Voice Mismatch in VP Ellipsis | Active and Passive Translation

How do you translate active–passive voice mismatch in verb-phrase ellipsis? First recover the missing action, then identify the agent and affected entity separately from their grammatical positions. English can sometimes link an unspoken verb phrase to an antecedent expressed in the other voice. A passive sentence can introduce a task, and a later active phrase such as “nobody did” can refer to performing that task. If a translation preserves only the surface subject or mechanically repeats the passive, it may change who was responsible, whether the action happened, and what was acted upon.

Take a natural example: “The records were supposed to be reviewed, but no one did.” The first clause is passive and concerns the records; the second contains an active subject, no one, and a missing verb phrase understood as reviewing the records. The sentence does not report that the records were reviewed. It says a review was expected and that no one performed it. An accurate translation keeps the intended review, the expectation and the negative statement about actual performance distinct.

This guide answers searches for voice mismatch in VP ellipsis, active and passive ellipsis, how to translate passive voice, ellipsis antecedent identity, and who did what in a passive sentence. It explains what linguistic research does and does not establish, contrasts verb-phrase ellipsis with pseudogapping and clausal ellipsis, supplies worked examples from professional writing, and gives translators a source-to-target checklist for controlling agency, obligation, time and outcome. The aim is readable language that stays faithful to the original event.

A quick answer: split the event from its grammatical packaging

“The invoices were meant to be audited, but the external reviewer did not.” The first clause uses a passive construction, describing invoices as the things meant to receive an audit. The second names a possible agent who did not audit them. The shared event concept is auditing the invoices, but the syntax presents it through different grammatical voices and different truth commitments.

Write a small interpretation record: action = audit; patient or affected item = invoices; expected event = an audit; possible agent = external reviewer; actual performance by that agent = negated. You can now translate the sentence by repeating audit in the target language: “The invoices were scheduled for an audit, but the external reviewer did not audit them,” provided scheduled accurately reflects meant to in context. Often a simpler equivalent of “were meant to be audited” will be safer than changing the force to a firm timetable.

The critical test is whether the finished target still lets readers say “an audit was expected but not done by that reviewer.” If it says the reviewer approved the invoices, it changes audit to approval. If it says the reviewer was audited, it reverses the participant role. If it says the invoices were never audited by anybody, that may overstate what the second clause tells us; the source specifically denies a named reviewer’s performance unless broader context says no one performed it.

For translation, voice is a tool for packaging an event. It is not an excuse to change the event itself. The safest practice is to record semantic participants before choosing target-language subjects, objects or passive morphology.

What linguists mean by voice mismatch in ellipsis

The term grammatical voice describes how a clause maps participant roles to grammatical positions. A typical active clause “The auditor checked the file” presents the auditor as subject and the file as object. Its passive counterpart “The file was checked by the auditor” presents the file as subject while preserving the underlying checking event and the auditor’s agent role.

Verb-phrase ellipsis (VPE) occurs when part of a verbal expression is unpronounced but recoverable: “The auditor checked the file, and the manager did too.” The second did too is interpreted through an earlier predicate. In voice-mismatch cases, the source of the interpretation and the missing phrase differ in active/passive realization. Linguists have used such cases to investigate what kind of identity licenses silence: should the grammar require the same syntactic structure, the same event meaning, or some constrained relation between them?

Jason Merchant’s paper Voice and Ellipsis discusses cases where VPE tolerates voice differences and contrasts them with other kinds of ellipsis. His earlier An Asymmetry in Voice Mismatches in VP-Ellipsis and Pseudogapping explores why not all superficially similar reductions behave alike. The analysis has prompted alternative theoretical proposals, and particular sentences vary in speaker acceptability.

A translator need not choose a theory of voice heads or deletion size. The practical lesson is more stable: an unspoken predicate can evoke an event even when that event was packaged in a different voice earlier. Recover the roles explicitly and select a grammatical target construction that preserves them. Do not treat every possible voice reversal as automatically natural English, or every unnatural source as a candidate for confident correction.

A recognized pattern: passive expectation, active denial

Linguistic discussions include a classic style of example in which a problem was expected to receive attention but no one actually undertook the task. A common structure is “The problem was supposed to be investigated, but nobody did.” The implicit action after did is investigate the problem. “Nobody” describes the absence of an agent performing the action; it is not the grammatical subject of the first clause.

The word supposed introduces an expectation, plan or obligation rather than necessarily a completed event. Translators who preserve investigate but drop supposed can wrongly claim that an investigation took place. Conversely, translators who preserve the passive but omit nobody can incorrectly suggest that responsibility was unspecified rather than that actual performance was denied. Both elements matter.

Compare three propositions: “An investigation was expected”; “Someone investigated”; “No one investigated.” They do not mean the same thing. The source can affirm the first and the third without contradiction. A target that combines them into “the problem was investigated unsuccessfully” changes nonperformance into performance with poor results. An investigation that produced no solution still occurred; an investigation that nobody undertook did not occur.

This distinction appears constantly in operational documents. Training was required but was not delivered. Files were to be checked but were not checked. Equipment was scheduled for inspection but inspection never happened. The translator’s job is to keep intended, required, scheduled, attempted and completed states from drifting into one another while resolving the missing verbal content.

Why active and passive cannot simply be swapped everywhere

Changing active into passive can preserve an event’s basic participants, but it can also change what a sentence foregrounds. “The engineer repaired the valve” names the agent directly. “The valve was repaired” highlights the affected object and leaves the agent unexpressed. If a translation freely adds “by the engineer” to the passive when the source omits the agent, it creates information. If it removes a known agent from an active clause in a document where accountability matters, it may conceal information readers need.

Different languages also impose different restrictions on passive formation. A straightforward English be-passive may have a periphrastic, morphological, impersonal or other natural counterpart. A translation may be more accurate with an active sentence in a target language that seldom uses the same passive construction. The requirement is preservation of roles, focus and truth conditions, not duplication of auxiliary morphology.

Voice interacts with agency and causation. “The message was sent” does not necessarily say who sent it. “The administrator sent the message” does. “The message sent itself” ordinarily implies a very different cause or an automated event. A machine translator might produce a grammatically fluent reflexive or middle-voice construction that alters the actor if it follows surface subjects too closely.

In ellipsis, these risks increase because the second predicate is silent. Before translating, fill in who would perform the action, what would be affected, and whether the second clause affirms, denies, predicts or merely permits the action. Only then choose the target’s active or passive structure. This two-stage approach guards meaning even when literal voice matching is impossible.

Different ellipsis constructions have different limits

Not all missing-verbal-material constructions tolerate voice mismatch in the same way. Ordinary VP ellipsis, pseudogapping, sluicing, stripping and right-node raising have distinct structural and discourse properties. Merchant’s research identifies a contrast between VPE cases that can permit active/passive alternation and some pseudogapping or clausal-ellipsis cases where analogous alternation is disfavored or unacceptable under the analysis.

Consider an awkward invented sentence: “Roses were delivered by the courier, and the assistant did lilies.” A translator should not declare this a normal voice-mismatched pseudogapping sentence merely because the intended meaning “the assistant delivered lilies” is recoverable. The surface form can be unacceptable or highly marked for reasons involving auxiliary structure, remnant licensing and identity. An expansion may still be an accurate paraphrase of intended discourse, but that is different from claiming a source construction is standard.

Now take an ordinary question: “Who signed the form?” A truncated answer “the manager” may be licensed by a different elliptical construction, whose conditions should not be inferred directly from what is true of VPE. A theory permitting one voice mismatch in VPE does not automatically certify a mismatch in all other kinds of ellipsis.

For translation, this distinction is practical. Use syntactic labels to guide interpretation, not to force a source to be elegant or grammatical when it is not. If the source is visibly nonstandard, decide whether its irregularity is meaningful dialogue, a drafting error or a genuine ambiguity. Preserve intent when it is clear and avoid unsupported repairs when the sentence’s actual content is uncertain.

Active antecedent and passive continuation: caution is essential

Suppose a draft reads, “The laboratory checked the samples, and the controls were too.” The intended expansion may be “the controls were checked too,” but the surface is not a uniformly accepted or equally natural construction in every context. The fact that a passive proposition can be understood from an active one does not guarantee that each auxiliary-and-remnant arrangement is licensed by speakers. Target-language writing should not imitate an awkward source merely to display ellipsis.

A cautious translator can translate the intended two complete propositions if context confirms them: the laboratory checked the samples; the controls were also checked. But notice that the second clause leaves the agent of checking controls unidentified. It would be wrong to add that the laboratory checked the controls too unless the source or wider document establishes that agency.

The same issue arises with plural and singular agreement. “The controls were” contains plural agreement, and a target whose passive inflection must agree with the affected entity needs to maintain that number. An English auxiliary can provide grammar even when a lexical predicate is missing; target languages may distribute that information across verbs and noun endings.

When the passive continuation is doubtful, explain the source’s intended meaning to the reviewer through an explicit paraphrase while preserving any uncertainties. A professional translation is evaluated by the event relationships the target supports, not by its willingness to reproduce every marked English reduction. Readable full clauses are often better than an unstable fragment whose interpretation depends on specialized linguistics knowledge.

Worked example: inspections and compliance duties

Source: “The stairwell was meant to be inspected before opening, but the contractor did not.” There is a planned inspection of the stairwell, tied to a before-opening condition. The second clause denies that the contractor performed the expected inspection. It does not necessarily prove nobody inspected the stairwell, because a different inspector could have done it unless the document explicitly rules that out.

The worst translation error is to write “The contractor opened the stairwell without inspection.” That inserts an opening event by the contractor, even though the sentence only locates the expected inspection in relation to opening. Another error is “The contractor failed the inspection,” which implies inspection occurred and received an adverse result. The source instead denies performance by that contractor.

A reliable expansion is: “The stairwell was supposed to undergo inspection before opening. The contractor did not inspect it.” The first clause still does not certify that the actual opening took place. In a formal report, keep any defined meaning of contractor, inspection and opening threshold. If the policy mandates a licensed inspector, use its controlled terminology rather than substituting a general contractor.

The review grid should include object = stairwell, required event = inspection, temporal condition = before opening, named agent = contractor, agent’s performance = negative, actual opening = not asserted, inspections by others = unknown. If the target claims that the stairwell was approved for use, it has invented a downstream status. Voice and ellipsis become an accountability issue when the source describes required-but-unperformed tasks.

Worked example: financial documents and an incomplete audit

A finance update says, “The year-end statements were scheduled for review, but the independent auditors have not yet.” The missing predicate is reviewed the statements. Have not yet reports that, up to the relevant present, the auditors have not completed that reviewing action. It does not claim that the audit has been cancelled, that the statements are wrong or that approval was denied.

One dangerous shortcut is “the independent auditors rejected the statements.” Rejection is a substantive judgment after some assessment, not mere noncompletion. Another is “the statements have not been audited by anybody.” The named auditors’ incomplete review is only one part of the document’s possible process. A third is to omit yet, turning temporary noncompletion into a more absolute denial.

A natural target may need to repeat the full review verb after the auditors’ subject. If the local accounting terminology distinguishes audited, reviewed, certified and signed, the translator should select the correct one from the source and surrounding institutional definitions. The word review does not automatically carry audit assurance obligations.

For publication quality control, build a timeline: planned review date, current date if given, performing party, progress state and resulting report if any. Populate only what the source actually states. A translator who preserves the passive object but invents completed audit results creates a false public impression. Accurate meaning recovery can be more important than stylistic fidelity to a tiny English omission.

Worked example: study data and a promised validation

A research manuscript states, “All raw measurements were to be validated independently, but the second team did not.” The passive clause describes a planned independent validation of measurements. The active elliptical continuation denies validation by the second team. It does not tell us whether the first team validated the measurements or whether the raw measurements were reliable.

A translator must distinguish collection, cleaning, validation, calibration and analysis. These may occur in sequence, but their names are not substitutes. “The second team did not analyze” would report a different omission from “did not validate.” The term independently matters because the method could require a separate assessor or process. If the target omits independence, readers may miss a limitation of the intended procedure.

Imagine a table with fifty observations. The second team might have reviewed ten and not completed the full validation; whether the sentence’s did not means no validation whatsoever or failure to perform the stated full procedure can depend on context. Rather than announce a percentage, preserve the source’s language and ask for clarification when the detail is consequential.

In scholarly documents, an editorial comment can state: “Does this mean no independent validation was attempted, or that the planned validation was not completed?” That is not an admission of translation weakness; it identifies a genuine scope question. Formal syntax gets the missing predicate; discourse and research methods determine exactly how broad the denial is.

Worked example: museum artifacts and conservation plans

Source: “The manuscript was supposed to be conserved, but the outside specialist did not.” The first clause is a passive statement of an intended conservation activity. The second denies performance by the outside specialist. The source does not say that the manuscript was destroyed, that another conservator did not work on it, or that the specialist was unqualified.

Conservation is not always restoration. A conservator may stabilize materials, slow deterioration or undertake preventive work rather than return an artifact to an earlier condition. Translating conserve as restore in a public archive record can overstate what was planned. It may also imply an outcome that the source never reports.

If the translator changes the passive clause into an active target construction, a cautious option is to retain an impersonal plan: “A conservation treatment was planned for the manuscript.” The second clause can name the outside specialist as a nonperforming party, but should not turn the plan into a contractual obligation unless the document actually says so.

The semantic checklist is simple: artifact identity, intended treatment, named specialist, actual performance and remaining unknowns. Dates, budgets, condition scores and preservation outcomes should not appear unless separately supported. The complexity here is not that the English has two voices; it is that an expected event and an absent event sit next to one another and can be confused by a too-confident translation.

Worked example: school reports and assessed work

A teacher writes, “The submitted essays were meant to be moderated, but the panel did not.” In school assessment, moderation may refer to reviewing grading standards and consistency. It does not necessarily mean marking each essay for the first time. The source indicates planned moderation and its nonperformance by the panel.

A translation saying “the panel failed all the essays” is a serious distortion. Another saying “the panel did not grade the essays” may be inaccurate if another assessor had already graded them and only moderation was outstanding. The difference between marking, reviewing, standardizing and moderating should follow the institution’s assessment vocabulary.

The target may express the first clause actively: “Moderation of the submitted essays had been planned.” That preserves the event without inventing an agent. The next clause can say the panel did not conduct moderation, subject to the grammatical conventions of the language. If the source names an official board or specific moderation policy, keep that body distinct from a class teacher or examination administrator.

In a parent-facing progress note, readers care about what is complete and what remains pending. A translated statement should not exaggerate poor outcomes or imply judgments that never occurred. The original reports process status, and the ellipsis does not convert an unperformed quality-control step into an assessment decision.

Worked example: public communications and translation delivery

A public information officer writes, “The emergency notice was due to be translated on Monday, but the contracted provider did not.” This describes an expected translation deliverable and a named provider that did not perform the action. It does not establish whether the notice was released untranslated, whether an in-house team translated it instead, or why the provider failed to act.

A rushed translation might report “the provider delivered a late translation.” That entails work was done, only behind schedule. The English did not says the contracted provider did not perform the understood action. Another rushed version could say “the public authority neglected to translate,” shifting responsibility from the named provider to the authority. Agency must remain explicit and correctly assigned.

Do not confuse translated with interpreted, posted, approved or distributed. The notice might require all those stages, but this sentence identifies translation. The word Monday supplies an intended date, not necessarily a contractual deadline with penalties. If the surrounding agreement establishes a binding obligation, the target should use the correct legal terminology; otherwise avoid adding sanctions or a contractual breach conclusion.

A practical target expansion is “Translation of the emergency notice was expected on Monday. The contracted provider did not translate it.” Then read the wider paragraph to see whether the event was cancelled, transferred or completed by others. Those are separate statements and should be translated separately when present, not inferred from an awkward ellipsis.

Worked example: software deployment and security review

A release checklist says, “The package was supposed to be signed before distribution, but the build pipeline did not.” In software engineering, signing a package can involve a cryptographic signature, not a handwritten approval. The passive construction identifies the package as the expected signed artifact; the active continuation identifies the automated pipeline as the nonperforming system.

A translator must not replace signed with deployed, validated or uploaded because those are neighboring steps. A package can be built and uploaded without having the required signature. The source does not tell us whether it was actually distributed or whether another process signed it. “Before distribution” describes a required sequence, not proof that distribution occurred.

If the target language uses a more explicit agentive construction, it can say the build pipeline did not sign the package. Avoid anthropomorphizing the pipeline into a human engineer or adding an unauthorized operator. If the document is security-sensitive, file names, hashes, versions and exact status labels should remain intact.

The quality check separates artifact, intended signature operation, system responsible for performing it, temporal condition and actual status. If the target makes readers think the package has an invalid signature rather than no reported signature from the pipeline, that too may be a meaningful difference. In technical documentation, failed validation and omitted signing are not equivalent events.

Worked example: transport maintenance and scope of responsibility

An operations note says, “The vehicle was scheduled for servicing, but the depot staff did not.” The missing action is servicing the vehicle. The phrase depot staff is the agent in the second clause. The sentence need not imply malicious neglect, lack of technical skill or completed deployment of the vehicle. A translation that adds blame or consequence may turn a schedule discrepancy into an allegation.

Servicing can encompass inspections, adjustments, fluid replacement or several related tasks. Without a detailed maintenance plan, the translator should not expand it into “replaced the brakes,” even if that would be a common service operation. The source’s generality may be deliberate.

If the same vehicle was serviced by an external provider instead, the initial report can still be true about depot staff. A target that says the vehicle was not serviced by anyone would be too strong. If the vehicle had not been serviced at all, that needs a separate source statement.

This is a useful test of polarity and agency: “depot staff did not service it” and “it was not serviced” differ when other agents are possible. An English voice shift can hide that difference in a short continuation. Accurate translation makes sure the negative applies to the right participant and the right action.

Worked example: legal disclosure and a missing procedural act

A case file says, “The correspondence was to be disclosed at the hearing, but counsel did not.” One reading is that counsel did not disclose the correspondence, despite the earlier expectation. The sentence itself does not establish whether counsel had legal authority, whether disclosure was mandatory under a court order, or whether the correspondence was withheld deliberately.

Disclosure is not the same as admission into evidence, publication or delivery to opposing counsel. A legal translator should choose the term associated with the actual procedural act and seek domain review if the stakes require it. A target such as “counsel concealed the correspondence” would add intent. “The court rejected the correspondence” would add a decision by an entirely different actor.

Time matters: “was to be disclosed at the hearing” identifies an expected hearing-stage event. If the target replaces that with “had been disclosed before the hearing,” it reports completed performance at a different time. Both temporal and aspectual distinctions must survive. The elliptical did not supplies a negative past event relative to the named counsel.

A reviewer should be able to answer who was expected to do what, what did not happen, and which consequences remain unasserted. Do not turn a grammar question into a legal finding. High-stakes procedural records need competent human review, but even basic actor and event tracking can prevent false translations before specialist review begins.

Voice mismatch and negation: three distinct negatives

Negative wording often creates the greatest practical risk in voice-mismatched ellipsis. Compare “The file was not reviewed,” “The manager did not review the file,” and “The file was not meant to be reviewed.” The first denies that the review event occurred, without naming any reviewer. The second denies review by a particular manager. The third denies an intended or authorized review, not necessarily actual performance. They are not interchangeable.

Now consider “The file was meant to be reviewed, but the manager did not.” The second clause denies an action by the manager; it does not in isolation tell us whether someone else reviewed the file. A target that combines these into a broad passive “the file was never reviewed” may erase the possibility of work by another person.

The same caution applies to not yet, no longer, not always and not necessarily. “The manager has not yet” indicates noncompletion up to a reference time with possible future performance. “The manager no longer does” indicates a cessation of a previously relevant practice. “The manager does not always” denies universal frequency without necessarily denying every instance. Each of these modifiers must remain attached to the right predicate and time.

A quick negative-scope test uses a world in which a colleague completed the task. If the original is true in that world but the proposed translation is false, the target probably broadened a participant-specific denial into a universal event denial. This check is useful across compliance, education, financial and research records.

What voice mismatches do not justify: added agents and inferred motives

When a passive antecedent omits an agent, the omission may be meaningful. “The notice was translated” reports translation but does not identify who performed it. If a later clause says “the editor did not,” the editor is simply excluded from an action; the source does not identify the actual translator. A target that assigns translation to the publisher because the publisher normally arranges it is speculative.

Conversely, if the source says “the notice was translated by the contracted linguist,” the named agent should remain visible when the target reorders the sentence. Dropping the linguist from a report of responsibility can be a substantive omission. Voice choice does not determine whether an agent can be discarded; the source determines that.

Motives are a separate risk. The phrase did not is not automatically refused to, neglected to or deliberately failed to. A task may go undone because it was reassigned, cancelled, impossible to perform or simply overlooked. Without evidence, the translator should not choose a target verb that assigns motive. The same applies to an affirmative did: performing a task does not prove enthusiasm, consent to a larger project or a particular degree of success.

Translators should make three checks: agent identity, action performance and mental-state attribution. If the source supplies only the first two, leave the third unknown. The combination of passive voice and ellipsis makes a sentence short, but it does not give the translator permission to fill the factual gaps with a coherent story.

From an English passive to an active target-language sentence

A target language may prefer a human subject even when English begins with an affected object. For example, “The reports were prepared last week” might be most naturally rendered through an impersonal construction, a passive affix or a contextually appropriate subject. If the source does not name the preparer, avoid selecting an invented human agent just to satisfy a preferred active template.

Where the target language has an impersonal plural or subject-omission strategy, the translation can state that preparation occurred without specifying who did it. Then a later active elliptical clause can name the relevant individual and describe participation or nonparticipation. This preserves the information distribution even if the grammar differs significantly.

Do not translate passive voice into a vague active verb that loses the affected item. If the report is what was prepared, that reference still matters. Conversely, do not make the item an agent when the source gives it a patient role. Languages with middle or reflexive constructions can be natural choices in some contexts, but their conventional meaning should be checked rather than assumed.

The decision procedure is consistent: build an event representation, select the target construction, and compare the result against the source. Linguistic naturalness is the goal after semantic roles have been stabilized. A fully active target can be more faithful than a word-for-word passive imitation if it reports exactly the same participants, events, modality and information focus.

Tense and aspect can make a voice shift deceptive

“The application has been checked, and the final receipt should be too.” The two clauses share the broad checking predicate, but their event statuses differ: the application’s checking is presented as completed with present relevance; checking the receipt is recommended or expected and not asserted as completed. A translation that uses the same completed form for both may falsely say the receipt has already been checked.

Now compare “The application was being checked, but the supervisor was not.” That sentence is ambiguous or awkward because the passive supervisor might be understood as an object undergoing checking rather than as a person not checking. The translator should not guess that the supervisor failed to check the application. Parse the grammatical roles and use context before expanding an ellipsis.

“The items will be inspected, and the external team may as well” carries a modal distinction: one part states a future inspection; the other says an external team may also inspect or be permitted to inspect, depending on may and context. The English may itself be stylistically marked. A target may benefit from explicit repetition of the verb while keeping will and may distinct.

Every modal and aspectual marker provides independent meaning. Does the source report a completed event, ongoing process, future plan, ability, permission, duty or prediction? A translation that reconstructs the hidden verb correctly but flattens these modes remains inaccurate. Voice mismatch is only one layer of the sentence’s logical commitments.

A source-to-target audit with an event ledger

Make a two-row ledger before drafting. Row one describes the event or expected event in the antecedent: affected entity, predicate, time, modality and any named agent. Row two describes the elliptical continuation: subject or other overt participant, recovered predicate, polarity and new modifiers. This ledger is deliberately independent of grammatical voice.

For “The data were to be verified before release, but the contractor did not,” row one is data / verify / before release / intended; row two is contractor / verify data / negative / no other modifier. The ledger does not include “data released,” “contractor intentionally refused,” “verification was legally required” or “data invalid,” because those are not asserted.

Translate each row explicitly into the target language. Then combine them if the result is clearer than two short sentences. A sentence that correctly makes one part passive and the other active can be kept; a target whose grammar requires different ordering can be rearranged freely. The invariant is the event ledger, not English word order.

For a high-stakes document, add a reviewer field recording whether the term verify has a controlled meaning and whether the source ambiguity needs clarification. A bilingual reviewer should reconstruct the same ledger from the finished target without being shown the English. If they disagree about agent, object, action or actual performance, the translation is not ready for publication.

Human and AI translation: the main error patterns

AI translation systems can produce an elegant active sentence by inventing a missing agent. They may also turn a passive plan into an accomplished event because that matches a familiar workflow narrative. A system that sees “the form was supposed to be signed, but the manager did not” may output “the manager declined to sign the form,” which introduces a refusal motive. The English only says that the manager did not perform the signing action.

Another error is predicate drift. “Was to be reviewed” becomes “was approved”; “was due for inspection” becomes “was certified.” These verbs describe different states. A third is global negation drift: a denial about one participant becomes a claim that nobody performed the action. A fourth is agent reversal: the target makes the patient of the passive into an acting person or makes the second named subject receive rather than perform the action.

A useful AI-assisted check asks for two full expansions and an event ledger, then compares them with the source. Ask what would be true if a different participant completed the action. If the original allows that possibility but the translation forbids it, the target may have overgeneralized. Ask also whether the source’s expected event actually happened; if not, the translation must not claim success.

Do not treat a model-generated grammatical explanation as legal, safety or scientific certification. For sensitive content, professional review remains necessary. AI is valuable for identifying competing interpretations, but the source text and verified institutional record determine which claims may appear in the final translation.

Twelve short practice cases with model recoveries

1. “The manuscript was to be proofread, but the editor did not.” Recovery: the editor did not proofread it. No publication outcome is asserted. 2. “The permit was meant to be checked, but the officer did not.” Recovery: the officer did not check the permit. This is not evidence that the permit was denied.

3. “The findings were scheduled for independent review, but the external team did not.” Recovery: external review was expected; the team did not conduct it. 4. “The images were to be captioned, but the contractor did not.” Recovery: contractor did not caption images. A target saying images were not photographed changes the task.

5. “The device should be inspected before use, but the operator did not.” Recovery: operator did not inspect before use. Do not infer the device was actually used. 6. “The invoices were supposed to be reconciled, but the accountant did not.” Recovery: no reconciliation by that accountant; actual payment remains unknown.

7. “The notices were meant to be translated, but the provider had not yet.” Recovery: translation by provider remained uncompleted up to the reference time; yet matters. 8. “The reports were to be anonymized, but the analyst did not.” Recovery: analyst did not anonymize reports. Do not invent a privacy breach unless release occurred.

9. “The samples were expected to be retested, but the laboratory did not.” Recovery: lab did not retest; previous tests may still exist. 10. “The contract was meant to be archived, but the clerk did not.” Recovery: clerk did not archive; no assertion that contract was destroyed.

11. “The equipment was scheduled for calibration, but the team did not.” Recovery: team did not calibrate. Do not convert into a failed calibration test. 12. “The plan should be explained to parents, but the school did not.” Recovery: the school did not explain the plan; do not assume families opposed it.

For each case, state the event, affected item, supposed agent and actual polarity. Then produce a target sentence with the full verb repeated. Compare a scenario where another agent completes the task. The best translation will preserve that distinction instead of turning participant-specific nonperformance into a global claim about the outcome.

Frequently asked questions

Does VP ellipsis always require the same voice as the antecedent? No. Linguistic research documents some voice mismatches, while acceptability depends on construction and context. Do not generalize a permissive case to all ellipsis. What is the simplest translation strategy? Recover the event’s action, agent and affected entity independently of English subjects, then use natural target-language grammar.

Can I change passive into active in a translation? Yes when the resulting sentence preserves event participants, agency information and focus adequately. Do not invent an agent where the source withholds one. Does “nobody did” always mean nobody performed the entire preceding workflow? It refers to a recoverable action; its exact scope depends on discourse. In a simple single-task case it can deny performance by anyone, but complex antecedents need review.

Is voice mismatch the same as pseudogapping? No. Pseudogapping has a distinct remnant structure and different constraints. Does a passive sentence mean the action is complete? No. Passives combine with all sorts of tense, aspect and modal expressions. “Was to be inspected” is not “was inspected.”

Why is a grammatically fluent target sometimes wrong? Because fluency does not ensure correct agent, predicate, negative scope or event status. Can machine translation resolve this automatically? Sometimes, but it may choose an unjustified agent or outcome. Source-to-target checking is still necessary.

Should a legal or medical document use this guide alone? No. Use qualified domain review where rights, safety or clinical outcomes could be affected. What related pages should I read? Continue with the guide to pseudogapping for remnant structure and the wider translation system for practical accuracy checks.

Translate Precisely | Active–Passive Voice Mismatch in VP Ellipsis — Preserve Who Acts, What Changes and the Missing Verb