Split ergativity in translation is the problem of preserving a grammatical system whose alignment changes under different conditions. A language may behave ergatively in one tense or aspect and accusatively in another; pronouns may follow one alignment while full noun phrases follow another; main clauses may differ from subordinate clauses; person, animacy, polarity or mood may also control the split. A translator who assumes one fixed subject–object pattern can misread case, agreement and participant roles.
This guide explains how to translate split ergativity accurately. It covers ergative–absolutive and nominative–accusative alignment, tense/aspect splits, person splits, pronoun versus noun-phrase splits, clause-type splits, agreement splits, perfective and imperfective contexts, participial origins, agent marking, patient tracking, translation into English, translation out of English, machine-translation risks and document-level quality assurance.
The core discipline is to treat alignment as conditional grammar. Do not ask only “Is this language ergative?” Ask “Under exactly which construction is this clause ergative, which arguments are grouped together, and which layer—case, agreement or syntax—shows the split?” WALS notes that split alignment is widely attested and that tense/aspect, mood, polarity, humanness, animacy, definiteness and clause type can condition changes in verbal person marking. Translation therefore begins by identifying the active alignment for this clause, not by applying one language-wide label.
The Three Roles: S, A and P
Alignment is easier to translate when the analyst stops using “subject” as the only starting point. Typology often distinguishes S, the sole participant of an intransitive verb; A, the more agent-like participant of a transitive verb; and P, the more patient-like participant of a transitive verb.
In nominative–accusative alignment, S and A are treated alike, while P is treated differently. In ergative–absolutive alignment, S and P are treated alike, while A is treated differently. A split-ergative system uses different alignments under different grammatical conditions.
These letters are analytical tools, not translations. They let the translator identify participant structure before deciding how English subject and object syntax should represent it.
What ‘Split’ Actually Means
A split does not necessarily mean that every part of grammar switches at once. Case marking may become ergative while verbal agreement remains accusative. Pronouns may stay accusative while nouns become ergative. Syntax can pattern differently from morphology. A translator therefore needs to ask which subsystem is split.
This is one reason broad statements such as “Language X is ergative” are insufficient for production work. They are useful orientation, not a complete parsing rule.
Tense and Aspect Splits
One of the best-known types of split ergativity is conditioned by tense or aspect. A language may use ergative case marking in perfective or completed-event constructions but nominative–accusative patterns in imperfective or non-completed contexts.
WALS notes that alignment of verbal person marking can vary by tense and aspect; it gives Itzaj as an example where completive and dependent contexts pattern ergatively while non-completive contexts pattern accusatively. The broader lesson is that temporal-aspectual morphology can change how participant roles are encoded.
For translation, never parse case before identifying tense/aspect. A noun ending that marks the transitive agent in one construction may not appear in another. If the translator ignores aspect, the role system can seem inconsistent when it is actually rule-governed.
Perfective Splits
Perfective contexts often describe bounded or completed events. In several languages with split ergativity, these contexts select ergative marking on A. The historical pathways differ by language, so the translator should avoid universal causal stories. Operationally, however, the workflow is clear: identify the perfective construction, expect the relevant ergative pattern, and map A and P accordingly.
English may translate both perfective and imperfective source clauses with ordinary subject-first syntax. The ergative split can disappear from the target surface while remaining essential to accurate role recovery.
Person-Based Splits
Some splits depend on person. First and second person pronouns can pattern differently from third-person nouns or pronouns. Person hierarchies often interact with alignment because speech-act participants are highly prominent and frequently have special paradigms.
A translator should therefore record person before interpreting ambiguous case or agreement. The same formal pattern may not extend from a third-person noun phrase to “I” or “you.”
Pronouns Versus Full Noun Phrases
Pronouns often preserve older or different case distinctions from full noun phrases. A language can show ergative alignment with lexical nouns but accusative-like behavior with pronouns. If a translator learns the noun paradigm and applies it mechanically to pronouns, errors follow immediately.
For project resources, keep separate tables for pronouns and full noun phrases whenever the grammar requires it. Do not hide the split under one generic “subject case” label.
Clause-Type Splits
Main clauses and subordinate clauses can pattern differently. Relative clauses, complement clauses, nominalized clauses, imperatives or dependent clauses may select a different alignment from ordinary positive declarative main clauses.
WALS explicitly notes that main-versus-subordinate status can condition alignment patterns in verbal person marking. For translators, this means clause identification must precede alignment analysis. A form that looks exceptional may simply belong to a different clause type.
Polarity and Mood
Negation and mood can also trigger alignment changes. A negative clause may neutralize a contrast, an imperative may use a special pattern, or a non-declarative construction may alter agreement. These systems are language-specific, but they share one translation lesson: alignment can be construction-dependent.
Do not build parsers—human or computational—that assume positive declarative behavior everywhere. Tag mood and polarity before assigning roles in difficult cases.
Case Alignment Versus Agreement Alignment
Case on noun phrases and person marking on verbs can follow different alignment systems. A clause may have ergative case on A but verb agreement that groups S with A. Another language may show the reverse combination. This mixed behavior is not noise; it is part of the grammar.
WALS maintains separate typological features for case alignment and verbal person marking precisely because the two dimensions do not always match. Translation analysis should do the same.
Worked Example 1: Perfective Ergative Clause
Imagine a language where the perfective transitive clause marks the agent ergatively. The source equivalent of “Mira opened the door” appears as “Mira-ERG door.ABS opened.PFV.” English still says “Mira opened the door.”
The ergative marker does not make Mira an indirect object or passive agent. It is the normal A marking in that aspectual construction. The translator must resist the temptation to translate case shape with English prepositions.
Worked Example 2: Imperfective Accusative Clause
The same language may express “Mira is opening the door” with Mira in the same case as an intransitive subject and the door in accusative case. A superficial comparison makes the two clauses look as though participant roles changed. They did not; alignment changed with aspect.
The target should preserve the aspect contrast and the same actor–patient relation.
Worked Example 3: Pronoun Split
Suppose lexical nouns use ergative marking in a construction, but first-person pronouns use an accusative-like pattern. A translator encounters “I saw the dog” and expects the pronoun to carry the same ergative marker as a noun agent. Its absence is not evidence that “I” is not the actor.
Paradigm knowledge prevents false role reassignment.
Worked Example 4: Agreement Does Not Match Case
A transitive agent carries ergative case, but the verb agrees with the absolutive patient. If the patient is plural, the verb shows plural morphology. A translator who assumes verbal agreement always identifies the agent may reverse the sentence.
Record case and agreement separately: A = ergative-marked noun; P = absolutive noun; verb agreement = P features. Then translate roles normally.
Split Ergativity Versus Active–Stative Alignment
Active–stative or split-S systems divide intransitive S arguments according to semantic properties such as volition, control or affectedness. Split ergativity divides alignment according to constructions such as tense/aspect, person or clause type. The two phenomena can coexist but should not be conflated.
The question is different: split ergativity asks when the grammar groups S with A or P; active–stative alignment asks whether different S arguments pattern with different transitive roles based on semantic class.
Split Ergativity Versus Differential Subject Marking
Differential subject marking concerns variation in how subjects are marked under factors such as animacy or volitionality. Split ergativity is broader alignment alternation. A subject can receive different case without the entire S/A/P alignment changing.
Translation should identify whether the construction represents a local case alternation or a genuine alignment split.
Split Ergativity Versus Antipassive
An antipassive changes transitivity and often demotes or backgrounds the patient, typically in ergative languages. It is not itself a split-ergativity condition. However, antipassives can appear in the same languages and may be used to satisfy syntactic or discourse requirements.
Do not treat an unmarked agent automatically as evidence that the language has switched to accusative alignment. The clause may have changed valency instead.
Historical Sources and Synchrony
Some split systems historically develop from perfect participles, possessive constructions or reanalysis of resultative structures. These histories can explain why perfective clauses pattern differently. But translation should prioritize the synchronic grammar used by current speakers.
Historical explanation is useful when it clarifies irregular morphology. It should not force modern sentences into archaic English structures.
Translation Into English
- Identify S, A and P before choosing English subject/object order.
- Determine the active alignment for the clause’s tense, aspect, person and type.
- Do not translate ergative case with a preposition by default.
- Check whether verbal agreement tracks A, P or another participant.
- Preserve aspect independently of case alignment.
- Use English passive only when source voice or discourse justifies it, not because ergative case looks “agentive.”
Translation Out of English
English usually provides no overt clue that a target language will switch alignment under a particular aspect or person combination. The translator must first choose the target tense/aspect construction, then apply the alignment rules that construction requires.
This means that grammatical decisions are ordered. Aspect choice can determine case marking, and case marking can determine agreement. Translating word by word before deciding aspect can produce a cascade of wrong forms.
Agreement and Auxiliary Systems
Split ergativity can involve auxiliary constructions where one element carries tense/aspect and another carries lexical meaning. Agreement may attach to the auxiliary, main verb or both. The distribution can differ across constructions.
For translation technology, segmenting auxiliary and lexical verb separately can obscure their joint role in alignment. Review the whole predicate complex.
Nominalized and Participial Constructions
Some ergative patterns appear in clauses with nominalized or participial properties. A source construction may historically or synchronically resemble possession or resultative predication. Translators should identify whether the clause is fully finite, dependent or nominalized before applying the main-clause alignment rule.
Narrative Translation
Narratives often alternate perfective foreground events with imperfective background description. If alignment switches with aspect, the grammar can change repeatedly as the story moves between event types. A translator may see different case patterns for the same character from sentence to sentence.
That alternation does not imply changing agency. Track the character’s semantic role and the aspectual layer separately.
Legal and Evidential Text
Role accuracy is critical where responsibility matters. A perfective ergative clause describing who signed, struck, transferred or authorized something must not be misread as passive simply because the agent bears marked case.
High-stakes review should include an S/A/P map for every consequential transitive clause whose alignment is non-familiar to the reviewer.
Machine Translation Failure Modes
1. Global-language-label error
The model learns that the language is “ergative” and applies ergative parsing in clauses where the grammar is accusative.
2. Aspect-blind parsing
Perfective and imperfective clauses receive the same participant analysis despite different case systems.
3. Agreement-as-agent assumption
The model assumes the participant controlling agreement is always the transitive agent.
4. Preposition hallucination
Ergative case is rendered as “by” even though the source is an ordinary active clause.
5. Pronoun overgeneralization
A noun-case rule is incorrectly applied to first- or second-person pronouns.
The Split-Alignment Worksheet
- Clause type
- Tense and aspect
- Polarity and mood
- Person and animacy of each participant
- Case on S/A/P
- Agreement controller
- Voice or valency operation
- Target subject/object structure
Complete the worksheet only for clauses that are genuinely uncertain. Once the pattern is internalized, most clauses can be translated normally. The worksheet is a diagnostic tool, not permanent bureaucracy.
QA Table
| Layer | Question | Risk |
|---|---|---|
| Aspect | Perfective, imperfective or another category? | Wrong alignment rule |
| Person | Do pronouns follow the same pattern as nouns? | Paradigm overgeneralization |
| Case | Which role bears ergative/absolutive marking? | Role reversal |
| Agreement | Which participant controls the verb? | Agreement misread as agency |
| Clause type | Main, dependent, relative or imperative? | Wrong construction rule |
| Voice | Is there antipassive/passive morphology? | Alignment confused with valency |
Training Exercise: Same Event, Different Aspect
Translate a pair equivalent to “Mira opened the door” and “Mira was opening the door” in a language where aspect conditions alignment. Label A and P in both. Observe how case changes while semantic roles remain constant.
Then reverse the exercise into English. If the English output changes actor or patient merely because source case changed, the learner has confused alignment with event structure.
Training Exercise: Noun Versus Pronoun
Use the same transitive verb with a third-person lexical noun agent, a first-person pronoun agent and a second-person pronoun agent. Compare case and agreement. Build separate mini-paradigms rather than one generalized rule.
Training Exercise: Main Versus Subordinate Clause
Translate the same event once as a main clause and once inside a complement or relative clause. If alignment changes, document the condition. This trains translators to recognize that clause status can activate a different grammar.
Corpus Review
Filter a corpus by perfective and imperfective forms, then compare A marking. Next filter by first/second person versus third person. Patterns that seem irregular in isolated sentences often become systematic under corpus sorting.
Use corpus evidence to refine project rules, especially in languages where reference grammars describe multiple dialects or historical stages.
Translation Memory Risk
A translation-memory match can be dangerous when the source lexical string is identical but aspect differs. The target may require different case, agreement or auxiliary structure. Treat tense/aspect morphology as context that affects match safety.
TM systems should not strip or normalize away grammatical markers that determine alignment.
Terminology and Glossary Practice
Do not store case-marked noun forms as universal preferred terms. Store lemmas plus inflection guidance. Alignment belongs to sentence grammar, not terminology identity.
Interpreting Split-Ergative Languages
Interpreters need fast role recovery. The practical sequence is: hear aspect/auxiliary, identify case, map A and P, then formulate the target. Beginning the target before alignment is known increases the risk of self-correction or role reversal.
Consecutive notes can use A→P arrows and aspect abbreviations. A compact “PFV A→P” note can preserve more structure than several source words.
High-Stakes Timeline and Role Audit
In medicine, law, policing and incident reporting, combine temporal and role review. Because aspect can trigger alignment, the same morphology that locates an event can also change participant marking. Verify both the event boundary and the actor–patient relation.
Evidence and Further Reading
The WALS chapter on alignment of verbal person marking documents accusative, ergative, active, hierarchical and split patterns and notes conditioning by tense/aspect, mood, polarity, animacy, definiteness and clause status. It also reports that many split patterns involve ergative/accusative alternations.
For the classic connection between person hierarchies and ergativity, WALS records Michael Silverstein’s influential 1976 work Hierarchy of Features and Ergativity. Translators should use such typological models as diagnostics while grounding final decisions in the grammar of the actual language and variety.
Production Operating Manual
At project intake, build an alignment matrix with rows for major tense/aspect constructions and columns for S, A, P case and agreement. Add separate rows for pronouns if they differ from nouns. Include subordinate and imperative patterns where relevant.
During translation, consult the matrix before role assignment in unfamiliar constructions. During review, sample one sentence from each alignment condition rather than checking only the most frequent main-clause pattern.
When an apparent exception appears, first verify whether the clause belongs to a different row of the matrix. Many “exceptions” are unrecognized split conditions.
Advanced Diagnostic: Mixed Alignment
A clause can show ergative case but accusative agreement, or the reverse. Call this mixed alignment in the project notes rather than forcing one global label. The translation question is always local: what does each morphological subsystem tell us about participant roles?
If case and agreement appear to contradict each other, do not choose one arbitrarily. Check whether one tracks grammatical role and the other information structure, person hierarchy or another feature.
Advanced Diagnostic: Optional Ergative Marking
Some languages allow variation in ergative marking under discourse or dialect conditions. Optionality should not be assumed to mean free interchangeability. Corpus evidence may show differences in emphasis, agent salience or register.
For translation, preserve any documented pragmatic contrast where it affects the message; otherwise follow the target’s natural grammar without overloading the sentence.
Extended QA Checklist
- S, A and P were identified before target drafting.
- The active split condition was identified.
- Case and agreement alignment were checked separately.
- Pronoun paradigms were not inferred from noun paradigms.
- Perfective/imperfective distinctions remain correct.
- Main/dependent clause status was verified.
- Antipassive or passive voice was not confused with alignment change.
- Ergative case was not translated as English “by” automatically.
- Translation-memory reuse respected aspect and alignment.
- High-stakes clauses passed an actor–patient audit.
Final Principle: Alignment Changes; Event Roles Do Not Automatically Change With It
Split ergativity is difficult only if the translator expects one language to use one permanent subject–object coding strategy. Once the split conditions are known, the system becomes predictable. The same actor can appear with different case across perfective and imperfective clauses without ceasing to be the actor.
The translator should therefore protect semantic roles first, grammatical alignment second, and target form third. Case and agreement are evidence for the parse; English word order is the result of that parse.
Identify the construction, activate the correct alignment rule, recover S/A/P, and then translate naturally. That is the reliable method for split ergativity.
Extended Project Guide: Keep Alignment Rules Versioned
Large projects should version alignment guidance when linguistic review changes a rule. A correction to perfective case marking can affect hundreds of segments. Record the date, source and affected constructions, then rerun targeted QA rather than relying on translators to remember the update.
This is especially important when a project spans dialects. One variety may preserve an ergative split that another has reduced or reorganized. Tag content by variety before applying bulk corrections.
Why Split Ergativity Matters for Translation Technology
Automatic systems often learn frequent correlations such as first noun equals subject or agreement equals agent. Split alignment breaks those shortcuts. Better systems need morphology, clause structure and context, while human reviewers need tests that specifically target the split conditions.
A robust evaluation set should contain the same lexical event across multiple aspects, persons and clause types. If the system preserves roles across the alternation, it is learning more than surface order. If roles flip when alignment changes, the system needs targeted repair.
Deep Dive: Aspect-Conditioned Splits in Real Translation Work
Aspect-conditioned split ergativity deserves special attention because it can change the entire surface shape of a clause while leaving the participant roles unchanged. A translator may see the same two nouns with different cases in perfective and imperfective contexts and incorrectly infer that the actor or patient has changed. The correct interpretation comes from the interaction between aspect and alignment, not from case shape alone.
In production work, build paired examples. One row contains the imperfective construction, one the perfective. Keep the lexical verb and noun phrases constant. Label A and P explicitly. Then record the case on each noun phrase and the agreement pattern on the verb. This pair becomes a reference model for hundreds of later sentences.
The point is to teach the eye to expect formal alternation without semantic panic. Once translators recognize that a completed event activates one alignment pattern and an ongoing event another, they stop treating each new case form as a fresh exception.
Split Ergativity and Auxiliary Constructions
Some split systems are tied to auxiliary-based constructions. A perfective or resultative auxiliary may participate in agreement differently from the lexical verb, and the historical development of the construction can leave mixed patterns. Translation software that separates the auxiliary from the main predicate may miss the alignment relation entirely.
Review the predicate as one grammatical unit before translating it as several target words. Ask which element carries tense, which carries agreement, which carries lexical meaning and how case is assigned to the participants. English may then reconstruct the whole clause with an ordinary finite verb and a simple subject–object order.
This is another example of a recurring translation principle: source complexity can legitimately collapse in the target, but only after the source system has been fully understood.
Person Splits and the Silverstein Hierarchy
Person-based splits often interact with hierarchies of discourse prominence. First and second persons are deeply anchored in the speech event, while third-person nouns can be less prominent. Silverstein’s influential work connected such hierarchies with ergative patterning in several languages. Translators do not need to reproduce the full theoretical model to benefit from it.
The operational lesson is to test pronouns separately. If “the man saw the dog” uses an ergative pattern, do not assume “I saw the dog” will use the same morphology. If first person switches to accusative-like marking, the absence of ergative morphology is expected, not evidence of a different event structure.
When building a project grammar card, place speech-act participant forms at the top rather than burying them among noun paradigms. They are frequent, structurally important and likely to trigger the split.
Split Ergativity and Differential Case Marking
Split ergativity can coexist with differential subject or object marking. A perfective clause may generally be ergative, yet highly animate objects receive additional marking. An imperfective clause may be accusative, yet some subjects behave differently because of volitionality or experiencer semantics. Translation errors multiply when several systems are flattened into one rule.
Keep separate annotations for alignment, differential marking and valency. Alignment answers how S, A and P are grouped. Differential marking answers why one member of a grammatical class receives special case. Valency answers how many arguments the verb licenses and how they participate in the event. These dimensions interact but are not interchangeable.
Split Ergativity and Relative Clauses
Relative clauses can activate special alignment rules or restrict which argument can be relativized directly. A translator who assumes the main-clause pattern may misinterpret a relative marker or promote the wrong participant. Before parsing a difficult relative clause, determine whether the language’s split includes subordinate or nominalized environments.
English relative clauses then reconstruct the intended role with “who,” “that,” “which” or a preposition. The target relative pronoun does not need to mirror the source case, but it must preserve whether the relativized participant is actor, patient or another role.
Split Ergativity and Coordination
Coordination can expose hidden alignment assumptions. Two clauses with different aspect values may coordinate the same participant under different case patterns. A translator may be tempted to harmonize them and accidentally change one clause. Treat each conjunct according to its own grammar before merging the English sentence.
If the target permits ellipsis of a repeated subject, verify that the omitted subject in English still corresponds to the same participant across both source clauses. Alignment alternation should not be mistaken for reference shift.
Narrative Foregrounding and Backgrounding
Aspect-conditioned splits often align with narrative structure. Perfective forms advance foregrounded events; imperfective forms describe background states, ongoing actions or circumstances. If alignment changes with aspect, narrative grammar and case morphology shift together.
This can actually help the translator. Once the story layer is recognized, the expected alignment becomes easier to predict. Conversely, an unexpected case pattern can signal that the source has shifted from background description to a bounded event or from one discourse mode to another.
Machine Translation Evaluation for Split Alignment
General accuracy scores rarely reveal whether a model handles split ergativity correctly. Build targeted evaluation pairs that keep the same lexical participants but change aspect, person or clause type. The target role assignment should remain stable while the source morphology changes according to the split.
A useful red-team set includes perfective versus imperfective, lexical noun versus first-person pronoun, main versus subordinate clause, positive versus negative clause where relevant, and transitive versus antipassive constructions. If the model flips actor and patient under any of these alternations, it is using surface shortcuts rather than robust grammar.
Human reviewers should see these contrast sets too. They train attention on exactly the environments where a familiar parsing strategy fails.
Localization and String-Level Risk
Short localization strings can be dangerous because the aspect or clause context that determines alignment may live outside the string. A button label, notification fragment or templated sentence may contain a case-marked noun whose form depends on variables inserted at runtime.
Localization architecture should expose the tense/aspect environment and argument roles to translators. Placeholders should be documented as A, P or S rather than generic “name1” and “name2” where grammatical case depends on role.
Reviewer Escalation Rules
Not every unusual case pattern requires a linguist. Establish escalation rules. If the clause matches a documented matrix row, translate normally. Escalate when case and agreement conflict with the known pattern, when a clause type is absent from the project grammar, when person-based behavior is unclear, or when a role decision affects high-stakes meaning.
This keeps expert attention focused where uncertainty is genuine rather than turning every ergative clause into a research project.
Project Governance: One Alignment Matrix, Many Translators
Shared alignment guidance should be version-controlled. Every rule needs an example, a source grammar reference or reviewer decision, and a scope note. If a new dialect or genre enters the project, do not silently reuse the old matrix. Validate the split conditions again.
A mature project eventually develops predictable checkpoints: aspect first, person second, clause type third, case fourth, agreement fifth. Translators move through them quickly because the order is stable. That procedural stability is what turns a typological complication into routine professional practice.
Extended Practice Lab
Exercise one: keep the nouns constant and alternate perfective/imperfective morphology. Label S/A/P and translate. Exercise two: keep aspect constant and replace a lexical noun with first person. Record how case changes. Exercise three: embed the clause under a complement verb and compare the alignment. Exercise four: convert the transitive clause to antipassive if the language permits it and identify which change belongs to valency rather than alignment.
The learner should explain every target role without referring to English word order. If the explanation depends on “the first noun looked like the subject,” the exercise is not yet complete.
Final Extended Checklist
- Has the reviewer identified the split condition before interpreting case?
- Are case alignment and agreement alignment recorded separately?
- Are pronouns checked against their own paradigm?
- Does the clause type activate a special pattern?
- Has aspect been preserved independently of role marking?
- Could an antipassive explain the surface pattern better than an alignment switch?
- Does coordination hide a reference shift?
- Have short localization strings preserved enough grammatical context?
- Has the project matrix been validated for the correct variety?
- Do all high-stakes clauses preserve the same actor–patient relation as the source?
Split ergativity does not make translation unstable. Unrecorded split conditions make translation unstable. Once the conditions are explicit, the system becomes another controlled mapping problem.