Namespaces

Introduction

An important idea that had been worked on was the possibility of establishing a frame lattice that would allow the definition of top frames. The network would be structured around these top frames. In the Lutma implementation, a lattice was also defined with the aim of facilitating the creation of new frames.

The problem with defining top frames is the difficulty of establishing inheritance or perspective relationships in many generic frames. Thus, the idea implemented from Webtool 4.0 was the definition of namespaces for frames. A namespace is basically a set of frames that share some basic semantics.

The frames sharing the same namespace do not need to have frame-to-frame relations to each other.

In the current version, the following namespaces are implemented:

Namespace	Description	Typical structure
@situation	This is a "catch-all" namespace for frames that do not fit into the other namespaces. Used for scenario frames and image-schema frames.	Variable structure
@eventive	Eventive frames profile a bare occurrence with no caused change: no agent, no cause acting on a patient, and no experiencer. This includes natural phenomena (rain, snow), spontaneous processes (physical, biological, social), disease as process, and existence. A natural force that produces a result is a non-intentional Cause and belongs to @causative, not here.	FE core ("event") incorporated by LU.
@causative	Eventive frames that have a cause (non intentional) or an agent (intentional, volitional, goal-directed actions).	FE core "agent" or "cause" (in excludes relation)
@inchoative	Eventive frames that exhibit inchoative alternation.	FE core indicating the affected element.
@action	Eventive frames that profiles the activity performed by agents.	FE core "agent"
@stative	Frames that represent states (no change implied) about entities.	FE core for entity, FE core-unexpressed for state.
@psychological	Frames foreground a sentient participant — an experiencer or a cognizer — undergoing a mental, perceptual, somatic, or attitudinal event.	FE core for entity associated to ("cognizer", experiencer", "perceiver", etc)
@transition	Eventive frames that represent changes in a situation (state, attribute, category, etc.) of an entity.	FE core for entity under transition; FEs for initial/final state or condition.
@attribute	Frames that represent attributes or attribute values.	FE core for attribute, FE core-unexpressed for attribute.
@entity	Frames that represent entities.	FE core for entity, incorporated by LU.
@relational	Frames that represent relations between entities or events.	FEs core for related concepts and FE core joining the concepts.
@pragmatic	Pragmatic frames.	Variable structure.

Purpose and orientation

This document specifies the semantic infrastructure that lets FrameNet Brasil (FN-Br) describe and evaluate the namespace classification of its frames. It has two practical aims:

To audit existing Lexical Units (LUs) and frames, surfacing assignments that are inconsistent with the namespace they sit in.
To prompt deeper analysis when a new frame is created — not by hard constraint, but by giving the frame creator a reference structure to relate their frame to.

The whole apparatus is descriptive, not prescriptive. As in FrameNet generally — and as in the DUL ontology this model borrows from — the goal is to encode the modeller's analysis and make inconsistencies visible, never to forbid a construal. Language is a complex system; full coverage is not the target. Useful, honest progress is.

The model has two layers, deliberately operating at different granularities:

Layer	Attaches to	Granularity	Job
Ontological type (see Ontological Types)	LU (concept identity)	Coarse — 5 classes	Catch cross-class errors (an attribute LU in an eventive frame)
Meta-frame (§2–§4 of Eventive Namespaces)	Frame (via the `Meta` relation, on its FEs)	Fine — per-namespace role signatures	Prompt and evaluate within-class perspectivization

The two layers are complementary and must not be conflated. The coarse layer runs automatically and tracks what a concept is. The fine layer is a relational reference structure that tracks which view a frame takes — and is populated by human judgement at frame-creation or audit time.

The organizing principle: Event vs. Situation (DUL)

The single idea that holds the whole model together is the DUL distinction between an Event and a Situation.

An Event is something that occurs, with one stable identity. The rock erosion in the Sinni valley is one event.
A Situation is a view of that event, consistent with a Description (a theory under which the view is taken). ErosionAsAccomplishment and ErosionAsTransition are two Situations over the same Event; neither changes the Event's identity.

This maps directly onto FN-Br's architecture:

DUL notion	FN-Br correlate
Event (stable identity)	the LU / concept and its ontological type
Description (theory licensing a view)	the frame
Situation (the view taken)	the namespace the frame belongs to

Two consequences follow, and they justify the most important design choices in this document:

The ontological type tracks Event identity; the namespace tracks the Situation/view. This is why destruição evoking a transition-frame is still destruição.event and never destruição.transition: transition is the view a particular frame takes; event is the concept's identity. One ontological type therefore maps to many compatible namespaces (see Ontological Types).
A single event can legitimately be viewed under more than one namespace. O gelo derreteu may be read as @inchoative (the resulting liquid state is profiled) or @eventive (the melting process is profiled). These are two Situations over one Event. The model must therefore expect such pairs and never treat them as errors (§4 of Eventive Namespaces). This is the FrameNet's native notion of perspective applied to the LU level.

This is also the realist/constructivist truce DUL is built for: the Event is not changed by how we describe it, while each description still earns its own identity. The model commits to neither metaphysics; it just records both the concept and the view.

Why a participant-based classification

Events can be classified by aspect (stative, accomplishment, achievement, punctual…), by agentivity (intentional, natural…), or by typical participants (human, physical, abstract…). This model classifies primarily by participants (via qualia roles, §1 of Eventive Namespaces), because participant structure is the most stable basis for the meta-frames and the least dependent on the modeller's momentary construal. Aspectual and agentive distinctions are recorded where they discriminate (as features on meta-FEs, §3 and §5 of Eventive Namespaces) but are not the primary axis.

The reason is the identity problem, which the Event/Situation distinction resolves. A single event admits several incompatible-looking views at once; classifying by the view would fracture one event into many identities. DUL avoids this by assigning the view its identity as a Situation, leaving the Event intact. The canonical illustration is rock erosion in the Sinni valley, which supports three kinds of alternative view:

View type	The alternative readings	Effect on Event identity
Aspectual	accomplishment (process toward a state) · achievement (the resulting state) · punctual (time interval collapsed to a point) · transition (state A → state B)	None. Each is a Situation (ErosionAsAccomplishment, ErosionAsTransition…) consistent with a different Description; the Event is unchanged.
Intentionality	natural forces alone vs. some intentional effort behind them	None — unless an Agent's actions are taken as parts of the event, which does change identity (a part is added). Whether intentionality is an identity criterion is the designer's choice.
Participant	which basic participants count (snow, slope, wind… vs. also water, human agents…)	Changes identity — the event depends on its participants, so adding or removing them yields a different event.

The contrast that matters for this model: aspectual and intentionality views are Situations over one Event (so a single LU/concept may be viewed under several namespaces — §4 of Eventive Namespaces), whereas participant choice individuates the Event itself (so it is the appropriate primary classifying axis, and the meta-frames are defined by participant/qualia structure — §1 and §3 of Eventive Namespaces). The common alternative — classifying events by aspect (cf. Levin's aspectual classes, or DOLCE Full's WonderWeb D18 axiomatization) — is rejected here precisely because it would give one event several identities on the basis of the modeller's attitude alone.

Stance

The realist position (events are not changed by how we describe them) and the constructivist position (no event can be modelled without the theoretical burden of how it is observed) are both, taken to extremes, only partly relevant to a working linguistic resource. Following DUL, the model admits Events, Situations, and Descriptions together, so the modeller's analysis is encoded regardless of metaphysical commitment. This is what makes the model descriptive, not prescriptive: it records the chosen view rather than adjudicating the correct one.