From language to
semantic graphs.

Extract events and semantic roles from English text, with source evidence.

Knowledge embedded in text.

01 / KNOWLEDGE GRAPHS

Relationships across documents

Extract events and participants into JSON-LD or RDF.

02 / RETRIEVAL

Context beyond keywords

Use semantic roles and source sentences in downstream retrieval or RAG.

03 / LANGUAGE RESEARCH

Evidence for interpretation

Inspect roles, word senses, and negation against the original text.

Linguistic analysis.

Example sentence

The bank approved the loan.

Semantic roles

agentthemebankapproveloanpredicate

Word senses

The bank approved the loan.Financial institution
She sat on the river bank.Land alongside a river

Ambiguity is retained when no sense meets the selection criteria.

From input to extraction.

Illustrative extraction
01
LINGUISTIC ANALYSIS

Analyze the text

Tokenization, dependencies, NER, and coreference.

Technical details

Stanford CoreNLP supplies tokenization, dependency parsing, named entity recognition, and coreference chains.

02
SEMANTIC INTERPRETATION

Interpret roles & senses

VerbNet rules and a local graded-WSD model.

Technical details

Dependency rules and VerbNet assign semantic roles. A local graded-WSD model scores candidate WordNet senses, retaining ambiguity when selection criteria are not met.

03
GRAPH CONSTRUCTION

Build the graph

Documents, frames, entities, and evidence spans.

Technical details

Frames preserve negation and modality. Role mentions retain their original text and character offsets, so you can inspect where the structure came from.

04
STRUCTURED OUTPUT

Export the extraction

JSON-LD, Turtle, RDF/XML, OWL, or N-Quads.

Technical details

Choose JSON-LD, Turtle, RDF/XML, OWL RDF/XML, or N-Quads. KExtractor exports the graph for your tools; embeddings and database indexing are handled downstream.

Explore an extraction.

Unstructured textINPUT

Marie Curie discovered radium in Paris.

Agent Predicate Theme Location

A semantic frame connects the action to its agent, theme, and location, while preserving the sentence it came from.

Structured knowledge
agentthemelocationsourcediscoverSEMANTIC FRAMEMarie CurieradiumParisSource sentence
Entity / participant Semantic frame Evidence

These simplified examples illustrate the graph structure. Run the CLI to extract knowledge from your own text

Run KExtractor.

Input
English strings or plain-text documents, processed through the CLI.
Setup
JDK 21+, Maven to build, WordNet, and a compatible ONNX WSD model with its tokenizer and configuration. Model resources are prepared separately.
Interpretation
Ambiguous senses are retained when selection criteria are not met. Model scores do not establish correctness; frames describe the text, including negation and modality.
Installation and setup
Illustrated portrait of Pasquale Esposito

Pasquale Esposito

PhD in Linguistics

Built by Pasquale Esposito, a linguist working with NLP and knowledge graphs. KExtractor connects language research with practical tools.

EXPLORE HIS WORKpasqualeesposito.net
Credits, citations & licenses Resource names are used for attribution, without implying affiliation or endorsement.