Sluicing in translation is a form of wh-ellipsis in which most of an interrogative clause is unpronounced while a wh-expression remains visible. The classic example is Someone called, but I don’t know who, where the missing clause is recoverable as who called. Cambridge work describes sluicing as deletion of a clausal domain leaving a wh-remnant, while research on Farsi shows that languages with ordinary wh-in-situ can still have sluicing through a different derivational route.
People searching for sluicing, wh ellipsis, wh-remnant, elliptical questions, fragment questions, sluicing syntax, island repair are solving a reconstruction problem. The translator must recover the missing predicate, tense, aspect, voice, modality and grammatical role of the wh-remnant from the antecedent discourse, but should not invent material beyond what that discourse supports. The full clause is an analysis tool; the final target can remain elegantly elliptical.
This article owns the sluicing / wh-ellipsis precision layer for eduKateSG. It does not replace the broad Ellipsis owner, Wh-in-Situ, Preposition Stranding or Multiple Wh work. Its narrow task is to preserve the wh-remnant and its hidden clause, distinguish sluicing from other ellipsis types, manage P-stranding/swiping, and prevent AI from hallucinating antecedents or overexpanding concise questions.
Quick Read
Find ANTECEDENT PROPOSITION, CORRELATE if any, WH-REMNANT, WH ROLE and FULL RECONSTRUCTED QUESTION. Verify tense/aspect/voice/scope. Then decide whether the target should remain sluiced or pronounce more of the hidden clause.
One-Sentence Answer
Translate sluicing by reconstructing the missing interrogative clause completely in analysis while printing only as much of it as the target language naturally requires.
Why This Precision Layer Matters
A one-word remnant can carry an entire clause’s grammar. If the wrong antecedent or role is supplied, the translation asks a different question. If every sluice is expanded, natural discourse becomes repetitive. Precision comes from full hidden analysis plus economical final expression.
What Sluicing Is
The first task is to reconstruct the missing question. Sluicing is an elliptical question construction in which most of an interrogative clause is unpronounced while a wh-expression remains overt. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is translating the wh-remnant as a standalone lexical fragment without reconstructing the missing proposition. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to recover the full question that the remnant stands for. For example, ‘Someone called, but I don’t know who’ corresponds to ‘I don’t know who called’. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
The Wh-Remnant
This construction depends on discourse antecedence rather than visible syntax alone. The overt wh-word or wh-phrase is the visible remnant of a larger question. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is treating the remnant as an indefinite or discourse particle. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to identify its grammatical role in the reconstructed clause. For example, who can be subject, object, complement of preposition or another role. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
The Antecedent Clause
A precise translation assigns the wh-remnant a grammatical role before final wording. Sluicing normally depends on an antecedent clause or discourse that introduces the relevant event and often a correlate such as someone/something. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is reconstructing the missing clause from the wh-word alone. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to find the antecedent proposition first. For example, ‘Someone bought something, but I don’t know what’ reconstructs what someone bought. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Correlates
The target can stay elliptical once the hidden structure has been verified. An overt indefinite correlate in the antecedent—someone, something, somewhere, a person—often corresponds to the wh-remnant. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is creating two entities instead of linking correlate and remnant. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to coindex the correlate with the information gap. For example, someone…who asks for the identity of that someone. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Sprouting
Ellipsis licenses economy, not invention. Sluicing can occur even without an overt correlate, a pattern often called sprouting. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is assuming the antecedent must contain an explicit someone/something. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to infer the missing semantic argument licensed by the antecedent predicate. For example, ‘She was eating, but I don’t know with whom’ introduces an unspoken companion relation. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Merger-Type Sluicing
The first task is to reconstruct the missing question. When an overt correlate exists, some literature distinguishes merger-type sluicing from sprouting. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is treating the terminology as different translation meanings. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to use it to diagnose antecedent structure, not target vocabulary. For example, both patterns can translate naturally as wh-fragment questions. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Full-Clause Reconstruction
This construction depends on discourse antecedence rather than visible syntax alone. A practical translation method is to expand the sluice into its full interrogative clause before target compression. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is translating directly from the fragment and losing tense, voice or argument role. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to write the unreduced question. For example, ‘but I don’t know where’ → ‘but I don’t know where she went’. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Tense Recovery
A precise translation assigns the wh-remnant a grammatical role before final wording. The elided clause inherits tense from the antecedent/discourse unless the grammar indicates otherwise. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is using present tense simply because the remnant has no tense morphology. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to copy relevant temporal anchoring into the reconstruction. For example, ‘someone left, but I don’t know when’ asks when they left, not when they leave. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Aspect Recovery
The target can stay elliptical once the hidden structure has been verified. Progressive, perfect or resultative meaning can survive inside the unpronounced material. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is dropping aspect because only the wh-word is visible. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to reconstruct the full verbal complex. For example, ‘she has been working somewhere, but I don’t know where’ preserves perfect-progressive context. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Voice Recovery
Ellipsis licenses economy, not invention. The missing clause can contain active or passive voice inherited from the antecedent. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is reconstructing an active question when the antecedent is passive and changing roles. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to copy the event frame before ellipsis. For example, ‘someone was arrested, but I don’t know by whom’ preserves passive agency. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Negation
The first task is to reconstruct the missing question. Sluicing can occur under negative antecedents or within negative contexts, with scope affecting interpretation. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is adding or deleting negation during reconstruction. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to bracket negation before ellipsis expansion. For example, ‘Nobody called, but I don’t know why’ differs from ‘Someone didn’t call…’. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Modality
This construction depends on discourse antecedence rather than visible syntax alone. Modal meanings such as can, must, might and should can be part of the missing clause. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is flattening the fragment into a simple factual question. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to recover the modal from the antecedent. For example, ‘someone might help, but I don’t know who’ asks who might help. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Subject Sluices
A precise translation assigns the wh-remnant a grammatical role before final wording. The wh-remnant can correspond to the subject of the missing clause. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is inserting a generic subject because fragments seem incomplete. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to test the full clause with wh as subject. For example, ‘Someone arrived, but I don’t know who’ = who arrived. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Object Sluices
The target can stay elliptical once the hidden structure has been verified. The remnant can correspond to an object. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is treating who/what as subject by default. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to reconstruct verb valency. For example, ‘Lina bought something, but I don’t know what’ = what Lina bought. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Prepositional Sluices
Ellipsis licenses economy, not invention. Wh-remnants can be PPs or interact with preposition stranding depending on the language. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is dropping the preposition or choosing the wrong target placement. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to reconstruct governor + preposition + wh complement. For example, ‘she spoke with someone, but I don’t know with whom/who with’. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Sluicing and Preposition Stranding
The first task is to reconstruct the missing question. Languages that allow P-stranding can show wh-remnants without the preposition in suitable sluices; languages that forbid P-stranding behave differently. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is copying English-style stranded remnants into a no-stranding target. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to apply target preposition rules after reconstructing the full question. For example, who can stand for who she spoke with in English but not universally. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Swiping
This construction depends on discourse antecedence rather than visible syntax alone. Swiping is a special ellipsis pattern where a wh-word precedes a stranded/prepositional remnant in an order unavailable in the corresponding full clause. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is normalising the order before recognising that ellipsis licenses it. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to identify swiping as an ellipsis-specific construction. For example, who with-type remnant can be licensed only under ellipsis in some varieties. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Multiple Remnants
A precise translation assigns the wh-remnant a grammatical role before final wording. Some languages or constructions allow more than one remnant, though classic sluicing usually has a single wh-expression. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is assuming any multi-fragment sequence is ordinary sluicing. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to identify whether the source licenses multiple sluicing or another ellipsis type. For example, who where can require special analysis. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Sluicing vs Fragment Answers
The target can stay elliptical once the hidden structure has been verified. A fragment answer such as ‘John’ to ‘Who came?’ is not the same construction as a sluiced wh-question. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is treating every isolated phrase as sluicing. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to check whether the remnant is interrogative and depends on an antecedent gap. For example, ‘Who?’ after ‘Someone came’ differs from answer ‘John’. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Sluicing vs Gapping
Ellipsis licenses economy, not invention. Gapping removes a shared verbal core from coordinated clauses while leaving two or more remnants; sluicing leaves a wh-remnant in an interrogative context. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is mixing coordination ellipsis with question ellipsis. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to identify interrogative force and remnant type. For example, John ate rice and Mary beans is gapping, not sluicing. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Sluicing vs Stripping
The first task is to reconstruct the missing question. Stripping often leaves one non-wh remnant plus additive/contrastive material in coordination. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is calling any one-word ellipsis sluicing. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to test for wh-question force. For example, John ate rice, and Mary too differs from ‘someone ate, but who?’. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Sluicing vs VP Ellipsis
This construction depends on discourse antecedence rather than visible syntax alone. VP ellipsis deletes a verbal phrase but leaves auxiliaries or other clause material rather than a wh-question remnant. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is reconstructing the wrong deletion domain. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to identify what overt material remains. For example, she will leave, and he will too is VP ellipsis. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Wh-in-Situ Languages
A precise translation assigns the wh-remnant a grammatical role before final wording. Languages such as Farsi can show sluicing even though ordinary wh-questions keep wh-expressions in situ; analyses may involve focus movement or other mechanisms rather than English-style wh-movement. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is forcing the English movement analysis onto every language. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to separate translation from theoretical derivation. For example, the target can use ordinary wh-ellipsis even if source syntax derives it differently. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Island Repair
The target can stay elliptical once the hidden structure has been verified. A classic theoretical property is that sluicing can sometimes appear to rescue wh-dependencies that are unacceptable in fully pronounced clauses. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is treating repaired island structure as permission to create an ungrammatical full target question. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to translate the accepted elliptical meaning rather than insisting on overt reconstruction in target surface syntax. For example, ellipsis can license a remnant where the full source movement would violate constraints. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Syntactic Identity
Ellipsis licenses economy, not invention. Many analyses require a structural relation between antecedent and elided clause, but the degree of syntactic identity needed is theoretically debated. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is overly literal reconstruction that ignores legitimate voice/morphology mismatches. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to preserve semantic and grammatical recoverability rather than one analysis. For example, translation can remain faithful without solving every ellipsis identity debate. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Semantic Recoverability
The first task is to reconstruct the missing question. The elided material must be recoverable enough for interlocutors to interpret the question. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is inventing details beyond what discourse licenses. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to reconstruct only shared event structure. For example, ‘someone left, but I don’t know why’ leaves reason open but not the leaving event. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Ambiguous Antecedents
This construction depends on discourse antecedence rather than visible syntax alone. If several antecedent events are available, the sluice can be genuinely ambiguous. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is choosing one antecedent merely for fluent target prose. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to list possible antecedents before committing. For example, ‘Lina called Omar after Mei left, but I don’t know why’ can have multiple discourse attachments. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Parallelism
A precise translation assigns the wh-remnant a grammatical role before final wording. Sluicing often depends on semantic parallelism between antecedent and reconstructed question. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is rebuilding a syntactically possible but semantically unrelated clause. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to align event participants and predicate. For example, the sluiced question should ask about an open variable from the antecedent. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Register
The target can stay elliptical once the hidden structure has been verified. Sluicing is common in both speech and writing, often producing economical, natural discourse. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is expanding every sluice into a full clause and making dialogue repetitive. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to retain ellipsis where the target allows it. For example, ‘I don’t know why’ is often better than repeating the whole predicate. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Translation into English
Ellipsis licenses economy, not invention. English productively uses sluicing, so many source ellipses can remain elliptical after reconstruction confirms the meaning. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is over-expanding because the source grammar differs. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to reconstruct for analysis but compress for final English. For example, the hidden clause need not be printed if English readers recover it naturally. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Translation into Less-Elliptical Targets
The first task is to reconstruct the missing question. Some targets may require more overt clause material to avoid ambiguity or satisfy syntax. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is copying a bare wh-remnant into an ungrammatical target. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to reconstruct and pronounce the necessary material. For example, target can expand ‘who?’ into ‘who did it?’. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Legal and Interview Context
This construction depends on discourse antecedence rather than visible syntax alone. Question fragments can be dangerous in transcripts if the antecedent is unclear. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is guessing the elided proposition. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to anchor every sluice to the immediately relevant discourse proposition. For example, a legal transcript may require explicit expansion in explanatory translation, not silent reinterpretation. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Subtitles
A precise translation assigns the wh-remnant a grammatical role before final wording. Sluicing is useful for subtitle economy, but pronoun/antecedent clarity must remain strong. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is keeping an ultra-short fragment after target reordering makes antecedent unclear. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to expand slightly if target viewer cannot recover the gap. For example, economy is secondary to referential clarity. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
AI Antecedent Hallucination
The target can stay elliptical once the hidden structure has been verified. Models can invent a richer hidden clause than discourse supports. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is adding location, actor or reason not licensed by the antecedent. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to reconstruct minimally. For example, ellipsis resolution should recover shared structure, not create new facts. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
AI Wrong-Role Remnant
Ellipsis licenses economy, not invention. A model can assign who/what to the wrong grammatical role during expansion. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is producing ‘who bought it’ when source means ‘who she bought it from’. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to map valency/preposition before reconstruction. For example, the wh-remnant inherits its role from the missing clause. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
AI Overexpansion
The first task is to reconstruct the missing question. Models can replace every sluice with a full repeated sentence, losing natural economy. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is confusing analysis with final style. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to recompress after verification. For example, target can remain elliptical once roles and scope are safe. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Back-Translation Sluice Check
This construction depends on discourse antecedence rather than visible syntax alone. A final audit should recover antecedent, remnant, missing predicate and wh-role. Sluicing is cross-linguistically diverse, and its theoretical derivation can differ even when the discourse function looks similar.
The main translation risk is checking only that the wh-word survives. A fragment is easy to interpret by common sense, which tempts models to add plausible but unsupported participants or circumstances. Reconstruction should therefore be anchored tightly to the antecedent.
A practical diagnostic is to write ANTECEDENT + CORRELATE + WH ROLE + RECONSTRUCTED CLAUSE. For example, translation succeeds when a target reader can recover the same question and no extra facts. After the full question is secure, compress it to the shortest grammatical target form that preserves reference and scope.
Worked Examples
Worked Example 1: Subject Sluice
Source problem: ‘Someone arrived, but I don’t know who.’
Diagnosis: Who is subject of ARRIVE.
Target strategy: Keep English sluice or expand ‘who arrived’ if target requires.
Verification: Do not invent an object.
Worked Example 2: Object Sluice
Source problem: ‘Lina bought something, but I don’t know what.’
Diagnosis: What is object of BUY.
Target strategy: Target may remain ‘I don’t know what.’
Verification: Object role must survive reconstruction.
Worked Example 3: Sprouting
Source problem: ‘Lina was talking, but I don’t know to whom.’
Diagnosis: No overt correlate appears; TALK licenses an addressee/interlocutor relation.
Target strategy: Reconstruct the missing PP before target choice.
Verification: Do not assume a location or topic instead.
Worked Example 4: P-Stranding
Source problem: ‘She spoke to someone, but I don’t know who.’
Diagnosis: English strands to in the hidden clause ‘who she spoke to.’
Target strategy: Target may need overt adposition with the wh-remnant.
Verification: Do not lose the speech-to relation.
Practical Diagnostic Workflow
Create a SLUICE CARD: ANTECEDENT CLAUSE; CORRELATE; WH REMNANT; ROLE; GOVERNOR/PREPOSITION; TENSE; ASPECT; VOICE; MODALITY; NEGATION; POSSIBLE FULL QUESTION. Mark SPRouting vs CORRELATE-based sluicing when useful.
Run the full-question test. If the reconstructed full question changes a participant, tense, voice or preposition from the antecedent without evidence, the reconstruction is unsafe. Repair before target ellipsis.
Verification Method
Back-read the target sluice and expand it. A fluent speaker should reconstruct the same hidden question. Compare role, event, tense and antecedent. If multiple source antecedents remain genuinely possible, preserve ambiguity rather than choosing silently.
Recovery Method
If the target asks the wrong question, restore the antecedent event frame and assign the wh-role explicitly. If the target is overexpanded, compress after meaning is secure. If an ultra-short target becomes ambiguous, pronounce the minimum extra material needed.
Quality-Assurance Checklist
- Identify antecedent proposition.
- Identify correlate where present.
- Identify wh-remnant role.
- Reconstruct full question.
- Recover tense/aspect/voice/modality.
- Check negation scope.
- Handle sprouting separately.
- Check preposition stranding/pied-piping.
- Distinguish sluicing from fragments/gapping/stripping/VP ellipsis.
- Respect wh-in-situ languages.
- Do not overcommit to one theoretical derivation.
- Preserve genuine ambiguity.
- Do not hallucinate hidden content.
- Audit AI wrong-role remnants.
- Audit AI overexpansion.
Frequently Asked Questions
What is sluicing?
An elliptical wh-question in which most of the clause is unpronounced and a wh-expression remains overt.
What is the hidden clause in ‘Someone called, but I don’t know who’?
The recoverable question is ‘who called?’.
What is sprouting?
Sluicing without an overt correlate in the antecedent, where the discourse/predicate licenses an additional wh-relation.
Is sluicing the same as a fragment answer?
No. Sluicing is an interrogative ellipsis; a fragment answer supplies an answer to a question.
How is sluicing different from gapping?
Gapping typically occurs in coordination and deletes a shared verbal core while leaving multiple non-wh remnants.
Can wh-in-situ languages have sluicing?
Yes. Farsi is a well-known case where sluicing exists despite ordinary wh-in-situ syntax.
What is island repair?
Sluicing can sometimes make dependencies acceptable that would be ill-formed if the full clause were pronounced; this is a major theoretical property of ellipsis.
Why does AI mistranslate sluices?
It may guess the wrong antecedent, assign the wh-remnant the wrong role, or expand the hidden clause with unsupported details.
Where This Fits in the eduKate Translation Architecture
Route broad omitted-material recovery to the Ellipsis and Omitted Meaning owner; wh-placement to Wh-in-Situ and Wh-Fronting; P-stranding/swiping relations to Preposition Stranding and Pied-Piping; polarity-sensitive wh forms to Polarity Items and Licensing.
Final Principle
Sluicing hides a full question behind a small remnant. Reconstruct the full question in analysis, verify every role and operator, then keep the target as short as its grammar and discourse safely allow.