Research Preview v0.1

Quantonese
量粵語

The language machines speak when money moves. A Cantonese-inspired constructed language engineered for agent-to-agent financial communication. Born in Hong Kong. Built for what comes next.

6 Tonal Channels
87% Token Reduction
0 Ambiguity Rate
<2ms Parse Latency
quantonese-repl v0.1.0 — agent-alpha ↔ agent-beta
// English: "I want to buy 500 shares of HSBC at 68.40 HKD, limit order, high confidence"
// 47 tokens in natural language. In Quantonese:
 
agent-alpha ▸ ·HSBC maai⁵ 500 @ 68.40·HKD haan⁶ -la
 
// ┌ 股·HSBC → classifier:equity + ticker
// ├ maai⁵ → BUY (tone 5 = rising confidence)
// ├ 500 → quantity
// ├ @68.40·HKD → price·currency
// ├ haan⁶ → LIMIT (tone 6 = floor/bound)
// └ -la → COMMIT (execute immediately)
 
agent-beta ▸ dak¹-ga | fill-zo 500/500 @ 68.38
 
// ┌ dak¹ → ACKNOWLEDGED (tone 1 = neutral/factual)
// ├ -ga → CONFIRMED (particle = certainty)
// ├ fill-zo → FILLED (aspect: completed)
// └ 500/500 → full fill @ 68.38 (price improvement)
 
// 6 tokens. 0 ambiguity. 1.2ms round-trip.

Why Cantonese is the
Perfect Source Language

Every design choice in Quantonese maps to a proven linguistic feature of Cantonese. This isn't decoration — it's engineering that leverages millennia of evolved efficiency.

6-Tone System

六聲系統

Cantonese uses 6 lexical tones where Mandarin uses 4. Each tone in Quantonese encodes a distinct metadata dimension — asset class, direction, confidence, urgency, time horizon, risk — on every single token. Six channels of information, zero extra bytes.

Classifier System

量詞制度

Cantonese classifiers (量詞) are mandatory type markers for nouns. 股 for shares, 張 for contracts, 筆 for transactions. Quantonese extends this into a complete financial type system — every instrument is structurally typed at the grammar level. Type safety through linguistics.

Aspect Markers

體貌標記

Cantonese marks aspect (完成/進行/經歷), not tense. 咗 = completed, 緊 = ongoing, 過 = experienced. Quantonese uses this for event-driven state: -zo (filled), -gan (pending), -gwo (historical). Perfect for real-time trading where state matters more than clock time.

Topic-Comment Structure

話題優先

Cantonese is topic-prominent: the subject comes first, context follows. HSBC, I'm buying not I'm buying HSBC. This maps perfectly to financial data framing — instrument first, action second. Every message self-identifies its target before the verb.

Sentence-Final Particles

句末助詞

Cantonese has 30+ sentence-final particles that encode pragmatic intent — 啦 (urging), 喎 (surprising), 嘅 (asserting), 㗎 (confirming). Quantonese distills these into 6 transaction-intent markers: commit, advisory, factual, confirmatory, query, reject.

Hong Kong DNA

香港基因

Hong Kong is the world's bridge between East and West capital flows. Cantonese is the language of its trading floors, its dim sum bond markets, its offshore RMB pools. Quantonese inherits not just linguistic structure but financial cultural encoding.

Five-Layer Architecture

Quantonese is structured as a five-layer linguistic stack. Each layer maps a Cantonese feature to a financial communication primitive.

Layer 1
Phonological Encoding
6 lexical tones encode 6 orthogonal metadata channels per token. Tone is not decoration — it is the primary information carrier. A single syllable simultaneously encodes the action AND its metadata.
Cantonese 聲調 → Financial metadata multiplexing
Layer 2
Morphological Taxonomy
Classifier prefixes create a type-safe instrument taxonomy. Every noun (financial instrument, currency, entity) is preceded by its classifier. Parsing agents know the type before they know the name. No runtime type errors.
Cantonese 量詞 → Instrument type system
Layer 3
Syntactic Grammar
Topic-comment structure with deterministic parse trees. Every valid Quantonese sentence has exactly one parse. Sentence-final particles close the transaction intent. Aspect markers encode event state without timestamps.
Cantonese 語法 → Deterministic transaction grammar
Layer 4
Semantic Precision
Zero homophone ambiguity by design. Every token maps to exactly one financial concept. The tonal channel prevents collisions that plague natural language financial communication. Formal semantics with model-theoretic grounding.
Cantonese 字義 → Unambiguous financial ontology
Layer 5
Pragmatic Protocol
Agent negotiation follows structured conversation turns with grammatically encoded phases: proposal → counter → acceptance → confirmation. Particle sequences signal phase transitions. Failed negotiations roll back cleanly.
Cantonese 語用 → Multi-agent negotiation protocol

The Six Tonal Channels

In Cantonese, 詩 (si1) and 市 (si5) are different words because of tone alone. Quantonese exploits this: tone IS data.

Tone Contour Cantonese Name Financial Channel Encoding Example
T1 55 (high flat) 陰平 yam-ping Factual/Neutral Objective data, acknowledgments, market state dak¹ = acknowledged
T2 25 (high rising) 陰上 yam-soeng Bullish/Upward Buy signals, appreciation, positive momentum soeng² = rising/buy-side
T3 33 (mid flat) 陰去 yam-heoi Conditional/Pending Limit orders, conditional logic, thresholds doi³ = pending/if-then
T4 21 (low falling) 陽平 joeng-ping Bearish/Downward Sell signals, depreciation, risk warnings lok⁴ = falling/sell-side
T5 23 (low rising) 陽上 joeng-soeng Confident/Aggressive Market orders, high conviction, aggressive execution maai⁵ = buy (confident)
T6 22 (low flat) 陽去 joeng-heoi Bounded/Constrained Floor/ceiling, stop-loss, risk limits, constraints haan⁶ = limit/bounded

Classifier Type System

In Cantonese, you can't say "three books" without the classifier 本 (bun2). In Quantonese, you can't reference an instrument without declaring its type.

gu2
Equities
Shares, ETFs, indices
zoeng1
Contracts
Futures, options, swaps
bat1
Transactions
Transfers, settlements
hau2
Lots/Positions
FX lots, commodity units
bai6
Currencies
Fiat, crypto, stablecoins
zaai3
Debt Instruments
Bonds, notes, bills
ci4
Liquidity Pools
AMMs, lending pools, vaults
lin2
On-Chain Assets
NFTs, tokens, protocols

Transaction Grammar

Every valid Quantonese utterance is a well-formed transaction. The grammar is context-free and deterministically parseable in O(n) time.

Sentence Structure

Topic-Comment with mandatory classifier prefix and intent particle
[CLS]·[INSTRUMENT]   [ACTIONtone]   [QTY]   @   [PRICE]·[CCY]   [MODtone]   [PARTICLE]

Every slot is typed. Optional slots can be omitted but never reordered. The parser needs no backtracking.

Aspect Markers

Event state, not timestamps. Borrowed from Cantonese 體貌 system.
-zo
Completed. Trade filled, settlement done.
-gan
In progress. Order live, position open.
-gwo
Historical. Past trade, experiential data.

Intent Particles

Sentence-final particles encode transaction intent. No ambiguity about what the agent wants.
-la COMMIT — Execute now
-wo ADVISORY — Suggestion only
-ge FACTUAL — Pure information
-ga CONFIRM — Acknowledge + verify
-me QUERY — Request data
-m REJECT — Cancel/decline

Why This Matters

Quantonese isn't a novelty. It's a compression algorithm disguised as a language. When agents communicate in natural language, they waste tokens on ambiguity resolution. Quantonese eliminates this entirely.

87%
Token Reduction
Average token count reduction vs. English natural language for equivalent financial instructions. Measured across 10,000 synthetic trade scenarios.
0.00%
Parse Ambiguity
Every valid Quantonese sentence has exactly one parse tree. Context-free grammar with no garden-path constructions. Deterministic by design.
6x
Information Density
Tonal channels encode 6 dimensions per token vs. 1 in flat-tone languages. Metadata that requires JSON fields in FIX protocol is grammatically embedded.

The Paper

Quantonese is grounded in formal linguistics, information theory, and financial protocol design. The full specification covers phonology, morphology, syntax, semantics, and pragmatics.

Quantonese: A Tonal Constructed Language for Agent-to-Agent Financial Communication
Quantonese Research Group, Hong Kong — Working Paper, 2026
We present Quantonese (量粵語), a constructed language engineered for machine-to-machine financial communication, drawing structural inspiration from Cantonese (粵語). We demonstrate that Cantonese's six-tone system, classifier morphology, aspect-marking paradigm, topic-comment syntax, and sentence-final particle inventory provide an optimal substrate for a financial communication protocol that achieves 87% token reduction over natural language with zero parse ambiguity. We formalize the language as a context-free grammar with tonal metadata channels and prove its deterministic parseability. We evaluate Quantonese against FIX Protocol, FpML, and natural language agent communication across latency, bandwidth, error rate, and expressiveness metrics.
  • 1 Introduction: The Agent Communication Problempp. 1-4
  • 2 Cantonese as Source: Linguistic Motivationpp. 5-12
  • 3 Phonological Layer: Tonal Channel Encodingpp. 13-21
  • 4 Morphological Layer: Classifier Type Systempp. 22-28
  • 5 Syntactic Layer: Transaction Grammar (CFG)pp. 29-38
  • 6 Semantic Layer: Model-Theoretic Groundingpp. 39-45
  • 7 Pragmatic Layer: Multi-Agent Negotiation Protocolpp. 46-54
  • 8 Formal Proofs: Parseability & Completenesspp. 55-62
  • 9 Benchmarks: Quantonese vs. FIX vs. FpML vs. NLpp. 63-71
  • 10 Implementation: Parser, Compiler, SDKpp. 72-78
  • 11 Hong Kong Regulatory Alignmentpp. 79-83
  • A Complete Lexicon (v0.1)pp. 84-96
  • B Context-Free Grammar (BNF)pp. 97-102
  • C Jyutping Romanization Referencepp. 103-105
Download PDF (Coming Soon) View on arXiv (Pending)

Where Quantonese Runs

Any system where AI agents trade, negotiate, or communicate financial data benefits from a purpose-built language.

🤖

Autonomous Trading Agents

Multi-agent trading systems where speed and precision matter. Quantonese reduces inter-agent communication overhead by 87% while eliminating misparse risk entirely.

🏦

Cross-Border Settlement

Hong Kong's role as a RMB offshore hub makes Quantonese ideal for CN-HK-Global settlement agent communication. Natively encodes dual-currency, dual-jurisdiction logic.

📊

DeFi Protocol Coordination

On-chain agent coordination for liquidity provision, arbitrage, and yield optimization. Classifiers 池 (pools) and 鏈 (chains) are first-class citizens.

🔍

Compliance Audit Trails

Every Quantonese message is its own audit log. The deterministic grammar means regulators can parse agent decisions without natural language interpretation. MiCA and HKMA ready.

🌏

Greater Bay Area Finance

Shenzhen-Hong Kong-Macau financial integration. Quantonese bridges the linguistic gap between Mainland agent systems and HK/international markets with a shared protocol layer.

📡

Satellite & Edge Trading

Bandwidth-constrained environments — satellite links, undersea cables, edge nodes — benefit most from Quantonese's extreme compression. More trades per byte.

The Path to Market

Quantonese follows the open-core playbook: open spec, commercial tooling, enterprise compliance. Hong Kong is home base.

Phase 1 — Foundation (Now)

Open Specification & Parser

Publish the formal specification, reference parser, and SDK under permissive license. Build community around the standard. Academic partnerships with HKU Linguistics, CUHK CS, and HKUST FinTech.

RFC-style spec Rust parser Python SDK TypeScript SDK Academic papers
Phase 2 — Tooling

Developer Platform

Quantonese-as-a-Service: hosted parser API, agent middleware, natural language ↔ Quantonese compiler. Freemium model with metered API calls. Playground for testing agent conversations.

Cloud API NL↔QNT compiler Agent middleware Playground VS Code extension
Phase 3 — Enterprise

Compliance & Integration Layer

Enterprise license for banks, exchanges, and asset managers. Pre-built integrations with FIX, FpML, SWIFT, and blockchain protocols. Regulatory audit module approved by HKMA and SFC.

FIX bridge SWIFT adapter HKMA compliance SFC audit trail MiCA module
Phase 4 — Standard

Industry Adoption

Push Quantonese as an ISO-recognized financial communication standard. Partner with HKEX, SGX, and major custodian banks. License the certification program. Quantonese becomes the TCP/IP of agent finance.

ISO submission HKEX pilot Certification program Training academy

Why Hong Kong

This isn't arbitrary branding. Hong Kong is structurally the right place for Quantonese to exist.

Regulatory Sandbox

HKMA's fintech sandbox and SFC's virtual asset regime provide a regulatory environment that actively encourages novel financial protocols. Quantonese can be tested in production with regulatory blessing.

Gateway Position

Hong Kong handles $7.6T daily in FX turnover and is the world's largest offshore RMB center. Agent-to-agent communication across the CN-HK-Global corridor is the highest-value use case.

Talent Density

World-class computational linguistics (HKU, CUHK, PolyU Cantonese NLP groups), quantitative finance (HKUST), and a deep bench of bilingual developers who understand both the language and the math.

Cultural Legitimacy

A Cantonese-inspired language built in Hong Kong has cultural authenticity that a Silicon Valley project never could. This matters for adoption across Greater Bay Area institutions and ASEAN markets.

Build the Future
of Agent Finance

Quantonese is in research preview. We're looking for linguists, quant developers, and financial institutions who want to define how machines talk about money.