Master Art of Translation | The Ellipsis Recovery Fidelity System — How Omitted Verbs, Fragments, Gapping and Context Preserve Meaning

How to translate ellipsis accurately: the quick answer

How do you translate ellipsis, missing words, omitted verbs, fragments, gapping or short answers without accidentally adding meaning? First reconstruct only the material that the source grammar and context license. Then decide whether the target language needs that material expressed, can leave it implicit, or must restructure the sentence. The goal is not to make every sentence longer. It is to preserve the same participants, event, contrast, certainty and unanswered questions.

In practical translation, ellipsis recovery means identifying what a listener or reader can recover from earlier words, the conversational situation or an established grammatical pattern. A sentence such as “Mira ordered tea; Luis, coffee” normally shares the verb “ordered”. Yet “Mira ordered tea; Luis, coffee” cannot safely become “Luis drank coffee”: ordering and drinking are different events. That small distinction is the foundation of accurate gapping translation, verb-phrase ellipsis translation and context-aware translation of dialogue.

This guide explains how to translate omitted subjects, answer fragments, sluicing, comparative deletion, zero anaphora, short subtitles and compressed technical text. It provides a repeatable source-first method, original worked diagnostics and an explicit rule for uncertainty: when the antecedent cannot be established, mark the ambiguity or ask the source owner rather than invent a fluent completion. It is a world-facing translation-method chapter, not a dictionary or a guide to a single language pair.

In one sentence: reconstruct to understand, express only what the target reader needs, and verify that no recovered word has become a new fact.

• Identify the gap: distinguish licensed ellipsis from poor writing, missing data, interruption or deliberate withholding
• Find the antecedent: point to the exact recoverable phrase and record competing candidates
• Preserve scope: carry negation, modality, question force, comparison and quantity into the reconstructed reading
• Choose target form: retain a natural ellipsis when possible; otherwise use minimal explicitness
• Review the evidence: test every supplied subject, verb, object, reason, timeframe and agent against the source
• Escalate uncertainty: request context or retain ambiguity if more than one recovery remains plausible

Contents

• The hidden problem: sounding complete while becoming unfaithful
• A precise map of linguistic ellipsis
• The 24-pass recovery and translation workflow
• 72 worked translation diagnostics
• Common failure modes and corrections
• Transfer laboratories and practice plans
• Frequently asked questions
• Authoritative references and next reading routes

The hidden problem: fluency can manufacture an event

Ellipsis is not simply the absence of a word. It is a relation between something expressed and something understood without being spoken again. English can say “Nora will attend, and Tomas will too.” The second clause inherits an event from the first. A target language may demand a different verb form, person agreement or explicit complement. Those requirements justify grammatical changes; they do not justify inventing the purpose of the attendance, the duration or the speaker’s enthusiasm.

An editor can make a translation look polished by filling every silence. That can be a mistake. A fragment such as “Tomorrow, perhaps” may deliberately sound tentative. Rewriting it as “We will definitely send the package tomorrow” replaces not only an omitted subject and predicate but also uncertainty with a promise. The failure is semantic, not stylistic. A professional revision asks which omissions are recoverable and which are purposeful gaps in knowledge.

The key distinction is between an antecedent and a guess. An antecedent is linguistic or contextual material that licenses recovery: for example, “inspect the north valve” in a preceding instruction. A guess is a plausible narrative that lacks such support: for example, assuming why the valve is being inspected or which worker is responsible. A plausible completion is not automatically an authorized completion.

Not all languages permit the same omissions. Some routinely leave subjects unpronounced when verb morphology or discourse makes them recoverable. Others allow entire noun phrases to disappear in a well-established topic chain. Still others prefer overt pronouns or repeat verbs to keep reference clear. Translation therefore requires a distinction between what the source explicitly says, what it grammatically implies and what the target language requires for idiomatic interpretation.

A second distinction concerns ellipsis as a structure versus a printing convention. A row of three dots may mark hesitation, omitted quoted text or an unfinished thought, while a grammatically elliptical sentence may contain no dots at all. Translators must not equate punctuation with deletion. A quotation with editorial omission must be handled under the citation rules of the project, whereas an ordinary conversational fragment requires discourse analysis.

The practical risk rises where fragments control action. In a support chat, “Only after approval” may qualify an earlier instruction. In an operating checklist, “If safe” may be a condition for the next step. In a report, “Not the second sample” may correct a mistaken referent. Expanding the material without tracking its connection can transfer the condition to the wrong act, or remove it entirely.

Another risk is overinterpretation of English auxiliaries. In “Lea has finished; Marco hasn’t”, the inherited predicate is “finished”, but the intended scope may be “finished the draft” if the first clause identifies that object. Adding “Marco has not started” changes the phase of the activity. It is possible to understand a short utterance without making it more explicit in the delivered translation.

When a translator sees two candidate antecedents, the answer may depend on prosody, prior turns, punctuation, visual evidence or a later correction. The first clause does not always win. A document may contain two scheduled releases and then say “The later one, if ready.” The target must not silently decide that ‘later’ means a particular calendar date unless the schedule identifies it.

Ellipsis is also an economy of attention. Speakers avoid re-stating material their audience already has. Forcing all recovered words into the finished translation can damage pacing, characterize a speaker as unusually formal, or misrepresent a concise interface label. Fidelity is not a mechanical word count. It is the preservation of recoverability, information status and the reader’s likely understanding.

Translation decisions should therefore leave a small audit trail. Record the source segment; name the ellipsis type; cite the antecedent; write the minimally expanded interpretation; label what remains underdetermined; and decide the best target formulation. This audit is particularly valuable when reviewing machine translation, where an apparently coherent output may hide an invented subject or action.

Ellipsis can travel across sentence boundaries. A brief reply often depends on a question several turns earlier, and a list item may inherit a heading rather than the previous line. For technical and legal work, the layout is part of the evidence. A table heading, bullet nesting, preceding definition and explicit correction may determine the recoverable predicate more reliably than the closest verb in running prose.

The final test is reader action and belief. Would the target reader infer the same person acted, the same condition applied, the same uncertainty remained and the same information was withheld? If the answer is no, a superficially smooth translation has failed. If the source is irreducibly ambiguous, a successful translation may preserve that ambiguity or flag it for resolution.

A precise map of linguistic ellipsis and what to recover

1. Verb-phrase ellipsis

Recognition. An auxiliary or comparable marker remains while a predicate is silent. A practical example is “Anika can swim, and David can too”. The source-first reconstruction is: Recover swim, not win a race. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. The auxiliary carries modality and often tense; changing can to will changes the claim. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

2. Gapping

Recognition. A coordinated clause retains contrasted participants while a shared verb disappears. A practical example is “Asha chose red; Noel, blue”. The source-first reconstruction is: Recover chose for Noel. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. Do not substitute preferred or received for chose. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

3. Stripping

Recognition. One contrasted remnant stands after a coordinator or contrastive marker. A practical example is “We need paper, not plastic”. The source-first reconstruction is: Recover need for the corrected object. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. A wrong attachment can turn an excluded object into a new order. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

4. Sluicing

Recognition. A question word remains while the remainder of its clause is unspoken. A practical example is “Someone called, but I don’t know who”. The source-first reconstruction is: Recover who called. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. Do not invent a caller or imply the speaker knows the identity. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

5. Fragment answer

Recognition. A reply supplies only the focus requested by a question. A practical example is “When did the package leave?” — “Yesterday”. The source-first reconstruction is: Recover it left yesterday. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. Do not infer the time zone, hour or carrier. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

6. Predicate answer

Recognition. A response omits subject material under discussion. A practical example is “Ready?” — “Almost”. The source-first reconstruction is: Recover nearly ready. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. Almost is a degree response, not an affirmative yes. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

7. Comparative deletion

Recognition. Shared material disappears from the comparison clause. A practical example is “Eva wrote more than Dan did”. The source-first reconstruction is: Recover wrote. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. More could refer to words, pages or reports depending on context. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

8. Nominal ellipsis

Recognition. The head noun is omitted inside a noun phrase. A practical example is “Take the blue one”. The source-first reconstruction is: Recover the previously established kind of object. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. The color does not establish a specific physical item in a crowded scene. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

9. Zero subject

Recognition. A language permits the recoverable subject to remain unpronounced. A practical example is A contextually known actor performs a second verb. The source-first reconstruction is: Recover only the actor supported by agreement or discourse. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. Defaulting every omitted actor to I or they may change responsibility. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

10. Zero object

Recognition. An understood object is not repeated. A practical example is “I brought the forms; please sign”. The source-first reconstruction is: Recover the relevant forms only if discourse licenses it. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. Other forms may be present and require a clarifying question. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

11. Shared auxiliary

Recognition. Parallel clauses omit repeated auxiliary material. A practical example is “She has checked the list; I, the receipts”. The source-first reconstruction is: Recover have checked when licensed. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. Aspect and completion matter. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

12. Shared preposition

Recognition. Parallel lists can omit a repeated relational marker. A practical example is “To the archive on Monday; the library on Tuesday”. The source-first reconstruction is: Recover to where the layout supplies it. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. Do not assume a date qualifies both destinations. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

13. Conjunction reduction

Recognition. A shared subject or predicate disappears across conjuncts. A practical example is “Nina arrived and unpacked”. The source-first reconstruction is: Recover Nina as subject of unpacked. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. A subject change requires explicit evidence. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

14. Subject gap in coordination

Recognition. Two verbs take one subject in a linked structure. A practical example is “The sensor measured and recorded pressure”. The source-first reconstruction is: Recover sensor as agent of both verbs. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. Measurement and recording need not be the same event time. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

15. Dialogue adjacency

Recognition. The answer obtains its proposition from the immediately relevant question. A practical example is “Which room?” — “The upstairs one”. The source-first reconstruction is: Recover room. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. Do not attach the fragment to an unrelated intervening remark. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

16. Topic-chain omission

Recognition. A multi-sentence passage keeps a participant active without renaming it. A practical example is A report follows one device across several operations. The source-first reconstruction is: Recover the tracked device until the focus shifts. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. A new heading or intervening noun may break the chain. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

17. Deictic fragment

Recognition. A demonstrative or time expression points to context rather than repeating text. A practical example is “There, after sunset”. The source-first reconstruction is: Resolve place and time from grounded context. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. There is not automatically a named country or street. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

18. Conditional fragment

Recognition. A condition appears without a repeated result clause. A practical example is “Only if approved”. The source-first reconstruction is: Recover exactly the earlier proposed act as conditioned. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. Do not convert a condition into an instruction. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

19. Corrective fragment

Recognition. A short utterance replaces one portion of a previous claim. A practical example is “No, the smaller box”. The source-first reconstruction is: Replace the box choice, not the entire project. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. The correction scope must be preserved. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

20. Partial quotation

Recognition. An excerpt leaves cited words out intentionally. A practical example is “We found…evidence” in an edited quotation. The source-first reconstruction is: Consult the complete original before interpreting the dots. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. Editorial ellipses are not grammatical license to supply missing evidence. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

21. Interrupted speech

Recognition. The speaker stops before delivering the proposition. A practical example is “If he finds out—”. The source-first reconstruction is: Do not complete the threatened consequence. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. An interruption can be meaningful precisely because it is unresolved. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

22. List-header inheritance

Recognition. Bullet items obtain a predicate or category from a heading. A practical example is “Required documents: identity record; signed form”. The source-first reconstruction is: Recover required documents for each item. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. A list may mix examples and requirements; verify header semantics. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

23. Table-cell inheritance

Recognition. A blank or short cell inherits context from row and column labels. A practical example is “Status: pending; Owner: —”. The source-first reconstruction is: Recover only the status and known owner field. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. A dash may mean unavailable, not the previous owner. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

24. Parallel title compression

Recognition. A heading omits words repeated elsewhere in the series. A practical example is “Safety: inspection, storage, disposal”. The source-first reconstruction is: Treat entries as topics or phases according to layout. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. Do not fabricate ordered steps from a topical list. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

25. Imperative subject omission

Recognition. An instruction omits the addressee as an ordinary grammatical feature. A practical example is “Save the file”. The source-first reconstruction is: Understand an implicit addressee. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. Do not invent a named employee. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

26. Passive agent omission

Recognition. A passive clause does not supply an actor. A practical example is “The sample was rejected”. The source-first reconstruction is: Preserve unknown or unstated agent. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. An agentless passive is not permission to name the laboratory. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

27. Generic agent omission

Recognition. An impersonal construction backgrounds who acts. A practical example is “Applications are reviewed weekly”. The source-first reconstruction is: Preserve the schedule and process. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. Do not claim a particular authority reviews them. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

28. Anaphoric ‘do so’

Recognition. A pro-form stands for a previously stated act. A practical example is “Return the device, and do so promptly”. The source-first reconstruction is: Recover return the device. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. Do so does not include later actions. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

29. Contrastive substitution

Recognition. One clause uses a replacement rather than repeating a phrase. A practical example is “I chose the early train; she, the late one”. The source-first reconstruction is: Recover chose. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. Neither departure city nor ticket class follows. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

30. Ellipsis under negation

Recognition. Negative markers interact with omitted material. A practical example is “He promised to go; she didn’t”. The source-first reconstruction is: Recover promise to go or go only after resolving the antecedent. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. Scope ambiguity is a reason to query. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

31. Ellipsis under modality

Recognition. Must, may, should, or can survives while action is omitted. A practical example is “We may proceed; they may not”. The source-first reconstruction is: Recover proceed under permission/possibility context. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. Do not turn prohibited into physically unable. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

32. Ellipsis across turns

Recognition. A repeated question can be omitted when returning to an earlier topic. A practical example is “And the second invoice?”. The source-first reconstruction is: Recover the prior invoice question. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. The reply need not assert an invoice exists. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

33. Ellipsis and humour

Recognition. An expected completion is withheld for effect. A practical example is “I could tell you, but…”. The source-first reconstruction is: Keep the withheld consequence. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. Explaining the joke can erase its effect. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

34. Ellipsis in captions

Recognition. A caption condenses a visible action into a noun phrase. A practical example is “After the storm”. The source-first reconstruction is: Use visual evidence for referent, not imagined history. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. A photo cannot establish an unsupported causal sequence. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

35. Ellipsis in subtitles

Recognition. Time and reading-speed constraints favor short recoverable lines. A practical example is “Never again”. The source-first reconstruction is: Preserve the refusal and context. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. Compactness cannot delete warning scope. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

36. Ellipsis in UI labels

Recognition. A short button or control inherits meaning from surrounding interface. A practical example is “Continue”. The source-first reconstruction is: Recover continue the current workflow. This reconstruction is a diagnostic gloss, not automatically the wording to publish in the target language.

Transfer decision. Do not translate as confirm payment absent evidence. Compare a minimally explicit target option with a natural target-language omission. Ask whether the reader obtains the same actor, proposition and scope, whether a contrast is still legible, and whether an apparently helpful addition has been promoted from inference to fact. If the two readings require different behavior, choose neither on intuition alone: document the evidence and seek context. A reviewer can mark the antecedent in the source, underline each supplied element in the proposed translation, and reject any element whose support cannot be named.

The 24-pass source-first workflow

Pass 01 — Freeze the source

Action. Save the complete source segment and its immediate neighbors before editing. Why this matters. A late change to a preceding line may change the antecedent. The translator should mark the specific source evidence and write a one-line statement of the target reader’s expected interpretation before finalizing wording.

Audit gate. Keep revision identifiers and do not rely on a copied fragment alone. Review the segment once with the reconstructed text visible and once without it. In the first reading, look for missing semantic components; in the second, look for a target sentence that has become over-explanatory or misleadingly certain. Record whether recovery is licensed by grammar, by a nearby turn, by a heading, or by an external source explicitly approved for this project. If support is merely plausible, that is a query rather than a fact.

Pass 02 — Identify the communicative job

Action. Decide whether the segment answers, corrects, requests, reports, jokes or instructs. Why this matters. The same fragment can be an answer in one exchange and a label in another. The translator should mark the specific source evidence and write a one-line statement of the target reader’s expected interpretation before finalizing wording.

Audit gate. Ask what the reader should believe or do afterward. Review the segment once with the reconstructed text visible and once without it. In the first reading, look for missing semantic components; in the second, look for a target sentence that has become over-explanatory or misleadingly certain. Record whether recovery is licensed by grammar, by a nearby turn, by a heading, or by an external source explicitly approved for this project. If support is merely plausible, that is a query rather than a fact.

Pass 03 — Mark visible predicates

Action. Underline each spoken verb and its auxiliaries. Why this matters. Auxiliaries can carry tense, aspect, negation and possibility independently. The translator should mark the specific source evidence and write a one-line statement of the target reader’s expected interpretation before finalizing wording.

Audit gate. Never recover only the lexical verb when the modal scope matters. Review the segment once with the reconstructed text visible and once without it. In the first reading, look for missing semantic components; in the second, look for a target sentence that has become over-explanatory or misleadingly certain. Record whether recovery is licensed by grammar, by a nearby turn, by a heading, or by an external source explicitly approved for this project. If support is merely plausible, that is a query rather than a fact.

Pass 04 — Locate every empty slot

Action. Identify missing subjects, objects, verbs or entire clauses. Why this matters. A single short sentence may contain several independent omissions. The translator should mark the specific source evidence and write a one-line statement of the target reader’s expected interpretation before finalizing wording.

Audit gate. Count recoverable positions separately instead of supplying one wholesale paraphrase. Review the segment once with the reconstructed text visible and once without it. In the first reading, look for missing semantic components; in the second, look for a target sentence that has become over-explanatory or misleadingly certain. Record whether recovery is licensed by grammar, by a nearby turn, by a heading, or by an external source explicitly approved for this project. If support is merely plausible, that is a query rather than a fact.

Pass 05 — Classify the pattern

Action. Distinguish gapping, sluicing, stripping, nominal ellipsis and ordinary answers. Why this matters. Each pattern places different limits on acceptable antecedents. The translator should mark the specific source evidence and write a one-line statement of the target reader’s expected interpretation before finalizing wording.

Audit gate. A construction label helps but cannot override actual context. Review the segment once with the reconstructed text visible and once without it. In the first reading, look for missing semantic components; in the second, look for a target sentence that has become over-explanatory or misleadingly certain. Record whether recovery is licensed by grammar, by a nearby turn, by a heading, or by an external source explicitly approved for this project. If support is merely plausible, that is a query rather than a fact.

Pass 06 — Collect candidate antecedents

Action. List the closest plausible phrases without immediately choosing. Why this matters. Proximity is one cue, not a universal decision rule. The translator should mark the specific source evidence and write a one-line statement of the target reader’s expected interpretation before finalizing wording.

Audit gate. A correction or topic switch can defeat the closest candidate. Review the segment once with the reconstructed text visible and once without it. In the first reading, look for missing semantic components; in the second, look for a target sentence that has become over-explanatory or misleadingly certain. Record whether recovery is licensed by grammar, by a nearby turn, by a heading, or by an external source explicitly approved for this project. If support is merely plausible, that is a query rather than a fact.

Pass 07 — Use agreement and structure

Action. Check grammatical compatibility with each candidate. Why this matters. Person, number, aspect and case narrow possible reconstructions. The translator should mark the specific source evidence and write a one-line statement of the target reader’s expected interpretation before finalizing wording.

Audit gate. Do not invent features when source form leaves them open. Review the segment once with the reconstructed text visible and once without it. In the first reading, look for missing semantic components; in the second, look for a target sentence that has become over-explanatory or misleadingly certain. Record whether recovery is licensed by grammar, by a nearby turn, by a heading, or by an external source explicitly approved for this project. If support is merely plausible, that is a query rather than a fact.

Pass 08 — Test discourse focus

Action. Check what the conversation currently treats as given. Why this matters. A heading, question or contrast often establishes recoverable content. The translator should mark the specific source evidence and write a one-line statement of the target reader’s expected interpretation before finalizing wording.

Audit gate. Do not import material from an old topic simply because it uses a matching noun. Review the segment once with the reconstructed text visible and once without it. In the first reading, look for missing semantic components; in the second, look for a target sentence that has become over-explanatory or misleadingly certain. Record whether recovery is licensed by grammar, by a nearby turn, by a heading, or by an external source explicitly approved for this project. If support is merely plausible, that is a query rather than a fact.

Pass 09 — Preserve polarity

Action. Attach not, never, no and negative particles at the right scope. Why this matters. Negation may apply to the inherited event, permission or particular constituent. The translator should mark the specific source evidence and write a one-line statement of the target reader’s expected interpretation before finalizing wording.

Audit gate. Create a minimal counterexample to see whether the reading reverses. Review the segment once with the reconstructed text visible and once without it. In the first reading, look for missing semantic components; in the second, look for a target sentence that has become over-explanatory or misleadingly certain. Record whether recovery is licensed by grammar, by a nearby turn, by a heading, or by an external source explicitly approved for this project. If support is merely plausible, that is a query rather than a fact.

Pass 10 — Preserve modality

Action. Trace can, may, must, should, might and comparable distinctions. Why this matters. Permission, possibility, obligation and prediction are not interchangeable. The translator should mark the specific source evidence and write a one-line statement of the target reader’s expected interpretation before finalizing wording.

Audit gate. If target grammar requires another construction, preserve the force. Review the segment once with the reconstructed text visible and once without it. In the first reading, look for missing semantic components; in the second, look for a target sentence that has become over-explanatory or misleadingly certain. Record whether recovery is licensed by grammar, by a nearby turn, by a heading, or by an external source explicitly approved for this project. If support is merely plausible, that is a query rather than a fact.

Pass 11 — Preserve tense and aspect

Action. Check time anchoring and whether the event is completed or ongoing. Why this matters. An omitted verb may inherit an auxiliary from earlier material. The translator should mark the specific source evidence and write a one-line statement of the target reader’s expected interpretation before finalizing wording.

Audit gate. Do not silently add completion or repetition. Review the segment once with the reconstructed text visible and once without it. In the first reading, look for missing semantic components; in the second, look for a target sentence that has become over-explanatory or misleadingly certain. Record whether recovery is licensed by grammar, by a nearby turn, by a heading, or by an external source explicitly approved for this project. If support is merely plausible, that is a query rather than a fact.

Pass 12 — Preserve questionhood

Action. Determine whether a fragment asks for information or asserts an answer. Why this matters. Sluicing retains an unresolved variable. The translator should mark the specific source evidence and write a one-line statement of the target reader’s expected interpretation before finalizing wording.

Audit gate. Never replace ‘who’ with a guessed participant. Review the segment once with the reconstructed text visible and once without it. In the first reading, look for missing semantic components; in the second, look for a target sentence that has become over-explanatory or misleadingly certain. Record whether recovery is licensed by grammar, by a nearby turn, by a heading, or by an external source explicitly approved for this project. If support is merely plausible, that is a query rather than a fact.

Pass 13 — Preserve comparison class

Action. Inspect comparative ellipsis for shared measure and baseline. Why this matters. More, less and better require a property or measurement. The translator should mark the specific source evidence and write a one-line statement of the target reader’s expected interpretation before finalizing wording.

Audit gate. Do not invent units or denominators. Review the segment once with the reconstructed text visible and once without it. In the first reading, look for missing semantic components; in the second, look for a target sentence that has become over-explanatory or misleadingly certain. Record whether recovery is licensed by grammar, by a nearby turn, by a heading, or by an external source explicitly approved for this project. If support is merely plausible, that is a query rather than a fact.

Pass 14 — Identify role assignment

Action. Write who did what to whom in the minimally expanded gloss. Why this matters. Pronoun choice and passive voice can obscure agency. The translator should mark the specific source evidence and write a one-line statement of the target reader’s expected interpretation before finalizing wording.

Audit gate. Ask who bears responsibility in the source and target. Review the segment once with the reconstructed text visible and once without it. In the first reading, look for missing semantic components; in the second, look for a target sentence that has become over-explanatory or misleadingly certain. Record whether recovery is licensed by grammar, by a nearby turn, by a heading, or by an external source explicitly approved for this project. If support is merely plausible, that is a query rather than a fact.

Pass 15 — Check exception boundaries

Action. Mark unless, except, only if and similar qualifiers. Why this matters. An elliptical exception may condition only one nearby act. The translator should mark the specific source evidence and write a one-line statement of the target reader’s expected interpretation before finalizing wording.

Audit gate. Keep nesting and list structure visible. Review the segment once with the reconstructed text visible and once without it. In the first reading, look for missing semantic components; in the second, look for a target sentence that has become over-explanatory or misleadingly certain. Record whether recovery is licensed by grammar, by a nearby turn, by a heading, or by an external source explicitly approved for this project. If support is merely plausible, that is a query rather than a fact.

Pass 16 — Check register and pacing

Action. Assess whether explicit recovery would change voice or relationship. Why this matters. Short replies can communicate speed, reluctance, warmth or impatience. The translator should mark the specific source evidence and write a one-line statement of the target reader’s expected interpretation before finalizing wording.

Audit gate. Preserve intended brevity when it remains intelligible. Review the segment once with the reconstructed text visible and once without it. In the first reading, look for missing semantic components; in the second, look for a target sentence that has become over-explanatory or misleadingly certain. Record whether recovery is licensed by grammar, by a nearby turn, by a heading, or by an external source explicitly approved for this project. If support is merely plausible, that is a query rather than a fact.

Pass 17 — Choose the target strategy

Action. Select retain, minimally expand, restructure or annotate. Why this matters. Different language pairs license different silent arguments. The translator should mark the specific source evidence and write a one-line statement of the target reader’s expected interpretation before finalizing wording.

Audit gate. Record why the chosen target option is necessary rather than decorative. Review the segment once with the reconstructed text visible and once without it. In the first reading, look for missing semantic components; in the second, look for a target sentence that has become over-explanatory or misleadingly certain. Record whether recovery is licensed by grammar, by a nearby turn, by a heading, or by an external source explicitly approved for this project. If support is merely plausible, that is a query rather than a fact.

Pass 18 — Draft a neutral paraphrase

Action. Write a full-source-language diagnostic sentence in a private note. Why this matters. The paraphrase should disclose hidden assumptions in the translator’s reading. The translator should mark the specific source evidence and write a one-line statement of the target reader’s expected interpretation before finalizing wording.

Audit gate. Do not publish the paraphrase as if it were the original text. Review the segment once with the reconstructed text visible and once without it. In the first reading, look for missing semantic components; in the second, look for a target sentence that has become over-explanatory or misleadingly certain. Record whether recovery is licensed by grammar, by a nearby turn, by a heading, or by an external source explicitly approved for this project. If support is merely plausible, that is a query rather than a fact.

Pass 19 — Draft the target version

Action. Write idiomatically with required morphology and word order. Why this matters. Literal repetition may be unnatural while omission may be ambiguous. The translator should mark the specific source evidence and write a one-line statement of the target reader’s expected interpretation before finalizing wording.

Audit gate. Use the smallest grammatical form that keeps meaning stable. Review the segment once with the reconstructed text visible and once without it. In the first reading, look for missing semantic components; in the second, look for a target sentence that has become over-explanatory or misleadingly certain. Record whether recovery is licensed by grammar, by a nearby turn, by a heading, or by an external source explicitly approved for this project. If support is merely plausible, that is a query rather than a fact.

Pass 20 — Perform a reverse-meaning test

Action. Paraphrase the target back into a language understood by the reviewer. Why this matters. Back-translation is a diagnostic, not proof of equivalence. The translator should mark the specific source evidence and write a one-line statement of the target reader’s expected interpretation before finalizing wording.

Audit gate. Look for extra actors, events, certainty and deadlines. Review the segment once with the reconstructed text visible and once without it. In the first reading, look for missing semantic components; in the second, look for a target sentence that has become over-explanatory or misleadingly certain. Record whether recovery is licensed by grammar, by a nearby turn, by a heading, or by an external source explicitly approved for this project. If support is merely plausible, that is a query rather than a fact.

Pass 21 — Check the context window

Action. Review adjacent lines, display, recording and document hierarchy. Why this matters. One segment can look accurate in isolation but contradict a later correction. The translator should mark the specific source evidence and write a one-line statement of the target reader’s expected interpretation before finalizing wording.

Audit gate. Test after integrating the target into the full passage. Review the segment once with the reconstructed text visible and once without it. In the first reading, look for missing semantic components; in the second, look for a target sentence that has become over-explanatory or misleadingly certain. Record whether recovery is licensed by grammar, by a nearby turn, by a heading, or by an external source explicitly approved for this project. If support is merely plausible, that is a query rather than a fact.

Pass 22 — Score severity

Action. Rank loss of agency, safety, obligation and evidence above cosmetic changes. Why this matters. A wrong implied subject may matter more than a clumsy word order. The translator should mark the specific source evidence and write a one-line statement of the target reader’s expected interpretation before finalizing wording.

Audit gate. Use a risk-weighted review queue. Review the segment once with the reconstructed text visible and once without it. In the first reading, look for missing semantic components; in the second, look for a target sentence that has become over-explanatory or misleadingly certain. Record whether recovery is licensed by grammar, by a nearby turn, by a heading, or by an external source explicitly approved for this project. If support is merely plausible, that is a query rather than a fact.

Pass 23 — Resolve or retain uncertainty

Action. Ask the source owner, record alternatives or keep ambiguity. Why this matters. The translator’s confidence does not create factual support. The translator should mark the specific source evidence and write a one-line statement of the target reader’s expected interpretation before finalizing wording.

Audit gate. Mark every unresolved gap with its exact reason. Review the segment once with the reconstructed text visible and once without it. In the first reading, look for missing semantic components; in the second, look for a target sentence that has become over-explanatory or misleadingly certain. Record whether recovery is licensed by grammar, by a nearby turn, by a heading, or by an external source explicitly approved for this project. If support is merely plausible, that is a query rather than a fact.

Pass 24 — Sign off and reuse learning

Action. Add validated choices to the project-specific style and issue log. Why this matters. The next occurrence may reuse the same decision only in the same context. The translator should mark the specific source evidence and write a one-line statement of the target reader’s expected interpretation before finalizing wording.

Audit gate. Do not turn an example-specific answer into a universal glossary rule. Review the segment once with the reconstructed text visible and once without it. In the first reading, look for missing semantic components; in the second, look for a target sentence that has become over-explanatory or misleadingly certain. Record whether recovery is licensed by grammar, by a nearby turn, by a heading, or by an external source explicitly approved for this project. If support is merely plausible, that is a query rather than a fact.

72 original worked ellipsis diagnostics

The following cases use invented English-language source passages as controlled semantic examples. They are not purported certified translations into any named language. Each case demonstrates the source interpretation a translator must preserve when writing idiomatic target-language text. Context varies deliberately: a successful reconstruction in one example may be unjustified in another.

Case 01 — Dialogue: Can you attend? If the train runs.

Source. “Can you attend?” “If the train runs.” Minimal recovery: I can attend if the train runs. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Train service is uncertain.

Tempting but unfaithful output. I will attend on time This would replace a licensed omission with a stronger, different or unsupported claim. Why: Conditional answer, not a timetable guarantee. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Retain condition with attendance. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 02 — Dialogue: Ready to send? Not quite.

Source. “Ready to send?” “Not quite.” Minimal recovery: Not quite ready to send. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: The reply is a status answer.

Tempting but unfaithful output. The message has been sent This would replace a licensed omission with a stronger, different or unsupported claim. Why: Degree of readiness, not completed action. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Keep incompleteness. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 03 — Dialogue: Tea or water? Water.

Source. “Tea or water?” “Water.” Minimal recovery: I would like water. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: There is one requested choice.

Tempting but unfaithful output. Bring both tea and water This would replace a licensed omission with a stronger, different or unsupported claim. Why: The question offers alternatives. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Preserve exclusive choice. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 04 — Dialogue: Did he sign? The first page.

Source. “Did he sign?” “The first page.” Minimal recovery: He signed the first page. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Only part of the document is confirmed.

Tempting but unfaithful output. He signed the entire contract This would replace a licensed omission with a stronger, different or unsupported claim. Why: The answer narrows what was signed. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Limit completion to first page. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 05 — Dialogue: Why leave early? Family reasons.

Source. “Why leave early?” “Family reasons.” Minimal recovery: I left early for family reasons. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Privacy may be intentional.

Tempting but unfaithful output. The departure was due to illness This would replace a licensed omission with a stronger, different or unsupported claim. Why: Cause is given only at a broad level. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Avoid specific invented cause. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 06 — Dialogue: Will they accept it? Maybe, after review.

Source. “Will they accept it?” “Maybe, after review.” Minimal recovery: They may accept it after review. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: No guarantee has been issued.

Tempting but unfaithful output. They are required to accept it This would replace a licensed omission with a stronger, different or unsupported claim. Why: Modal uncertainty and review condition coexist. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Keep both constraints. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 07 — Verb phrase: Rina will revise the summary; Omar will too.

Source. “Rina will revise the summary; Omar will too.” Minimal recovery: Omar will revise the summary. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Two workers share a proposed act.

Tempting but unfaithful output. Omar will publish the summary This would replace a licensed omission with a stronger, different or unsupported claim. Why: Will inherits revise, not publish. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. No extra stage of publication. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 08 — Verb phrase: The camera can record, but the microphone cannot.

Source. “The camera can record, but the microphone cannot.” Minimal recovery: The microphone cannot record. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: No reason for inability is given.

Tempting but unfaithful output. The microphone is broken This would replace a licensed omission with a stronger, different or unsupported claim. Why: Ability and equipment failure differ. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Keep negative ability only. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 09 — Verb phrase: I have checked the figures, and she has as well.

Source. “I have checked the figures, and she has as well.” Minimal recovery: She has checked the figures. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Completion is asserted for both actors.

Tempting but unfaithful output. She has approved the figures This would replace a licensed omission with a stronger, different or unsupported claim. Why: Checking is not approval. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Retain the same predicate. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 10 — Verb phrase: They might appeal; we might not.

Source. “They might appeal; we might not.” Minimal recovery: We might not appeal. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: The scope of might must survive.

Tempting but unfaithful output. We have no right to appeal This would replace a licensed omission with a stronger, different or unsupported claim. Why: A possible future decision is not a legal bar. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Avoid prohibition. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 11 — Verb phrase: The editor has read the draft; the reviewer hasn’t.

Source. “The editor has read the draft; the reviewer hasn’t.” Minimal recovery: The reviewer has not read the draft. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: The auxiliary carries perfect aspect.

Tempting but unfaithful output. The reviewer rejected the draft This would replace a licensed omission with a stronger, different or unsupported claim. Why: Non-reading is not rejection. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Protect event identity. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 12 — Verb phrase: I can open the gate, and you can too.

Source. “I can open the gate, and you can too.” Minimal recovery: You can open the gate. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: A preceding access policy is needed to disambiguate.

Tempting but unfaithful output. You have permission to open the gate This would replace a licensed omission with a stronger, different or unsupported claim. Why: Can may express ability or permission. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Do not force one modal reading. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 13 — Gapping: Amina ordered soup; Ben, salad.

Source. “Amina ordered soup; Ben, salad.” Minimal recovery: Ben ordered salad. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Nothing confirms consumption.

Tempting but unfaithful output. Ben ate salad This would replace a licensed omission with a stronger, different or unsupported claim. Why: The shared verb is ordered. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Maintain act of ordering. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 14 — Gapping: The first team measured length; the second, width.

Source. “The first team measured length; the second, width.” Minimal recovery: The second team measured width. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: This is a parallel experimental assignment.

Tempting but unfaithful output. The second team estimated width This would replace a licensed omission with a stronger, different or unsupported claim. Why: Measurement and estimation are different methods. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Repeat measure only as required. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 15 — Gapping: Mara sent the invoice on Monday; Ellis, the receipt on Tuesday.

Source. “Mara sent the invoice on Monday; Ellis, the receipt on Tuesday.” Minimal recovery: Ellis sent the receipt on Tuesday. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Separate day and object belong to Ellis.

Tempting but unfaithful output. Ellis received the receipt on Tuesday This would replace a licensed omission with a stronger, different or unsupported claim. Why: Shared sent determines action. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. No role reversal. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 16 — Gapping: One route crosses the valley; another, the ridge.

Source. “One route crosses the valley; another, the ridge.” Minimal recovery: Another route crosses the ridge. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Geometry of the route matters.

Tempting but unfaithful output. Another route begins at the ridge This would replace a licensed omission with a stronger, different or unsupported claim. Why: The inherited predicate is crosses. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Preserve crossing relation. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 17 — Gapping: The trial group saw a change; the control group, none.

Source. “The trial group saw a change; the control group, none.” Minimal recovery: The control group saw no change. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Both groups are compared.

Tempting but unfaithful output. The control group was not tested This would replace a licensed omission with a stronger, different or unsupported claim. Why: None negates observed change, not participation. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Keep negative observation. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 18 — Gapping: The north panel showed green; the south, amber.

Source. “The north panel showed green; the south, amber.” Minimal recovery: The south panel showed amber. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: A diagnostic display is reported.

Tempting but unfaithful output. The south panel requires replacement This would replace a licensed omission with a stronger, different or unsupported claim. Why: Display color does not authorize a repair. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Do not escalate status. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 19 — Sluicing: Someone moved the box, but we do not know who.

Source. “Someone moved the box, but we do not know who.” Minimal recovery: We do not know who moved the box. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: The source reports the movement but not mover.

Tempting but unfaithful output. The warehouse team moved the box This would replace a licensed omission with a stronger, different or unsupported claim. Why: Identity remains unknown. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Retain open variable. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 20 — Sluicing: They met somewhere; she won’t say where.

Source. “They met somewhere; she won’t say where.” Minimal recovery: She will not disclose where they met. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: The refusal to disclose is communicative.

Tempting but unfaithful output. They met at headquarters This would replace a licensed omission with a stronger, different or unsupported claim. Why: Location has been withheld. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Do not infer location. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 21 — Sluicing: The schedule changed; no one explained why.

Source. “The schedule changed; no one explained why.” Minimal recovery: No one explained why it changed. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: A change is known, cause is not.

Tempting but unfaithful output. It changed because of staffing This would replace a licensed omission with a stronger, different or unsupported claim. Why: The reason is unreported. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Keep causal question unresolved. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 22 — Sluicing: He bought a tool; I forgot which.

Source. “He bought a tool; I forgot which.” Minimal recovery: I forgot which tool he bought. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: The speaker knows purchase occurred.

Tempting but unfaithful output. He bought a drill This would replace a licensed omission with a stronger, different or unsupported claim. Why: Tool identity is indeterminate. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Preserve forgotten detail. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 23 — Sluicing: A file was deleted, but we can’t tell when.

Source. “A file was deleted, but we can’t tell when.” Minimal recovery: We cannot tell when the file was deleted. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: A forensic review may still be in progress.

Tempting but unfaithful output. The file was deleted overnight This would replace a licensed omission with a stronger, different or unsupported claim. Why: Unknown time must remain unknown. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Do not manufacture timestamp. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 24 — Sluicing: The witness mentioned a price; the notes omit how much.

Source. “The witness mentioned a price; the notes omit how much.” Minimal recovery: The notes do not specify the price amount. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Price value is missing.

Tempting but unfaithful output. The cost was high This would replace a licensed omission with a stronger, different or unsupported claim. Why: How much asks quantity, not evaluation. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Preserve missing number. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 25 — Stripping: The manager requested photos, not recordings.

Source. “The manager requested photos, not recordings.” Minimal recovery: The manager did not request recordings. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: A project policy is not stated.

Tempting but unfaithful output. Recordings are prohibited everywhere This would replace a licensed omission with a stronger, different or unsupported claim. Why: The correction is about this request. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Keep local scope. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 26 — Stripping: Send the larger envelope, not the narrow one.

Source. “Send the larger envelope, not the narrow one.” Minimal recovery: Send larger envelope rather than narrow envelope. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Object identity rests on context.

Tempting but unfaithful output. Send both envelopes This would replace a licensed omission with a stronger, different or unsupported claim. Why: Corrective contrast selects one option. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Do not merge choices. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 27 — Stripping: Lina approved the design, but not the budget.

Source. “Lina approved the design, but not the budget.” Minimal recovery: Lina did not approve the budget. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Other authorities may act later.

Tempting but unfaithful output. The budget is invalid This would replace a licensed omission with a stronger, different or unsupported claim. Why: The shared approving predicate governs budget. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Avoid global conclusion. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 28 — Stripping: We are meeting on Thursday, not Wednesday.

Source. “We are meeting on Thursday, not Wednesday.” Minimal recovery: The meeting is Thursday. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: The calendar should reflect corrected day.

Tempting but unfaithful output. Wednesday remains the confirmed date This would replace a licensed omission with a stronger, different or unsupported claim. Why: A correction substitutes the day. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Honor scope of correction. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 29 — Stripping: He asked for guidance, not a guarantee.

Source. “He asked for guidance, not a guarantee.” Minimal recovery: He did not ask for a guarantee. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Nothing proves what he received.

Tempting but unfaithful output. He received a guarantee This would replace a licensed omission with a stronger, different or unsupported claim. Why: Request type is corrected. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Keep question of outcome open. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 30 — Stripping: Offer plain text, not a password.

Source. “Offer plain text, not a password.” Minimal recovery: Offer plain text instead of a password. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Sensitive-data scope is critical.

Tempting but unfaithful output. Collect the user’s password This would replace a licensed omission with a stronger, different or unsupported claim. Why: The contrast prohibits changing object of offer. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Do not add collection instruction. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 31 — Answers: Which version passed? The earlier one.

Source. “Which version passed?” “The earlier one.” Minimal recovery: The earlier version passed. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Version IDs are not stated.

Tempting but unfaithful output. Both versions passed This would replace a licensed omission with a stronger, different or unsupported claim. Why: One version is identified relationally. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Do not invent a release number. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 32 — Answers: How many returned? Only four.

Source. “How many returned?” “Only four.” Minimal recovery: Only four returned. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Exactness depends on count being complete.

Tempting but unfaithful output. At least four returned This would replace a licensed omission with a stronger, different or unsupported claim. Why: Only creates upper-bound/exact focus in this context. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Preserve restrictive particle. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 33 — Answers: Is the bridge open? Not yet.

Source. “Is the bridge open?” “Not yet.” Minimal recovery: The bridge is not open yet. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: No date is promised.

Tempting but unfaithful output. The bridge will never reopen This would replace a licensed omission with a stronger, different or unsupported claim. Why: Not yet leaves future open. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Avoid permanent closure. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 34 — Answers: Who noticed the error? The intern.

Source. “Who noticed the error?” “The intern.” Minimal recovery: The intern noticed the error. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: The question asks identity of notice.

Tempting but unfaithful output. The intern caused the error This would replace a licensed omission with a stronger, different or unsupported claim. Why: Observer and cause are different roles. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. No blame attribution. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 35 — Answers: What did they store? The backups.

Source. “What did they store?” “The backups.” Minimal recovery: They stored the backups. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Question fixes the predicate.

Tempting but unfaithful output. They restored the backups This would replace a licensed omission with a stronger, different or unsupported claim. Why: Store and restore differ by one letter but not meaning. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Retain storage. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 36 — Answers: When should we retry? After confirmation.

Source. “When should we retry?” “After confirmation.” Minimal recovery: Retry after confirmation. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Confirmation has not yet occurred.

Tempting but unfaithful output. Retry now This would replace a licensed omission with a stronger, different or unsupported claim. Why: The answer supplies a temporal condition. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Preserve prerequisite. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 37 — Instructions: Check the seal. If damaged, replace.

Source. “Check the seal. If damaged, replace.” Minimal recovery: Replace the seal if the seal is damaged. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: A maintenance instruction controls action.

Tempting but unfaithful output. Replace every seal routinely This would replace a licensed omission with a stronger, different or unsupported claim. Why: Conditional applies to the preceding item. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Minimal explicitness acceptable. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 38 — Instructions: Inspect valve A; valve B if pressure rises.

Source. “Inspect valve A; valve B if pressure rises.” Minimal recovery: Inspect valve B if pressure rises. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Condition applies only to valve B unless stated otherwise.

Tempting but unfaithful output. Inspect valve B immediately This would replace a licensed omission with a stronger, different or unsupported claim. Why: The second instruction inherits inspect. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Preserve list scoping. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 39 — Instructions: Upload the draft, then the signed version.

Source. “Upload the draft, then the signed version.” Minimal recovery: Upload the signed version after the draft. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Signature state already exists in object name.

Tempting but unfaithful output. Sign the draft online immediately This would replace a licensed omission with a stronger, different or unsupported claim. Why: Second action inherits upload. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Do not add step. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 40 — Instructions: Backup first. Restart only afterward.

Source. “Backup first. Restart only afterward.” Minimal recovery: Restart only after backup. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Order is safety-critical.

Tempting but unfaithful output. Restart before backup This would replace a licensed omission with a stronger, different or unsupported claim. Why: Afterward refers to completed backup action. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Keep temporal gate. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 41 — Instructions: Review two samples; more if required.

Source. “Review two samples; more if required.” Minimal recovery: Review more than two samples if required. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Additional quantity conditional.

Tempting but unfaithful output. Review exactly two in all cases This would replace a licensed omission with a stronger, different or unsupported claim. Why: Second clause inherits review. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Preserve minimum plus option. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 42 — Instructions: Set the dial to low; the second to medium.

Source. “Set the dial to low; the second to medium.” Minimal recovery: Set the second dial to medium. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: There are two dials.

Tempting but unfaithful output. Set both dials to medium This would replace a licensed omission with a stronger, different or unsupported claim. Why: Shared verb set, distinct targets. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Don’t overwrite first setting. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 43 — Formal: The applicant may respond; the sponsor may not.

Source. “The applicant may respond; the sponsor may not.” Minimal recovery: The sponsor may not respond. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: A normative rule is at issue.

Tempting but unfaithful output. The sponsor failed to respond This would replace a licensed omission with a stronger, different or unsupported claim. Why: Permission or entitlement differs from past failure. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Preserve modal category. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 44 — Formal: Receipt by Friday; supporting material by Tuesday.

Source. “Receipt by Friday; supporting material by Tuesday.” Minimal recovery: Supporting material must be received by Tuesday if heading establishes receipt requirement. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Heading determines noun-predicate inheritance.

Tempting but unfaithful output. Supporting material is due Friday This would replace a licensed omission with a stronger, different or unsupported claim. Why: Two deadlines must not be merged. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Check document hierarchy. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 45 — Formal: The committee examined the claim; the panel, the evidence.

Source. “The committee examined the claim; the panel, the evidence.” Minimal recovery: The panel examined the evidence. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Institutional roles remain separate.

Tempting but unfaithful output. The panel adopted the evidence This would replace a licensed omission with a stronger, different or unsupported claim. Why: Examination is not acceptance. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Use same evidentiary verb. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 46 — Formal: One signature suffices for draft review; two for release.

Source. “One signature suffices for draft review; two for release.” Minimal recovery: Two signatures suffice for release. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Draft and release gates are distinct.

Tempting but unfaithful output. One signature suffices for final release This would replace a licensed omission with a stronger, different or unsupported claim. Why: Quantifier differs by stage. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Maintain threshold split. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 47 — Formal: The amendment applies to the schedule, not the fee.

Source. “The amendment applies to the schedule, not the fee.” Minimal recovery: It does not apply to the fee. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Scope restricts change.

Tempting but unfaithful output. The fee was waived This would replace a licensed omission with a stronger, different or unsupported claim. Why: Non-application is not forgiveness. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Do not generate new financial entitlement. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 48 — Formal: The supplier warrants quality; the carrier, timely delivery.

Source. “The supplier warrants quality; the carrier, timely delivery.” Minimal recovery: The carrier warrants timely delivery. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Parallel legal structures may be consequential.

Tempting but unfaithful output. The carrier guarantees product quality This would replace a licensed omission with a stronger, different or unsupported claim. Why: Separate subjects and obligations. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Review governing definitions. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 49 — Narrative: Mina watched the door. Still locked.

Source. “Mina watched the door. Still locked.” Minimal recovery: The door remained locked. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Viewpoint remains with Mina.

Tempting but unfaithful output. Mina locked the door This would replace a licensed omission with a stronger, different or unsupported claim. Why: A state fragment is not an action. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Protect event chronology. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 50 — Narrative: He reached for the letter. Too late.

Source. “He reached for the letter. Too late.” Minimal recovery: The reaching was too late for its implicit purpose. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Narrative withholding builds suspense.

Tempting but unfaithful output. He destroyed the letter This would replace a licensed omission with a stronger, different or unsupported claim. Why: The purpose is not specified. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Avoid adding event. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 51 — Narrative: Two lights in the distance; one fading.

Source. “Two lights in the distance; one fading.” Minimal recovery: One of the two lights was fading. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: One remains uncharacterized.

Tempting but unfaithful output. Both lights went out This would replace a licensed omission with a stronger, different or unsupported claim. Why: The verbless description tracks one light. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Retain visual contrast. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 52 — Narrative: She wanted to ask. Couldn’t.

Source. “She wanted to ask. Couldn’t.” Minimal recovery: She could not ask. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Reason for inability is unstated.

Tempting but unfaithful output. She chose not to ask This would replace a licensed omission with a stronger, different or unsupported claim. Why: Inability and refusal differ. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Retain distinction. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 53 — Narrative: A pause, then laughter.

Source. “A pause, then laughter.” Minimal recovery: There was a pause and then laughter. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Sequence is a narrative effect.

Tempting but unfaithful output. Everyone agreed This would replace a licensed omission with a stronger, different or unsupported claim. Why: Sound is not agreement. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Do not assign shared intent. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 54 — Narrative: He said yes. She, nothing.

Source. “He said yes. She, nothing.” Minimal recovery: She said nothing. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Contrast is about response.

Tempting but unfaithful output. She consented silently This would replace a licensed omission with a stronger, different or unsupported claim. Why: Silence may not be consent. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Keep silence. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 55 — Subtitles: Again? Not today.

Source. “Again?” “Not today.” Minimal recovery: The proposed activity is declined today. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Prior action may be visible on screen.

Tempting but unfaithful output. It will never happen again This would replace a licensed omission with a stronger, different or unsupported claim. Why: Short reply restricts time only. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Preserve no-for-today. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 56 — Subtitles: The red one!

Source. “The red one!” Minimal recovery: Choose or notice the red one depending on scene. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Without video the command is ambiguous.

Tempting but unfaithful output. Buy the red one This would replace a licensed omission with a stronger, different or unsupported claim. Why: Visual context determines speech act. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Request screen context. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 57 — Subtitles: Over there—now!

Source. “Over there—now!” Minimal recovery: Act or look over there now as scene licenses. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Gesture and scene may supply action.

Tempting but unfaithful output. Move permanently to that location This would replace a licensed omission with a stronger, different or unsupported claim. Why: Urgency is explicit, event may be underdetermined. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Don’t invent motive. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 58 — Subtitles: Too many. Stop.

Source. “Too many. Stop.” Minimal recovery: There are too many; stop the established activity. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Compression must not create new instruction.

Tempting but unfaithful output. Evacuate the entire building This would replace a licensed omission with a stronger, different or unsupported claim. Why: The referent and act depend on prior line. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Ask for preceding clip. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 59 — Subtitles: I thought you would.

Source. “I thought you would.” Minimal recovery: I thought you would perform the previously discussed action. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Surprise or disappointment may be pragmatic.

Tempting but unfaithful output. You already performed it This would replace a licensed omission with a stronger, different or unsupported claim. Why: Expected future action differs from completed action. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Preserve temporal attitude. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 60 — Subtitles: No, the left!

Source. “No, the left!” Minimal recovery: The intended object or direction is the left one. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Screen geometry may resolve left.

Tempting but unfaithful output. Turn right This would replace a licensed omission with a stronger, different or unsupported claim. Why: Corrective fragment requires spatial anchor. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Keep corrective force. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 61 — Health: Dose unchanged; timing adjusted.

Source. “Dose unchanged; timing adjusted.” Minimal recovery: The dose remains unchanged; the timing was adjusted. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Health-related wording needs specialist review.

Tempting but unfaithful output. The dose was increased This would replace a licensed omission with a stronger, different or unsupported claim. Why: Two distinct attributes are reported. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Do not infer prescribed regimen. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 62 — Health: The first sample was adequate; the second, not.

Source. “The first sample was adequate; the second, not.” Minimal recovery: The second sample was not adequate. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Both cases are sample quality judgments.

Tempting but unfaithful output. The second sample was positive This would replace a licensed omission with a stronger, different or unsupported claim. Why: Adequacy is not a test result. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Separate quality from finding. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 63 — Health: No symptoms overnight; mild discomfort in morning.

Source. “No symptoms overnight; mild discomfort in morning.” Minimal recovery: No symptoms were reported overnight; mild discomfort was present in morning. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Record may be observation not diagnosis.

Tempting but unfaithful output. There was continuous severe pain This would replace a licensed omission with a stronger, different or unsupported claim. Why: Temporal contrast is essential. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Don’t intensify severity. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 64 — Health: Review after testing, earlier if symptoms worsen.

Source. “Review after testing, earlier if symptoms worsen.” Minimal recovery: Review earlier than planned if symptoms worsen. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Clinical judgment requires qualified interpreter.

Tempting but unfaithful output. Never review until testing This would replace a licensed omission with a stronger, different or unsupported claim. Why: Exception changes review timing. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Preserve conditional override. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 65 — Health: Three reports completed; one pending.

Source. “Three reports completed; one pending.” Minimal recovery: One report remains pending. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Status differs across counted items.

Tempting but unfaithful output. All four reports were completed This would replace a licensed omission with a stronger, different or unsupported claim. Why: The noun report is inherited. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Maintain completed/pending split. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 66 — Health: A negative screen in one group; an inconclusive one in another.

Source. “A negative screen in one group; an inconclusive one in another.” Minimal recovery: The other group’s screen was inconclusive. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Result categories are not synonyms.

Tempting but unfaithful output. The second group tested negative This would replace a licensed omission with a stronger, different or unsupported claim. Why: Inconclusive is not negative. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Keep uncertainty. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 67 — Analytics: Visits rose in June; conversions, not.

Source. “Visits rose in June; conversions, not.” Minimal recovery: Conversions did not rise in June. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Two metrics are compared.

Tempting but unfaithful output. Conversions fell in June This would replace a licensed omission with a stronger, different or unsupported claim. Why: Not rising includes flat or falling. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Do not force decline. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 68 — Analytics: Costs: stable. Revenue: higher.

Source. “Costs: stable. Revenue: higher.” Minimal recovery: Costs stayed stable; revenue was higher relative to the established baseline. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Baseline may need clarification.

Tempting but unfaithful output. Profit necessarily doubled This would replace a licensed omission with a stronger, different or unsupported claim. Why: Two metrics lack arithmetic relation without amounts. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. No inferred profits. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 69 — Analytics: The first model achieved 80%; the second, 83%.

Source. “The first model achieved 80%; the second, 83%.” Minimal recovery: The second model achieved 83% under the same metric if specified. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Comparison needs methodological context.

Tempting but unfaithful output. The second model is universally better This would replace a licensed omission with a stronger, different or unsupported claim. Why: Same metric and sample must be verified. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Do not invent evaluation design. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 70 — Analytics: Region A used 12 units; B, 15.

Source. “Region A used 12 units; B, 15.” Minimal recovery: Region B used 15 units. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Quantities are not invoices.

Tempting but unfaithful output. Region B bought 15 units This would replace a licensed omission with a stronger, different or unsupported claim. Why: Used differs from purchased. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Preserve usage. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 71 — Analytics: Baseline: 100. After change: 92.

Source. “Baseline: 100. After change: 92.” Minimal recovery: The measured quantity moved from 100 to 92. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: A decrease is not automatically financial savings.

Tempting but unfaithful output. The intervention saved 8 dollars This would replace a licensed omission with a stronger, different or unsupported claim. Why: The unit and causal relation are unspecified. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Require measure label. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Case 72 — Analytics: More respondents preferred version B; fewer, version A.

Source. “More respondents preferred version B; fewer, version A.” Minimal recovery: Fewer respondents preferred A than preferred B. This is the meaning hypothesis to test, not a license to introduce every recovered word into the final target sentence. The crucial antecedent is established by the words already supplied and, where needed, by the documented setting: Other options may exist.

Tempting but unfaithful output. All respondents preferred B This would replace a licensed omission with a stronger, different or unsupported claim. Why: Comparative distribution is not unanimity. Compare the original and proposed reading by asking whether the subject, event, quantity, obligation, time or uncertainty has changed. If a reader could now make a decision that the original did not justify, the expansion fails even when the sentence is fluent.

Better translation decision. Keep comparative only. In a target language that allows the same omission, a concise reconstruction may be unnecessary; in a language that requires overt arguments, express the minimum licensed material. Check the full passage before adding pronouns or repeated predicates. A reviewer should record an antecedent span and list any elements supplied purely for grammar, then separate those from factual content. This yields an auditable resolution rather than a guess disguised as clarity.

Diagnostic challenge. Remove the surrounding question, heading or prior clause and read the short fragment alone. What now becomes ambiguous? Restore the context and state exactly which ambiguity disappeared. If a supposed recovery remains equally plausible without the context, it may be a conventional interpretation rather than a context-proven fact. Finally draft a different but faithful wording and compare what a new reader infers. The pair is equivalent only if the same semantic commitments survive.

Failure modes: what to inspect before approving a translation

Failure 01 — Invented agent

How it happens. A passive has no named actor, but the translation supplies an institution or employee. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Separate grammatical subject requirements from factual identification. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 02 — Unlicensed object

How it happens. An imperative inherits an object from one of several candidate nouns. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Re-open the antecedent list and seek the correct item identifier. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 03 — Over-explicit paraphrase

How it happens. The target explains a motive that the source deliberately leaves tacit. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Trim back to documented roles and events. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 04 — Wrong gapped verb

How it happens. A second coordinate clause gets a verb related in theme but not identical in predicate. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Copy the precise event label into the private reconstruction. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 05 — Wrong tense

How it happens. A missing predicate receives a completed-past form when the source remains prospective. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Track the surviving auxiliary or time phrase. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 06 — Wrong aspect

How it happens. A process becomes a finished event. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Distinguish ongoing, habitual, attempted and completed predicates. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 07 — Wrong polarity

How it happens. A negative fragment is attached to the wrong alternative. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Mark the constituent that not or no actually rejects. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 08 — Wrong modality

How it happens. The translator turns may into must or could into did. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Preserve possibility, permission, ability and obligation separately. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 09 — Wrong question variable

How it happens. Who becomes where or vice versa. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Keep the interrogative remnant tied to the right participant or location slot. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 10 — Unjustified answer

How it happens. A sluiced question is replaced with a guessed proper name. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Retain the unknown variable. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 11 — Suppressed hesitation

How it happens. An intentionally unfinished sentence is completed. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Respect interrupted speech and its rhetorical function. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 12 — Treating dots as syntax

How it happens. Three dots in an edited quotation are mistaken for ordinary grammar. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Check whether omissions are editorial, literary or conversational. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 13 — Misread table dash

How it happens. A dash or blank cell is copied as the previous value. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Consult column conventions and metadata. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 14 — Incorrect list inheritance

How it happens. A bullet inherits a previous bullet’s verb rather than the heading. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Recover from structural parents, not merely nearest words. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 15 — Collapsed comparison

How it happens. One shared measure is replaced by another. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Write a baseline and measurement note. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 16 — Overstated only

How it happens. A restricted answer is translated as a minimum. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Test upper/lower-bound implications. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 17 — False certainty

How it happens. Maybe or perhaps disappears from a fragment. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Reinsert the modal uncertainty or select an equivalent target construction. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 18 — False impossibility

How it happens. Not yet becomes never. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Keep temporal openness. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 19 — Blame through pronouns

How it happens. A subject is assigned to the wrong person in a topic chain. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Track named entities across every switch. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 20 — Silence as consent

How it happens. A non-response is rewritten as agreement. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Represent silence as silence. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 21 — Screen-dependent guess

How it happens. A caption or subtitle is translated without the crucial image. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Request the visual source or flag uncertainty. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 22 — Invented date

How it happens. After review becomes a fixed weekday. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Preserve relative temporal anchor. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 23 — Negation attachment

How it happens. Not a budget approval becomes budget rejection in general. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Restrict the negation to the correct proposition. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 24 — Double recovery

How it happens. Both a pronoun and named noun are added so the target appears to contain two actors. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Perform referent identity check. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 25 — Artificial repetition

How it happens. A sentence repeats a full antecedent when target-language natural ellipsis would be clear. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Compare meaning and register, not length. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 26 — Unnatural overcompression

How it happens. A target fragment becomes incomprehensible because its language requires an overt subject. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Use minimal grammatical completion. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 27 — Policy inflation

How it happens. Not requested becomes prohibited. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Keep local preference separate from institution-wide rule. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 28 — Stage inflation

How it happens. Checked becomes approved, or approved becomes published. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Track workflow phase verbs. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 29 — Count inflation

How it happens. A fragment describing four items is expanded to all items. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Protect numeric scope. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 30 — Wrong procedural order

How it happens. Afterward is attached to the wrong prior step. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Represent procedure as ordered events. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 31 — Context leakage

How it happens. A detail from another project or prior conversation is injected. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Scope contextual evidence to the current source. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 32 — Machine output confidence

How it happens. Fluent MT hides an unsupported completion. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Require alignment to source predicates and antecedents. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 33 — Glossary misuse

How it happens. A once-correct substitution is applied automatically in a new context. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Treat resolved ellipsis as context-bound rather than a term replacement. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 34 — Audience mismatch

How it happens. A clipped intimate reply is rendered as a formal lecture. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Preserve pragmatic register with target conventions. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 35 — No escalation channel

How it happens. The translator resolves multiple candidate antecedents privately. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Record uncertainty and query the source owner. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Failure 36 — Unreviewed change

How it happens. Source revision alters the antecedent but translation is not rechecked. A sentence can pass spelling, grammar and terminology review while failing this test, because the error lies in the relationship between the fragment and its surrounding evidence. Such errors should be reviewed at the discourse level, not only as isolated segments.

Repair and test. Re-run all dependent gap resolutions after source changes. The editor should compare source and target in a side-by-side view, highlight every element that was not overtly present in the fragment, and identify the exact construction or context that licenses it. If the addition alters what readers can infer about agency, time, conditions or certainty, mark it as a substantive error. Where more than one valid completion remains, retain the source ambiguity or request clarification rather than choosing a convenient narrative.

Transfer laboratories: applying the method outside the classroom

Laboratory 1 — Customer support chat

Task. Review a six-turn exchange in which the final answer is “Only after verification”. Before writing the target, mark what is expressed, what is withheld and what is available through the source setting. Acceptance criterion: Identify the request qualified by after and confirm that verification is not represented as complete. A good solution is neither maximally short nor maximally explicit; it is the least intervention that preserves the source commitments.

Perform a three-version test: first write a literal fragment, then a minimally expanded target, then a polished version suitable for the actual audience. Give each version an antecedent pointer and a risk note. Reject any version that changes the triggering condition, the participant, the event stage or the confidence level. If a version communicates the source accurately but sounds unnatural, improve word order or morphology before adding facts. Record what evidence would change the decision, so another reviewer can reproduce the reasoning.

Laboratory 2 — Product help interface

Task. Translate button text “Try again” beside an error message. Before writing the target, mark what is expressed, what is withheld and what is available through the source setting. Acceptance criterion: Determine the action retried from the surrounding screen rather than assuming a purchase. A good solution is neither maximally short nor maximally explicit; it is the least intervention that preserves the source commitments.

Perform a three-version test: first write a literal fragment, then a minimally expanded target, then a polished version suitable for the actual audience. Give each version an antecedent pointer and a risk note. Reject any version that changes the triggering condition, the participant, the event stage or the confidence level. If a version communicates the source accurately but sounds unnatural, improve word order or morphology before adding facts. Record what evidence would change the decision, so another reviewer can reproduce the reasoning.

Laboratory 3 — Film dialogue

Task. A character says “Not this time” after an offer. Before writing the target, mark what is expressed, what is withheld and what is available through the source setting. Acceptance criterion: Retain the local refusal without portraying a permanent decision. A good solution is neither maximally short nor maximally explicit; it is the least intervention that preserves the source commitments.

Perform a three-version test: first write a literal fragment, then a minimally expanded target, then a polished version suitable for the actual audience. Give each version an antecedent pointer and a risk note. Reject any version that changes the triggering condition, the participant, the event stage or the confidence level. If a version communicates the source accurately but sounds unnatural, improve word order or morphology before adding facts. Record what evidence would change the decision, so another reviewer can reproduce the reasoning.

Laboratory 4 — Meeting notes

Task. Two agenda items inherit “requires approval” from a heading. Before writing the target, mark what is expressed, what is withheld and what is available through the source setting. Acceptance criterion: Do not transfer approval requirement to items in the following ‘informational’ section. A good solution is neither maximally short nor maximally explicit; it is the least intervention that preserves the source commitments.

Perform a three-version test: first write a literal fragment, then a minimally expanded target, then a polished version suitable for the actual audience. Give each version an antecedent pointer and a risk note. Reject any version that changes the triggering condition, the participant, the event stage or the confidence level. If a version communicates the source accurately but sounds unnatural, improve word order or morphology before adding facts. Record what evidence would change the decision, so another reviewer can reproduce the reasoning.

Laboratory 5 — Scientific abstract

Task. The second experiment is described only as “the same, except at a lower temperature”. Before writing the target, mark what is expressed, what is withheld and what is available through the source setting. Acceptance criterion: Write out exactly which design features are shared and identify the sole stated exception. A good solution is neither maximally short nor maximally explicit; it is the least intervention that preserves the source commitments.

Perform a three-version test: first write a literal fragment, then a minimally expanded target, then a polished version suitable for the actual audience. Give each version an antecedent pointer and a risk note. Reject any version that changes the triggering condition, the participant, the event stage or the confidence level. If a version communicates the source accurately but sounds unnatural, improve word order or morphology before adding facts. Record what evidence would change the decision, so another reviewer can reproduce the reasoning.

Laboratory 6 — Quality report

Task. One row says “passed” and the next “not tested”. Before writing the target, mark what is expressed, what is withheld and what is available through the source setting. Acceptance criterion: Avoid converting absence of testing into failure. A good solution is neither maximally short nor maximally explicit; it is the least intervention that preserves the source commitments.

Perform a three-version test: first write a literal fragment, then a minimally expanded target, then a polished version suitable for the actual audience. Give each version an antecedent pointer and a risk note. Reject any version that changes the triggering condition, the participant, the event stage or the confidence level. If a version communicates the source accurately but sounds unnatural, improve word order or morphology before adding facts. Record what evidence would change the decision, so another reviewer can reproduce the reasoning.

Laboratory 7 — Contract schedule

Task. A date appears under “filing deadlines” without a verb. Before writing the target, mark what is expressed, what is withheld and what is available through the source setting. Acceptance criterion: Recover the filing obligation only if the heading and governing clause support it. A good solution is neither maximally short nor maximally explicit; it is the least intervention that preserves the source commitments.

Perform a three-version test: first write a literal fragment, then a minimally expanded target, then a polished version suitable for the actual audience. Give each version an antecedent pointer and a risk note. Reject any version that changes the triggering condition, the participant, the event stage or the confidence level. If a version communicates the source accurately but sounds unnatural, improve word order or morphology before adding facts. Record what evidence would change the decision, so another reviewer can reproduce the reasoning.

Laboratory 8 — Research interview

Task. The speaker answers “My sister” to a question about who helped. Before writing the target, mark what is expressed, what is withheld and what is available through the source setting. Acceptance criterion: Preserve helper identity without turning help into sole responsibility. A good solution is neither maximally short nor maximally explicit; it is the least intervention that preserves the source commitments.

Perform a three-version test: first write a literal fragment, then a minimally expanded target, then a polished version suitable for the actual audience. Give each version an antecedent pointer and a risk note. Reject any version that changes the triggering condition, the participant, the event stage or the confidence level. If a version communicates the source accurately but sounds unnatural, improve word order or morphology before adding facts. Record what evidence would change the decision, so another reviewer can reproduce the reasoning.

Laboratory 9 — Museum caption

Task. An artwork description begins “After the fire”. Before writing the target, mark what is expressed, what is withheld and what is available through the source setting. Acceptance criterion: Maintain time relation without inventing a restoration story. A good solution is neither maximally short nor maximally explicit; it is the least intervention that preserves the source commitments.

Perform a three-version test: first write a literal fragment, then a minimally expanded target, then a polished version suitable for the actual audience. Give each version an antecedent pointer and a risk note. Reject any version that changes the triggering condition, the participant, the event stage or the confidence level. If a version communicates the source accurately but sounds unnatural, improve word order or morphology before adding facts. Record what evidence would change the decision, so another reviewer can reproduce the reasoning.

Laboratory 10 — Subtitling

Task. An off-screen voice says “The other one”. Before writing the target, mark what is expressed, what is withheld and what is available through the source setting. Acceptance criterion: Use visual referents and preceding dialogue, and query if multiple objects qualify. A good solution is neither maximally short nor maximally explicit; it is the least intervention that preserves the source commitments.

Perform a three-version test: first write a literal fragment, then a minimally expanded target, then a polished version suitable for the actual audience. Give each version an antecedent pointer and a risk note. Reject any version that changes the triggering condition, the participant, the event stage or the confidence level. If a version communicates the source accurately but sounds unnatural, improve word order or morphology before adding facts. Record what evidence would change the decision, so another reviewer can reproduce the reasoning.

Laboratory 11 — News copy

Task. A quote reads “We could, but won’t”. Before writing the target, mark what is expressed, what is withheld and what is available through the source setting. Acceptance criterion: Separate ability from willingness and preserve the speaker’s refusal. A good solution is neither maximally short nor maximally explicit; it is the least intervention that preserves the source commitments.

Perform a three-version test: first write a literal fragment, then a minimally expanded target, then a polished version suitable for the actual audience. Give each version an antecedent pointer and a risk note. Reject any version that changes the triggering condition, the participant, the event stage or the confidence level. If a version communicates the source accurately but sounds unnatural, improve word order or morphology before adding facts. Record what evidence would change the decision, so another reviewer can reproduce the reasoning.

Laboratory 12 — Business email

Task. A reply says “Tuesday works, Friday not”. Before writing the target, mark what is expressed, what is withheld and what is available through the source setting. Acceptance criterion: Carry the same scheduling predicate to both days. A good solution is neither maximally short nor maximally explicit; it is the least intervention that preserves the source commitments.

Perform a three-version test: first write a literal fragment, then a minimally expanded target, then a polished version suitable for the actual audience. Give each version an antecedent pointer and a risk note. Reject any version that changes the triggering condition, the participant, the event stage or the confidence level. If a version communicates the source accurately but sounds unnatural, improve word order or morphology before adding facts. Record what evidence would change the decision, so another reviewer can reproduce the reasoning.

Laboratory 13 — Medical note

Task. “Stable overnight; improved by morning”. Before writing the target, mark what is expressed, what is withheld and what is available through the source setting. Acceptance criterion: Preserve observations without inventing diagnosis or medication. A good solution is neither maximally short nor maximally explicit; it is the least intervention that preserves the source commitments.

Perform a three-version test: first write a literal fragment, then a minimally expanded target, then a polished version suitable for the actual audience. Give each version an antecedent pointer and a risk note. Reject any version that changes the triggering condition, the participant, the event stage or the confidence level. If a version communicates the source accurately but sounds unnatural, improve word order or morphology before adding facts. Record what evidence would change the decision, so another reviewer can reproduce the reasoning.

Laboratory 14 — Survey reporting

Task. “More yes answers in group A; fewer in B”. Before writing the target, mark what is expressed, what is withheld and what is available through the source setting. Acceptance criterion: Preserve comparative relation without assuming absolute sample sizes. A good solution is neither maximally short nor maximally explicit; it is the least intervention that preserves the source commitments.

Perform a three-version test: first write a literal fragment, then a minimally expanded target, then a polished version suitable for the actual audience. Give each version an antecedent pointer and a risk note. Reject any version that changes the triggering condition, the participant, the event stage or the confidence level. If a version communicates the source accurately but sounds unnatural, improve word order or morphology before adding facts. Record what evidence would change the decision, so another reviewer can reproduce the reasoning.

Laboratory 15 — Software release notes

Task. “Fixed on mobile; desktop pending”. Before writing the target, mark what is expressed, what is withheld and what is available through the source setting. Acceptance criterion: Do not portray the desktop issue as resolved. A good solution is neither maximally short nor maximally explicit; it is the least intervention that preserves the source commitments.

Perform a three-version test: first write a literal fragment, then a minimally expanded target, then a polished version suitable for the actual audience. Give each version an antecedent pointer and a risk note. Reject any version that changes the triggering condition, the participant, the event stage or the confidence level. If a version communicates the source accurately but sounds unnatural, improve word order or morphology before adding facts. Record what evidence would change the decision, so another reviewer can reproduce the reasoning.

Laboratory 16 — Recipe instruction

Task. “Bake for ten minutes; five more if pale”. Before writing the target, mark what is expressed, what is withheld and what is available through the source setting. Acceptance criterion: Recover bake for the additional period only if target language requires it. A good solution is neither maximally short nor maximally explicit; it is the least intervention that preserves the source commitments.

Perform a three-version test: first write a literal fragment, then a minimally expanded target, then a polished version suitable for the actual audience. Give each version an antecedent pointer and a risk note. Reject any version that changes the triggering condition, the participant, the event stage or the confidence level. If a version communicates the source accurately but sounds unnatural, improve word order or morphology before adding facts. Record what evidence would change the decision, so another reviewer can reproduce the reasoning.

Laboratory 17 — Transport alert

Task. “Platform changed; time unchanged”. Before writing the target, mark what is expressed, what is withheld and what is available through the source setting. Acceptance criterion: Preserve separate facts about location and departure time. A good solution is neither maximally short nor maximally explicit; it is the least intervention that preserves the source commitments.

Perform a three-version test: first write a literal fragment, then a minimally expanded target, then a polished version suitable for the actual audience. Give each version an antecedent pointer and a risk note. Reject any version that changes the triggering condition, the participant, the event stage or the confidence level. If a version communicates the source accurately but sounds unnatural, improve word order or morphology before adding facts. Record what evidence would change the decision, so another reviewer can reproduce the reasoning.

Laboratory 18 — Sports report

Task. “First half: three chances; second: one”. Before writing the target, mark what is expressed, what is withheld and what is available through the source setting. Acceptance criterion: Do not equate chances with goals. A good solution is neither maximally short nor maximally explicit; it is the least intervention that preserves the source commitments.

Perform a three-version test: first write a literal fragment, then a minimally expanded target, then a polished version suitable for the actual audience. Give each version an antecedent pointer and a risk note. Reject any version that changes the triggering condition, the participant, the event stage or the confidence level. If a version communicates the source accurately but sounds unnatural, improve word order or morphology before adding facts. Record what evidence would change the decision, so another reviewer can reproduce the reasoning.

Laboratory 19 — Fiction translation

Task. “She turned. Too late.”. Before writing the target, mark what is expressed, what is withheld and what is available through the source setting. Acceptance criterion: Preserve compressed viewpoint rather than specifying an unsupported disaster. A good solution is neither maximally short nor maximally explicit; it is the least intervention that preserves the source commitments.

Perform a three-version test: first write a literal fragment, then a minimally expanded target, then a polished version suitable for the actual audience. Give each version an antecedent pointer and a risk note. Reject any version that changes the triggering condition, the participant, the event stage or the confidence level. If a version communicates the source accurately but sounds unnatural, improve word order or morphology before adding facts. Record what evidence would change the decision, so another reviewer can reproduce the reasoning.

Laboratory 20 — Academic review

Task. “The first argument is clear; the second less so”. Before writing the target, mark what is expressed, what is withheld and what is available through the source setting. Acceptance criterion: Recover clarity scale, not factual truth. A good solution is neither maximally short nor maximally explicit; it is the least intervention that preserves the source commitments.

Perform a three-version test: first write a literal fragment, then a minimally expanded target, then a polished version suitable for the actual audience. Give each version an antecedent pointer and a risk note. Reject any version that changes the triggering condition, the participant, the event stage or the confidence level. If a version communicates the source accurately but sounds unnatural, improve word order or morphology before adding facts. Record what evidence would change the decision, so another reviewer can reproduce the reasoning.

Laboratory 21 — Insurance notice

Task. “Supporting documents now; originals on request”. Before writing the target, mark what is expressed, what is withheld and what is available through the source setting. Acceptance criterion: Preserve request condition, not mandatory immediate originals. A good solution is neither maximally short nor maximally explicit; it is the least intervention that preserves the source commitments.

Perform a three-version test: first write a literal fragment, then a minimally expanded target, then a polished version suitable for the actual audience. Give each version an antecedent pointer and a risk note. Reject any version that changes the triggering condition, the participant, the event stage or the confidence level. If a version communicates the source accurately but sounds unnatural, improve word order or morphology before adding facts. Record what evidence would change the decision, so another reviewer can reproduce the reasoning.

Laboratory 22 — Inventory list

Task. “Blue boxes: sixteen; green: seven”. Before writing the target, mark what is expressed, what is withheld and what is available through the source setting. Acceptance criterion: Carry quantity of boxes without assuming contents. A good solution is neither maximally short nor maximally explicit; it is the least intervention that preserves the source commitments.

Perform a three-version test: first write a literal fragment, then a minimally expanded target, then a polished version suitable for the actual audience. Give each version an antecedent pointer and a risk note. Reject any version that changes the triggering condition, the participant, the event stage or the confidence level. If a version communicates the source accurately but sounds unnatural, improve word order or morphology before adding facts. Record what evidence would change the decision, so another reviewer can reproduce the reasoning.

Laboratory 23 — Emergency message

Task. “Evacuate if instructed; otherwise remain”. Before writing the target, mark what is expressed, what is withheld and what is available through the source setting. Acceptance criterion: Expand only to remain in the established safe location if source context supports it. A good solution is neither maximally short nor maximally explicit; it is the least intervention that preserves the source commitments.

Perform a three-version test: first write a literal fragment, then a minimally expanded target, then a polished version suitable for the actual audience. Give each version an antecedent pointer and a risk note. Reject any version that changes the triggering condition, the participant, the event stage or the confidence level. If a version communicates the source accurately but sounds unnatural, improve word order or morphology before adding facts. Record what evidence would change the decision, so another reviewer can reproduce the reasoning.

Laboratory 24 — Event invitation

Task. “Dinner at seven; introductions at eight”. Before writing the target, mark what is expressed, what is withheld and what is available through the source setting. Acceptance criterion: Do not turn the sequence into mandatory attendance at both. A good solution is neither maximally short nor maximally explicit; it is the least intervention that preserves the source commitments.

Perform a three-version test: first write a literal fragment, then a minimally expanded target, then a polished version suitable for the actual audience. Give each version an antecedent pointer and a risk note. Reject any version that changes the triggering condition, the participant, the event stage or the confidence level. If a version communicates the source accurately but sounds unnatural, improve word order or morphology before adding facts. Record what evidence would change the decision, so another reviewer can reproduce the reasoning.

Laboratory 25 — Multilingual questionnaire

Task. “Agree strongly / agree / neither / disagree”. Before writing the target, mark what is expressed, what is withheld and what is available through the source setting. Acceptance criterion: Carry the common survey proposition consistently across fragments. A good solution is neither maximally short nor maximally explicit; it is the least intervention that preserves the source commitments.

Perform a three-version test: first write a literal fragment, then a minimally expanded target, then a polished version suitable for the actual audience. Give each version an antecedent pointer and a risk note. Reject any version that changes the triggering condition, the participant, the event stage or the confidence level. If a version communicates the source accurately but sounds unnatural, improve word order or morphology before adding facts. Record what evidence would change the decision, so another reviewer can reproduce the reasoning.

Laboratory 26 — Court transcript

Task. “Witness: at the entrance. Counsel: which one?”. Before writing the target, mark what is expressed, what is withheld and what is available through the source setting. Acceptance criterion: Keep question open without determining the entrance. A good solution is neither maximally short nor maximally explicit; it is the least intervention that preserves the source commitments.

Perform a three-version test: first write a literal fragment, then a minimally expanded target, then a polished version suitable for the actual audience. Give each version an antecedent pointer and a risk note. Reject any version that changes the triggering condition, the participant, the event stage or the confidence level. If a version communicates the source accurately but sounds unnatural, improve word order or morphology before adding facts. Record what evidence would change the decision, so another reviewer can reproduce the reasoning.

Laboratory 27 — Press release

Task. “First shipment dispatched; second due next week”. Before writing the target, mark what is expressed, what is withheld and what is available through the source setting. Acceptance criterion: Do not change scheduled shipment to confirmed dispatch. A good solution is neither maximally short nor maximally explicit; it is the least intervention that preserves the source commitments.

Perform a three-version test: first write a literal fragment, then a minimally expanded target, then a polished version suitable for the actual audience. Give each version an antecedent pointer and a risk note. Reject any version that changes the triggering condition, the participant, the event stage or the confidence level. If a version communicates the source accurately but sounds unnatural, improve word order or morphology before adding facts. Record what evidence would change the decision, so another reviewer can reproduce the reasoning.

Laboratory 28 — Archive transcription

Task. An author writes “If only…” and stops. Before writing the target, mark what is expressed, what is withheld and what is available through the source setting. Acceptance criterion: Do not complete the sentiment or motivation. A good solution is neither maximally short nor maximally explicit; it is the least intervention that preserves the source commitments.

Perform a three-version test: first write a literal fragment, then a minimally expanded target, then a polished version suitable for the actual audience. Give each version an antecedent pointer and a risk note. Reject any version that changes the triggering condition, the participant, the event stage or the confidence level. If a version communicates the source accurately but sounds unnatural, improve word order or morphology before adding facts. Record what evidence would change the decision, so another reviewer can reproduce the reasoning.

Laboratory 29 — Instructional diagram

Task. “Open; then close” beside two distinct levers. Before writing the target, mark what is expressed, what is withheld and what is available through the source setting. Acceptance criterion: Check arrows and labels before choosing whether each command shares the same object. A good solution is neither maximally short nor maximally explicit; it is the least intervention that preserves the source commitments.

Perform a three-version test: first write a literal fragment, then a minimally expanded target, then a polished version suitable for the actual audience. Give each version an antecedent pointer and a risk note. Reject any version that changes the triggering condition, the participant, the event stage or the confidence level. If a version communicates the source accurately but sounds unnatural, improve word order or morphology before adding facts. Record what evidence would change the decision, so another reviewer can reproduce the reasoning.

Laboratory 30 — Data dashboard

Task. “Costs down; complaints unchanged”. Before writing the target, mark what is expressed, what is withheld and what is available through the source setting. Acceptance criterion: Keep metric separation and avoid inventing a causal relationship. A good solution is neither maximally short nor maximally explicit; it is the least intervention that preserves the source commitments.

Perform a three-version test: first write a literal fragment, then a minimally expanded target, then a polished version suitable for the actual audience. Give each version an antecedent pointer and a risk note. Reject any version that changes the triggering condition, the participant, the event stage or the confidence level. If a version communicates the source accurately but sounds unnatural, improve word order or morphology before adding facts. Record what evidence would change the decision, so another reviewer can reproduce the reasoning.

A four-week practice and quality-control plan

Week 1 — Recognize

Mark ellipsis in dialogue, parallel clauses, lists and question answers. Produce minimally expanded source-language glosses. Compare with a colleague and catalogue disagreements.

Completion test: Success is correctly locating recoverable material, not making the output verbose. Keep a short log of source sentence, candidate antecedents, chosen strategy, target wording and reviewer decision. The same worksheet can be used during human revision or machine-translation post-editing, although service requirements and responsibilities must follow the applicable project specification.

Week 2 — Resolve

Add multi-turn dialogue, competing antecedents, negative fragments and modal auxiliaries. Require explicit notes whenever subject, time or scope depends on context.

Completion test: Success is the ability to explain exactly why one recovery is licensed and another is not. Keep a short log of source sentence, candidate antecedents, chosen strategy, target wording and reviewer decision. The same worksheet can be used during human revision or machine-translation post-editing, although service requirements and responsibilities must follow the applicable project specification.

Week 3 — Translate

Work between actual language pairs under an agreed brief. Compare a natural ellipsis, a minimal expansion and an alternative restructuring. Review register and target-language grammar with a qualified reader.

Completion test: Success is equal reader understanding with no new assertions. Keep a short log of source sentence, candidate antecedents, chosen strategy, target wording and reviewer decision. The same worksheet can be used during human revision or machine-translation post-editing, although service requirements and responsibilities must follow the applicable project specification.

Week 4 — Audit

Sample a complete document. Prioritize instructions, contracts, health information and anything containing numbers, permissions or conditions. Track severe errors separately from minor style issues.

Completion test: Success is a repeatable evidence trail and fewer unsupported completions at the release gate. Keep a short log of source sentence, candidate antecedents, chosen strategy, target wording and reviewer decision. The same worksheet can be used during human revision or machine-translation post-editing, although service requirements and responsibilities must follow the applicable project specification.

A practical scoring rubric

Score a candidate translation on five independent dimensions, each from zero to two: antecedent accuracy (is the right phrase recovered?), semantic boundary (are there no invented actors, causes or facts?), scope (are negation, modal, quantity and condition attached correctly?), target-language naturalness (does the sentence work for a real reader?), and uncertainty management (are unresolved alternatives handled transparently?). A score of ten is an excellent local check, but it is not a certification or substitute for expert review.

A zero on safety-critical scope, actor identity or invented instruction triggers revision regardless of aggregate score. Two translators can disagree about elegant surface form while agreeing on the underlying reconstruction. This separation of linguistic judgment from semantic obligation prevents a style debate from concealing a serious error. The scoring tool is an original pedagogical diagnostic, not a claimed ISO rating scale.

Frequently asked questions

FAQ 1. What is ellipsis in translation?

It is the transfer of a meaning in which recoverable linguistic material is left unspoken or unwritten in the source. The translator identifies what context licenses and chooses whether the target language should also omit it.

A useful check is to write the smallest possible source-language expansion, underline which words were actually present in the fragment, and mark which elements came from a preceding turn, grammatical agreement or visible document structure. Compare the final target against this evidence map. If the target commits the reader to an extra fact, revise. If it remains unclear whether a fact is recoverable, preserve the uncertainty or escalate the question rather than guessing.

FAQ 2. Do translators have to fill every missing word?

No. They may need a private reconstruction to verify meaning, but publishing that reconstruction can make a natural fragment unwieldy or introduce unsupported precision.

A useful check is to write the smallest possible source-language expansion, underline which words were actually present in the fragment, and mark which elements came from a preceding turn, grammatical agreement or visible document structure. Compare the final target against this evidence map. If the target commits the reader to an extra fact, revise. If it remains unclear whether a fact is recoverable, preserve the uncertainty or escalate the question rather than guessing.

FAQ 3. What is the difference between ellipsis and omission error?

Licensed ellipsis relies on shared information. An omission error loses information the source intended to communicate. The distinction depends on grammar, context and reader recoverability.

A useful check is to write the smallest possible source-language expansion, underline which words were actually present in the fragment, and mark which elements came from a preceding turn, grammatical agreement or visible document structure. Compare the final target against this evidence map. If the target commits the reader to an extra fact, revise. If it remains unclear whether a fact is recoverable, preserve the uncertainty or escalate the question rather than guessing.

FAQ 4. What is gapping?

Gapping occurs when a coordinated clause omits a verb shared with the preceding clause, as in “Lia bought apples; Hugo, pears.” The second action is still bought.

A useful check is to write the smallest possible source-language expansion, underline which words were actually present in the fragment, and mark which elements came from a preceding turn, grammatical agreement or visible document structure. Compare the final target against this evidence map. If the target commits the reader to an extra fact, revise. If it remains unclear whether a fact is recoverable, preserve the uncertainty or escalate the question rather than guessing.

FAQ 5. What is sluicing?

Sluicing is a question construction with a wh-remnant, such as “Someone called, but I don’t know who.” The unresolved identity is part of the meaning.

A useful check is to write the smallest possible source-language expansion, underline which words were actually present in the fragment, and mark which elements came from a preceding turn, grammatical agreement or visible document structure. Compare the final target against this evidence map. If the target commits the reader to an extra fact, revise. If it remains unclear whether a fact is recoverable, preserve the uncertainty or escalate the question rather than guessing.

FAQ 6. What is verb-phrase ellipsis?

A clause may retain an auxiliary while leaving its verbal complement unspoken: “She can swim, and he can too.” The auxiliary can carry important modal or tense information.

A useful check is to write the smallest possible source-language expansion, underline which words were actually present in the fragment, and mark which elements came from a preceding turn, grammatical agreement or visible document structure. Compare the final target against this evidence map. If the target commits the reader to an extra fact, revise. If it remains unclear whether a fact is recoverable, preserve the uncertainty or escalate the question rather than guessing.

FAQ 7. Is an ellipsis always marked by three dots?

No. Grammatical ellipsis can occur without punctuation. Three dots may signal editorial omission, hesitation or incomplete speech instead.

A useful check is to write the smallest possible source-language expansion, underline which words were actually present in the fragment, and mark which elements came from a preceding turn, grammatical agreement or visible document structure. Compare the final target against this evidence map. If the target commits the reader to an extra fact, revise. If it remains unclear whether a fact is recoverable, preserve the uncertainty or escalate the question rather than guessing.

FAQ 8. How do I translate a one-word answer?

Read the question and the active discourse topic, recover only the necessary proposition, and preserve the answer’s focus and degree of certainty.

A useful check is to write the smallest possible source-language expansion, underline which words were actually present in the fragment, and mark which elements came from a preceding turn, grammatical agreement or visible document structure. Compare the final target against this evidence map. If the target commits the reader to an extra fact, revise. If it remains unclear whether a fact is recoverable, preserve the uncertainty or escalate the question rather than guessing.

FAQ 9. How do I translate omitted subjects?

Use agreement and discourse tracking to identify the actor. If several candidates remain, do not silently assign one simply because the target language often requires pronouns.

A useful check is to write the smallest possible source-language expansion, underline which words were actually present in the fragment, and mark which elements came from a preceding turn, grammatical agreement or visible document structure. Compare the final target against this evidence map. If the target commits the reader to an extra fact, revise. If it remains unclear whether a fact is recoverable, preserve the uncertainty or escalate the question rather than guessing.

FAQ 10. Can machine translation resolve ellipsis?

Sometimes, but systems can add the wrong pronoun or predicate. Human or qualified domain review should verify antecedent selection and consequential semantic claims.

A useful check is to write the smallest possible source-language expansion, underline which words were actually present in the fragment, and mark which elements came from a preceding turn, grammatical agreement or visible document structure. Compare the final target against this evidence map. If the target commits the reader to an extra fact, revise. If it remains unclear whether a fact is recoverable, preserve the uncertainty or escalate the question rather than guessing.

FAQ 11. Does a fuller sentence mean a more accurate translation?

Not necessarily. Accuracy is measured by what the target reader understands, not by the number of words added.

A useful check is to write the smallest possible source-language expansion, underline which words were actually present in the fragment, and mark which elements came from a preceding turn, grammatical agreement or visible document structure. Compare the final target against this evidence map. If the target commits the reader to an extra fact, revise. If it remains unclear whether a fact is recoverable, preserve the uncertainty or escalate the question rather than guessing.

FAQ 12. How do I handle two possible antecedents?

Write both minimal interpretations, document which source evidence favors each, and request clarification or preserve ambiguity when neither can be justified.

A useful check is to write the smallest possible source-language expansion, underline which words were actually present in the fragment, and mark which elements came from a preceding turn, grammatical agreement or visible document structure. Compare the final target against this evidence map. If the target commits the reader to an extra fact, revise. If it remains unclear whether a fact is recoverable, preserve the uncertainty or escalate the question rather than guessing.

FAQ 13. Should I translate a deliberately unfinished sentence?

Yes, but preserve the incompleteness where it is meaningful. Filling a threatening or emotional pause can damage authorial intent.

A useful check is to write the smallest possible source-language expansion, underline which words were actually present in the fragment, and mark which elements came from a preceding turn, grammatical agreement or visible document structure. Compare the final target against this evidence map. If the target commits the reader to an extra fact, revise. If it remains unclear whether a fact is recoverable, preserve the uncertainty or escalate the question rather than guessing.

FAQ 14. What if target grammar cannot omit the subject?

Use the least committed grammatical expression the target allows, resolving the subject only to the degree supported by source morphology and discourse.

A useful check is to write the smallest possible source-language expansion, underline which words were actually present in the fragment, and mark which elements came from a preceding turn, grammatical agreement or visible document structure. Compare the final target against this evidence map. If the target commits the reader to an extra fact, revise. If it remains unclear whether a fact is recoverable, preserve the uncertainty or escalate the question rather than guessing.

FAQ 15. How should negation be handled?

Locate exactly which predicate or alternative is negated. “Not yet” is different from “never”; “not approved” is different from “rejected”.

A useful check is to write the smallest possible source-language expansion, underline which words were actually present in the fragment, and mark which elements came from a preceding turn, grammatical agreement or visible document structure. Compare the final target against this evidence map. If the target commits the reader to an extra fact, revise. If it remains unclear whether a fact is recoverable, preserve the uncertainty or escalate the question rather than guessing.

FAQ 16. How should conditional fragments be handled?

Identify the proposition the condition qualifies. “Only after review” must remain a dependency, not become a free-standing claim that review has finished.

A useful check is to write the smallest possible source-language expansion, underline which words were actually present in the fragment, and mark which elements came from a preceding turn, grammatical agreement or visible document structure. Compare the final target against this evidence map. If the target commits the reader to an extra fact, revise. If it remains unclear whether a fact is recoverable, preserve the uncertainty or escalate the question rather than guessing.

FAQ 17. How do list headings affect ellipsis?

A bullet may inherit ‘required documents’, ‘optional attachments’ or a common verb from a heading. The heading’s semantics control interpretation.

A useful check is to write the smallest possible source-language expansion, underline which words were actually present in the fragment, and mark which elements came from a preceding turn, grammatical agreement or visible document structure. Compare the final target against this evidence map. If the target commits the reader to an extra fact, revise. If it remains unclear whether a fact is recoverable, preserve the uncertainty or escalate the question rather than guessing.

FAQ 18. How do tables affect ellipsis?

Row and column headers provide context, but blanks and dashes may mean missing, not applicable or repeated. Check table conventions before repeating values.

A useful check is to write the smallest possible source-language expansion, underline which words were actually present in the fragment, and mark which elements came from a preceding turn, grammatical agreement or visible document structure. Compare the final target against this evidence map. If the target commits the reader to an extra fact, revise. If it remains unclear whether a fact is recoverable, preserve the uncertainty or escalate the question rather than guessing.

FAQ 19. Are subtitles different?

Subtitles impose time and space pressures, but compression still must preserve consequential events, conditions, speaker intent and referents.

A useful check is to write the smallest possible source-language expansion, underline which words were actually present in the fragment, and mark which elements came from a preceding turn, grammatical agreement or visible document structure. Compare the final target against this evidence map. If the target commits the reader to an extra fact, revise. If it remains unclear whether a fact is recoverable, preserve the uncertainty or escalate the question rather than guessing.

FAQ 20. Does context include a video frame?

Yes, where the translation brief authorizes audiovisual evidence. The frame can identify an object, but it should not be used to invent an unseen motive or history.

A useful check is to write the smallest possible source-language expansion, underline which words were actually present in the fragment, and mark which elements came from a preceding turn, grammatical agreement or visible document structure. Compare the final target against this evidence map. If the target commits the reader to an extra fact, revise. If it remains unclear whether a fact is recoverable, preserve the uncertainty or escalate the question rather than guessing.

FAQ 21. How does a translator handle a passive without an agent?

The translation should preserve the unstated agent unless the language forces a grammatical restructuring. Naming a specific actor requires evidence.

A useful check is to write the smallest possible source-language expansion, underline which words were actually present in the fragment, and mark which elements came from a preceding turn, grammatical agreement or visible document structure. Compare the final target against this evidence map. If the target commits the reader to an extra fact, revise. If it remains unclear whether a fact is recoverable, preserve the uncertainty or escalate the question rather than guessing.

FAQ 22. How does ellipsis affect comparative sentences?

The missing comparative material must match the relevant property or action. A translator should not invent units, groups or baselines.

A useful check is to write the smallest possible source-language expansion, underline which words were actually present in the fragment, and mark which elements came from a preceding turn, grammatical agreement or visible document structure. Compare the final target against this evidence map. If the target commits the reader to an extra fact, revise. If it remains unclear whether a fact is recoverable, preserve the uncertainty or escalate the question rather than guessing.

FAQ 23. How can I detect hallucinated recoveries?

Underline every added subject, predicate and explanation in the target. Ask where in the source or approved context each addition is licensed.

A useful check is to write the smallest possible source-language expansion, underline which words were actually present in the fragment, and mark which elements came from a preceding turn, grammatical agreement or visible document structure. Compare the final target against this evidence map. If the target commits the reader to an extra fact, revise. If it remains unclear whether a fact is recoverable, preserve the uncertainty or escalate the question rather than guessing.

FAQ 24. Can a glossary solve ellipsis?

A glossary helps with repeated terminology but cannot select context-dependent antecedents. Recovery needs passage-level reasoning.

A useful check is to write the smallest possible source-language expansion, underline which words were actually present in the fragment, and mark which elements came from a preceding turn, grammatical agreement or visible document structure. Compare the final target against this evidence map. If the target commits the reader to an extra fact, revise. If it remains unclear whether a fact is recoverable, preserve the uncertainty or escalate the question rather than guessing.

FAQ 25. What is the best translation of ‘Not yet’?

It depends on the question being answered and the target language. Preserve incompleteness at the reference time and openness about what may happen later.

A useful check is to write the smallest possible source-language expansion, underline which words were actually present in the fragment, and mark which elements came from a preceding turn, grammatical agreement or visible document structure. Compare the final target against this evidence map. If the target commits the reader to an extra fact, revise. If it remains unclear whether a fact is recoverable, preserve the uncertainty or escalate the question rather than guessing.

FAQ 26. Is ‘I don’t know who’ equivalent to a person’s name?

No. The source explicitly leaves the identity unknown. Supplying a name turns a question into a claim.

A useful check is to write the smallest possible source-language expansion, underline which words were actually present in the fragment, and mark which elements came from a preceding turn, grammatical agreement or visible document structure. Compare the final target against this evidence map. If the target commits the reader to an extra fact, revise. If it remains unclear whether a fact is recoverable, preserve the uncertainty or escalate the question rather than guessing.

FAQ 27. When should the translator ask the author?

When competing recoveries would cause materially different action or understanding and context does not settle them.

A useful check is to write the smallest possible source-language expansion, underline which words were actually present in the fragment, and mark which elements came from a preceding turn, grammatical agreement or visible document structure. Compare the final target against this evidence map. If the target commits the reader to an extra fact, revise. If it remains unclear whether a fact is recoverable, preserve the uncertainty or escalate the question rather than guessing.

FAQ 28. How can I train junior translators?

Start with controlled short pairs, require a written minimal reconstruction, then compare fluent target options against source-preserved commitments.

A useful check is to write the smallest possible source-language expansion, underline which words were actually present in the fragment, and mark which elements came from a preceding turn, grammatical agreement or visible document structure. Compare the final target against this evidence map. If the target commits the reader to an extra fact, revise. If it remains unclear whether a fact is recoverable, preserve the uncertainty or escalate the question rather than guessing.

FAQ 29. How do I measure improvement?

Record antecedent selection accuracy, unsupported additions, scope errors and unresolved queries. Review severity by likely consequence rather than counting only typos.

A useful check is to write the smallest possible source-language expansion, underline which words were actually present in the fragment, and mark which elements came from a preceding turn, grammatical agreement or visible document structure. Compare the final target against this evidence map. If the target commits the reader to an extra fact, revise. If it remains unclear whether a fact is recoverable, preserve the uncertainty or escalate the question rather than guessing.

FAQ 30. Does ISO prescribe this exact 24-pass method?

No. This 24-pass workflow is an original instructional method. ISO standards inform project specifications and service processes but do not endorse the numbered sequence here.

A useful check is to write the smallest possible source-language expansion, underline which words were actually present in the fragment, and mark which elements came from a preceding turn, grammatical agreement or visible document structure. Compare the final target against this evidence map. If the target commits the reader to an extra fact, revise. If it remains unclear whether a fact is recoverable, preserve the uncertainty or escalate the question rather than guessing.

FAQ 31. Where does this topic sit in the larger translation architecture?

It is a focused treatment of recoverable omissions. The translation root connects it to broad meaning transfer, while neighboring chapters handle discourse coherence, reference and grammatical features.

A useful check is to write the smallest possible source-language expansion, underline which words were actually present in the fragment, and mark which elements came from a preceding turn, grammatical agreement or visible document structure. Compare the final target against this evidence map. If the target commits the reader to an extra fact, revise. If it remains unclear whether a fact is recoverable, preserve the uncertainty or escalate the question rather than guessing.

Research foundation and references

This article uses established descriptive linguistic terminology but presents its own workflow, educational cases, rubric and audit steps. The sources below support the broader concepts and quality approach; they do not individually endorse every example or guarantee a particular translation.

• ISO 17100:2015 — Translation services: Requirements. The published 2015 edition describes core service processes and specifications. As checked in November 2026, a replacement project is under development; do not confuse a working draft with the published standard.
• ISO 11669:2024 — Translation projects: General guidance. Published March 2024; addresses project specifications, communication, risk and workflows rather than certifying the local teaching rubric used here.
• Rios Gonzales and Tuggener (2017), Co-reference Resolution of Elided Subjects and Possessive Pronouns in Spanish-English Statistical Machine Translation. Research on the importance of reference information for elided subjects.
• Frazier and Clifton (2005), The Syntax–Discourse Divide: Processing Ellipsis. Research on the syntactic and discourse processing of verb-phrase ellipsis and sluicing.
• Journal of Linguistics (2020), Nuclear Prominence in Ellipsis. A linguistic discussion of sluicing and the behavior of question remnants.
• Gerber-Morón and Szarkowska, Line Breaks in Subtitling: An Eye Tracking Study on Viewer Preferences. Research illustrating that segmentation and viewing conditions matter in audiovisual language presentation.
• Fischer and Volk (2025), Name Consistency in LLM-based Machine Translation of Historical Texts. A complementary study showing why coherent-looking machine translation still needs careful entity review.

Next reading routes: from ellipsis into the wider system

Return to Master Art of Translation | The Complete System for Moving Meaning Between Languages for the root framework, briefs, decisions and links to other transfer problems. The preceding reserved-tens chapter, The Implicit Coherence Transfer System, explains how readers recover relations such as causes and contrasts; the present chapter addresses missing linguistic constituents. These are different problems and require different checks.

For related language foundations without reproducing or replacing existing owners, visit the protected Vocabulary Learning Hub for word meanings and reference distinctions, and How English Works | Translation and Meaning Across Languages for grammatical patterns and English-language explanation. These links are outward routes, not invitations to modify those pages.

Reader’s takeaway: an omitted word is not a blank cheque for a translator. Recover what the source makes recoverable, leave what the source leaves open, and give the reader the same opportunity to understand the intended message.