@cite{beavers-udayana-2022} Middle voice as generalized argument suppression #
Beavers, John and I Nyoman Udayana. 2022. Middle voice as generalized argument suppression: The case from Indonesian. Natural Language & Linguistic Theory 41:51–102.
Core claim #
Indonesian ber- middles — reflexive, dispositional/passive, anticausative, and incorporation — all derive from one underspecified argument-suppression operation. The surface variation comes from independent argument realization strategies (functional application vs. noun incorporation) and lexical-semantic/pragmatic constraints on the suppressed variable.
Formal analysis #
ber- suppresses one direct argument of a dyadic VP by leaving it as an open variable, while preserving truth conditions:
⟦ber-⟧ = λP_{<e,α>}[P(z̲)] (the paper's (43))
The suppressed argument z is interpreted via lexical and pragmatic conventions: for naturally reflexive roots the convention is coreferent interpretation; for other roots the convention is disjoint interpretation. An alternative formulation existentially binds z with a contextual constraint function f:
⟦ber-⟧ = λP_{<e,α>} ∃z[f_{C,P}(z) ∧ P(z)] (the paper's (47))
The authors maintain (43) for consistency with @cite{beavers-zubair-2013}.
2×2 typology (the paper's (31)) #
The four middle types are classified along two independent dimensions: (a) whether the suppressed variable is interpreted as coreferent or disjoint from the surface subject, and (b) whether the base object is realized as a full DP (functional application) or an incorporated NP:
| non-reflexive (disjoint) | reflexive (coreferent) | |
|---|---|---|
| No incorporation | dispositional/passive middle | inherent reflexive |
| Incorporation | ber-V=lexical NP | ber-V=diri |
Key generalizations (the paper's (32)) #
(a) The base V is transitive (dyadic), taking subject and object DPs in the active meN- form. (b) ber- forms always take a subject DP but never a canonical object DP. (c) The underlying object is always expressed lexically (as an NP or DP). (d) The base subject can be the surface subject if and only if the object is an incorporated NP.
Anticausatives (§5) #
Anticausatives (ter- forms) are outside the core 2×2 typology but derive from the same mechanism applied to causer-unspecified verb roots. They have a unique diagnostic profile: no oleh, no rationale clauses, but YES dengan sendiri=nya.
Cross-linguistic predictions #
The core of argument suppression may underlie middles in other languages, but language-specific argument realization strategies (and their absence) determine which middle types surface.
Dispositional/passive middle: Mobil itu ber-jual dengan mudah. 'The car sells easily.' (the paper's (2b)) / Mobil itu ber-jual kemarin. 'The car sold yesterday.' (the paper's (7))
Surface subject = base patient; agent suppressed with disjoint interpretation. Whether the reading is dispositional (generic) or passive (episodic) depends on temporal/modal context, not on the suppression operation itself.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Natural reflexive: Ali ber-dandan. 'Ali dressed.' (the paper's (2a))
Surface subject = base patient (formally); agent suppressed with coreferent interpretation. The root class (body care/grooming verbs) conventionally expects self-action, triggering coreferent reading.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Incorporation middle (non-reflexive): Orang itu ber-cuci=mata. 'The man washed his eyes.' (the paper's (18))
Surface subject = base agent; patient = incorporated NP. The agent is the surface subject because incorporation satisfies the object's structural role, leaving the agent as the sole DP argument. The suppressed variable (patient) receives a disjoint interpretation.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Incorporation middle (reflexive): Orang itu ber-jual=diri. 'The man sold himself.' (the paper's (26b))
Incorporated diri 'self' triggers coreferent interpretation of the suppressed variable.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Which argument surfaces as subject depends on object realization.
(32d): The base subject can be the surface subject IFF the object is an incorporated NP. When the object is a full DP, the patient surfaces as subject (agent suppressed). When the object is an incorporated NP, the agent surfaces as subject (patient incorporated).
This is now also derived compositionally from suppressArg in § 5
above: the Montague type of the VP after FA vs. incorporation
determines which argument remains for the surface subject.
Instances For
Incorporation middles have the agent as surface subject.
Non-incorporation middles have the patient as surface subject.
Agent surfacing depends ONLY on object realization, not on suppressed variable interpretation. This captures the paper's key insight that the same ber- operation yields different surface argument structures through independent object realization.
Diagnostics that distinguish ber- middles from di- passives and meN- actives.
licensesOleh: can the agent be expressed via oleh 'by' PP?licensesRationale: can a rationale clause (purpose untuk PRO V) be controlled by the implicit agent?licensesDenganSendiriNya: is dengan sendiri=nya 'by itself' compatible?
These are the key tests from §§2.1–2.4 and §5 of the paper.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- One or more equations did not get rendered due to their size.
- Phenomena.ArgumentStructure.Studies.BeaversUdayana2022.instBEqMiddleDiagnostics.beq x✝¹ x✝ = false
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
ber- dispositional/passive middles: no oleh, no rationale clause, no dengan sendiri=nya (the paper's (11), (13), (10c)).
The suppressed agent is syntactically inaccessible: it cannot be expressed as a by-phrase, cannot control rationale PRO, and since it is not the surface subject, dengan sendiri=nya is ruled out.
Equations
Instances For
di- passives: oleh OK, rationale clause OK (controlled by implicit causer), dengan sendiri=nya blocked.
The implicit argument in di- passives is syntactically stronger (a weak pronoun e[-D]) than in ber- middles, licensing oleh and rationale control.
Equations
Instances For
ber- reflexives: no oleh, rationale clause OK (controlled by surface subject = agent), dengan sendiri=nya OK (the paper's (17)).
Equations
Instances For
ber- incorporation middles: no oleh, rationale clause OK (surface subject IS the agent), dengan sendiri=nya OK (the paper's (23)).
Equations
Instances For
Anticausative (ter-) middles: no oleh, no rationale clause, YES dengan sendiri=nya (the paper's §5, examples (67a–c)).
This is a UNIQUE diagnostic profile: anticausatives differ from dispositional/passive middles (which block dengan sendiri=nya) and from reflexives (which license rationale clauses). The paper argues dengan sendiri=nya is licensed because anticausatives have causative variants, creating a paradigmatic expectation of a causer — the modifier introduces the information that this causer is not the surface subject.
Equations
Instances For
The three-way diagnostic contrast: di- passives allow oleh; ber- dispositionals allow none; ber- reflexives/incorporation allow rationale + dengan sendiri=nya but not oleh.
dengan sendiri=nya is licensed when the event has a causer (whether the surface subject itself or a paradigmatic expectation from the verb's causative variant). It is blocked only for dispositional/passive middles and di- passives, where the implicit participant is not accessible as an effector.
The licensing condition is NOT simply "agent surfaces as subject" — anticausatives license dengan sendiri=nya even though the agent does not surface. Rather, the condition involves the availability of a causer interpretation (from the causative paradigm).
Four of the five profiles are pairwise distinct. The exception: reflexive and incorporation middles share the same diagnostic fingerprint ({¬oleh, rationale, sendirinya}). The paper distinguishes them by ARGUMENT STRUCTURE (which DP surfaces as subject), not by these syntactic diagnostics.
Reflexive and incorporation middles are diagnostically IDENTICAL — they are distinguished by argument structure (agent vs. patient as surface subject), not by oleh/rationale/sendirinya tests.
Grounding the 2×2 typology in Montague composition #
The four middle types arise from applying ONE operation (`suppressArg`)
to VPs of different Montague types. The VP type is determined by
independent argument realization (FA vs. incorporation), not by ber-.
We prove this for an arbitrary model `m` and transitive verb `V`.
Dispositional middle derivation (the paper's (54)):
FA saturates the object → VP has type e ⇒ t → ber- suppresses
the remaining (agent) argument → result is type t.
The patient (FA-applied argument) is the surface subject.
Incorporation middle derivation (the paper's (51)):
Incorporation narrows but preserves the object → VP has type
e ⇒ e ⇒ t → ber- suppresses the first (object) argument →
result is type e ⇒ t → agent fills the remaining position.
The agent is the surface subject.
Active voice derivation (contrast): Active (meN-) applies the identity, preserving both arguments. The subject is the agent; the object is the patient.
The argument structure difference between dispositional and
incorporation middles is a TYPE difference, not an operation
difference. In both cases ber- = suppressArg z.
Agent surfaces as subject iff the VP retains both arguments (incorporation case). This is the paper's (32d), now derived from Montague composition rather than stipulated.
ber-'s underspecification means it is compatible with the Minimalist Voice parameters of EVERY other Indonesian voice. This is the formal content of "generalized argument suppression": the morpheme doesn't commit to ±D or ±λx.
meN- and di- are NOT compatible with each other — they differ on ±λx. This is why they are distinct voices, while ber- can behave like either.
linglib bridge (not formalized in the paper):
Dispositional middles are restricted to change-of-state verbs. The paper notes (§2.1) that dispositional ber- forms "are only possible with verbs that describe change-of-state or at least some degree of affectedness."
We formalize this by connecting @cite{beavers-2010}'s affectedness hierarchy to the middle typology: verbs licensing dispositional middles must have affectedness degree ≥ nonquantized.
This independently connects to @cite{levin-1993}'s prediction
that the middle alternation requires changeOfState (see
predictedAlternation in Core/RootDimensions.lean) — the same
verb class restriction viewed through different theoretical lenses.
Equations
Instances For
Change-of-state verbs (quantized/nonquantized) license dispositionals.
Non-CoS verbs (potential/unspecified) do NOT license dispositionals.
Bridge: Levin's middle alternation diagnostic and Beavers 2010's
affectedness constraint make the same prediction. Verbs that
participate in the middle alternation (i.e., have changeOfState)
are exactly those whose patients are affected enough (≥ nonquantized)
to license dispositional middles.
The paper predicts that which middle types surface in a language depends on its argument realization inventory. Languages lacking incorporation can only have the no-incorporation row of the typology (reflexives and dispositional/passive middles).
This is a testable prediction: if a language has incorporation, it should (ceteris paribus) have incorporation middles. If it lacks incorporation, it should lack them.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Indonesian has incorporation → all four types predicted.
A language without incorporation → incorporation middles blocked.
A language without incorporation → reflexive + dispositional still OK. This is the predicted pattern for Spanish se middles.