- Understand what GALEN is and why it exists
- Memorise the 5-layer architecture
- Know all 12 intent tokens
- Use modifiers to qualify statements
- Complete all practice exercises
- Write 3 GALEN statements from memory
Every time you write a message to an AI in plain English, you are sending far more words than the machine needs. Articles, auxiliary verbs, hedge phrases, repeated context — all of it adds overhead without adding meaning.
GALEN is a compact structured language that strips that overhead away. It is designed so that a human can write it quickly, and a machine can parse it without guessing.
GALEN does not replace language — it replaces the overhead around language. You still think in English. You write in GALEN. The machine reads GALEN. Everyone saves time.
The Name
GALEN stands for George Alex Language Encoding Network. It also references Galen of Pergamon (129–216 AD) — the physician who systematised medical knowledge into a precise language that lasted fifteen centuries. The same principle applies here: precision and structure make knowledge transmissible.
GALEN works with any AI — Claude, ChatGPT, Gemini. Paste the system prompt from the specification page and any model will understand GALEN input immediately.
Every GALEN statement has up to five layers. Position carries meaning — the order never changes. Only layers 1 and 3 are mandatory.
Layer by Layer
| Layer | Separator | Purpose | Example |
|---|---|---|---|
| 1 · Intent | none — opens statement | What kind of response do you need? | G |
| 2 · Domain | dot after intent | Content type or sub-category | G.email |
| 3 · Subject | colon : | The core topic or target | G.email: @surgeon |
| 4 · Modifier | tilde ~ | Qualifiers, adjustments, values | ~ angle=compliance |
| 5 · Format | angle > | Shape of the response | >txt |
The Separators Are the Grammar
You do not need punctuation, tense, or articles. The separators carry the grammar:
Domain Context — The Session Saver
Declare your context once at the start of a session. Everything that follows inherits it automatically — you never repeat yourself.
Never repeat your context. Set [CTX] once. Every statement after it inherits the domain, product, and user role automatically. This one habit accounts for most of GALEN's token reduction.
The intent token is the first character of every statement. It declares what class of response is required — before the machine reads anything else. There are 12 intent tokens in GALEN v1.0.
Always start with intent. It is the difference between the machine guessing what you want and knowing what you want.
The 12 Intent Tokens
| Token | Intent | Example |
|---|---|---|
| Q | Query — ask for information | Q: GLP-1.mechanism |
| C | Command — instruct an action | C: send.report @director |
| G | Generate — create new content | G.email: @patient |
| E | Explain — educate or clarify | E: insulin.resistance |
| S | Summarise — compress content | S: #prev3 >brief |
| D | Data input — structured data | D.med: BMI=42 sex=F age=45 |
| A | Affirm — approve or confirm | A: proceed |
| N | Negate — reject | N: wrong.approach |
| R | Revise — modify previous output | R: #prev -length |
| X | Cancel — discard previous | X: ignore.last |
| V | Verify — validate or check | V: dosage=correct |
| T | Translate — convert format/language | T: #prev >french |
Sub-types with Dot Notation
Intent tokens may be qualified with a content sub-type. This avoids ambiguity in generate-class commands:
Memory Tip
Create: Q G E S D — you are putting something into or getting something out of the system
Control: C R X — you are directing the machine
Respond: A N V T — you are reacting to output
Modifiers are symbolic qualifiers that adjust the subject or output without adding word overhead. They sit after the ~ separator and can be stacked in any order.
Modifier Symbol Library
| Symbol | Meaning | Example |
|---|---|---|
| ! | Negation — not, non, without | !surgical = non-surgical |
| + | Increase / more / intensify | +detail = add more detail |
| - | Decrease / less / reduce | -length = make it shorter |
| @ | Target recipient | @surgeon = directed at surgeon |
| # | Reference prior output | #prev · #prev3 |
| ^ | High priority / urgent | ^urgent |
| ? | Uncertain / approximate | ?3kg.loss = approx 3kg |
| = | Defined value | BMI=42 · age=45 |
| | | Conditional — if/then | BMI>40 | C: refer |
| >> | Ordered sequence | login >> search >> export |
| & | Parallel / simultaneous | @surgeon & @director |
| +past | Past tense | surgery+past |
| +next | Future tense | appointment+next |
| ++ | High degree | ++confident |
| +++ | Maximum / critical | +++urgent |
Compression Rules for Subjects
| Rule | English | GALEN |
|---|---|---|
| Drop articles | the patient, a surgeon | patient · surgeon |
| Drop auxiliaries | is, are, was, were, have | (omit entirely) |
| Plural by number prefix | three patients | 3patient |
| Possession by ~ | surgeon's outcome | surgeon~outcome |
| Compound by dot | rate of weight loss | weight.loss.rate |
| Negation by ! | non-surgical, not invasive | !surgical · !invasive |
Stacking Modifiers
Output Format Tokens
Place a format token at the end of the statement after > to specify the response shape.
Convert each of these plain English instructions into GALEN. Use everything you have learned — intent, context, subject, modifiers, and format.
Without looking at the tables — write one G statement, one Q statement, and one D statement about something relevant to your own work. This is the most effective way to consolidate what you have learned.
Everything you need in one place. Bookmark this page or print this section.
Use the GALEN Converter to practice converting your own English sentences.
Read the full specification for domain packs and advanced grammar.
Paste the system prompt into your AI of choice — and start writing in GALEN today.