Eventive Namespaces

Qualia roles as the shared FE vocabulary

FrameNet assigns frame-specific Frame Elements (FEs) by design — Cook in Apply_heat is not literally the same FE as Agent in Killing. This specificity is a strength and the model does not seek to abolish it. But it is plainly true that many FEs share a high-level semantics: the Cook, the Killer, the Builder are all, at a higher level, agentive instigators.

The model exploits this shared high-level semantics using qualia roles, adapted (roughly) from the Qualia Structure of the Generative Lexicon — a theory already in use in FN-Br for LU description. Qualia roles are not new FEs and do not replace frame-specific FEs. They are a small classifying vocabulary used to (a) group FEs across frames and (b) define the core meta-FEs of each meta-frame (§2).

The four qualia roles:

Qualia role	Intuition	Typical FE members
AGENTIVE	what brings the event about	Agent, Cause
TELIC	the endpoint / purpose / affected participant	Patient, Experiencer, Goal
CONSTITUTIVE	what the event is made of / generic participants	Theme, Instrument, Source, Path, Part, Material
FORMAL	the condition that holds / classifies	State, Attribute, Category, Condition

The generic event structure, showing all four roles and their relation to the Event, is the backdrop against which each meta-frame (§3) profiles a subset:

graph TB
    subgraph AQ["AGENTIVE ROLE"]
        Agent[Agent<br/>+intentional +volitional]
        Cause[Cause<br/>+causation -intentional]
    end

    subgraph TQ["TELIC ROLE"]
        Patient[Patient<br/>+affected +endpoint +change]
        Experiencer[Experiencer<br/>+perceiver +psychological]
        Goal[Goal<br/>+recipient +destination]
    end

    subgraph CQ["CONSTITUTIVE ROLE"]
        Theme[Theme<br/>+participant -specific_role]
        Instrument[Instrument<br/>+means +used_by_agent]
        Source[Source<br/>+origin]
        Part[Part<br/>+component]
        Material[Material<br/>+substance]
    end

    subgraph FQ["FORMAL ROLE"]
        Condition[Condition]
        State[State]
        Attribute[Attribute]
        Category[Category]
    end

    Condition -->|predicates| Theme
    Theme -->|participates_in| Event
    Agent -->|intentionally_causes| Event
    Cause -->|causes| Event

    State -->|is_a| Condition
    Attribute -->|is_a| Condition
    Category -->|is_a| Condition

    Instrument -->|is_a| Theme
    Source -->|is_a| Theme
    Part -->|is_a| Theme
    Material -->|is_a| Theme

    Event -->|affects| Patient
    Event -->|stimulus_for| Experiencer
    Event -->|directed_to| Goal
    Event -->|affects_reflexively| Agent
    Event -->|mutual| Agent

    Patient -->|from| Condition
    Patient -->|to| Condition
    Experiencer -->|to| Condition

FrameNet names the roles by frame context; the labels above are the high-level classifying vocabulary, not FE names. Each meta-frame in §3 is a profiling of this structure — a selection of which roles are core to a given namespace.

Two cautions, carried over from prior analysis:

A frame-specific FE may not fit exactly one qualia role. Qualia roles are a grouping criterion, not a partition. Where an FE spans roles (the notorious Theme; see below), the model records the profiled role and notes the others.
Qualia roles answer "what role does this FE play in the event". They are distinct from semantic types (§5), which answer "what kind of thing fills this FE". The two axes are orthogonal and the model uses both.

Note on cross-role FEs

Patient, Experiencer, and Theme are intrinsically multi-role and the model treats this as expected, not as a defect:

Patient combines TELIC (affected endpoint), CONSTITUTIVE (participant), and FORMAL (the entity that ends in a state). Classified as TELIC (its profiled aspect), with the others noted.
Experiencer combines TELIC (endpoint of a stimulus), AGENTIVE (when under control, e.g. olhar), and FORMAL (bearer of a mental state, e.g. saber). Classified as TELIC, with the agentive variant flagged (§3.5, Psychological).
Theme is the most generic CONSTITUTIVE participant and is deliberately underspecified for affectedness. It is the default role for a participant with no sharper characterization.

Layer 2 — Meta-frames

Concept

For each namespace there is one meta-frame: an abstract frame whose core FEs ("meta-FEs") are the qualia-role participants that define the namespace. Ordinary frames relate to their namespace's meta-frame through the dedicated Meta relation (§2.2), mapping their own frame-specific core FEs onto the meta-FEs.

The meta-frame is a reference structure, not a constraint:

Relating a frame to its meta-frame is the act in which the creator states which view the frame takes — choosing the Causative meta-frame over the Inchoative one is the disambiguation, performed by a human at the right moment.
Unmapped frame-specific FEs are expected and fine. The meta-frame captures only the shared high-level core, never the full FE roster.
The intended effect is to make the creator think structurally about the new frame. Its value is as guidance and post-hoc evaluation, held to the guidance standard: success is "obvious mismatches drop and creators analyse more deeply," not "misclassification is impossible."

This is squarely within the FrameNet tradition: relating specific frames to abstract reference frames is what FrameNet already does. The novelty is only that the reference frames are organised per namespace and connected by a new, dedicated relation so their semantics stays clean.

The `Meta` relation

Meta is a new FrameNet-Brasil frame-to-frame relation, defined here from scratch so its semantics is not conflated with Inheritance, Using, or Perspective_on.

Definition. Meta(F, M) asserts that ordinary frame F realises the core meta-FEs of meta-frame M, where M is the meta-frame of F's namespace. The relation carries, as its payload, a set of FE-FE mappings from core FEs of F to meta-FEs of M.

What Meta does not assert (the reason it is a separate relation):

It is not Inheritance: F is not an is-a subtype of M, and F does not inherit M's full FE set, semantic types, or relations. Only the explicitly mapped core FEs are related.
It is not Using: F does not presuppose M as background.
It is not Perspective_on: M is not a neutral frame of which F is one perspective. (Perspective among views is carried by the meta-meta relations, §2.4, not by the frame↔meta-frame link.)

Properties:

Property	Value
Direction	from meta-frame `M` to ordinary frame `F`
Cardinality	each `F` relates to one meta-frame (its namespace's)
Payload	FE→meta-FE mappings on core FEs (peripheral FEs optional)
Coverage	partial by design — unmapped FEs are normal
Force	descriptive — no creation is blocked by absence or mismatch

Mapping FEs to meta-FEs

The FE-FE mappings carried by Meta are the mechanism by which a frame's specific FEs are connected to the shared qualia vocabulary. Example (illustrative):

Meta(Causar_mudança_de_temperatura → Causative_meta) :
    Agente              ↦ Agent         (meta-FE)
    Item                ↦ Patient       (meta-FE)
    Temperatura_alvo    ↦ Result        (meta-FE)
    Temperatura_inicial ↦ (unmapped — peripheral)

The presence or absence of a mapping is itself informative. A frame that has no FE mapping to the Causative meta-frame's Result meta-FE has thereby signalled that it is not result-profiling — i.e. Action-like rather than Causative-like. The structural relation replaces an explicit feature. This is why the model needs no "±result" feature: the meta-FE is either realised or it is not.

Meta-meta relations (`Meta` among meta-frames)

The meta-frames are themselves related to each other using the same Meta relation, applied at the meta level (Meta(M₁, M₂)). This encodes, once and at the top, the perspective and alternation structure that otherwise would have to be restated for every specific frame pair. Specific frames inherit the relationship through their own Meta link to a meta-frame.

The meta-meta layer is where the legitimate cross-namespace relationships live (the "perspective pairs" of §4). Stating them once here is the most FrameNet-native way to record that, e.g., Causative and Inchoative are two views of one event. See §4 for the specific pairs.

Note on overloading. A single relation Meta does two jobs: frame→ meta-frame (§2.2) and meta-frame→meta-frame (§2.4). The level of the relata disambiguates the use; the semantics ("realises / relates-to the core structure of") is uniform. If implementation reveals these need to diverge, split them then — but the uniform treatment is the simpler starting point.

The meta-frames and their core meta-FEs

Each meta-frame below lists its core meta-FEs, the qualia role of each, and any semantic-type expectation on the filler (§5). Roles come from the meta-FE. The ±volitional feature comes for free from the Sentient semantic type on the agentive meta-FE, not from a bespoke feature — but note it splits Agent from Cause within Causative; it does not separate Causative from Eventive (a non-Sentient natural force is still a Cause). What separates the namespaces is the presence/absence of meta-FEs: a result-bearing meta-FE marks the result-profiling namespaces, and a Patient+Result caused-change core marks Causative/Inchoative off from residual Eventive.

Causative_meta — namespace `@causative`

Meta-FE	Qualia role	Semantic type	Notes
Agent	AGENTIVE	`Sentient`	the intentional, volitional instigator
Cause	AGENTIVE	non-`Sentient` (a natural/physical force, or a `State_of_affairs`)	the non-intentional instigator — a natural force (o vento, o calor) or an abstract force/event (disease, policy)
Patient	TELIC	—	the entity that changes
Result	FORMAL	—	the resultant state; its presence is the causative signature

Profiles the AGENTIVE role and a Result. The agentive role is realised as either an Agent (Sentient, intentional) or a Cause (non-Sentient: a natural/physical force or an abstract State_of_affairs, non-intentional). The Sentient type does the Agent-vs-Cause split within this namespace — it does not gate Causative vs. Eventive. Distinguished from Action by the presence of Result + Patient; from Eventive by profiling a caused change at all — a Cause/Agent acting on a Patient that reaches a Result. A natural force is simply a non-Sentient Cause, so o vento quebrou a janela is Causative, not Eventive.

graph LR
    Agent["Agent<br/><i>AGENTIVE</i><br/>Sentient"]
    Cause["Cause<br/><i>AGENTIVE</i><br/>non-Sentient: force / State_of_affairs"]
    Event(("Event"))
    Patient["Patient<br/><i>TELIC</i>"]
    Result["Result<br/><i>FORMAL</i>"]

    Agent -->|intentionally_causes| Event
    Cause -->|causes| Event
    Event -->|affects| Patient
    Patient -->|reaches| Result

    classDef core fill:#8B0000,stroke:#8B0000,color:#fff
    class Agent,Cause,Patient,Result core

Action_meta — namespace `@action`

Meta-FE	Qualia role	Semantic type	Notes
Agent	AGENTIVE	`Sentient`	volitional performer
Activity	CONSTITUTIVE	—	the activity itself

No Result meta-FE and no Patient meta-FE. Their absence is the signal that distinguishes Action from Causative. Any object present maps to a CONSTITUTIVE Instrument/Theme (used, not changed), never to Patient.

graph LR
    Agent["Agent<br/><i>AGENTIVE</i><br/>Sentient"]
    Event(("Event"))
    Activity["Activity<br/><i>CONSTITUTIVE</i>"]

    Agent -->|performs| Event
    Event -->|consists_of| Activity

    classDef core fill:#8B0000,stroke:#8B0000,color:#fff
    class Agent,Activity core

Inchoative_meta — namespace `@inchoative`

Meta-FE	Qualia role	Semantic type	Notes
Patient	TELIC	—	affected entity, in subject position
Result	FORMAL	—	the achieved state (built-in endpoint)

No Agent or Cause meta-FE. Same Patient+Result core as Causative, minus the agentive role — this shared core is exactly the causative/inchoative alternation, recorded at the meta level (§4).

graph LR
    Event(("Event"))
    Patient["Patient<br/><i>TELIC</i><br/>subject position"]
    Result["Result<br/><i>FORMAL</i><br/>achieved endpoint"]

    Event -->|affects| Patient
    Patient -->|reaches| Result

    classDef core fill:#8B0000,stroke:#8B0000,color:#fff
    class Patient,Result core

Eventive_meta — namespace `@eventive` (residual)

Meta-FE	Qualia role	Semantic type	Notes
Event	(the occurrence)	—	the central happening
Theme	CONSTITUTIVE	—	optional, generic, underspecified participant

The residual meta-frame: a frame relates here when it profiles a dynamic occurrence but maps cleanly to no more specialised meta-frame. It covers occurrences with no caused change profiled — natural phenomena (choveu, o vento soprou), spontaneous processes (a fruta amadureceu), disease as process or progression, and existence (surgir, desaparecer). Choose Eventive only when nothing better fits.

A natural force that produces a result is not Eventive. Membership turns on the absence of a caused change, never on the nature of the causer. A natural/physical force acting on a Patient to bring about a Result is a non-Sentient Cause, so o vento quebrou a janela relates to Causative_meta (§3.1), not here — exactly as João quebrou a janela does. The Sentient vs. non-Sentient contrast only splits Agent from Cause inside Causative; it does not gate Causative vs. Eventive. What lands in Eventive is the intransitive counterpart with no causer profiled (o vento soprou, a janela quebrou-as-occurrence).

graph LR
    Event(("Event"))
    Theme["Theme<br/><i>CONSTITUTIVE</i><br/>optional, generic"]

    Event -.->|optional| Theme

    classDef core fill:#8B0000,stroke:#8B0000,color:#fff
    classDef opt fill:#fff,stroke:#8B0000,stroke-dasharray:4 3,color:#8B0000
    class Event core
    class Theme opt

Multiple participants in Eventive. Eventive's core is deliberately thin — Event plus an optional, generic Theme. Real eventive frames often have more than one participant, which raises two recurring questions; the answers follow from §1 and §2.2:

Do not map two distinct participants to the same Theme. A meta-FE is one role slot; mapping two FEs onto Theme asserts they play the same generic role and erases the distinction the qualia vocabulary exists to record (§2.3). Reserve many-to-one mapping for genuinely co-typed participants (a reciprocal pair, X e Y colidiram).
Record a sharper participant on a sharper CONSTITUTIVE sub-role, not on a second Theme. In §1, Instrument, Source, Path, Part, and Material are all is_a Theme — refinements of the generic participant. Map a participant to the sub-role that fits; map to bare Theme only when it has no sharper characterization.
Meta coverage is partial by design (§2.2). Genuinely peripheral participants may be left unmapped; the absence is itself informative.

A participant that is clearly affected and reaches a result state is a smell: that Patient + Result signature is the Causative/Inchoative signature, not Eventive's. Such a frame is a candidate for the Inchoative ↔ Eventive perspective pair (§4), or — if a causer is profiled in subject position — for Causative outright. (Recall that, under the corrected boundary above, a natural-force causer makes the frame Causative, not Eventive.)

Psychological_meta — namespace `@psychological`

Meta-FE	Qualia role	Semantic type	Notes
Experiencer	TELIC (may be AGENTIVE when controlled)	`Sentient`	the sentient participant for perception/sensation/emotion (undergoes a stimulus); core, Excludes `Cognizer`
Cognizer	TELIC (may be AGENTIVE when controlled)	`Sentient`	the sentient participant for cognition/belief/attitude (entertains a content); core, Excludes `Experiencer`
Stimulus	CONSTITUTIVE	—	optional; the trigger/object of a perception or emotion (entity or event)
Content	CONSTITUTIVE	`State_of_affairs`	optional; the object of a cognitive or propositional-attitude member (a proposition/event), as distinct from a `Stimulus` trigger
Mental_condition	FORMAL	—	the resulting/ongoing mental, perceptual, or affective state

Experiencer ⊕ Cognizer (Excludes). FrameNet keeps Experiencer (undergoes a stimulus/sensation) distinct from Cognizer (entertains a content/proposition). Rather than erase the distinction under a coined cover term, the meta-frame carries both as core meta-FEs in an Excludes relation — the native FrameNet Core-Set pattern. A specific frame's Meta mapping picks up whichever its own FE corresponds to. Likewise, members whose object is a trigger map it to Stimulus; members whose object is a proposition/event (cognition, desiderative attitude — saber, acreditar, querer, esperar) map it to Content (semantic type State_of_affairs).

Membership gate — the domain test (Test 1). The frame must require a sentient participant (Experiencer or Cognizer) undergoing a mental, perceptual, affective, or attitudinal event, with no physical change in the world required. The domain (psychological) is the membership criterion and takes precedence over aspect and over agentivity:

The sentient participant's variable role is recorded on the meta-FE, not used to reclassify: controlled olhar = agentive; uncontrolled ver = telic endpoint; saber = formal state-bearer. None of these leave @psychological.
Stimulus-subject psych frames (O filme emocionou Maria) relate here, not to Causative, because the profiled endpoint is the Experiencer's state, not a physical Result. (These are the most causative-like members and may warrant a meta-meta link to Causative_meta; see §4.)

The Senser gate also decides modal-adjacent frames (see §3.8): a frame predicating certainty of a thinker (Certeza, whose definition requires the #Pensador/Cognizer to be expressed) enters @psychological as a cognition member; a frame predicating possibility/probability of a proposition with no sentient participant does not — it is a modal attribute (§3.8).

graph LR
    Stimulus["Stimulus<br/><i>CONSTITUTIVE</i><br/>optional, trigger"]
    Content["Content<br/><i>CONSTITUTIVE</i><br/>optional, State_of_affairs"]
    Event(("Event"))
    Experiencer["Experiencer<br/><i>TELIC</i> (AGENTIVE if controlled)<br/>Sentient"]
    Cognizer["Cognizer<br/><i>TELIC</i> (AGENTIVE if controlled)<br/>Sentient"]
    Mental["Mental_condition<br/><i>FORMAL</i>"]

    Stimulus -.->|optional| Event
    Content -.->|optional| Event
    Event -->|experienced_by| Experiencer
    Event -->|known_by| Cognizer
    Experiencer ---|Excludes| Cognizer
    Experiencer -->|reaches| Mental
    Cognizer -->|reaches| Mental

    classDef core fill:#8B0000,stroke:#8B0000,color:#fff
    classDef opt fill:#fff,stroke:#8B0000,stroke-dasharray:4 3,color:#8B0000
    class Experiencer,Cognizer,Mental core
    class Stimulus,Content opt

Transition_meta — namespace `@transition`

Meta-FE	Qualia role	Semantic type	Notes
Theme	TELIC	—	the entity that moves/changes
Source	CONSTITUTIVE	—	starting point/state (often backgrounded)
Path	CONSTITUTIVE	—	trajectory
Goal	TELIC	—	destination/target state (often profiled)

Profiles the Source-Path-Goal schema. Distinguished from Inchoative by the presence of Path/Goal (the path is profiled, not merely the achieved state). Note FN-Br Portuguese is verb-framed: path is in the verb root, manner is an adjunct — a manner verb with no Source/Goal mapping is Action, not Transition.

graph LR
    Source["Source<br/><i>CONSTITUTIVE</i><br/>optional"]
    Theme["Theme<br/><i>TELIC</i>"]
    Path["Path<br/><i>CONSTITUTIVE</i>"]
    Goal["Goal<br/><i>TELIC</i><br/>often profiled"]

    Theme -->|moves_from| Source
    Theme -->|along| Path
    Theme -->|moves_to| Goal

    classDef core fill:#8B0000,stroke:#8B0000,color:#fff
    classDef opt fill:#fff,stroke:#8B0000,stroke-dasharray:4 3,color:#8B0000
    class Theme,Path,Goal core
    class Source opt

Stative_meta — namespace `@stative`

Meta-FE	Qualia role	Semantic type	Notes
Entity	CONSTITUTIVE	—	the bearer of the condition
Condition	FORMAL	—	covers State / Attribute / Category sub-roles

No Event meta-FE — nothing happens; a condition holds. The crucial boundary: ficar + predicate and estar + past participle (está quebrado) encode entry into a state and therefore relate to Inchoative_meta, not here. Pure ser/estar property-holding relates here.

graph LR
    Entity["Entity<br/><i>CONSTITUTIVE</i><br/>bearer"]
    Condition["Condition<br/><i>FORMAL</i><br/>State / Attribute / Category"]

    Entity -->|holds| Condition

    classDef core fill:#8B0000,stroke:#8B0000,color:#fff
    class Entity,Condition core

Note the absence of any Event node — the defining contrast with all six dynamic meta-frames above.

Modality is cross-cutting, not a namespace

There is no @modal meta-frame, by deliberate decision. Modality (possibility, probability, necessity, obligation, permission, capability) is a feature that cuts across namespaces, not a namespace of its own. Inspection of the candidate frames shows they do not form one class — they distribute by modal flavor (epistemic / deontic / dynamic) and by the Senser gate (§3.5):

Modal flavor	Senser in core?	Routes to	Example frames
Epistemic — likelihood of a proposition	No (predicated of a `State_of_affairs`)	`@attribute` (modal-attribute sub-kind, see below)	Possibilidade, Probabilidade
Epistemic — a thinker's confidence	Yes (Cognizer required)	`@psychological` (cognition)	Certeza (its definition requires the `#Pensador`)
Deontic — obligation, as a scenario	n/a	`@situation`	Cenário_de_obrigação
Deontic — imposing a duty	Yes (implicit imposer acts)	`@action`	Impor_obrigação
Deontic — obligation holding of a party	No	`@stative`	Ser_obrigado, Ser_obrigatório (profiling alternation)
Dynamic — capability/disposition of an entity	No (predicated of an `Entity`)	`@attribute`	Capacidade_ação, Capacidade_volume

Two consequences:

Deontic modality decomposes by perspective into @situation (the scenario), @action (imposing), and @stative (the obligation holding) — using the same features the event model already uses. A @modal namespace would compete with these and force an arbitrary choice; the perspective decomposition is cleaner and is recorded as a meta-meta relation among those three meta-frames.
The only residue without a pre-existing home is Senser-less epistemic predication of a proposition (Group 1: Possibilidade, Probabilidade). This is a narrow class and is housed in @attribute, distinguished from ordinary entity-attributes by the bearer's semantic type:

@attribute sub-kind Bearer (semantic type) Examples

Entity-attribute Entity inteligência, cor, tamanho, Capacidade_volume

Modal-attribute State_of_affairs Possibilidade, Probabilidade

The semantic-type layer (§5) carries this distinction; no new namespace is needed. Revisit only if the modal-attribute class grows large enough to justify the overhead — current evidence (two frames) says it will not.

`@attribute` sub-kind	Bearer (semantic type)	Examples
Entity-attribute	`Entity`	inteligência, cor, tamanho, Capacidade_volume
Modal-attribute	`State_of_affairs`	Possibilidade, Probabilidade

Worked example — the possibility lemma family splits by the Senser gate. Possibilidade [atributo] / Probabilidade [atributo] (no sentient participant; possibility predicated of a #Evento_possível/#Evento_hipotético) → @attribute, modal-attribute sub-kind. But a frame foregrounding an #Agente who considers which of several possible events will happen has a Cognizer in its core → @psychological (cognition). One lemma family, correctly distributed across two namespaces by the Senser test — not a defect, but the gate doing its job.

Scope note. This document specifies the event-related namespaces (the seven meta-frames above). The remaining namespaces in the master table — @attribute, @entity, @relational, @pragmatic, @situation — are deferred to a later revision. @attribute and @entity already appear in Layer 1 (see Ontological Types) as ontological-type targets, and @attribute is referenced in §3.8 as the home of modal-attributes; their full meta-frames, and those of @relational, @pragmatic, and @situation, are out of scope here. There is deliberately no @modal namespace (§3.8).

Shared-signature / perspective pairs

Some namespaces share a core meta-FE signature. Frames in these pairs are distinguished by profiling, judged by a human, not by FE structure — and the model must therefore never auto-flag a frame merely because it could also fit the partner namespace. These relationships are recorded once, at the meta level, via meta-meta Meta relations (§2.4).

Pair	Shared core	Distinguished by	Recorded as
Causative ↔ Inchoative	`Patient` + `Result`	is the `Agent`/`Cause` profiled? (subject = causer vs. affected entity)	`Meta(Causative_meta, Inchoative_meta)` — the alternation
Inchoative ↔ Eventive	affected `Theme`/`Patient`, no agent	is the resultant state profiled (Inchoative) or the process/occurrence (Eventive)?	`Meta(Inchoative_meta, Eventive_meta)`
Inchoative ↔ Transition	`Theme` undergoes change	is the achieved state profiled (Inchoative) or the path (Transition)?	`Meta(Inchoative_meta, Transition_meta)`
Psychological ↔ Causative	stimulus acts on a participant	is the profiled endpoint a mental state (Psychological) or a physical Result (Causative)?	`Meta(Psychological_meta, Causative_meta)`
Psychological ↔ Inchoative	entry into a (mental) state	is the psychological domain primary (Psychological) or the change of state (Inchoative)?	`Meta(Psychological_meta, Inchoative_meta)`
Action ↔ Transition	a (possibly moving) agent	is manner profiled (Action) or a directed path (Transition)?	`Meta(Action_meta, Transition_meta)`

The governing principle (from the Event/Situation section in Namespaces): each pair is two Situations over one Event. The alternation is real and expected; recording it at the meta level means every specific frame pair inherits the relationship through its own Meta link, and the evaluation tooling treats partner-namespace membership as legitimate, not as an error to flag.

Worked case. derreter: O sol derreteu o gelo relates to Causative_meta (o sol is a non-Sentient Cause, profiled in subject position); O gelo derreteu relates to Inchoative_meta (Patient+Result, no agent). The two are connected by the meta-meta Meta(Causative_meta, Inchoative_meta) alternation. Could O gelo derreteu instead relate to Eventive_meta as derreter.event? Yes — but only as the residual choice, taken when the Inchoative reading (which has a causative counterpart) is judged not to apply. Inchoative is preferred precisely because the causative counterpart exists.

Semantic types (to be confirmed)

Semantic types constrain what kind of thing fills an FE. They are orthogonal to qualia roles (which constrain what role the FE plays). The model needs only a very small set; the role distinctions are carried by the meta-FEs, not by types. FrameNet has a native semantic-type facility and a small ontology; these should be reused or extended rather than duplicated.

Types required by this specification:

Semantic type	Short definition	Used by
`Sentient`	a filler capable of volition, intention, and perception	the agentive split (Agent vs. Cause) in Causative_meta; the obligatory participant in Action_meta and Psychological_meta
`State_of_affairs`	a filler that is itself an event, situation, or proposition (rather than a concrete entity)	abstract/non-intentional Cause in Causative_meta; the `Content` of cognitive/attitudinal members and propositional Stimulus in Psychological_meta; the bearer of modal-attributes in `@attribute` (§3.8)

Only these two are needed to make the model's distinctions work. The ±volitional distinction — obtained simply as Sentient vs. not on the agentive meta-FE — separates @action from @eventive (an action needs a Sentient performer) and Agent from Cause within @causative. It does not separate @causative from @eventive: that boundary is the presence of a caused change (a Cause/Agent + Patient + Result), because a non-Sentient natural force is still a Cause. No bespoke feature is introduced. Add further types only if the data requires; keep the set minimal.

How the two layers work together

	Layer 1 — Ontological type	Layer 2 — Meta-frames
Attaches to	LU (concept identity)	Frame (via `Meta`, on its FEs)
Granularity	Coarse (5 classes)	Fine (per-namespace role signatures)
Validates	Cross-class category errors	Within-class perspectivization
Mode	Automatic, retrospective audit	Human relational mapping; prospective guidance + post-hoc evaluation
Catches `aquecer` causative/inchoative split?	No	Yes
Depends on creator's cooperation?	No	Yes

Layer 1 is the retrospective audit: cheap, automatic, runs over the whole base without anyone's cooperation, and surfaces gross cross-class mistakes.
Layer 2 is the prospective nudge: it asks the frame creator to relate their FEs to a meta-frame, prompting deeper analysis, and lets the modeller evaluate existing classifications empirically by inspecting how cleanly real frames map to the meta-FEs.

Neither layer is a classifier that reproduces a full decision procedure, and neither blocks frame creation. Together they make the namespace classification inspectable — which, for a descriptive resource in the FrameNet tradition, is the appropriate and sufficient goal.

Appendix A — Glossary

LU (Lexical Unit) — a lemma + sense pairing; the unit that evokes a frame.
Ontological type — coarse semantic class of an LU's concept identity (.event/.entity/.state/.attribute/.relation); see Ontological Types.
Namespace — the classification of a frame by the view it takes of an event; the Situation, in DUL terms.
Qualia role — high-level classifying vocabulary for FEs (AGENTIVE/TELIC/CONSTITUTIVE/FORMAL); §1.
Meta-frame — the abstract reference frame of a namespace, whose core meta-FEs are its defining qualia participants; §2–§3.
Meta relation — the dedicated FN-Br frame-to-frame relation linking a frame to its meta-frame, and meta-frames to each other; §2.2, §2.4.
Semantic type — constraint on an FE's filler (Sentient, State_of_affairs); orthogonal to qualia role; §5.
Event / Situation / Description — DUL triple: stable occurrence / a view of it / the theory licensing the view; see the Event/Situation section in Namespaces.

Appendix B — Open points for the operational document

The following are deliberately out of scope here and belong to the future operational document:

Annotator step-by-step for creating a frame and relating it via Meta.
Audit-script logic for Layer 1 (the class-compatibility pass) over legacy data.
Procedure for adjudicating wrong-sense vs. wrong-class flags.
Meta-frames for the deferred namespaces (@attribute, @entity, @relational, @pragmatic, @situation).
Empirical evaluation method: how cleanly real frames map to meta-FEs, and what mapping-failure rates indicate about a namespace's coherence.