Differential subject marking in translation occurs when subjects are not all encoded alike. A language may mark some subjects differently because of animacy, volitionality, control, thematic role, person, tense–aspect, clause type or other grammatical conditions. The phenomenon—often abbreviated DSM—means that a special case ending or agreement pattern on a subject does not automatically indicate that the noun has stopped being a subject or become an object. Instead, the source grammar may be signalling that this subject is atypical, non-agentive, embedded, low-control or otherwise belongs to a particular construction.
People searching for differential subject marking, DSM, differential case marking on subjects, noncanonical subjects, dative subjects, ergative subjects, experiencer case, volitionality and subject marking are solving an argument-coding problem. Typological work shows that DSM can be conditioned by properties of the noun phrase itself, such as animacy or thematic role, and by properties of the clause, including tense/aspect or main-versus-dependent status. It can surface through case, agreement, voice or inverse-like systems. There is therefore no safe translation rule such as “marked subject = object” or “dative subject = recipient.”
This article owns the differential-subject-marking precision layer for eduKateSG. It complements the existing Differential Object Marking article, Ergative–Absolutive Alignment, Grammatical Case and Verb Valency. Its narrow job is to explain why the source gives one subject a special form, how to determine whether semantic agency, volitionality, experiencer status, animacy or clause-level features are responsible, and how to translate that subject naturally when the target language uses ordinary nominative subjects or a different noncanonical-subject system.
Quick Read
First prove that the noun is the source subject using agreement, syntax and predicate structure. Then record WHY it is marked differently: ANIMACY, VOLITION/CONTROL, EXPERIENCER ROLE, PERSON, TAM, CLAUSE TYPE, ERGATIVE/SPLIT ALIGNMENT or another language-specific condition. Translate the participant role and construction; do not translate the case marker as if it were automatically an object preposition.
One-Sentence Answer
Translate DSM by preserving the subject’s semantic role and the conditioning contrast that triggered special marking, even when the target expresses that contrast somewhere other than case.
Why This Translation Problem Deserves Its Own Precision Layer
Differential subject marking matters because subjects are often assumed to be grammatically uniform. When an experiencer receives dative case, an inanimate causer gets special marking, or an embedded subject appears in accusative, translators can over-trust familiar case names and reassign the participant to object status. The result may be fluent but conceptually wrong: the person who feels cold becomes someone to whom cold is given, or an embedded subject becomes an object of the reporting verb.
What DSM Is
DSM is systematic variation in how subject arguments are marked or indexed under identifiable semantic, referential or clausal conditions.
The main translation risk is calling every irregular subject form an exception rather than looking for the conditioning system. A target sentence can stay fluent while the event structure changes, so the error may survive ordinary proofreading. A precise reviewer therefore asks who performs the action, who receives or undergoes it, what grammatical privileges each participant has, and what discourse assumptions the source construction carries.
A practical diagnostic is to compare subjects across minimally different clauses. For example, animate versus inanimate or volitional versus nonvolitional subjects can receive different marking Once the structure is visible, the target language can use its own case, word order, prepositions, agreement, pronouns or clause shape. Formal similarity is optional; the participant relation is not.
Before release, confirm the contrast is repeatable and grammatical. If the source leaves a feature genuinely unspecified, keep it unspecified where the target allows. If the target forces a choice, use contextual evidence rather than choosing whichever version merely sounds smoother.
Subjecthood Before Case Labels
Subject status must be established from the language’s own diagnostics rather than from nominative-looking morphology alone.
The main translation risk is assuming a dative, ergative or accusative-marked noun cannot be subject. A target sentence can stay fluent while the event structure changes, so the error may survive ordinary proofreading. A precise reviewer therefore asks who performs the action, who receives or undergoes it, what grammatical privileges each participant has, and what discourse assumptions the source construction carries.
A practical diagnostic is to check agreement, syntactic privileges, predicate frame and language-specific diagnostics. For example, a dative experiencer can behave subject-like even though dative often marks recipients elsewhere Once the structure is visible, the target language can use its own case, word order, prepositions, agreement, pronouns or clause shape. Formal similarity is optional; the participant relation is not.
Before release, preserve subject-like semantic role in target. If the source leaves a feature genuinely unspecified, keep it unspecified where the target allows. If the target forces a choice, use contextual evidence rather than choosing whichever version merely sounds smoother.
Atypical Subjects
DSM often marks subjects that are less prototypically agentive, animate, volitional or discourse-expected.
The main translation risk is treating atypical marking as a completely different participant role. A target sentence can stay fluent while the event structure changes, so the error may survive ordinary proofreading. A precise reviewer therefore asks who performs the action, who receives or undergoes it, what grammatical privileges each participant has, and what discourse assumptions the source construction carries.
A practical diagnostic is to compare the participant with a canonical intentional human agent. For example, an inanimate force or involuntary experiencer can be a less typical subject Once the structure is visible, the target language can use its own case, word order, prepositions, agreement, pronouns or clause shape. Formal similarity is optional; the participant relation is not.
Before release, keep the participant central while preserving reduced agency. If the source leaves a feature genuinely unspecified, keep it unspecified where the target allows. If the target forces a choice, use contextual evidence rather than choosing whichever version merely sounds smoother.
Animacy
Some differential systems condition subject marking on human, animate or inanimate status.
The main translation risk is turning an inanimate marked subject into an object because it is conceptually unlikely as agent. A target sentence can stay fluent while the event structure changes, so the error may survive ordinary proofreading. A precise reviewer therefore asks who performs the action, who receives or undergoes it, what grammatical privileges each participant has, and what discourse assumptions the source construction carries.
A practical diagnostic is to separate grammatical subjecthood from animacy. For example, wind can be a causal subject even if specially marked Once the structure is visible, the target language can use its own case, word order, prepositions, agreement, pronouns or clause shape. Formal similarity is optional; the participant relation is not.
Before release, preserve causation without adding human agency. If the source leaves a feature genuinely unspecified, keep it unspecified where the target allows. If the target forces a choice, use contextual evidence rather than choosing whichever version merely sounds smoother.
Volitionality
Intentional versus involuntary participation can condition subject case or agreement in some languages.
The main translation risk is translating a nonvolitional subject as though the person deliberately acted. A target sentence can stay fluent while the event structure changes, so the error may survive ordinary proofreading. A precise reviewer therefore asks who performs the action, who receives or undergoes it, what grammatical privileges each participant has, and what discourse assumptions the source construction carries.
A practical diagnostic is to ask whether the participant intended or controlled the event. For example, accidental falling and deliberate jumping can differ in marking Once the structure is visible, the target language can use its own case, word order, prepositions, agreement, pronouns or clause shape. Formal similarity is optional; the participant relation is not.
Before release, preserve control and intentionality in target. If the source leaves a feature genuinely unspecified, keep it unspecified where the target allows. If the target forces a choice, use contextual evidence rather than choosing whichever version merely sounds smoother.
Control and Agentivity
Degree of control can form a continuum rather than a simple binary.
The main translation risk is reducing all marked subjects to passive victims. A target sentence can stay fluent while the event structure changes, so the error may survive ordinary proofreading. A precise reviewer therefore asks who performs the action, who receives or undergoes it, what grammatical privileges each participant has, and what discourse assumptions the source construction carries.
A practical diagnostic is to rank control over initiation and outcome. For example, an experiencer controls less than an intentional agent but remains event centre Once the structure is visible, the target language can use its own case, word order, prepositions, agreement, pronouns or clause shape. Formal similarity is optional; the participant relation is not.
Before release, match target agency without over-passivising. If the source leaves a feature genuinely unspecified, keep it unspecified where the target allows. If the target forces a choice, use contextual evidence rather than choosing whichever version merely sounds smoother.
Experiencer Subjects
Psychological, perception, bodily-state and emotion predicates commonly show noncanonical subject marking.
The main translation risk is translating an experiencer dative as a recipient. A target sentence can stay fluent while the event structure changes, so the error may survive ordinary proofreading. A precise reviewer therefore asks who performs the action, who receives or undergoes it, what grammatical privileges each participant has, and what discourse assumptions the source construction carries.
A practical diagnostic is to map EXPERIENCER and STIMULUS separately. For example, a source equivalent of ‘to Lina is cold’ means Lina feels cold Once the structure is visible, the target language can use its own case, word order, prepositions, agreement, pronouns or clause shape. Formal similarity is optional; the participant relation is not.
Before release, use natural target experiencer syntax. If the source leaves a feature genuinely unspecified, keep it unspecified where the target allows. If the target forces a choice, use contextual evidence rather than choosing whichever version merely sounds smoother.
Dative Subjects
Dative case can mark subject-like experiencers, obligations or states in some languages.
The main translation risk is forcing to/for into the target because the case is called dative. A target sentence can stay fluent while the event structure changes, so the error may survive ordinary proofreading. A precise reviewer therefore asks who performs the action, who receives or undergoes it, what grammatical privileges each participant has, and what discourse assumptions the source construction carries.
A practical diagnostic is to analyse predicate semantics before lexicalising case. For example, I need, I like or I am cold can have dative-subject counterparts Once the structure is visible, the target language can use its own case, word order, prepositions, agreement, pronouns or clause shape. Formal similarity is optional; the participant relation is not.
Before release, translate the state naturally. If the source leaves a feature genuinely unspecified, keep it unspecified where the target allows. If the target forces a choice, use contextual evidence rather than choosing whichever version merely sounds smoother.
Ergative Subject Marking
Ergative case on transitive A is a structural alignment pattern and can participate in split/differential systems.
The main translation risk is treating ergative A as object or instrument. A target sentence can stay fluent while the event structure changes, so the error may survive ordinary proofreading. A precise reviewer therefore asks who performs the action, who receives or undergoes it, what grammatical privileges each participant has, and what discourse assumptions the source construction carries.
A practical diagnostic is to map A/O first and identify split condition. For example, perfective transitive subjects may be ergative in some languages Once the structure is visible, the target language can use its own case, word order, prepositions, agreement, pronouns or clause shape. Formal similarity is optional; the participant relation is not.
Before release, preserve A role and TAM separately. If the source leaves a feature genuinely unspecified, keep it unspecified where the target allows. If the target forces a choice, use contextual evidence rather than choosing whichever version merely sounds smoother.
Accusative Embedded Subjects
Some languages mark subjects of embedded clauses with accusative or other non-nominative case.
The main translation risk is treating the embedded subject as a direct object of the matrix verb. A target sentence can stay fluent while the event structure changes, so the error may survive ordinary proofreading. A precise reviewer therefore asks who performs the action, who receives or undergoes it, what grammatical privileges each participant has, and what discourse assumptions the source construction carries.
A practical diagnostic is to draw clause boundaries and assign the noun to the lower predicate. For example, an accusative NP can be subject of a subordinate clause Once the structure is visible, the target language can use its own case, word order, prepositions, agreement, pronouns or clause shape. Formal similarity is optional; the participant relation is not.
Before release, translate lower subject with target finite or nonfinite syntax. If the source leaves a feature genuinely unspecified, keep it unspecified where the target allows. If the target forces a choice, use contextual evidence rather than choosing whichever version merely sounds smoother.
Tense–Aspect Conditioning
DSM can be conditioned by perfective/imperfective, tense or aspectual distinctions.
The main translation risk is changing participant roles merely because case changes with TAM. A target sentence can stay fluent while the event structure changes, so the error may survive ordinary proofreading. A precise reviewer therefore asks who performs the action, who receives or undergoes it, what grammatical privileges each participant has, and what discourse assumptions the source construction carries.
A practical diagnostic is to hold participant identity constant and compare clauses across TAM. For example, the same actor can receive different case in perfective versus imperfective contexts Once the structure is visible, the target language can use its own case, word order, prepositions, agreement, pronouns or clause shape. Formal similarity is optional; the participant relation is not.
Before release, preserve actor identity and aspect separately. If the source leaves a feature genuinely unspecified, keep it unspecified where the target allows. If the target forces a choice, use contextual evidence rather than choosing whichever version merely sounds smoother.
Main vs Dependent Clause
Clause type can trigger different subject marking.
The main translation risk is assuming subordinate-case marking changes semantic subjecthood. A target sentence can stay fluent while the event structure changes, so the error may survive ordinary proofreading. A precise reviewer therefore asks who performs the action, who receives or undergoes it, what grammatical privileges each participant has, and what discourse assumptions the source construction carries.
A practical diagnostic is to identify matrix versus embedded clause before interpreting case. For example, embedded subjects may receive special case unavailable in main clauses Once the structure is visible, the target language can use its own case, word order, prepositions, agreement, pronouns or clause shape. Formal similarity is optional; the participant relation is not.
Before release, rebuild target subordination while keeping the same subject. If the source leaves a feature genuinely unspecified, keep it unspecified where the target allows. If the target forces a choice, use contextual evidence rather than choosing whichever version merely sounds smoother.
Person Splits
First/second-person subjects can pattern differently from third-person lexical nouns.
The main translation risk is applying full-NP case rules to pronouns mechanically. A target sentence can stay fluent while the event structure changes, so the error may survive ordinary proofreading. A precise reviewer therefore asks who performs the action, who receives or undergoes it, what grammatical privileges each participant has, and what discourse assumptions the source construction carries.
A practical diagnostic is to compare person paradigms. For example, speech-act participants can have special forms or zero marking Once the structure is visible, the target language can use its own case, word order, prepositions, agreement, pronouns or clause shape. Formal similarity is optional; the participant relation is not.
Before release, preserve person and subject role independently. If the source leaves a feature genuinely unspecified, keep it unspecified where the target allows. If the target forces a choice, use contextual evidence rather than choosing whichever version merely sounds smoother.
Definiteness and Referentiality
Some DSM systems interact with definiteness, specificity or discourse status, though conditions are language-specific.
The main translation risk is copying DOM assumptions directly onto DSM. A target sentence can stay fluent while the event structure changes, so the error may survive ordinary proofreading. A precise reviewer therefore asks who performs the action, who receives or undergoes it, what grammatical privileges each participant has, and what discourse assumptions the source construction carries.
A practical diagnostic is to record referential features separately. For example, a marked subject can be discourse-new or atypical under one system Once the structure is visible, the target language can use its own case, word order, prepositions, agreement, pronouns or clause shape. Formal similarity is optional; the participant relation is not.
Before release, translate reference status only when source supports it. If the source leaves a feature genuinely unspecified, keep it unspecified where the target allows. If the target forces a choice, use contextual evidence rather than choosing whichever version merely sounds smoother.
Information Structure
Topic/focus can interact with subject marking in some constructions.
The main translation risk is turning special marking into contrastive focus automatically. A target sentence can stay fluent while the event structure changes, so the error may survive ordinary proofreading. A precise reviewer therefore asks who performs the action, who receives or undergoes it, what grammatical privileges each participant has, and what discourse assumptions the source construction carries.
A practical diagnostic is to check independent discourse evidence. For example, a marked subject can be prominent without contrastive focus Once the structure is visible, the target language can use its own case, word order, prepositions, agreement, pronouns or clause shape. Formal similarity is optional; the participant relation is not.
Before release, preserve topic/focus through target discourse devices if relevant. If the source leaves a feature genuinely unspecified, keep it unspecified where the target allows. If the target forces a choice, use contextual evidence rather than choosing whichever version merely sounds smoother.
DSM and DOM
DSM and differential object marking are related as differential argument marking but often respond differently to prototypicality.
The main translation risk is assuming the same animacy rule applies in the same direction. A target sentence can stay fluent while the event structure changes, so the error may survive ordinary proofreading. A precise reviewer therefore asks who performs the action, who receives or undergoes it, what grammatical privileges each participant has, and what discourse assumptions the source construction carries.
A practical diagnostic is to write subject and object conditioning separately. For example, animate objects can be atypical while inanimate subjects are atypical Once the structure is visible, the target language can use its own case, word order, prepositions, agreement, pronouns or clause shape. Formal similarity is optional; the participant relation is not.
Before release, use role-specific diagnostics. If the source leaves a feature genuinely unspecified, keep it unspecified where the target allows. If the target forces a choice, use contextual evidence rather than choosing whichever version merely sounds smoother.
DSM and Split Ergativity
A TAM/person split in ergative alignment can look like DSM because only some transitive subjects carry ergative case.
The main translation risk is creating a new DSM explanation where alignment split is more informative. A target sentence can stay fluent while the event structure changes, so the error may survive ordinary proofreading. A precise reviewer therefore asks who performs the action, who receives or undergoes it, what grammatical privileges each participant has, and what discourse assumptions the source construction carries.
A practical diagnostic is to compare S/O/A paradigms across the condition. For example, perfective A marking can belong to split ergativity Once the structure is visible, the target language can use its own case, word order, prepositions, agreement, pronouns or clause shape. Formal similarity is optional; the participant relation is not.
Before release, route full alignment analysis to the ergative owner. If the source leaves a feature genuinely unspecified, keep it unspecified where the target allows. If the target forces a choice, use contextual evidence rather than choosing whichever version merely sounds smoother.
DSM and Split-S
Intransitive subjects can divide by agentivity or affectedness, producing split-S or fluid-S systems.
The main translation risk is forcing all S arguments into one canonical case. A target sentence can stay fluent while the event structure changes, so the error may survive ordinary proofreading. A precise reviewer therefore asks who performs the action, who receives or undergoes it, what grammatical privileges each participant has, and what discourse assumptions the source construction carries.
A practical diagnostic is to compare agentive RUN with patientive FALL or BE-SICK predicates. For example, S can align differently according to control/state Once the structure is visible, the target language can use its own case, word order, prepositions, agreement, pronouns or clause shape. Formal similarity is optional; the participant relation is not.
Before release, preserve event semantics and differential marking effect. If the source leaves a feature genuinely unspecified, keep it unspecified where the target allows. If the target forces a choice, use contextual evidence rather than choosing whichever version merely sounds smoother.
DSM and Voice
Differential subject patterns can surface through voice or agreement rather than only overt case suffixes.
The main translation risk is searching only for case markers. A target sentence can stay fluent while the event structure changes, so the error may survive ordinary proofreading. A precise reviewer therefore asks who performs the action, who receives or undergoes it, what grammatical privileges each participant has, and what discourse assumptions the source construction carries.
A practical diagnostic is to check whether the same semantic subject appears under different voice morphology. For example, an atypical actor can trigger special voice in some systems Once the structure is visible, the target language can use its own case, word order, prepositions, agreement, pronouns or clause shape. Formal similarity is optional; the participant relation is not.
Before release, translate participant relation while respecting source construction. If the source leaves a feature genuinely unspecified, keep it unspecified where the target allows. If the target forces a choice, use contextual evidence rather than choosing whichever version merely sounds smoother.
Noncanonical Subjects
A noncanonical subject is subject-like but bears case or agreement unlike the default nominative pattern.
The main translation risk is reclassifying every noncanonical subject as object. A target sentence can stay fluent while the event structure changes, so the error may survive ordinary proofreading. A precise reviewer therefore asks who performs the action, who receives or undergoes it, what grammatical privileges each participant has, and what discourse assumptions the source construction carries.
A practical diagnostic is to use multiple subjecthood diagnostics. For example, a dative experiencer can control some subject behaviours Once the structure is visible, the target language can use its own case, word order, prepositions, agreement, pronouns or clause shape. Formal similarity is optional; the participant relation is not.
Before release, choose target nominative subject if that is the natural equivalent. If the source leaves a feature genuinely unspecified, keep it unspecified where the target allows. If the target forces a choice, use contextual evidence rather than choosing whichever version merely sounds smoother.
Weather and Impersonal Predicates
Weather and impersonal constructions can interact with special marking or expletive subjects.
The main translation risk is inventing a concrete actor where the source is impersonal. A target sentence can stay fluent while the event structure changes, so the error may survive ordinary proofreading. A precise reviewer therefore asks who performs the action, who receives or undergoes it, what grammatical privileges each participant has, and what discourse assumptions the source construction carries.
A practical diagnostic is to separate formal/expletive subject from semantic participants. For example, English it rains has no referential it Once the structure is visible, the target language can use its own case, word order, prepositions, agreement, pronouns or clause shape. Formal similarity is optional; the participant relation is not.
Before release, use natural target impersonal grammar. If the source leaves a feature genuinely unspecified, keep it unspecified where the target allows. If the target forces a choice, use contextual evidence rather than choosing whichever version merely sounds smoother.
Inanimate Causers
Tools, forces and circumstances can function as subjects or causers while being nonhuman and nonvolitional.
The main translation risk is adding a human agent because an inanimate subject sounds odd. A target sentence can stay fluent while the event structure changes, so the error may survive ordinary proofreading. A precise reviewer therefore asks who performs the action, who receives or undergoes it, what grammatical privileges each participant has, and what discourse assumptions the source construction carries.
A practical diagnostic is to map CAUSER separately from intentional AGENT. For example, the wind opened the door can be causal without intention Once the structure is visible, the target language can use its own case, word order, prepositions, agreement, pronouns or clause shape. Formal similarity is optional; the participant relation is not.
Before release, preserve causal source without invented human agency. If the source leaves a feature genuinely unspecified, keep it unspecified where the target allows. If the target forces a choice, use contextual evidence rather than choosing whichever version merely sounds smoother.
Accidental Events
DSM can encode accidental or involuntary participation in some languages.
The main translation risk is turning accidental events into deliberate actions. A target sentence can stay fluent while the event structure changes, so the error may survive ordinary proofreading. A precise reviewer therefore asks who performs the action, who receives or undergoes it, what grammatical privileges each participant has, and what discourse assumptions the source construction carries.
A practical diagnostic is to test volitional paraphrases. For example, a special subject form can indicate unintentional involvement Once the structure is visible, the target language can use its own case, word order, prepositions, agreement, pronouns or clause shape. Formal similarity is optional; the participant relation is not.
Before release, use accidental or involuntary target wording when needed. If the source leaves a feature genuinely unspecified, keep it unspecified where the target allows. If the target forces a choice, use contextual evidence rather than choosing whichever version merely sounds smoother.
Obligation and Necessity
Noncanonical subjects are common with obligation, possession and necessity predicates.
The main translation risk is translating case as transfer instead of modal relation. A target sentence can stay fluent while the event structure changes, so the error may survive ordinary proofreading. A precise reviewer therefore asks who performs the action, who receives or undergoes it, what grammatical privileges each participant has, and what discourse assumptions the source construction carries.
A practical diagnostic is to identify MODAL HOLDER or EXPERIENCER. For example, a dative-like subject can mean Lina must leave Once the structure is visible, the target language can use its own case, word order, prepositions, agreement, pronouns or clause shape. Formal similarity is optional; the participant relation is not.
Before release, preserve who bears the obligation. If the source leaves a feature genuinely unspecified, keep it unspecified where the target allows. If the target forces a choice, use contextual evidence rather than choosing whichever version merely sounds smoother.
Agreement Mismatches
Marked subjects may have reduced or unusual agreement privileges.
The main translation risk is using lack of canonical agreement as proof the noun is not subject. A target sentence can stay fluent while the event structure changes, so the error may survive ordinary proofreading. A precise reviewer therefore asks who performs the action, who receives or undergoes it, what grammatical privileges each participant has, and what discourse assumptions the source construction carries.
A practical diagnostic is to check the whole paradigm. For example, some noncanonical subjects do not control full agreement Once the structure is visible, the target language can use its own case, word order, prepositions, agreement, pronouns or clause shape. Formal similarity is optional; the participant relation is not.
Before release, preserve semantic role even if target agreement normalises. If the source leaves a feature genuinely unspecified, keep it unspecified where the target allows. If the target forces a choice, use contextual evidence rather than choosing whichever version merely sounds smoother.
Word Order
Specially marked subjects may occupy positions associated with experiencers/topics rather than default agents.
The main translation risk is taking position as decisive over morphology and semantics. A target sentence can stay fluent while the event structure changes, so the error may survive ordinary proofreading. A precise reviewer therefore asks who performs the action, who receives or undergoes it, what grammatical privileges each participant has, and what discourse assumptions the source construction carries.
A practical diagnostic is to identify role before linear order. For example, a dative experiencer can occur before or after stimulus Once the structure is visible, the target language can use its own case, word order, prepositions, agreement, pronouns or clause shape. Formal similarity is optional; the participant relation is not.
Before release, generate target order from target grammar. If the source leaves a feature genuinely unspecified, keep it unspecified where the target allows. If the target forces a choice, use contextual evidence rather than choosing whichever version merely sounds smoother.
Clinical and Sensation Predicates
Pain, cold, nausea, fear and bodily states are common environments for experiencer-like subjects.
The main translation risk is making the condition a transferred object or recipient event. A target sentence can stay fluent while the event structure changes, so the error may survive ordinary proofreading. A precise reviewer therefore asks who performs the action, who receives or undergoes it, what grammatical privileges each participant has, and what discourse assumptions the source construction carries.
A practical diagnostic is to write EXPERIENCER + STATE/STIMULUS. For example, Lina feels cold may correspond to a dative-subject construction Once the structure is visible, the target language can use its own case, word order, prepositions, agreement, pronouns or clause shape. Formal similarity is optional; the participant relation is not.
Before release, preserve patient/experiencer identity clinically. If the source leaves a feature genuinely unspecified, keep it unspecified where the target allows. If the target forces a choice, use contextual evidence rather than choosing whichever version merely sounds smoother.
Legal and Responsibility Language
Volitionality-sensitive marking can matter where responsibility is interpreted.
The main translation risk is adding intention because target nominative subjects sound agentive. A target sentence can stay fluent while the event structure changes, so the error may survive ordinary proofreading. A precise reviewer therefore asks who performs the action, who receives or undergoes it, what grammatical privileges each participant has, and what discourse assumptions the source construction carries.
A practical diagnostic is to separate legal responsibility from grammatical subjecthood. For example, a marked involuntary actor should not become a deliberate perpetrator Once the structure is visible, the target language can use its own case, word order, prepositions, agreement, pronouns or clause shape. Formal similarity is optional; the participant relation is not.
Before release, preserve source agency strength. If the source leaves a feature genuinely unspecified, keep it unspecified where the target allows. If the target forces a choice, use contextual evidence rather than choosing whichever version merely sounds smoother.
AI Case-Name Bias
AI can translate dative subject as recipient or accusative subject as object because case labels dominate familiar patterns.
The main translation risk is accepting a fluent role reassignment. A target sentence can stay fluent while the event structure changes, so the error may survive ordinary proofreading. A precise reviewer therefore asks who performs the action, who receives or undergoes it, what grammatical privileges each participant has, and what discourse assumptions the source construction carries.
A practical diagnostic is to run subjecthood and predicate-role diagnostics before generation. For example, embedded accusative NP can still be lower-clause subject Once the structure is visible, the target language can use its own case, word order, prepositions, agreement, pronouns or clause shape. Formal similarity is optional; the participant relation is not.
Before release, restore participant role and treat case constructionally. If the source leaves a feature genuinely unspecified, keep it unspecified where the target allows. If the target forces a choice, use contextual evidence rather than choosing whichever version merely sounds smoother.
AI Agency Inflation
Models may normalise atypical subjects into canonical active agents.
The main translation risk is adding intention or control absent from source. A target sentence can stay fluent while the event structure changes, so the error may survive ordinary proofreading. A precise reviewer therefore asks who performs the action, who receives or undergoes it, what grammatical privileges each participant has, and what discourse assumptions the source construction carries.
A practical diagnostic is to compare target agency with source volitionality. For example, an involuntary experiencer can become an active decider in a bad paraphrase Once the structure is visible, the target language can use its own case, word order, prepositions, agreement, pronouns or clause shape. Formal similarity is optional; the participant relation is not.
Before release, reduce target agency to source level. If the source leaves a feature genuinely unspecified, keep it unspecified where the target allows. If the target forces a choice, use contextual evidence rather than choosing whichever version merely sounds smoother.
Worked Examples
Worked Example 1: Dative Experiencer
A source uses a dative-marked person with a predicate meaning cold.
Diagnosis: The person is experiencer and subject-like centre, not a recipient of an object called cold.
Target strategy: English naturally says ‘Lina is cold’ or ‘Lina feels cold.’
Verification: Lina must remain experiencer without transfer semantics.
Worked Example 2: Accusative Embedded Subject
A source subordinate clause marks its subject accusative.
Diagnosis: Case is licensed by embedded-clause grammar; the noun remains semantic subject of the lower predicate.
Target strategy: English can use a finite that-clause or infinitival construction.
Verification: The noun must perform or undergo the lower predicate.
Worked Example 3: Volitionality Split
A language marks an intentional actor differently from an accidental participant.
Diagnosis: Subject role is central in both clauses, but control differs.
Target strategy: Use lexical, aspectual or adverbial target resources to preserve deliberate versus accidental involvement.
Verification: The accidental source must not become intentional.
Worked Example 4: TAM-Conditioned Subject Marking
The same actor changes case between perfective and imperfective clauses.
Diagnosis: The conditioning factor is TAM, not participant identity.
Target strategy: English can keep the same subject while expressing aspect elsewhere.
Verification: Actor and aspect must both survive.
Practical Diagnostic Workflow
Build a DSM card: SUBJECT CANDIDATE; PREDICATE; SEMANTIC ROLE; CASE/AGREEMENT; ANIMACY; VOLITION/CONTROL; PERSON; TAM; CLAUSE TYPE; INFORMATION STATUS. Compare marked and unmarked subjects. The conditioning factor usually becomes visible.
Run subjecthood diagnostics appropriate to the source language. Evidence can include agreement, syntactic position, control, binding, extraction or predicate structure. No one test is universal, so rely on the language’s documented grammar rather than English expectations.
Verification Method
Read the target and identify who experiences, acts, causes, possesses or bears the obligation. Rate agency and volitionality. Compare with the source DSM card. If case morphology vanished but participant and agency profile remain, the translation can still be precise.
Recovery Method
If a marked subject has become an object, restore the predicate’s semantic centre first. Remove target prepositions introduced merely from the case label, rebuild the experiencer/actor/causer relation, then express TAM, clause-type or agency differences in their natural target location.
Practice Sequence
Create pairs such as deliberate jump/accidental fall, human actor/inanimate causer, feel cold/cause cold, main-clause subject/embedded subject, perfective/imperfective. Predict the source marking and state exactly what condition changed.
Quality-Assurance Checklist
- Prove subjecthood before trusting case labels.
- Identify semantic role.
- Record animacy.
- Record volitionality/control.
- Check experiencer constructions.
- Check TAM-conditioned marking.
- Check main vs subordinate clause.
- Check person/NP-type splits.
- Distinguish DSM from DOM.
- Distinguish DSM from split ergativity when appropriate.
- Handle noncanonical agreement cautiously.
- Avoid invented agency with inanimate or accidental subjects.
- Preserve obligation/experiencer holder.
- Audit AI case-name bias.
- Audit AI agency inflation.
Frequently Asked Questions
What is differential subject marking?
Systematic variation in how subjects are marked or indexed depending on semantic, referential or clausal conditions.
What can condition DSM?
Animacy, thematic role, volitionality, person, tense/aspect, clause type and other language-specific factors.
Is a dative-marked noun necessarily an indirect object?
No. Many languages have dative or other noncanonical subjects.
Can an accusative noun be a subject?
In some embedded or special constructions, yes; case name alone does not determine subjecthood.
How is DSM different from DOM?
DSM concerns differential subject marking; DOM concerns objects. Their conditioning patterns can differ.
Is split ergativity DSM?
They can overlap descriptively, but A/S/O alignment analysis is often more informative when an ergative split explains the pattern.
Does special subject marking mean less agency?
Sometimes systems are sensitive to control or volitionality, but not every special case means reduced agency.
Why does AI misread DSM?
Models often associate dative/accusative case names with familiar object roles and normalise noncanonical subjects.
Internal Routes Through the eduKate Translation Architecture
Route the broader case system to Grammatical Case: https://edukatesg.com/2026/09/18/translate-grammatical-case-who-did-what-to-whom-language-roles/. Route object-side differential marking to Differential Object Marking: https://edukatesg.com/2026/09/18/translate-precisely-differential-object-marking-animacy-definiteness-specificity-case/. Route A/S/O alignment and splits to the prepared Ergative–Absolutive Alignment and Split Ergativity lane. Route predicate roles to Verb Valency and Argument Structure: https://edukatesg.com/2026/09/18/translate-precisely-verb-valency-argument-structure-transitivity-complements/.
Final Principle
A specially marked subject is still a subject if the source grammar makes it one. Identify the participant’s semantic role and the condition that triggered special marking, then let the target express agency, experience, TAM and clause status in its own way.