Internally headed relative clauses in translation keep the lexical head noun inside the relative clause instead of placing it outside the clause as English normally does. Recent 2026 research in Language revisits how internal heads are positioned and interpreted, arguing that some internally headed relatives do not require the head to move all the way to an external position. Japanese and Korean are major languages in the literature, and Korean is described as having both head-external and head-internal relative constructions.
People searching for internally headed relative clauses, IHRC, head-internal relative clause, Japanese internally headed relative, Korean head internal relative, Navajo relative clauses are solving a reference-and-restructuring problem. The noun that becomes the English head may be an object, subject, possessor or oblique inside the source relative clause, while the entire nominalised IHRC then fills a separate role in the matrix clause. Translation precision therefore requires two role maps for one referent.
This article owns the internally-headed-relative precision layer for eduKateSG. It does not replace the existing Relative Clauses owner, Correlative Constructions, Nominalisation or Pronouns and Reference. Its narrow task is to identify the internal lexical head, preserve its local grammatical role, recover definiteness/maximality where relevant, and externalise the head naturally in a target language that prefers external-headed relatives.
Quick Read
Write HEAD NOUN, INTERNAL ROLE, MATRIX ROLE, NOMINALISER, CASE, DEFINITENESS and QUANTIFICATION. Then build the English external head and reconstruct the relative-clause gap or prepositional relation. Translate the head once.
One-Sentence Answer
Translate internally headed relatives by preserving one noun’s role inside the relative clause while allowing that noun to become the external head required by the target language.
Why This Precision Layer Matters
IHRCs tempt translators to count nouns rather than dependencies. The lexical noun is inside one clause but semantically supplies the referent used by another. If it is translated twice, duplicated reference appears; if the wrong noun is promoted, the matrix clause becomes about the wrong entity. Structural diagnosis must precede reordering.
What an Internally Headed Relative Clause Is
The first task is to identify the lexical internal head. An internally headed relative clause places the lexical head noun inside the relative clause rather than outside it as in the most familiar English pattern. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is pulling the internal noun out as though it were already an external English head and then also translating its clause-internal role. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to identify the noun’s local role inside the relative clause before building the target head. For example, a source equivalent to ‘[Lina bought a book] that one was expensive’ can mean ‘the book that Lina bought was expensive’. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
External vs Internal Heads
This construction requires two role maps for one referent. In an externally headed relative, the noun modified by the relative clause appears outside the clause; in an internally headed relative, the lexical noun appears within the relative clause itself. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is assuming both structures can be translated word-for-word with the same noun placement. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to draw the source bracket structure before reordering. For example, English ‘the book [that Lina bought]’ externalises book, whereas an IHRC keeps book inside the bracketed clause. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
The Internal Head’s Local Role
A precise translation separates the head’s local role from the matrix role of the whole nominal. The internal head still has an ordinary grammatical role inside the relative clause: subject, object, oblique, possessor or another role. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is losing that role when the noun is promoted to target head position. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to label the head’s internal case/role before externalising it. For example, a noun that is object of buy must remain semantically the bought item after English restructuring. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
The Matrix Role
The target may externalise a source-internal head without changing meaning. The entire IHRC construction then fills a role in the matrix clause, often through nominalisation or another clause-level mechanism. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is confusing the internal head’s relative-clause role with the matrix role of the whole construction. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to write two roles: INTERNAL ROLE and MATRIX ROLE. For example, the book can be object inside the relative clause but the whole IHRC can be subject of ‘was expensive’. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
Head Identification
Nominalisation and definiteness should be resolved before English relative syntax is generated. Translation begins by identifying which noun inside the relative clause supplies the referent for the larger construction. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is choosing the most prominent or nearest noun rather than the grammar’s actual head. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to use case, semantic fit, nominalisation and discourse cues. For example, a clause can contain several nouns but only one functions as the internal head. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
Nominalisation
The first task is to identify the lexical internal head. Many IHRC constructions are closely linked to nominalisation, with the relative clause packaged as a nominal expression. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is translating the nominaliser as an extra noun such as thing even when it is grammatical scaffolding. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to separate nominalising morphology from the lexical head noun. For example, Japanese no/koto/kes-type elements may require different English strategies than lexical nouns. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
Japanese IHRCs
This construction requires two role maps for one referent. Japanese is a major language in the literature on internally headed relativisation, and current analyses focus on how the internal head is interpreted within a nominalised clause. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is treating Japanese IHRCs as ordinary external noun modification with unusual word order. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to identify the internal noun and the nominaliser before generating English. For example, a noun inside the clause can correspond to an English head noun outside it. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
Korean IHRCs
A precise translation separates the head’s local role from the matrix role of the whole nominal. Korean has both head-external and head-internal relative constructions; head-internal relatives keep the lexical head inside the clause and use nominalising/complementiser material such as kes in well-studied patterns. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is translating the clause-internal noun and kes as two separate referents. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to coindex the lexical noun and the nominalised construction. For example, the English target often needs one noun referent, not noun + thing. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
Navajo and Other Head-Internal Languages
The target may externalise a source-internal head without changing meaning. Head-internal relativisation has also been documented in Navajo and numerous other languages, though structural details vary substantially. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is copying Japanese/Korean diagnostics into every head-internal language. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to analyse each language’s relative-clause morphology and word order independently. For example, head-internal status is a typological property, not one universal construction. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
Quantificational Readings
Nominalisation and definiteness should be resolved before English relative syntax is generated. IHRCs often interact with quantification, definiteness or maximality because the clause introduces a set from which the head referent is interpreted. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is translating the internal noun as a simple indefinite and losing an all/the relevant one(s) reading. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to check whether the matrix construction selects a maximal/specific set. For example, ‘the students who came’ can differ from ‘some students came and…’. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
Definiteness
The first task is to identify the lexical internal head. Some IHRCs receive definite-like interpretations despite an internal noun that lacks an overt definite article. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is adding an indefinite article automatically because the noun appears bare. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to derive target definiteness from the whole construction and discourse. For example, a bare internal noun can map to English the N who…. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
Specificity
This construction requires two role maps for one referent. An IHRC can refer to a specific discourse entity even when the internal head is morphologically indefinite. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is treating internal indefiniteness as target nonspecificity. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to track discourse identity and matrix reference. For example, one particular book can be internally introduced without a definite determiner. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
Maximality
A precise translation separates the head’s local role from the matrix role of the whole nominal. Some analyses derive the IHRC referent as the maximal individual or set satisfying the internal clause. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is weakening a source all/the relevant set interpretation to some. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to test whether the construction includes all contextually qualifying members. For example, a plural IHRC may denote the entire relevant set. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
Case on the Internal Head
The target may externalise a source-internal head without changing meaning. The internal head receives case according to its role inside the relative clause. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is assigning target case/article from the matrix role only. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to decode local case first. For example, an accusative-marked internal noun can become the nominative English head after externalisation while remaining semantic object of the embedded verb. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
Case on the IHRC as a Whole
Nominalisation and definiteness should be resolved before English relative syntax is generated. The nominalised relative construction can receive matrix case independently of the head’s internal case. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is treating two case markers as contradictory or as two noun phrases. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to separate inner and outer case domains. For example, internal object case plus outer subject/topic case can coexist structurally. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
Internal Subject Heads
The first task is to identify the lexical internal head. The internal head can be subject of the relative-clause predicate. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is assuming subject heads are easier and skipping matrix-role analysis. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to still distinguish local subject from matrix function. For example, ‘[child arrived] NOM…’ can correspond to ‘the child who arrived…’. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
Internal Object Heads
This construction requires two role maps for one referent. Object-headed IHRCs are especially easy to mistranslate because the noun appears deep inside a transitive clause. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is making the external English head the actor because it moves to noun-head position. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to preserve object role through the restructuring. For example, ‘[Lina bought book]’ → ‘the book Lina bought,’ not ‘the book that bought Lina’. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
Oblique Heads
A precise translation separates the head’s local role from the matrix role of the whole nominal. Locations, instruments, recipients and other oblique nominals can function as internal heads in some languages/constructions. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is dropping the source adposition/case relation when externalising the noun. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to reconstruct the gap/preposition relation in the target relative. For example, internal location can become ‘the place where…’ or ‘the place in which…’. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
Possessor Heads
The target may externalise a source-internal head without changing meaning. Possessors inside noun phrases can sometimes participate in head-internal relative interpretations depending on language constraints. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is promoting the possessed noun rather than the possessor or vice versa. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to map the exact dependency. For example, a possessor head may require ‘the person whose…’ in English. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
Agreement and Reference
Nominalisation and definiteness should be resolved before English relative syntax is generated. Agreement within the relative clause can help identify the internal head’s grammatical role and features. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is ignoring agreement because English relative heads do not show the same morphology. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to use agreement as evidence before target reordering. For example, number/class features can distinguish competing noun candidates. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
Anaphoric Continuity
The first task is to identify the lexical internal head. The matrix clause can contain pronouns or agreement that continue the IHRC referent. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is resolving them to another noun inside the clause. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to coindex the IHRC head with later reference. For example, a later plural pronoun can confirm that the IHRC denotes a plural head set. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
Head Movement Analyses
This construction requires two role maps for one referent. Current theory debates whether the internal head moves covertly, moves to a sentence-internal position, or is interpreted through other mechanisms; recent 2026 work argues some IHRCs need not move the head to an external position. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is turning a theoretical movement analysis into a translation requirement. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to translate the interpreted relation rather than choosing sides in the syntactic debate. For example, English can externalise the head even if the source analysis keeps it clause-internal. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
No Need to Reproduce Source Head Position
A precise translation separates the head’s local role from the matrix role of the whole nominal. A target language with external-headed relatives does not need to imitate source head-internal order. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is preserving source order and producing a confusing English nominalised clause. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to generate the target’s normal relative construction after role mapping. For example, IHRC source can become ordinary ‘the N who/that…’. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
IHRC vs Free Relative
The target may externalise a source-internal head without changing meaning. A free relative such as ‘what Lina bought’ has no overt lexical head noun, while an IHRC contains a lexical head inside the source relative clause. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is translating the lexical internal head away and accidentally creating an underspecified free relative. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to check whether the noun contributes lexical classification. For example, book-head IHRC should usually retain book rather than generic what. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
IHRC vs Correlative
Nominalisation and definiteness should be resolved before English relative syntax is generated. Correlative constructions link a relative clause and matrix demonstrative across clauses; IHRCs package the head inside one relative/nominalised structure. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is merging the two categories because both can lack an external head. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to look for a separate matrix correlate. For example, IHRC need not contain a that-one demonstrative pair. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
IHRC vs Headless Relative
The first task is to identify the lexical internal head. Headless/free relatives lack a lexical noun head; IHRCs do not. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is assuming any relative without an external head is headless. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to inspect clause-internal nouns and nominalisation. For example, the head can be present but internal. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
IHRC vs Nominal Complement
This construction requires two role maps for one referent. A nominalised clause describing an event is not automatically an internally headed relative. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is selecting one internal noun as head when the whole clause denotes an event/fact. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to test what the matrix clause refers to. For example, ‘the fact that Lina bought a book’ denotes a fact, not the book. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
Translation into English
A precise translation separates the head’s local role from the matrix role of the whole nominal. English usually prefers external-headed relative clauses or free relatives, so IHRC translation often requires structural reorganisation. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is copying source internal noun placement and nominaliser. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to externalise the lexical head and recreate its internal gap/role. For example, ‘[Lina bought book]-NMLZ was expensive’ → ‘The book Lina bought was expensive’. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
Translation into Another IHRC Language
The target may externalise a source-internal head without changing meaning. A target with IHRCs may still differ in nominalisation, definiteness, head restrictions or case patterns. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is mapping source IHRC morphology directly to target morphology. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to transfer referent structure before construction choice. For example, two head-internal languages need not license the same head roles. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
Literary and Discourse Effects
Nominalisation and definiteness should be resolved before English relative syntax is generated. Head-internal structures can package information in a way that delays identification of the matrix referent until the clause has unfolded. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is always converting to a short English prenominal/head-first structure and losing pacing. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to consider whether a cleft, free relative or delayed-head paraphrase better preserves discourse flow. For example, translation can preserve informational suspense without copying grammar. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
Technical/Legal Reference
The first task is to identify the lexical internal head. Dense IHRCs can hide which entity the matrix clause legally or technically refers to. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is choosing the wrong internal noun as head. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to annotate every internal NP and matrix predicate before drafting. For example, the obligation may apply to the document signed, not the signer. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
AI External-Head Hallucination
This construction requires two role maps for one referent. Models can invent an external noun and retain the internal noun, creating duplicated referents. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is translating one source head twice. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to coindex and count target referents. For example, one internal source head should normally correspond to one target head. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
AI Wrong-Head Selection
A precise translation separates the head’s local role from the matrix role of the whole nominal. When several nouns occur inside the relative clause, AI can promote the semantically salient but grammatically wrong one. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is producing a fluent relative clause about the wrong entity. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to use case/nominaliser/discourse evidence to choose head. For example, matrix predicate selection can help identify the intended referent. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
AI Nominaliser Literalism
The target may externalise a source-internal head without changing meaning. Models can translate a nominaliser as thing/one and keep the lexical internal head, creating awkward duplication. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is turning grammatical packaging into an extra noun. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to decide whether the nominaliser has independent lexical reference. For example, many nominalisers disappear in English relative translation. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
Back-Translation IHRC Check
Nominalisation and definiteness should be resolved before English relative syntax is generated. A final audit should recover the lexical head, its role inside the relative clause and the matrix role of the entire construction. IHRC systems differ cross-linguistically, so Japanese, Korean and Navajo should not be forced into one structural template.
The main translation risk is checking only that the noun appears somewhere. Because several nouns may occur inside the relative clause, a fluent target can choose the wrong head or duplicate the correct one. Case, agreement, nominalisation and matrix selection provide the decisive evidence.
A practical diagnostic is to write HEAD + INTERNAL ROLE + MATRIX ROLE + DEFINITENESS/QUANTIFICATION. For example, translation succeeds when one referent occupies the correct two structural domains. Once the source referent is secure, generate the target’s ordinary headed or free-relative construction.
Worked Examples
Worked Example 1: Internal Object Head
Source problem: Source packages ‘Lina bought a book’ as an IHRC and the matrix says it was expensive.
Diagnosis: Book is object inside the relative clause and matrix subject referent.
Target strategy: English: ‘The book Lina bought was expensive.’
Verification: Book remains bought item, not buyer.
Worked Example 2: Internal Subject Head
Source problem: Source clause contains a child as subject of arrive and the whole IHRC is matrix object.
Diagnosis: Child is local subject but matrix object referent.
Target strategy: English can use ‘I met the child who arrived.’
Verification: Local and matrix roles must remain separate.
Worked Example 3: Nominaliser
Source problem: Source has internal noun plus a nominaliser often glossed one/thing.
Diagnosis: Nominaliser packages the clause and may not denote a second object.
Target strategy: English usually translates one lexical head only.
Verification: Do not output ‘the book thing that…’.
Worked Example 4: Case Mismatch
Source problem: Internal head carries object case while the whole IHRC bears matrix topic/subject case.
Diagnosis: Two cases belong to two structural layers.
Target strategy: English externalises the noun and uses ordinary target syntax.
Verification: Do not treat case mismatch as two nouns.
Practical Diagnostic Workflow
Create an IHRC CARD: ALL INTERNAL NPs; CANDIDATE HEAD; INTERNAL CASE/ROLE; RELATIVE PREDICATE; NOMINALISER; WHOLE-IHRC CASE; MATRIX PREDICATE; DEFINITENESS; QUANTIFICATION; DISCOURSE ANTECEDENTS. Select the head only after the full card is populated.
Run the externalisation test. Move the candidate head outside the clause in a neutral paraphrase and leave a gap in its original source role. If the result matches the matrix predicate and discourse reference, the head analysis is likely correct.
Verification Method
Back-read the target and identify exactly one head referent. Then reconstruct its role in the relative clause and the role of the entire relative NP in the matrix clause. Check definiteness and maximality separately. If two target entities appear where source had one, collapse duplication.
Recovery Method
If the wrong noun was externalised, return to source case/agreement and matrix selection. If the nominaliser became an extra noun, remove it unless independently referential. If English word order became awkward, rebuild a standard external-headed relative after preserving source roles.
Quality-Assurance Checklist
- Identify lexical internal head.
- Map internal grammatical role.
- Map matrix role of whole IHRC.
- Separate inner and outer case.
- Identify nominaliser function.
- Check definiteness/specificity.
- Check maximal/free-choice readings.
- Use agreement/reference as evidence.
- Distinguish IHRC from free relatives.
- Distinguish IHRC from correlatives.
- Distinguish IHRC from nominal complement clauses.
- Do not duplicate lexical head and nominaliser.
- Externalise only after role mapping.
- Audit AI wrong-head selection.
- Audit AI duplicate-reference and nominaliser literalism.
Frequently Asked Questions
What is an internally headed relative clause?
A relative construction where the lexical noun functioning as the semantic head appears inside the relative clause rather than outside it.
How is it different from an English relative clause?
English normally places the head noun outside the relative clause: the book [that Lina bought]. An IHRC keeps book inside the clause.
Does the internal head keep its case?
It carries case for its role inside the source relative clause, while the whole IHRC can receive a separate matrix case.
Are Japanese and Korean IHRCs identical?
No. Both are major sites of IHRC research, but their nominalisation, interpretation and syntactic constraints differ.
Is every nominalised clause an IHRC?
No. A clause nominalisation can denote an event or fact rather than select one internal noun as its head.
Is an IHRC the same as a free relative?
No. A free relative lacks a lexical noun head; an IHRC contains one internally.
Can English preserve head-internal order?
Usually English externalises the head, though literary or free-relative paraphrases can sometimes preserve information flow.
Why does AI mistranslate IHRCs?
Models can select the wrong internal noun as head, translate the head twice, or literalise the nominaliser as an extra thing/one.
Where This Fits in the eduKate Translation Architecture
Route restrictive/supplementary relative meaning to the live Relative Clauses owner; cross-clause referent tracking to Anaphora, Cataphora and Bridging Reference; ordinary case decoding to Grammatical Case; paired relative–demonstrative structures to Correlative Constructions.
Final Principle
The head noun may live inside the source relative clause and outside the target one. Preserve its identity and local role first, then let the target place that one referent where its own relative-clause grammar requires.