Expand description
Structured natural-language analysis types and traits.
The trait is designed around multi-head cascade models (e.g.
dragonscale-ai/kniv-deberta-nlp-base-en-xsmall) that produce
token-level POS / NER / DEP tags, sentence segmentation, optional SRL
frames, and optional dialog-act classification in a single forward pass.
Structs§
- DepLink
- A dependency-parse arc attached to an
NlpToken. - Extraction
Schema - The relations and entity types to extract from one request.
- Label
Spec - One schema label: a name, and a description that tells the model what it means and which side of a relation is the head.
- NerEntity
- One merged named-entity span, a BIO-collapsed view of per-token NER tags.
- NlpCalibration
- The decoding thresholds a model ships in
calibration.json. - NlpCalibration
Overrides - Per-request overrides of
NlpCalibration. Unset fields keep the model’s shipped value. - NlpEntity
- One typed entity of an
NlpWindow, grouped across its mentions. - NlpFact
- One extracted fact: a relation between two entities, with its attributes.
- NlpLabel
Maps - The label vocabularies a model decodes against, one list per head.
- NlpRequest
- One NLP analysis request: an input plus the heads to populate.
- NlpResult
- Structured output of an NLP analysis call.
- NlpSentence
- One sentence in an
NlpResult. - NlpTasks
- Selects which NLP heads a caller wants populated.
- NlpToken
- One token in an
NlpResult. - NlpWindow
- One analysis window: a run of whole sentences of up to the model’s word budget, analysed in a single pass.
- Speech
Act - One sentence-level classification result.
- SrlFrame
- One semantic-role-labeling frame: a predicate and its argument spans.
- SrlRole
- One argument span of an
SrlFrame. - Turn
- One turn of a conversation passed as
NlpRequest::turns. - Window
Frame - One semantic-role frame of an
NlpWindow. - Window
Sentence - One sentence of an
NlpWindow. - Word
Analysis - Per-word linguistic analysis.
- Word
Span - A labelled half-open
[start, end)span of words.
Enums§
- Fact
Modality - How a
NlpFactis held by its holder. - Fact
Polarity - Whether a
NlpFactis asserted or denied.
Traits§
- NlpModel
- A multi-head structured-NLP model.