self-interested in this scene.

Now the textual mechanism is visible.

Or:

A reader may interpret the refusal as selfish because the character withholds information that would help the group.

The interpreter is now explicit.

Mathematics: “an error was made” can hide the first wrong operation

An error was made in the solution.

Too broad for diagnosis.

Better:

At line three, Ethan distributed the negative sign to the first term but not the second.

Now the student can repair the exact operation.

After diagnosis, compression becomes useful:

This sign-distribution error then propagated through the remaining calculation.

The event noun now has a clear referent.

Science: “contamination occurred” may need a process trace

Contamination occurred during preparation.

What contaminated what?

Was the source airborne exposure, an unclean tool, cross-sample transfer or a storage problem?

If the source is unknown, the agentless event is honest.

If the process trace finds that one pipette tip was reused, the final report should say so.

The point is not blame.

The point is reproducibility.

Humanities: policy nouns can hide decision-makers

The introduction of the rule changed participation.

Who introduced it?

Who had authority?

Who was affected?

A more analytical sentence is:

When the organisers introduced the registration rule, participation fell among people who could not attend the earlier sign-up period.

Now institutional action and affected group are visible.

The same principle applies to historical reading: abstract nouns such as expansion, resistance, reform, industrialisation, migration are powerful analytical categories, but students should still be able to unpack who moved, who resisted, who changed institutions and who experienced the consequences.

The agency-density test

Take a paragraph and count how many events appear as nouns or agentless passives.

Example:

Following implementation, an improvement in consistency was observed. A reduction in processing time was also recorded. Concerns were raised about the loss of flexibility, leading to further modification.

The paragraph contains many events and almost no actors.

Unpack:

After the team introduced the new workflow, reviewers applied the checklist more consistently and average processing time fell. Several staff members then reported that the workflow reduced their flexibility, so the team revised two steps.

The second version is longer but easier to inspect.

A final polished paragraph might reintroduce one or two useful nominalisations:

After the team introduced the new workflow, checklist use became more consistent and average processing time fell. Staff concerns about reduced flexibility then led the team to revise two steps.

This is the target.

Not zero nominalisation.

Enough visible agency to understand the system.

Extended case files: restore agency without turning every report into blame

Case file 1. The late homework story

Constructed source packet. Ethan begins homework at 7 p.m. The instructions say, Explain how the evidence supports your conclusion. He interprets supports as “agrees with” and writes a descriptive answer. His parent notices the problem at 8:15 and explains the command word. Ethan rewrites the answer but finishes after the family’s planned bedtime.

Summary A:

The homework was completed late.

True but shallow.

Summary B:

Ethan failed to manage his time and completed the homework late.

Agency is visible, but the causal diagnosis is poor. The packet shows a language misunderstanding that delayed the work.

Summary C:

Ethan began on time but misread the command word supports, which led him to produce the wrong answer type. After the language problem was identified, he rewrote the response and finished later than planned.

This version restores actor, mechanism and timeline without turning the report into excuse-making.

The family can now intervene at the right place: academic command vocabulary, not generic time-management punishment.

Case file 2. The missing file

Constructed packet. Maya exports a document at 4:03 p.m. The system saves it automatically to a downloads folder. Jia Jun opens the shared folder, sees only the previous version and assumes the export failed. He tells Hana, “The file wasn’t saved.” Ethan later finds the new file in downloads.

Agentless report:

The file was not saved correctly.

But the file was saved.

The error was in location expectation, not storage.

Blame-heavy report:

Jia Jun wrongly reported that Maya had failed to save the file.

This identifies his mistaken report but can over-personalise a workflow ambiguity.

Mechanism report:

The export succeeded, but the software saved the file to downloads rather than the shared folder. Jia Jun checked only the shared folder and therefore concluded that the export had failed.

Now both system behaviour and human inference are visible.

The best corrective action might be changing the export path or naming workflow, not telling Jia Jun to “be more careful”.

Case file 3. The science result that becomes too impersonal

Constructed experiment. Four groups test the same reaction at 40°C. Three groups record similar times. One group records a much shorter time because they start the stopwatch before adding the final reagent rather than after, contrary to the intended procedure.

Report:

An anomalous result was obtained.

This identifies the outcome but not the mechanism.

Better:

Group 4 recorded a shorter reaction time because they started timing before adding the final reagent, while the other groups started timing afterwards.

The event is now traceable.

For a final concise discussion:

The anomalous Group 4 result can be explained by a timing-procedure difference rather than by the temperature condition itself.

The nominalisation timing-procedure difference is useful because the mechanism has already been established.

Case file 4. The parent message where agency needs tact

Constructed situation. A group project is submitted without one student’s section because two students believe the other has merged it. The teacher returns the incomplete file.

Message A:

Your child failed to include the section.

Too strong if the merging responsibility was shared or unclear.

Message B:

The section was omitted from the submitted file.

Accurate event description, but too vague if the purpose is repair.

Message C:

The submitted file did not contain the final section. The two students responsible for merging the document each believed the other had inserted it, so neither verified the final version before submission. We are adding a final-file check to prevent the same handoff error next time.

This message names responsibility without inventing malicious intent or collapsing a shared-process failure into one child.

Agency is visible.

Blame is proportionate.

The repair is concrete.

Case file 5. The academic summary where the noun becomes stronger than the source

Constructed source. One small study reports fewer editing errors after a checklist is introduced. The study uses a different passage at post-test and no comparison group.

Student summary:

The effectiveness of the checklist led to an improvement in editing accuracy.

The nominalisations sound academic:

• effectiveness;
• improvement;
• accuracy.

They also overstate the evidence.

Effectiveness assumes the checklist worked.

Led to adds causality.

Improvement in editing accuracy broadens the outcome beyond the specific error count.

Better:

In the study, participants made fewer editing errors on a later passage after using the checklist. Because the passages differed and there was no comparison group, the result does not establish that the checklist caused the reduction.

Less noun-heavy.

Much more faithful.

Case file 6. The operational report where passive is exactly right

Constructed process. A lab requires samples to be labelled, sealed and refrigerated within ten minutes of collection. Several trained technicians can perform the procedure.

Strong method sentence:

Each sample was labelled, sealed and refrigerated within ten minutes of collection.

The agent is irrelevant.

The important information is whether the procedure occurred correctly and on time.

An actor-heavy rewrite:

A technician labelled each sample, a technician sealed each sample, and a technician refrigerated each sample within ten minutes.

adds nothing useful unless technician identity itself affects reliability.

This case prevents the audit from becoming anti-passive dogma.

Good agency control includes knowing when the agent should stay hidden.

Independent practice: thirty hidden-agent audits

The following items mix good passive voice, harmful vagueness, useful nominalisation, missing sources, invented agents and over-personalised blame. The aim is not to convert everything into active voice. The aim is to decide what the reader needs to see.

For each item, do four things:

1. identify the event;
2. identify whether an agent exists in the sentence;
3. decide whether the agent matters for this task;
4. rewrite only if the current sentence hides information the reader needs.

Practice set A: should the agent be restored?

1. The timetable was changed on Tuesday. The reader is deciding who approved the change.

2. The samples were stored at 4°C overnight. The reader is reproducing the laboratory method; technician identity is irrelevant.

3. Concerns were raised about the earlier start time. The article is trying to describe how widespread the concern was.

4. The classroom window was broken overnight. Nobody yet knows what caused the break.

5. The proposal was rejected. The question asks which group had authority to reject it.

6. All forms must be submitted by Friday. The instruction is addressed directly to applicants.

7. The wrong file was uploaded. The team is trying to identify why the workflow failed.

8. It is believed that the second method is more reliable. Two studies disagree about reliability.

9. The labels were checked by Maya. The paragraph topic is the labels, and the checker matters.

10. The door was opened at 8 a.m. The schedule report only needs to establish opening time, not who opened it.

Practice set B: unpack the nominalisation without inventing facts

11. The cancellation caused families to rearrange their plans. The source does not say who cancelled the session.

12. The committee’s rejection of the proposal delayed the project. The committee is explicitly identified.

13. The improvement in accuracy was significant. The source gives scores of 61% and 72% but no statistical test.

14. The implementation of the new system reduced processing time. The source says the operations team automated duplicate data entry and processing time fell from nine minutes to six.

15. The failure led to a review. The programme missed its enrolment target but met its attendance target.

16. The discovery of the error prevented publication. Ethan found the error during his final check.

17. The reduction in practice time affected fluency. The source only shows that practice time fell; it does not establish the cause of lower fluency.

18. The approval of the change allowed the event to proceed. The approval was given by the programme coordinator and that authority matters.

19. The resistance to the proposal increased. The source says six parents objected after a new condition was announced.

20. The testing of the system was completed before launch. Several engineers performed the tests; their individual identities do not matter.

Practice set C: agency, blame and mechanism

21. A student begins homework on time, misunderstands one command word, rewrites the answer and finishes late. The report says: The homework was completed late because the student managed time poorly.

22. A webpage times out repeatedly before the deadline. The report says: The student failed to submit on time.

23. Two students each believe the other has merged the final section. The report says: The section was omitted from the final file. The purpose is to prevent recurrence.

24. A spreadsheet imports one column as text. Ethan copies the displayed subtotal without noticing the omitted values. The report says: Ethan caused the calculation failure.

25. A laboratory group starts timing before adding the final reagent. The report says: An anomalous result was obtained. The purpose is to explain why the result differed.

26. A team deliberately uses passive wording in an early debrief because the exact cause is not yet known. Is that automatically evasive?

27. A final investigation has identified who performed each step and how the workflow produced the error. The published report still says only: Mistakes were made. Is the agentless form still adequate?

28. An AI rewrite changes The tutor changed the sequence after reviewing the paper to The sequence was adjusted following a review. The article’s topic is how the tutor diagnoses and responds.

29. An AI rewrite changes The cancellation caused inconvenience to The school cancelled the session, causing inconvenience. The source does not identify the cancelling agent.

30. An editor changes The samples were refrigerated overnight to The researchers refrigerated the samples overnight. The method section is concerned only with storage conditions. Is the active rewrite an improvement?

Answer commentary: restore only what the evidence and task require

1. Restore the decision-maker. The question is about authority, so The organising team changed the timetable on Tuesday or The timetable was changed on Tuesday by the organising team is more useful, assuming the source identifies that group.

2. Keep the passive. The storage condition matters; technician identity does not. Active voice would add a person without improving reproducibility.

3. Restore the source and scope. Write Three parents raised concerns about the earlier start time if that is what the evidence shows. The original passive hides the population needed for interpretation.

4. Keep the unknown agent unknown. The classroom window was broken overnight is appropriate. Do not invent a person, accident or cause before evidence arrives.

5. Restore authority. If the committee rejected it, say so. When the reader is judging decision legitimacy, the actor is central.

6. Keep the passive instruction. Applicants already know who must submit the forms. The affected object and deadline are the useful focus.

7. Expand to actor plus mechanism. Jia Jun uploaded the earlier draft because both files displayed the same shortened name is better than either pure blame or pure vagueness. Then fix the workflow.

8. Name the source. When studies disagree, It is believed is too vague. Write Study A reports greater reliability, whereas Study B reports no difference.

9. Keep the passive with agent. The sentence successfully combines draft-focused information structure with visible responsibility.

10. Keep the passive. Opening time is the relevant event. Naming the staff member would distract from the schedule unless the person matters for another reason.

11. Unpack without inventing an agent. When the session was cancelled, families had to rearrange their plans. Do not write The school cancelled… unless the source says so.

12. Nominalisation is fine. The committee’s rejection keeps the agent visible while packaging the decision as a noun. It is a useful formal structure.

13. Replace vague evaluation with the measurement. Accuracy rose from 61% to 72%. Do not use significant technically unless statistical significance was actually established.

14. Unpack the mechanism. After the operations team automated duplicate data entry, processing time fell from nine minutes to six. This is more informative than the generic implementation sentence.

15. Avoid packaging a disputed overall judgement as fact. State the actual criterion: After the programme missed its enrolment target, the team reviewed its outreach strategy.

16. Restore the discoverer if the checking process matters. Ethan found the error during his final check, preventing publication of the incorrect version.

17. Separate change from causation. If the evidence only shows reduced practice time and lower fluency, write the two observations without claiming one caused the other unless the design supports that inference.

18. Restore the approving authority. The programme coordinator approved the change, allowing the event to proceed. The validity of approval depends on who had that authority.

19. Name the actors and event. Six parents objected after the new condition was announced. Resistance increased is less informative when the count is available.

20. Keep the nominalised process if the testing procedure has already been established. If clarity for beginners is the goal, The engineers completed system testing before launch may be simpler, but individual tester identity is unnecessary.

21. The report invents the wrong mechanism. The student began on time. The delay arose from a command-word misunderstanding. State that language bottleneck rather than blaming time management.

22. The sentence is incomplete for diagnosis. The student attempted to submit before the deadline, but repeated upload-page timeouts delayed completion. Human action and system failure both matter.

23. The event sentence is a useful starting point but insufficient for prevention. Add the handoff mechanism: both students assumed the other had inserted the section and nobody verified the final file.

24. Too blame-heavy. Ethan’s action contributed, but the spreadsheet’s import behaviour also mattered. Explain both rather than turning a system-human chain into one-person causation.

25. Restore the procedural actor and difference. Group 4 started timing before adding the final reagent, unlike the other groups. This explains why the result differed.

26. Not automatically evasive. Early agentless language can be appropriate when the cause is genuinely unknown and the team is still investigating. The problem arises if the final report never restores agency after evidence becomes available.

27. No. Once the process and responsible actions are known, Mistakes were made is too vague for an accountability report. Name the actions and relevant actors, then include system conditions.

28. Reject the agency loss. The tutor is part of the mechanism the article is explaining. A better rewrite is After reviewing the marked paper, the tutor adjusted the sequence.

29. Reject the invented actor. AI has supplied the school without evidence. Unpack the event while keeping the agent unknown.

30. Not necessarily. The passive is already well suited to a methods section focused on storage conditions. Active voice adds an agent that the reader does not need.

The hidden-agent independence test

A learner is developing advanced agency control when they can increasingly:

• identify nominalisations and event nouns without treating every abstract noun as an error;
• turn event nouns back into verbs for diagnostic purposes;
• identify missing agent slots without inventing actors;
• distinguish event recovery from causal attribution;
• distinguish active and passive focus;
• recognise when a by-phrase restores the agent while preserving passive topic structure;
• keep passive voice when the agent is unknown, obvious or irrelevant;
• restore the agent when responsibility, authority, causation or source credibility depends on it;
• notice impersonal reporting such as it is believed and ask who owns the belief;
• notice abstract evaluative nouns such as failure, improvement, success and ask for criteria;
• avoid turning temporary event descriptions into accusations of motive;
• map event, actor and mechanism separately;
• distinguish personal action from system conditions;
• write family feedback that is accountable without being needlessly accusatory;
• use process-focused passives naturally in scientific and technical writing;
• unpack dense academic prose when noun phrases hide sequence;
• repackage established events into useful nominalisations after the mechanism is clear;
• audit AI rewrites for removed agents;
• audit AI rewrites for invented agents;
• protect source ownership when summarising studies;
• use nominalisation for cohesion rather than prestige;
• keep measurements visible instead of hiding them inside broad nouns such as improvement;
• identify when agency is physical rather than human;
• understand that zero passive voice is not a sensible writing goal;
• decide what the paragraph should foreground—the actor, action, affected object or process;
• shift between event-level, actor-level and mechanism-level reporting;
• use agency analysis across English, Mathematics, Science and Humanities;
• stop unpacking when the missing agent is genuinely irrelevant;
• restore accountability after an investigation moves from uncertainty to evidence;
• write formal prose that remains inspectable.

One evening in Punggol

The four students are looking at a sentence:

A mistake was made during submission.

Maya says, “Passive voice. Change it.”

Ethan says, “To what?”

Jia Jun says, “Someone made a mistake.”

Hana says, “We still don’t know who.”

They read the rest of the incident log.

The final document was correct in the shared folder.

Jia Jun downloaded an earlier copy by mistake because both versions had the same visible filename.

He uploaded that copy.

No final-version check existed.

Maya rewrites:

Jia Jun uploaded an earlier copy because the two versions displayed the same filename, and the team had no final-version check before submission.

Ethan nods.

“Now we know who acted.”

Hana adds:

“And why the system let the mistake through.”

That is the destination.

Not blame without mechanism.

Not vagueness without accountability.

The real goal: make agency visible at the level the reader needs

Sometimes the doer matters.

Name them.

Sometimes the doer is unknown.

Do not invent them.

Sometimes the doer is obvious.

Do not clutter the sentence.

Sometimes the process matters more than the person.

Use the passive.

Sometimes a whole event needs to become a noun so the next sentence can reason about it.

Nominalise it.

But know what you compressed.

Know who disappeared.

Know what causal relationship you may have hidden.

Know whether the reader can recover the event from context.

That is advanced agency control.

Not active voice as a slogan.

Not formal nouns as decoration.

Language that keeps the right actor, action and mechanism visible for the job at hand.

Research notes and further reading

The sources below support the grammar and style distinctions. The Hidden-Agent Audit, resident-character examples, case files and practice sequence are original instructional material.

• Cambridge Grammar: Nouns — forming nouns from other words — explains nominalisation and its use in formal writing.
• Cambridge Grammar: Detached impersonal style — discusses noun phrases, nominalisation, passive voice and impersonal formal style.
• Cambridge Grammar: Passives with and without an agent — explains when agents are included or omitted.
• Cambridge Grammar: Passive voice — active/passive information structure and agent omission.
• British Council LearnEnglish: Advanced passives review — passive focus, agent omission, formal/scientific uses and impersonal passives.
• British Council LearnEnglish: Passives — core active/passive contrast and focus.
• Purdue OWL: Sentence Clarity — discusses active voice, passive voice and keeping doers/actions clear while acknowledging disciplinary conventions.
• eduKateSG: How English Works | Nominalisation and Event Nouns — specialist owner for the broad mechanism.
• eduKateSG: Nominalisation in English — specialist owner for academic-writing uses and cautions.

Continue through the advanced eduKatePunggol vocabulary chain

• Advanced Vocabulary Synthesis | The Contrast Audit
• Advanced Vocabulary Argument | The Qualification Audit
• Advanced Vocabulary Reading | The Assumption Audit
• Advanced Vocabulary Editing | The Counterexample Test
• Vocabulary and Connotation
• Vocabulary and Register
• Vocabulary and Collocation
• Education: Regarding Vocabulary