FOVEA
Overview
FOVEA is a video annotation system built around persona-based ontology construction, rich temporal and spatial modeling, entity/situation world models, claim extraction, and collaborative annotation with role-based access control. Layers was directly inspired by FOVEA's persona and ontology system.
Type-by-Type Mapping
Ontology System
| FOVEA Type | Layers Equivalent | Notes |
|---|---|---|
Persona | pub.layers.persona (record) | Direct mapping. name, description, domain all transfer. FOVEA's role and informationNeed map to features. ontologyRefs links to specific ontologies this persona uses. |
EntityType | pub.layers.ontology#typeDef with typeKind="entity-type" | name → name; gloss → gloss; constraints → features; wikidataId → knowledgeRefs[].identifier with source="wikidata". |
EventType | pub.layers.ontology#typeDef with typeKind="situation-type" | roles → allowedRoles (array of roleSlot references). parentEventId → parentTypeRef. |
RoleType | pub.layers.ontology#typeDef with typeKind="role-type" | allowedFillerTypes → allowedValues or features. |
EventRole | pub.layers.ontology#roleSlot | roleTypeId → roleSlot reference; optional → required (inverted); minOccurrences/maxOccurrences → features. |
RelationType | pub.layers.ontology#typeDef with typeKind="relation-type" | sourceTypes/targetTypes → features (standardized keys: domain, range); symmetric/transitive/reflexive → features (standardized boolean keys); inverse → features (AT-URI of the inverse relation typeDef). |
OntologyRelation | pub.layers.graph#graphEdge | Instance of a typed relation between ontology-level objects. |
TypeConstraint | pub.layers.ontology#typeDef.allowedValues + features | allowedTypes, requiredProperties, valueRange all expressible via features. |
GlossItem | pub.layers.ontology#typeDef.gloss (string) + knowledgeRefs | FOVEA's rich-text gloss with embedded references to types/objects is represented as a string gloss plus structured knowledgeRefs for cross-references. For richer inline markup, features can store structured gloss data. |
World Model (Domain Objects)
FOVEA separates ontology types (what kinds of things exist) from world instances (specific entities/situations discovered in data). Layers handles world instances differently:
| FOVEA Type | Layers Equivalent | Notes |
|---|---|---|
Entity (world object) | pub.layers.graph#graphNode with nodeType="entity" + pub.layers.annotation#clusterSet | FOVEA's world-level entities are modeled in Layers as: (1) graphNode records representing the entity in the property graph, with label, properties, and knowledgeRefs for grounding to external KBs (Wikidata, etc.); (2) clusterSet records that group all mentions of an entity across annotations; (3) graphEdge records linking mentions to graph nodes via edgeType="grounding" or edgeType="instance-of". |
Event (world object) | pub.layers.graph#graphNode with nodeType="situation" + pub.layers.annotation#clusterSet with kind="situation-coreference" | Same pattern as entities. Situation instances are graphNode records; annotations link to them via graphEdge. |
EntityTypeAssignment | pub.layers.annotation#annotation.ontologyTypeRef | Persona-specific type assignments are handled by creating separate annotationLayer records per persona, each with its own ontologyRef. |
EventInterpretation | pub.layers.annotation#annotation with ontologyTypeRef + arguments | Per-persona situation interpretation with typed participants via argumentRef. |
Location | pub.layers.defs#spatialExpression with type="location" or pub.layers.defs#boundingBox | Geographic coordinates (GPS, cartesian) use spatialExpression with spatialEntity.geometry as WKT POINT and crs="wgs84". For spatial annotations in images/video, Layers uses boundingBox and spatioTemporalAnchor. |
EntityCollection | pub.layers.annotation#clusterSet with appropriate kind | FOVEA's collection types (group, kind, functional, stage, portion, variant) map to clusterSet.kind (community-expandable via kindUri). aggregateProperties → features. |
EventCollection | pub.layers.annotation#clusterSet + features | Situation groupings with structure. EventStructureNode hierarchies can be represented via nested features or graph relations. |
Temporal Model
| FOVEA Type | Layers Equivalent | Notes |
|---|---|---|
Time (instant/interval) | pub.layers.defs#temporalEntity (for calendar time) or pub.layers.defs#temporalSpan (for media time) | FOVEA's TimeInstant maps to temporalEntity.instant; TimeInterval maps to temporalEntity.intervalStart/intervalEnd. For media-anchored time, temporalSpan with start/ending in milliseconds. |
Time.videoReferences | pub.layers.defs#temporalSpan | Direct mapping — frame numbers convert to milliseconds via frame rate. |
Time.vagueness | pub.layers.defs#temporalModifier + pub.layers.defs#temporalEntity.earliest/latest | Temporal vagueness maps to temporalModifier.mod (approximate, early, late, etc.) and uncertainty bounds (earliest/latest). granularity is a first-class field on temporalEntity. |
Time.deictic | pub.layers.defs#temporalExpression.anchorRef | Deictic temporal references use anchorRef pointing to the deictic center (document creation time, speech time, etc.). |
RecurrenceRule (RFC 5545) | pub.layers.defs#temporalEntity.recurrence | ISO 8601 repeating intervals (e.g., R/P1W for weekly). More complex RFC 5545 rules use features. |
HabitualPattern | pub.layers.defs#temporalEntity.recurrence + features | Habitual frequency as repeating interval; typicality in features. |
CyclicalPattern | pub.layers.defs#temporalEntity.recurrence + features | Phase-based temporal patterns. Recurrence captures the cycle; phase metadata in features. |
TimeCollection | pub.layers.annotation#clusterSet + features | Collections of temporal references with pattern metadata. |
Spatial Model and Video Annotation
| FOVEA Type | Layers Equivalent | Notes |
|---|---|---|
BoundingBox | pub.layers.defs#boundingBox | Direct mapping: x, y, width, height. FOVEA adds frameNumber, confidence, isKeyframe — Layers handles frame number via keyframe.timeMs (converted from frame number) and confidence via annotation.confidence. |
BoundingBoxSequence | pub.layers.defs#spatioTemporalAnchor | boxes → keyframes (each keyframe has timeMs and bbox); interpolationSegments → interpolation (linear, step, cubic); visibilityRanges → representable via features. |
InterpolationSegment | pub.layers.defs#spatioTemporalAnchor.interpolation + features | Layers supports linear, step, and cubic interpolation. FOVEA's bezier, ease-in, ease-out, parametric are representable as cubic with control points in features, or via community-defined interpolationUri values. |
ObjectAnnotation | pub.layers.annotation#annotation with anchor.spatioTemporalAnchor | Links a spatial region to a world object. linkedEntityId → annotation.knowledgeRefs or clusterSet membership. |
TypeAnnotation | pub.layers.annotation#annotation with ontologyTypeRef | Assigns an ontology type to a spatial region. Persona-specificity is achieved by having separate annotation layers per persona. |
Claims and Propositions
| FOVEA Type | Layers Equivalent | Notes |
|---|---|---|
Claim | pub.layers.annotation#annotation with kind="span" and appropriate subkind | Claims are text spans with structured metadata. The subkind can be community-defined (e.g., "claim", "proposition", "hypothesis"). text → annotation.text; gloss → annotation.features; confidence → annotation.confidence. |
Claim.parentClaimId | annotation.parentId + annotation.childIds | Claim hierarchies use the same tree structure as constituency parses. |
Claim.claimerType/claimerGloss | annotation.arguments with role="claimer" | The claimer is a semantic argument of the claim, modeled via argumentRef. |
Claim.textSpans (discontiguous) | anchor.tokenRefSequence | Discontiguous spans use tokenRefSequence.tokenIndexes. |
Claim (as graph node) | pub.layers.graph#graphNode with nodeType="claim" | Claims can also be modeled as standalone graph nodes when they represent propositions that exist independently of specific text spans. Properties store claim metadata. |
ClaimRelation | pub.layers.graph#graphEdge or pub.layers.annotation#annotation with kind="relation" | Typed relations between claims (supports, contradicts, refines, generalizes) map to graph edges or relation-type annotations. Edge types include supports, contradicts, and community-defined types via edgeTypeUri. |
VideoSummary | pub.layers.expression + pub.layers.annotation layers | A summary is itself an expression with annotations linking it to the source video and extracted claims. |
Collaboration Model
| FOVEA Type | Layers Equivalent | Notes |
|---|---|---|
User | ATProto DID | Decentralized identity. No centralized user database. |
Project | Corpus (pub.layers.corpus) | A project is a named collection of expressions with shared ontologies. |
ProjectMembership | ATProto social graph | Access control is handled by ATProto's decentralized identity layer, not by Layers records. |
ResourceShare | ATProto record permissions | Sharing is native to ATProto — records are public by default in user PDSes. |
RBAC | Appview-level access control | Layers delegates access control to the appview implementation, not the data model. |
Features Not in FOVEA (Layers Extensions)
- Text annotation: FOVEA is video-first; Layers provides full stand-off text annotation (character spans, tokenizations, linguistic layers)
- Interlinear glossing: Multiple tokenizations + alignment records
- Syntactic parsing: Constituency, dependency, CCG parse representations
- Discourse annotation: RST, PDTB, SDRT relation types
- Linguistic judgment experiments: Structured experiment framework with agreement metrics
- Eprint linkage: Academic paper references
- W3C Web Annotation interoperability: Selector types for web annotation ecosystem
- Community-expandable enums: URI+slug pattern across all fields