Copy raising in translation is the pattern illustrated by John seems like he is tired and They seem like they’ve missed the bus. The matrix subject is paired with a finite lower clause containing a pronominal copy of that same subject. Comparative Germanic research describes copy raising in English, German and Dutch and argues that, where licensed, it can carry an evidential function associated with direct evidence more strongly than ordinary raising.
People searching for copy raising, seem like syntax, John seems like he is tired, pronoun copy raising, appear as if grammar are solving a one-referent/two-position plus evidentiality problem. The translator must keep the matrix subject and lower pronoun coreferential, preserve the lower clause’s tense/aspect, and retain the weak appearance-based commitment expressed by seem/appear.
This article owns the copy-raising precision layer for eduKateSG. It does not replace ordinary Raising/Control, ECM, Evidentiality, Left Dislocation or Pronouns and Reference. Its narrow task is to preserve the copy relation and the construction’s appearance/evidence meaning without creating a second participant or flattening seems like into a factual assertion.
Quick Read
Write MATRIX SUBJECT, LOWER COPY, LOWER FINITE PROPOSITION, EVIDENCE SOURCE/TYPE and CERTAINTY. Coindex subject and copy. Then decide whether the target uses copy raising, standard raising, an evidential particle or another appearance construction.
One-Sentence Answer
Translate copy raising by keeping the matrix subject and lower pronoun as one referent while preserving the finite lower proposition and the speaker’s evidence-based uncertainty.
Why This Precision Layer Matters
The construction looks repetitive—John … he—but the repetition is grammatical, not participant duplication. It also contributes stance: John seems like he is tired presents evidence for tiredness rather than asserting tiredness outright. Both reference and evidential force need protection.
What Copy Raising Is
The first task is to coindex the matrix subject and lower copy. Copy raising is a construction in which a matrix subject appears with a verb such as seem or appear, while a finite complement introduced by like, as if or as though contains a pronoun or other expression referring back to that matrix subject. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is translating the matrix subject and lower pronoun as two independent participants. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to coindex the matrix subject and the lower pronominal copy. For example, ‘John seems like he is tired’ refers to one John. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
The Copy
This construction combines raising-like syntax with a finite pronominal complement. The lower pronoun is called a copy because it resumes or corresponds to the matrix subject inside the finite complement. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is treating the pronoun as ordinary discourse coreference without recognizing the constructional dependency. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to test whether the lower pronoun obligatorily matches the matrix subject in the intended reading. For example, John…he forms the characteristic copy relation. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Raising-Like Matrix Subject
A precise translation preserves both reference identity and evidential stance. The matrix subject behaves in important respects like the subject of a raising construction rather than an ordinary thematic subject selected by seem/appear. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is assigning an independent semantic role such as experiencer or agent to the matrix subject. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to identify where the descriptive content about the subject is expressed. For example, tired is predicated in the lower clause. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Finite Lower Clause
The target may replace the copy construction while retaining the same appearance judgment. Unlike standard subject-to-subject raising with an infinitive, copy raising uses a finite clause containing an overt pronominal subject. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is deleting the lower pronoun as if it were an infinitive gap. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to preserve or appropriately restructure the finite lower predication. For example, he is tired remains a complete finite proposition. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Seem Like
Lower-clause tense and aspect must remain independent enough to preserve chronology. English seem like is a central copy-raising frame. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is translating like as ordinary similarity between two nouns. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to recognize like as introducing a clausal predicative/evidential complement. For example, John seems like he is tired ≈ available evidence makes John appear tired. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Appear As If/Though
The first task is to coindex the matrix subject and lower copy. Appear as if/as though can create closely related constructions. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is translating as if only as hypothetical comparison and missing the evidential appearance meaning. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to identify whether the phrase evaluates evidence for a proposition. For example, John appears as though he has been crying. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Copy Raising vs Standard Raising
This construction combines raising-like syntax with a finite pronominal complement. Standard raising uses an infinitive: John seems to be tired. Copy raising uses a finite lower clause with a pronominal copy: John seems like he is tired. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is treating the two as morphologically identical. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to classify finite copy clause versus infinitival raising complement. For example, both can describe appearance but differ syntactically and sometimes semantically. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Direct Evidence
A precise translation preserves both reference identity and evidential stance. Comparative Germanic work has argued that copy raising often marks direct perceptual or evidential access more strongly than standard raising. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is flattening the construction to a neutral epistemic ‘probably’ statement. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to check whether the source implies observable evidence. For example, John looks/seems like he is sick can suggest visible or otherwise direct cues. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Evidential Nuance
The target may replace the copy construction while retaining the same appearance judgment. The construction can signal that the judgment is based on appearance, sound, behavior or another perceptible indication. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is turning evidential stance into certainty. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to preserve the weak evidence-based commitment. For example, seems like is less categorical than is. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Epistemic Weakness
Lower-clause tense and aspect must remain independent enough to preserve chronology. Seem/appear constructions generally hedge commitment to the embedded proposition. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is translating them as factual assertions. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to keep the evidential/modal distance. For example, John seems like he is tired does not assert tiredness with full certainty. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
One Referent, Two Positions
The first task is to coindex the matrix subject and lower copy. The matrix subject and lower copy normally denote the same referent. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is creating two Johns/they entities in target text. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to coindex before lexical generation. For example, John…he remains one discourse participant. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Pronoun Features
This construction combines raising-like syntax with a finite pronominal complement. The lower copy carries person, number and gender features matching the matrix subject. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is ignoring mismatches that may reveal another analysis. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to compare features explicitly. For example, the girls seem like they are tired requires plural matching. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
First Person
A precise translation preserves both reference identity and evidential stance. Copy raising can occur with first-person subjects: I seem like I’m overreacting. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is treating lower I as a quoted speaker or separate deictic center. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to keep first-person coreference stable. For example, both I forms denote the same speaker. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Second Person
The target may replace the copy construction while retaining the same appearance judgment. You seem like you’re upset uses a second-person copy. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is translating the lower you as generic you. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to preserve direct-addressee reference. For example, one addressee fills both positions. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Plural Subjects
Lower-clause tense and aspect must remain independent enough to preserve chronology. Plural subjects take plural copies. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is normalizing the lower pronoun to singular because seem is singular/plural-neutral in some languages. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to carry number through the dependency. For example, they seem like they are…. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Copy Raising with Expletive It
The first task is to coindex the matrix subject and lower copy. It seems like it is raining contains expletive subjects rather than an ordinary referential copy. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is treating the two it forms as concrete entities. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to classify expletives independently. For example, weather it has no referent. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
There-Expletives
This construction combines raising-like syntax with a finite pronominal complement. There seems like there is a problem is accepted by many speakers and creates a special expletive-copy pattern. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is translating there as a location twice. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to recognize existential expletives. For example, both there forms support existential structure, not spatial deixis. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Expletive Restrictions
A precise translation preserves both reference identity and evidential stance. Not all expletive combinations are equally acceptable, which is important for theoretical analysis but also cautions against mechanical copying. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is generating copy raising with any expletive pair by analogy. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to use attested target constructions. For example, there seems like it’s raining is markedly worse for many speakers. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Lower-Clause Predicate
The target may replace the copy construction while retaining the same appearance judgment. The substantive descriptive predicate appears inside the lower finite clause. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is assigning it to the matrix verb only. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to map the lower proposition separately. For example, he is tired / she has left / there is trouble. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Tense in the Lower Clause
Lower-clause tense and aspect must remain independent enough to preserve chronology. The lower finite clause carries its own tense/aspect. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is collapsing it into a tenseless adjective or infinitive and losing temporal relation. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to preserve lower tense/aspect. For example, John seems like he was tired yesterday differs from he is tired now. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Perfect Aspect
The first task is to coindex the matrix subject and lower copy. The lower copy can occur with perfect forms. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is dropping completion/anteriority. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to keep the lower aspect. For example, She seems like she has finished. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Progressive Aspect
This construction combines raising-like syntax with a finite pronominal complement. Progressive lower clauses describe ongoing evidence-related states/events. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is converting them to simple present. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to preserve ongoingness. For example, He seems like he is working too hard. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Negation
A precise translation preserves both reference identity and evidential stance. Negation can occur in the matrix or lower clause with different meanings. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is moving not across the boundary. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to bracket matrix vs lower negation. For example, John doesn’t seem like he is tired differs from John seems like he isn’t tired. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Questions
The target may replace the copy construction while retaining the same appearance judgment. Copy-raising clauses can be questioned through the matrix: Does John seem like he is tired? Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is treating the lower copy as the interrogative focus automatically. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to identify the questioned proposition and evidence relation. For example, the question asks about John’s apparent state. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Conditionals
Lower-clause tense and aspect must remain independent enough to preserve chronology. Copy raising can occur under conditional or modal contexts. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is stripping away outer modality because seem already looks modal. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to preserve each operator. For example, If John seems like he is tired, let him rest. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Source of Evidence
The first task is to coindex the matrix subject and lower copy. The experiencer/evaluator is often contextually supplied rather than overtly encoded as a separate argument. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is adding ‘to me’ or ‘to observers’ without evidence. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to only make the judge explicit if the source does. For example, seem like can leave perspective implicit. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Copy Raising vs Perception Verb
This construction combines raising-like syntax with a finite pronominal complement. Look/sound/feel constructions overlap semantically with copy raising but differ lexically and syntactically. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is replacing seem like with look like in all contexts. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to match evidence source and matrix predicate. For example, sounds like may specifically indicate auditory evidence. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Copy Raising vs Control
A precise translation preserves both reference identity and evidential stance. Control verbs select a semantic subject/object who participates in the matrix relation; copy raising does not work like persuade/try. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is assigning John an independent role under seem. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to test whether seem semantically acts on John. For example, John is the apparent subject, not an object of a matrix action. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Copy Raising vs ECM
The target may replace the copy construction while retaining the same appearance judgment. ECM uses an object-like embedded subject in a nonfinite clause; copy raising uses a matrix subject and finite lower copy. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is translating copy raising as believe him to…. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to keep the distinct clause architecture. For example, John seems like he is tired is not I believe John to be tired. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Copy Raising vs Left Dislocation
Lower-clause tense and aspect must remain independent enough to preserve chronology. A matrix NP plus lower pronoun can superficially resemble a topic-resumptive structure, but copy raising is selected by seem/appear-like predicates. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is classifying the matrix subject as merely a discourse topic. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to test matrix predicate and evidential semantics. For example, John seems like he… is not just ‘John, he…’. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Translation into Standard English
The first task is to coindex the matrix subject and lower copy. English itself supports copy raising productively in informal and neutral styles. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is unnecessarily normalizing every instance to standard raising and losing evidential nuance. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to retain seem like/as if where natural. For example, John seems like he is tired can be the best target. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Translation into More Formal English
This construction combines raising-like syntax with a finite pronominal complement. Formal prose may prefer seems to be or appears to be, but this can reduce direct-evidence nuance. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is standardizing register without considering semantics. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to decide whether evidential flavor matters. For example, appears to be may be more formal but less conversational. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Translation into Languages Without Copy Raising
A precise translation preserves both reference identity and evidential stance. A target may use a perception verb, evidential particle, standard raising, finite complement or adverb. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is copying the pronoun pattern into an ungrammatical target. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to translate evidential meaning and coreference rather than surface architecture. For example, target can say ‘apparently John is tired’. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Dialogue and Character Voice
The target may replace the copy construction while retaining the same appearance judgment. Copy raising is common in speech and conversational writing. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is replacing it with stiff formal raising and flattening voice. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to preserve conversational register where appropriate. For example, you seem like you’re upset sounds natural and interpersonal. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Journalistic/Analytical Writing
Lower-clause tense and aspect must remain independent enough to preserve chronology. Seems like can sound less formal or more subjective than appears to/appears that. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is using it in a register where the source is impersonal and formal. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to calibrate evidential phrasing. For example, formal analysis may prefer appears that or evidence suggests. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
AI Pronoun Duplication
The first task is to coindex the matrix subject and lower copy. Models can treat the lower copy as a second participant or resolve it to the wrong antecedent. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is creating participant drift. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to coindex matrix subject and copy before generation. For example, John/he remains one chain. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
AI Evidential Flattening
This construction combines raising-like syntax with a finite pronominal complement. Models can paraphrase seem like as simple be or think, losing direct-evidence/hedging nuance. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is turning an appearance judgment into fact or pure belief. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to preserve evidential distance. For example, the target should remain less certain than a bare assertion. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
AI Over-Normalisation
A precise translation preserves both reference identity and evidential stance. Models often replace copy raising with seems to automatically because it is stylistically cleaner. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is erasing colloquial/evidential distinctions. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to standardize only when the task calls for it. For example, surface variation can carry voice. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Back-Translation Copy-Raising Check
The target may replace the copy construction while retaining the same appearance judgment. A final audit should recover matrix subject, lower copy, lower proposition and evidential stance. Copy raising is not universal and differs across English, German and Dutch, so target grammar should be generated independently rather than copied mechanically.
The main translation risk is checking only that seem/appear survives. A fluent paraphrase can still be wrong if it creates two referents or strengthens a tentative appearance into fact. The dependency and stance should therefore be annotated before stylistic editing.
A practical diagnostic is to write MATRIX SUBJECT = LOWER COPY + LOWER PREDICATE + EVIDENCE STANCE. For example, translation succeeds when one referent remains one and the appearance judgment survives. Once one-reference identity and lower proposition are secure, choose the target’s most natural evidential construction.
Worked Examples
Worked Example 1: Canonical Copy Raising
Source problem: ‘John seems like he is tired.’
Diagnosis: John and he are one referent; tired is predicated in the lower clause.
Target strategy: Keep the construction or use ‘John seems to be tired’ if register allows.
Verification: Do not create two male participants.
Worked Example 2: Plural Copy
Source problem: ‘They seem like they’ve missed the bus.’
Diagnosis: Matrix they and lower they match; perfect aspect indicates prior missing.
Target strategy: Preserve one group plus anterior event.
Verification: Do not reduce to ‘they seem late’ unless source says that.
Worked Example 3: Expletive It
Source problem: ‘It seems like it is raining.’
Diagnosis: Both it forms are weather expletives, not entities.
Target strategy: Target may use one impersonal weather/evidential construction.
Verification: Do not translate it referentially.
Worked Example 4: Lower Negation
Source problem: ‘John seems like he isn’t ready.’
Diagnosis: Seem remains positive; NOT belongs to the lower readiness proposition.
Target strategy: Preserve not-ready under appearance judgment.
Verification: Do not change to ‘John doesn’t seem…’ unless equivalent in context.
Practical Diagnostic Workflow
Create a COPY-RAISING CARD: MATRIX SUBJECT; MATRIX VERB; LIKE/AS-IF COMPLEMENT; LOWER SUBJECT/COPY; PERSON/NUMBER/GENDER; LOWER TENSE/ASPECT; NEGATION; EVIDENCE TYPE; REGISTER. Mark expletives explicitly as nonreferential.
Run the standard-raising paraphrase test: compare ‘John seems like he is tired’ with ‘John seems to be tired.’ If both share the core proposition but the source has stronger direct-evidence/conversational flavor, preserve that difference where relevant.
Verification Method
Back-read the target and count participants. Matrix subject and lower copy should normally count as one. Then identify whether the target presents the lower proposition as observed/appearing rather than certain fact. Finally check lower tense/aspect.
Recovery Method
If two referents appear, coindex and remove the duplication. If evidential hedging disappeared, restore seem/appear/apparently/evidence-based wording. If register was overformalized, return to a natural copy-raising or conversational equivalent when the source voice requires it.
Quality-Assurance Checklist
- Coindex matrix subject and lower copy.
- Check person/number/gender match.
- Identify expletives separately.
- Preserve lower tense/aspect.
- Preserve matrix vs lower negation.
- Preserve evidence-based uncertainty.
- Distinguish copy raising from standard raising.
- Distinguish copy raising from control.
- Distinguish copy raising from ECM.
- Distinguish copy raising from left dislocation.
- Check direct-evidence nuance.
- Match target register.
- Do not create extra participants.
- Audit AI evidential flattening.
- Audit AI over-normalisation.
Frequently Asked Questions
What is copy raising?
A construction where a matrix subject with seem/appear-like predicate corresponds to a pronoun inside a finite lower clause, as in ‘John seems like he is tired.’
Why is it called copy raising?
Because the lower clause contains a pronominal copy corresponding to the matrix subject.
How is it different from ‘John seems to be tired’?
The latter is standard raising with an infinitive; copy raising uses a finite lower clause and often carries a more direct-evidence/conversational flavor.
Does the pronoun introduce a second person?
No. In the intended copy-raising reading, the matrix subject and lower pronoun are coreferential.
Can expletives occur in copy raising?
Yes. Examples such as ‘It seems like it is raining’ and, for many speakers, ‘There seems like there is a problem’ are discussed in the literature.
Is copy raising grammatical in all Germanic languages?
No. Comparative work finds important differences among English, German and Dutch.
Should translation always preserve ‘seems like’?
No. Standard raising, evidential adverbs or other appearance constructions can be better targets depending language and register.
Why does AI mistranslate copy raising?
It can resolve the lower pronoun incorrectly, treat the subject and copy as separate participants or flatten the evidential hedge into a factual statement.
Where This Fits in the eduKate Translation Architecture
Route object-like nonfinite raising to Exceptional Case Marking; information-source meaning to Evidentiality and Information Source; ordinary pronoun coindexing to Pronouns and Reference; topic+pronoun constructions to Left Dislocation and Hanging Topics.
Final Principle
Copy raising repeats a referent syntactically but not semantically. Keep one participant, keep the finite lower proposition and keep the appearance-based uncertainty. Surface cleanup should never erase those three invariants.