語義作成語言(Semantic Composition Language, SCL)是一種類似Lispエ³語言。雖然LLM広尹內底無SCLエ³詳細資料，ㆬ³コㅗㇷ¹尹ロㆁ有法度處理チッ¹種類Lisp語言。用EBNF來表示表達式文法(expression grammar)嘛是無問題。有一クァ模型エ³広語義作成チッ¹エ⁷作成性(Compositionality)エ³概念是Frege代先提出エ³，Wikipedia內底コㅗㇷ¹有コㆁ是コㅗㇷ¹卡早エ³Boole所提出エ³。
ラㇴエ³タㆁ²カ³一个散文(prose)prompt翻譯做一个SCL prompt。嘛エ³タㆁ²カ³一个SCL prompt翻倒뚜ㆁ²ラィㇷ做一个散文prompt，也着是自然語言エ⁷prompt。散文適合快速AI輸入，SCL適合可再使用エ⁷工課流程。
Shorthand是速記アㇷ³是簡略表記エ³意思。譬喻広，漢字着是一種速記符號，像南ジㇷ，極ジㇷ。注音符號用ㄇ來表示規个音節ㄇㄚ，用ㄎ來表示音節ㄎㄜ，着是一種簡略表記。使用者若無想ベㇷ¹用速記ネㇷ？使用者猶原是エ³用エ作成一个類Lispエ³樹仔，不免用着速記表記，只要有カ³動詞エ³項構造保留키ㇷラィㇷ着好。
下腳是無用着速記エ³一个例:
recommend user
filter (recommend user)
select (filter (recommend user))
(let ((x (filter (recommend user)))
(select x))
動詞エ³項構造對類Lisp作成來広是核心イェㇴ⁵ジㇴ²。類Lisp作成治基礎上倚靠動詞エ³項數，伊接受有型別的項，カㇷ¹巢狀logic(例， filter(recommend(user)))。
速記表記カㇴ⁷ナ⁷是一種構文上快速輸入カㇷ¹使用者操作利便エ⁷一个方便エ⁷層次-syntactic sugar，構文糖仔也。不是一个核心依存。
一个速記表達式
Movies → Sci-fi → Thought-provoking → Netflix
一个速記成員、部件
Sci-fi
所以速記表記有增加啥乜好處ネㇷ？
let + recommend + user + filter + select，取代巢形碼。治SCL語境內底エ³速記表記是代表一个卡長エ⁷表達式アㇷ³是表達鏈エ³實腹エ³別名アㇷ³是省略語。伊カ³結構複雜性藏키ㇷラィㇷ，但是保留語義上エ⁷clarity。
伊是一个語義上抽象化層次，一个方便エ⁷提示記述構文。
カ³下腳エ³列表下ルェㇷ¹AI對話內底，エ³得着詳細解說。アㇴ¹ツァ˜カ³表達map到速記。
ChatGPT産生
How Expressions Map to Shorthand.
1. Direct Verb Mapping
2. Chained Pipeline Mapping
3. Parameterized Mapping
4. Let-bound Structure
カ³下腳エ³列表下ルェㇷ¹AI對話內底，エ³得着詳細解說。SCL內底有啥乜款エ³速記。
ChatGPT産生
Types of Shorthand in SCL
1. Verb-Noun Shorthand
2. Attribute Shorthand
3. Chain Shorthand
4. Let-bound Shorthand
5. Custom Semantic Tokens
一个速記表達式的象徵形(symbolic form)實際上着是一个操作式(operation formula)。治轉換做最終解析化エ³表達式進前-治解析エ³時，伊提供語義logic。
速記表達式治解析エ³時ツㇴ⁷，エ³去ホㆁ⁵翻譯做象徵形，治轉換做Lisp樹仔進前去띠ッ¹띠ㅗㇷ³ヒッ¹个結構化formatエ³操作logic。
速記表達式卡倚使用者，伊エ³象徵形是卡倚Lisp語言。
ラㇴベㇷ¹用下腳エ³用例來看解析エ³step。伊是一个表示AND logicエ³序列式filtering，後壁接一个表示OR logicエ³分岐式filtering。
ChatGPT産生
🧩 4. Interactive Exploration (Branch + Merge)
Breadcrumb Flow:
Home → Movies → Sci-fi → Thought-provoking → Netflix
↘ Runtime: Short (<90m)
↘ Tone: Dark
↘ Compare with: Apple TV+
🔗 Prompt Chain:
Each refinement is its own step:
* “List sci-fi movies on Netflix.”
* “Filter by tone: thought-provoking.”
* “Now filter by runtime < 90m.”
* “Among those, highlight the ones with darker tones.”
* “Find 3 on Apple TV+ with similar tone and runtime.”
最後カ³feedback loop接키ㇷリㇷ，着變做下腳アㇴ¹ネ，有解析step，コㅗㇷ¹有Lisp code:
Copilot産生
🔶 1. Sequential Filtering
Start by narrowing the movie pool step by step:
1. recommend + user → candidate_movies
2. filter + genre("sci-fi") + platform("netflix") → netflixscifi
3. filter + tone("thought-provoking") + netflixscifi → thoughtful_pool
---
🔶 2. Branching & Merge
Explore two refinement paths in parallel and recombine:
4. filter + runtime<90 + thoughtfulpool → shortbranch
5. filter + tone("dark") + thoughtfulpool → darkbranch
6. merge(shortbranch, darkbranch) → mergedfocuspool
---
🔶 3. Feedback Loop: Actor Preference
Incorporate user affinity for Jennifer Lawrence:
7. rate(mergedfocuspool) → detect user preference
8. filter + actor("Jennifer Lawrence") + mergedfocuspool → jlawpreferredpool
---
🧠 Final Lisp-Style Composition
`lisp
(let* (
(candidates (recommend user))
(netflix-sci-fi (filter genre("sci-fi") platform("netflix") candidates))
(thoughtful (filter tone("thought-provoking") netflix-sci-fi))
(short-branch (filter runtime<90 thoughtful))
(dark-branch (filter tone("dark") thoughtful))
(merged (merge short-branch dark-branch))
(jlaw-filtered (filter actor("Jennifer Lawrence") merged))
(apple-tv (filter platform("apple tv+") (filter tone("dark") (filter runtime<90 (recommend user)))))
(result (compareTo apple-tv jlaw-filtered))
)
result)
`
利用터ㆁ³煞尾エ³frameworkカ³任何一个速記表達式エ³解析step印추ァィㇷ。
Show the parsing steps for this shorthand expression
各層次エ³運作。
ChatGPT産生
[[ 1. Natural Language Expression ]
"Show me some thought-provoking sci-fi movies on Netflix—
either short or dark— and compare them to what's on Apple TV+."
↓ (interpreted by NL-to-shorthand module)
[ 2. Shorthand Expression ]
Home → Movies → Sci-fi → Thought-provoking → Netflix →
Branch 1: Runtime: Short (<90m) →
Branch 2: Tone: Dark →
Merge →
Compare with: Apple TV+
↓ (parsed into structured symbolic operations)
[ 3. Symbolic Form ] ← 🧩 Operation formulas are defined and applied here
let + recommend + user
+ filter + genre("sci-fi")
+ filter + tone("thought-provoking")
+ filter + platform("netflix")
+ branch(filter + runtime<90, filter + tone("dark"))
+ merge
+ compareTo(platform("apple tv+"))
↓ (converted into executable computation structure)
[ 4. Parsed Expression (Lisp-style Tree) ]
(let ((base (platform "netflix"
(tone "thought-provoking"
(genre "sci-fi"
(recommend user)))))
(short (runtime<90 base))
(dark (tone "dark" base))
(merged (merge short dark)))
(compareTo (platform "apple tv+") merged))
譬喻広有一个無完整速記:
Home → Movies → Sci-fi
カㇷ¹伊エ³象徵形(操作式):
let + genre("sci-fi") +
象徵形터ㆁ³煞尾是一个加號。用伊做輸入。ヒッ¹个尾溜エ⁷加號是指出チッ¹條鏈仔アㇷ¹未完整。伊カ³建議器提示広アィ²提出有效エ⁷續行。
解析器建立一个Lisp風格エ³樹仔:
(genre "sci-fi")
쩨代表一組科幻電影，對續ロㅗㇷラィㇷエ³操作嘛準備好ア³。
建議器エ³産生쩨四款エ³建議，附加エ⁷filter，分岐，合併/比較，選擇。以目前エ³速記表達式來看，有下腳7種選項타ㆁ⁷選:
Grok産生
Suggestions for 'let + genre("sci-fi") +':
- platform("netflix")
- tone("thought-provoking")
- runtime<90
- actor("Keanu Reeves")
- award("oscar")
- branch(runtime<90, tone("dark"))
- select
若準広使用者カ³鏈仔訓ホ⁷長，
速記:
Home → Movies → Sci-fi → Netflix
象徵形:
let + genre("sci-fi") + platform("netflix") +
解析器カ³樹仔update做アㇴ¹ネ:
(platform "netflix" (genre "sci-fi"))
建議器コㅗㇷ¹根據updateエ³上下文(sci-fi movies on Netflix)獻出精鍊エ⁷しあげ:
- tone("dark")
- runtime<90
- actor("Keanu Reeves")
- award("oscar")
- branch(runtime<90, tone("dark"))
- select
チッ¹个過程反覆運作，建議器エ³根據鏈仔狀態カㇷ¹項數規則カ³建議創ホ⁷ハㇷ゚³ス。
建議器是運作治操作式チッ¹緣。
ChatGPT加上Grok産生
# Movie-Selecting Framework
## Folder Structure
```
/Framework/
/Verbs/
recommend.md
select.md
filter.md
rate.md
merge.md
compare.md
/Nouns/
user.md
movie.md
genre.md
actor.md
rating.md
platform.md
tone.md
runtime.md
award.md
/Rules/
operation_formula.md
/Templates/
prompt_templates.md
/SemanticParsingLayer/
suggestor.md
parser.md
```
## /Verbs/recommend.md
Verb: recommend
Valency: 1 or 2
Used to retrieve candidate movies for a user.
Examples:
recommend + user
recommend + user + filter
## /Verbs/select.md
Verb: select
Valency: 1
Selects final choice from a filtered list.
Examples:
select from recommendation
recommend + user + filter + select
## /Verbs/filter.md
Verb: filter
Valency: 1
Narrows down movie sets by attributes.
Filterable Attributes:
- genre
- actor
- tone
- runtime
- platform
- award
Examples:
filter + genre
recommend + user + filter + runtime<90
filter + tone:dark
## /Verbs/rate.md
Verb: rate
Valency: 1
Assigns a score or evaluation to a movie.
Examples:
rate movie
rate movie with user
## /Verbs/merge.md
Verb: merge
Combines results from parallel filter branches.
Example:
merge(short, dark)
## /Verbs/compare.md
Verb: compareTo
Used to compare result sets (e.g., between platforms or tones).
Example:
compareTo("Palm Springs", candidates)
## /Nouns/user.md
Noun: user
Represents an individual or profile with preferences.
Examples:
user:John
user:anonymous
## /Nouns/movie.md
Noun: movie
Represents a film entity.
Examples:
movie:"Inception"
movie:"The Matrix"
## /Nouns/genre.md
Noun: genre
Examples:
sci-fi
drama
comedy
## /Nouns/actor.md
Noun: actor
Examples:
Leonardo DiCaprio
Emma Stone
Keanu Reeves
## /Nouns/rating.md
Noun: rating
Examples:
8.5
PG-13
R
## /Nouns/platform.md
Noun: platform
Examples:
netflix
hulu
apple tv+
prime video
## /Nouns/tone.md
Noun: tone
Examples:
thought-provoking
dark
uplifting
## /Nouns/runtime.md
Noun: runtime
Expressed as numeric duration or constraints
Examples:
runtime<90
runtime<120
## /Nouns/award.md
Noun: award
Examples:
oscar
bafta
cannes
## /Rules/operation_formula.md
Operation Formulas
Sequential Recommendation Chain:
genre("sci-fi") + platform("netflix") → filtered_movies
recommend + user + filter → filtered_movies
recommend + user + filter + select → final_choice
Filtering Logic:
filter + genre → subset by genre
filter + tone → subset by mood
filter + runtime<90 → short movies
Branch + Merge:
let + A + branch(B, C) + merge → intersected result
let + base + compareTo(other_set)
Symbolic Chain Composition:
let + A + B + C + branch(...) + merge + compareTo(...) → Lisp-equivalent nested evaluation
## /Templates/prompt_templates.md
User-Facing Shorthand:
Home → Movies → Sci-fi → Thought-provoking → Netflix →
Branch 1: Runtime: Short (<90m) →
Branch 2: Tone: Dark →
Merge →
Compare with: Apple TV+
Framework Shorthand:
let + recommend + user
recommend + user + filter + select
Symbolic Chain Composition:
genre(...) + platform(...) + tone(...)
let + moodToGenre("uplifting") + platform(...) + runtime<90
let + analyze("Palm Springs") + themeMatch + toneShift(...) + platform(...) + compareTo(...)
Branch + Merge Templates:
let + A + branch(B, C) + merge
let + A + platform(...) + compareTo(...)
## /SemanticParsingLayer/suggestor.md
Suggestor Logic
Parses symbolic chain fragments
Uses valency + rule matching from operation_formula.md
Suggests completions based on current chain state
Example:
Input: genre("sci-fi") +
Suggestions:
- platform(...)
- tone(...)
- filter(...)
- compareTo(...)
## /SemanticParsingLayer/parser.md
Semantic Parser
Converts symbolic + composition into Lisp-style tree
Supports binding, branching, and merge evaluation
Also generates human-readable parsing steps
Example:
Input: recommend + user + filter + select
Parsing Steps:
1. recommend + user → candidate_movies
2. filter + (recommend + user) → filtered_movies
3. select + (filter + recommend + user) → final_choice
Lisp Translation:
(let ((x (filter (recommend user))))
(select x))
## Parsing Steps (Symbolic → Lisp)
Each + implies function application (right-to-left nesting)
Valency rules from /Verbs/ determine how verbs compose
Intermediate results may be named (e.g., candidate_movies, filtered_movies)
Use let to bind reusable subresults
Refer to /Rules/operation_formula.md for chaining logic and semantics