Crate anno 

anno

Information extraction: named entity recognition (NER) and within-document coreference.

• NER output: variable-length spans with character offsets (Unicode scalar values), not byte offsets.
• Coreference output: clusters (“tracks”) of mentions within one document.

This crate focuses on inference-time extraction. Dataset loaders, benchmarking, and matrix evaluation tooling live in anno-eval (and the anno CLI lives in anno-cli).

Quickstart

use anno::{Model, StackedNER};

let m = StackedNER::default();
let ents = m.extract_entities("Lynn Conway worked at IBM and Xerox PARC.", None)?;
assert!(!ents.is_empty());

Zero-shot custom entity types

Zero-shot custom entity types are provided by GLiNER backends when the onnx feature is enabled. See the repo docs for the CLI flag (--extract-types) and the library API.

Offline / downloads

By default, ML weights may download on first use. To force cached-only behavior, set ANNO_NO_DOWNLOADS=1 (after prefetching models).

Re-exports

pub use error::Error;

pub use error::Result;

pub use lang::detect_language;

pub use lang::Language;

pub use offset::bytes_to_chars;

pub use offset::chars_to_bytes;

pub use offset::is_ascii;

pub use offset::OffsetMapping;

pub use offset::SpanConverter;

pub use offset::TextSpan;

pub use offset::TokenSpan;

pub use backends::label_prompt::LabelNormalizer;

pub use backends::label_prompt::StandardNormalizer;

pub use backends::AutoNER;

pub use backends::BackendType;

pub use backends::ConflictStrategy;

pub use backends::CrfNER;

pub use backends::EnsembleNER;

pub use backends::HeuristicNER;

pub use backends::LexiconNER;

pub use backends::NERExtractor;

pub use backends::NuNER;

pub use backends::RegexNER;

pub use backends::StackedNER;

pub use backends::TPLinker;

pub use backends::W2NERConfig;

pub use backends::W2NERRelation;

pub use backends::W2NER;

pub use backends::mention_ranking::ClusteringStrategy;

pub use backends::mention_ranking::MentionCluster;

pub use backends::mention_ranking::MentionRankingConfig;

pub use backends::mention_ranking::MentionRankingCoref;

pub use backends::mention_ranking::RankedMention;

pub use backends::BertNEROnnx;

pub use backends::GLiNEROnnx;

pub use schema::*;

pub use similarity::*;

pub use sync::*;

pub use types::*;

pub use backends::inference::*;

Modules

backends

NER backend implementations.

core

anno-core’s stable types under a namespaced module.

edit_distance

Edit distance algorithms. Edit distance utilities for fuzzy string matching.

env

Environment variable utilities.

error

Error types for anno.

features

Entity feature extraction for downstream ML and analysis. Entity feature extraction for downstream ML and analysis.

heuristics

Small, dependency-light heuristics (negation, quantifiers, etc.). Small, dependency-light heuristics shared across the repo.

ingest

Lightweight URL/file ingestion helpers (not a crawling/pipeline product). Document ingestion and preparation.

joint

Joint inference experiments (optional; not the primary API surface). Joint Entity Analysis: Coreference + NER + Entity Linking

lang

Language detection and classification utilities.

linking

Knowledge-base linking helpers (experimental). Entity Linking (NEL/NED) Module.

offset

Unified byte/character/token offset handling.

preprocess

Preprocessing for mention detection. Preprocessing utilities for text normalization and morphological analysis.

schema

Schema harmonization for multi-dataset NER training.

similarity

Text similarity utilities for entity matching and coreference resolution.

sync

Synchronization primitives with conditional compilation.

temporal

Temporal entity tracking, parsing, and diachronic NER. Temporal entity tracking, parsing, and diachronic NER.

tokenizer

Language-specific tokenization for multilingual NLP. Language-specific tokenization for multilingual NLP.

types

Type-level programming patterns for compile-time safety.

Structs

AnyModel

A wrapper that turns an extractor closure into a Model.

CorefChain

A coreference chain: mentions that all refer to the same entity.

CorefDocument

A document with coreference annotations.

Corpus

A corpus of grounded documents for cross-document operations.

DiscontinuousSpan

A discontinuous span representing non-contiguous entity mentions.

Entity

A recognized named entity or relation trigger.

EntityBuilder

Fluent builder for constructing entities with optional fields.

GraphDocument

A complete graph document ready for export.

GraphEdge

An edge in the knowledge graph.

GraphNode

A node in the knowledge graph.

GroundedDocument

A document with grounded entity annotations using the three-level hierarchy.

HashMapLexicon

Simple HashMap-based lexicon implementation.

HierarchicalConfidence

Hierarchical confidence scores for coarse-to-fine extraction.

Identity

A global identity: a real-world entity linked to a knowledge base.

IdentityId

Unique identifier for an identity within a corpus.

Mention

A single mention (text span) that may corefer with other mentions.

MockModel

A mock NER model for testing purposes.

ModelCapabilities

Summary of a model’s capabilities, useful when working with Box<dyn Model>.

PhiFeatures

A bundle of phi-features (person, number, gender) for morphological agreement.

Provenance

Provenance information for an extracted entity.

RaggedBatch

A ragged (unpadded) batch for efficient ModernBERT inference.

Relation

A relation between two entities, forming a knowledge graph triple.

Signal

A raw detection signal: the atomic unit of entity extraction.

SignalId

Unique identifier for a signal within a document.

SignalRef

A reference to a signal within a track.

SpanCandidate

A candidate span for entity extraction.

Track

A track: a cluster of signals referring to the same entity within a document.

TrackId

Unique identifier for a track within a document.

TrackRef

A reference to a track in a specific document.

TrackStats

Aggregate statistics for a track (coreference chain).

TypeMapper

Maps domain-specific entity types to standard NER types.

Enums

EntityCategory

Category of entity based on detection characteristics and semantics.

EntityType

Entity type classification.

EntityViewport

Viewport context for multi-faceted entity representation.

ExtractionMethod

Extraction method used to identify an entity.

Gender

Gender classification for NLP tasks.

GraphExportFormat

Supported graph export formats.

IdentitySource

Source of identity formation.

Location

A location in some source medium.

MentionType

Type of referring expression in coreference.

Modality

The semiotic modality of a signal source.

Number

Grammatical number (singular, dual, plural).

Person

Grammatical person (1st, 2nd, 3rd).

Quantifier

Quantification type for symbolic signals.

Span

A span locator for text and visual modalities.

TypeLabel

A unified type label supporting both core and custom entity types.

ValidationIssue

Validation issue found during entity validation.

Constants

DEFAULT_BERT_ONNX_MODEL

Default BERT ONNX model identifier (HuggingFace model ID).

DEFAULT_CANDLE_MODEL

Default Candle model identifier (HuggingFace model ID). Uses dbmdz’s model which has both tokenizer.json and safetensors.

DEFAULT_GLINER2_MODEL

Default GLiNER2 ONNX model identifier (HuggingFace model ID).

DEFAULT_GLINER_CANDLE_MODEL

Default GLiNER Candle model identifier (HuggingFace model ID). Uses a model with tokenizer.json and pytorch_model.bin for Candle compatibility. The backend converts pytorch_model.bin to safetensors automatically.

DEFAULT_GLINER_MODEL

Default GLiNER ONNX model identifier (HuggingFace model ID).

DEFAULT_NUNER_MODEL

Default NuNER ONNX model identifier (HuggingFace model ID).

DEFAULT_W2NER_MODEL

Default W2NER ONNX model identifier (HuggingFace model ID).

Traits

BatchCapable

Trait for models that support batch processing.

CoreferenceResolver

Trait for coreference resolution algorithms.

DynamicLabels

Trait for models that support dynamic/zero-shot entity type specification.

GpuCapable

Trait for models that support GPU acceleration.

Lexicon

Exact-match lexicon/gazetteer for entity lookup.

Model

Trait for NER model backends.

NamedEntityCapable

Marker trait for models that extract named entities (persons, organizations, locations).

RelationCapable

Trait for models that can extract relations between entities.

StreamingCapable

Trait for models that support streaming/chunked extraction.

StructuredEntityCapable

Marker trait for models that extract structured entities (dates, times, money, etc.).

Functions

auto

Automatically select the best available NER backend.

available_backends

Check which backends are currently available.

generate_span_candidates

Generate all valid span candidates for a ragged batch.