A machine-readable Cantonese lexicon based on cognitive semantics, encoding lexemes and sense networks in JSON for linguistic research, ASR, and AI training.
This repository is an open Cantonese semantic lexicon that represents lexemes (not just words or characters) and their sense networks in a machine-readable JSON format.
Instead of organizing entries by traditional parts of speech, this project models Cantonese vocabulary around:
cognitive core meanings
image schemas and event structures
semantic extensions (metonymy, metaphor, abstraction, grammaticalization)
lexeme-level disambiguation (homophones, constructionalized forms, loan blends)
The goal is to build a resource that is linguistically grounded, computationally usable, and scalable for AI systems.
Build a lexeme-based semantic lexicon for Cantonese
Model sense networks instead of flat word lists
Support:
semantic disambiguation
ASR and NLP pipelines
grammar and construction modeling
cognitively informed AI training
Provide data in a clean, explicit JSON structure
Encourage open and community-oriented development
Lexemes, not characters Homophones with unrelated meanings are split into separate lexemes.
Cognitive semantics first Each lexeme is anchored by a core schema (event, control, motion, information, evaluation, etc.).
Sense networks, not sense lists Senses are linked through extension mechanisms:
prototype
metonymy
metaphor
abstraction
grammaticalization
Machine-readable by design All entries are encoded in structured JSON to support:
feature extraction
clustering
embedding alignment
grammar induction
cantonese-semantic-lexicon/
│
├── lexemes/
│ ├── zapj.json
│ ├── svj.json
│ ├── lukj.json
│ └── ...
│
├── schema/
│ └── msfs.json # Minimal Semantic Feature Set
│
├── docs/
│ ├── methodology.md
│ ├── annotation-guide.md
│ └── roadmap.md
│
├── examples/
│ └── annotated-sentences.json
│
└── README.md
Each lexeme entry is encoded using a Minimal Semantic Feature Set (MSFS), including:
lexeme-level features
cognitive domain
core schema
prototype type
sense-level features
ontological type (object, event, process, state, relation)
semantic dimensions (force, motion, control, information, evaluation, etc.)
participant structure
directionality and result states
extension mechanisms
constructional behavior
Example (simplified):
{ "lexeme_id": "zapj1", "homophone_group": "zapj", "core_schema": "manual control and acquisition of an entity", "senses": [ { "sense_id": "zapj1-1", "gloss": "pick up", "ontological_type": "event", "semantic_dimensions": ["force","control","motion"], "extension_type": "prototype" } ] }
This lexicon is designed for:
Cantonese linguistic research
cognitive-semantic modeling
ASR and speech technology
NLP and LLM pretraining / fine-tuning
grammar induction and construction analysis
semantic annotation projects
Contributions are welcome, especially in:
adding new lexeme entries
expanding sense networks
refining semantic features
improving documentation
providing annotated examples
If you contribute:
please keep entries lexeme-based
clearly state core schemas
indicate semantic extension paths
include authentic Cantonese examples where possible
A detailed annotation guide will be provided in docs/annotation-guide.md.
Formalize MSFS schema
Publish annotation guidelines
Seed lexicon with high-frequency Cantonese lexemes
Build cross-lexeme semantic indices
Provide example applications (disambiguation, grammar modeling)
Explore links to corpora and ASR datasets
To be determined. (Open data licenses such as CC BY 4.0 or CC BY-SA 4.0 are recommended.)
This project treats Cantonese not as a list of words, but as a structured semantic system.
The aim is not only to record meanings, but to model how meanings are organized, extended, and cognitively grounded.