FrameNet terminology

This glossary defines the terms used by Framex and the Berkeley FrameNet 1.7 artifact it queries. FrameNet is the canonical spelling; it is a lexical semantic resource, not a general ontology or an authorization system.

Core semantic model

Frame

A frame is a structured description of a situation, event, state, relation, attribute, or entity. It names the background knowledge needed to understand a word sense.

Commerce_buy, for example, describes a buying situation rather than merely the English word buy.

Frame Element (FE)

A Frame Element is a participant, property, or other meaningful component of a frame. In Commerce_buy, Buyer and Goods are FEs.

An FE may carry a semantic-type constraint. For example, a locative role may be typed as Locative_relation; not every FE has an explicit semantic type.

Core, peripheral, and extra-thematic FE

An FE's coreness describes its relation to the frame definition:

• Core — conceptually central to the frame, such as Buyer and Goods.
• Core-Unexpressed — central but normally not syntactically expressed.
• Peripheral — broadly reusable circumstance, such as Time, Place, or Manner.
• Extra-Thematic — meaningful event information that is not a defining participant of the frame.

Coreness does not mean that a role must occur in every sentence.

Lexical Unit (LU)

A Lexical Unit is a word sense or multiword expression paired with a part of speech and a frame. It is the unit that evokes a frame.

purchase.v in Commerce_buy and run.v in Self_motion are distinct LUs. The same spelling may occur in several frames; run.v is therefore not one meaning but several frame-specific senses.

Frame-evoking element and target

The Frame-Evoking Element (FEE) is the expression in a sentence that evokes the frame. In FrameNet annotation it appears in the Target layer.

Matheson [purchased] a plot of land.
          ^ Target / FEE

Corpus annotation

Exemplar

An exemplar is an attested corpus sentence annotated for one LU. It is evidence for an LU's frame and its syntactic realizations.

FrameNet exemplar counts are counts in the selected annotation corpus. They are not estimates of general English frequency.

Annotation set

An annotation set is one annotation analysis over a sentence. A sentence may contain more than one annotation set because it can contain multiple targets, senses, or auxiliary annotation layers.

Layer and label

A layer groups one kind of annotation. A label marks a character span inside that layer. Framex joins labels by their shared start and end span.

Important layers:

Phrase Type (PT)

Phrase Type records the constituent form realizing an FE:

• NP — noun phrase
• PP[from] — prepositional phrase headed by from
• AVP — adverb phrase
• VPto — infinitival verb phrase

Grammatical Function (GF)

Grammatical Function records the syntactic function of that constituent:

• Ext — external argument, often the subject
• Obj — object
• Dep — dependent

Together, PT and GF explain how a semantic role is realized.

Matheson purchased the land from a seller.

Buyer   = Matheson       NP / Ext
Goods   = the land       NP / Obj
Seller  = from a seller  PP[from] / Dep

Valence and omitted arguments

Valence pattern

A valence pattern records an attested FE realization for an LU, including its FE, phrase type, grammatical function, and annotation count.

For purchase.v, FrameNet attests patterns including:

Buyer   → NP / Ext
Goods   → NP / Obj
Seller  → PP[from] / Dep

Framex exposes the source data with Framex.Lu.valence/2.

Frame Element Configuration (FEC)

A Frame Element Configuration is a combination of FEs realized together in one valence pattern. It is richer than an individual FE realization: it models the configuration of roles that co-occur in a construction.

Null instantiation

Null instantiation means an FE is semantically understood but has no text span in the sentence. This is data, not an annotation defect.

Framex.Annotation.omitted_roles/2 returns sentence-level omitted roles from span-less FE labels. Framex.Lu.valence/2 retains the LU-level aggregate null-instantiation patterns.

Network structure

Frame relation

A frame relation links frame meanings. It is typed and directed as stored by FrameNet: sub_frame ──type──> super_frame.

Common relation types:

For example:

Self_motion ──Inheritance──> Motion
Motion ──Inheritance──> Event
Motion ──Causative_of──> Cause_motion

Relations support navigation and structured comparison. They must not be treated as automatic equivalence, application permission, or policy inference.

Frame Element mapping

An FE mapping links roles across a related pair of frames. It lets a client compare how a role in one conceptualization corresponds to a role in another.

Semantic type (SemType)

A semantic type is a FrameNet type label arranged in a hierarchy. It can annotate a frame, FE, or LU and provides a more general semantic constraint.

Framex.SemType.for_fe(corpus, "Motion", "Theme")
# => {:ok, ["Physical_object"]}

Semantic types are evidence supplied by FrameNet; an absent type is not an inference that the role has no real-world restriction.

Framex artifact terms

Catalog

catalog.json is the artifact's index. It contains frames, LU metadata, relations, semantic-type hierarchy, and each LU's annotation-shard reference. Framex loads the catalog when Framex.open/1 succeeds.

Annotation shard

An annotation shard is a separate JSON file containing exemplars for one or more LUs. Framex reads a shard only when an annotation query requires it.

Artifact manifest

An artifact manifest identifies the source, language, release, schema, exporter version, and integrity checksum of a distributed dataset. It belongs to the data release, not to the framex Hex code package.

Reviewed artifact

A reviewed artifact is an offline-exported and validated dataset accepted as the semantic evidence used at runtime. XML parsing, corpus acquisition, and artifact construction remain outside the runtime library.

Framex API map

Non-goals

Framex does not by itself:

• parse an arbitrary new sentence into frames and roles;
• claim that FrameNet annotation counts are population frequencies;
• turn semantic relations into business rules or permissions;
• alter the reviewed artifact at runtime;
• claim English FrameNet frames are automatically universal across languages.