English Prompt Vocabulary
Day 5-এ শিখেছো prompt-এর গঠন। আজ Day 6 — Module 2-এর dedicated vocabulary day: ৮টা verb family (~৪০টা verb), progressive-specificity ladder, ৬টা politeness-trap fix, আর ~২৫টা reusable English phrase — সব মিলিয়ে যেকোনো vague idea-কে ৩০ second-এ production-ready prompt বানানোর টুলকিট।
- •৮টা verb family চেনা — Analysis, Extraction, Transformation, Reasoning, Generation, Classification, Editing, Critique। প্রতিটায় ~৫টা verb মুখস্থ। মোট ~৪০টা production-ready English action verb।
- •The imperative rule মাথায় বসিয়ে দেওয়া: "Summarize this." not "Can you summarize this?" প্রম্পট-এ verb first।
- •৬টা common politeness trap চিনতে পারা — যেগুলো Bengali speaker-রা instinctively লেখে — এবং তাদের production fix।
- •Progressive-specificity ladder apply করতে পারা: vague one-line প্রম্পট-কে ৫ ধাপে production-ready ৪-বাক্যের প্রম্পট-এ rebuild করা।
- •৮ category-তে organized ~২৫টা reusable English phrase fragment চেনা — output format, length, tone, audience, persona, negation, uncertainty, no-preamble — যেগুলো কোনো prompt-এ paste করা যায়।
- •"No preamble" trick জানা — কেন এটা শত-শত wasted token save করে, এবং কোথায় add করতে হয়।
- •Day 4, Day 5 শেষ — prompts as software + 6-part anatomy বসে গেছে।
- •ChatGPT/Claude-এ অন্তত ১০-১৫টা real prompt পাঠানোর অভিজ্ঞতা।
- •English reading comfortable, কিন্তু from-scratch writing-এ hesitation — এই lesson-এর প্রধান audience।
The Big Picture
Verb দিয়ে prompt শুরু করা — single highest-leverage habit দিয়ে শুরু, তারপর ৮টা verb family।
Day 4-এ শিখেছো prompt একটা software। Day 5-এ শিখেছো তার ভেতরের ৬-অংশের anatomy আর system/user/assistant role। আজ Day 6 — Module 2-এর সবচেয়ে hands-on lesson, পুরো ৩ ঘণ্টা এক জায়গায় investing: English prompt vocabulary। Concept সব বোঝো, কিন্তু English-এ prompt লিখতে গেলে আটকে যাও? আজ সেই আটকানো ভাঙবে — ৮টা verb family, একটা progressive-specificity ladder, ~২৫টা reusable phrase, আর ৬টা politeness-trap fix। শেষে তুমি যেকোনো vague idea ৩০ second-এ production-ready English prompt-এ বদলাতে পারবে।
The imperative ruleverb first, always
Single highest-leverage habit। Prompt শুরু imperative verb দিয়ে। "Summarize this article." — not "Can you summarize this article?" — not "Could you please possibly summarize this article?" Imperative form shorter, clearer, এবং production prompts-এ standard। Bengali speaker-রা polite English-এ default করে — production-এ এই politeness clarity-কে hurt করে।
Production prompt-engineering guidance explicit: "command verb should be placed in the first sentence ... imperative mood should be used rather than questions or suggestions।" এক rule, daily prompts-এ massive improvement।
শুরু সবচেয়ে basic building block দিয়ে — verb। প্রতিটা prompt একটা verb দিয়ে শুরু হয়; কোন verb, সেটাই output-এর কাজ ঠিক করে। নিচে ৮টা family, প্রতিটায় ৫টা verb — family select করো, verb click করো, definition + real example দেখো।
verb পেলে। কিন্তু শুধু verb দিয়ে prompt production-ready হয় না — সেটাকে ঘিরে আরও কিছু ছোট piece লাগে: কতটা specific, কী length, কার জন্য। পরের অংশে বাকি concept-গুলো।
Core Concepts
Specificity ladder থেকে no-preamble trick পর্যন্ত — বাকি ৬টা ধারণা।
The 8 verb families~40 verbs you actually need
Vocabulary-টা finite। ৮টা category × ~৫টা verb = ~৪০টা verb। সব production prompt এই ৪০-র মধ্যে fit করে। Categories: Analysis (analyze, examine, evaluate, assess, review), Extraction (extract, identify, list, find, pull out), Transformation (summarize, paraphrase, translate, convert, rephrase), Reasoning (explain, justify, argue, compare, contrast), Generation (write, draft, create, generate, compose), Classification (classify, categorize, label, tag, sort), Editing (edit, refactor, fix, polish, rewrite), Critique (critique, audit, check, validate, verify)।
এই ~৪০ verb মুখস্থ থাকলে প্রতিটা prompt-এর প্রথম word-টা নিজে-নিজে আসে — "কী verb use করব" ভাবতে হবে না। ৩০ second-এ clear prompt লেখা সম্ভব।
Specificity ladder5-step climb from vague to production-ready
Vague প্রম্পট-কে production-ready করার ৫টা ধাপ। প্রতি ধাপে একটা specificity marker যোগ: Step 1 — verb দিয়ে শুরু ("Summarize this.")। Step 2 — input scope যোগ ("this article")। Step 3 — length constraint যোগ ("in 3 bullet points")। Step 4 — focus filter যোগ ("focusing on financial implications")। Step 5 — audience + format যোগ ("for an executive audience. Output as markdown.")। প্রতি step measurably output quality improve করে।
Reader-কে "vague প্রম্পট কীভাবে production-grade হবে" সেটার একটা mechanical recipe দেয়। Ladder মনে রাখলে তুমি কখনো "কী যোগ করব?" বলতে আটকাবে না।
নিচে ladder-টা নিজে চালিয়ে দেখো — এক ধাপ এক ধাপ "Add next marker" চেপে একটা vague প্রম্পট-কে production-ready বানাও।
Structural markersthe phrase fragments
Verb-এর সাথে যোগ করার ছোট ছোট English phrase — যেগুলো natural language-কে precise instruction-এ convert করে। ৮ category-তে organized: Output format ("Output as JSON"), Length ("in 3 bullet points"), Tone ("in a professional tone"), Audience ("for an executive audience"), Persona ("Act as a senior reviewer"), Negation ("Do not include speculation"), Uncertainty ("If unsure, say 'I don't know'"), No preamble ("Output directly. No preamble.")। মোট ~২৫টা reusable phrase।
এই phrasebook তোমার পকেট-reference। যেকোনো prompt-এ marker-গুলো combine করে লাগানো যায়। Lego block-এর মতো — কয়েকটা block জানলেই অসংখ্য combination বানানো যায়।
The politeness trapthe Bengali-speaker anti-pattern
আমরা Bengali থেকে English-এ politeness translate করি instinctively — "একটু" → "could you please possibly", "কষ্ট হলে" → "if you don't mind", "একটু চেষ্টা করেন" → "try to"। প্রতিটা production prompt-এ এই hedging weakening signals। ৬টা common trap: (১) "Could you please possibly", (২) "Can you" (question form), (৩) "It would be great if", (৪) "I was wondering if maybe", (৫) "Try to", (৬) "If you don't mind"। সবগুলোর fix: imperative + specific।
এটা hidden barrier। Bengali speaker-রা বুঝতেই পারে না কেন তাদের প্রম্পট weak output produce করছে — অথচ সমস্যা grammar-এ না, politeness-এ। এই trap-গুলো একবার চিনলে production-quality writing instinct হয়ে যায়।
No preamble tricksave tokens, get cleaner output
Model default-এ introduction জুড়ে দেয়: "Sure! Here's the summary you asked for:" — তারপর আসল content। সেই preamble অকেজো আর ~৩০-৮০ token নষ্ট করে। Fix: prompt-এর শেষে যোগ করো — "Output the result directly. No preamble." — বা "Begin your response with the answer itself." এক phrase, production-এ বড় savings।
Production scale-এ (১০,০০০+ calls/day) এই preamble strip token cost + latency দু'টোই কমায়। আর downstream parsing easier — "Sure! Here's..." strip করতে হয় না।
The 30-second promptthe goal
Day 6-এর promise: ৮ verb family + progressive-specificity ladder + ~২৫টা structural marker + politeness fixes — সব internalize হলে তুমি ৩০ second-এ যেকোনো clear production prompt লিখতে পারবে। "আমি English-এ slow" — এই অজুহাত আজকের পর আর চলবে না।
এটাই Day 6-এর ultimate deliverable — speed আর confidence। Module 2-এর বাকি lesson (eval harness, CoT, tool use) — সব তুমি প্রম্পট লিখতে পারলেই সম্ভব। এই vocabulary ছাড়া প্রতিটা Day ৫× বেশি সময় নেবে।
চারটা জিনিস আজই apply করো — (১) verb দিয়ে prompt শুরু করো, প্রশ্ন দিয়ে না; (২) scope → length → focus → audience → format — এই ৫ ধাপ ladder মনে রাখো; (৩) politeness ছেড়ে direct imperative লেখো — Bengali-instinct hedging বাদ; (৪) prompt-এর শেষে "No preamble" যোগ করো। চারটাই আজকের lesson-এর সরাসরি ফল।
Try It Yourself
পড়া বন্ধ — এবার নিজের হাতে।
পড়া শেষ — এবার হাতে-কলমে। প্রথম demo — politeness trap। Bengali-instinct hedging আর তার production fix পাশাপাশি দেখো, token-count পার্থক্যও।
মার্কার-গুলো নিজে toggle করে দেখো — প্রতিটা addition প্রম্পট-কে কীভাবে shape দেয়। দুইটা marker (verb + scope) দিয়েই usable প্রম্পট; সব ৮টা active করলে production-grade।
শেষ demo — পুরো phrasebook। ৮ category, ~২৫টা copy-paste-ready English phrase। এই page bookmark করে রাখো — পরের lesson-গুলোতে বারবার লাগবে।
Check Your Understanding
ভুল হলে সমস্যা নেই — প্রতিটার ব্যাখ্যা আছে।
Polite English-এ লিখলে model better respond করে
জটিল English use করলে professional শোনায় — output ভালো হবে
Question form ("Can you summarize?") imperative-এর সমান
Prompt vocabulary memorize করা rote learning — overrated
Bengali-তে prompt লিখে model-কে translate করতে বললে সহজ
"No preamble" trick gimmick — real impact নেই
"Try to" বা "Please" prompt-কে polite করে — সমস্যা নেই
নিচের কোনটা সবচেয়ে production-ready phrasing?
ভুল হলেও সমস্যা নেই — প্রতিটা option-এর সাথে ব্যাখ্যা আছে।
৮টা verb family-র মধ্যে "refactor" কোন family-তে?
ভুল হলেও সমস্যা নেই — প্রতিটা option-এর সাথে ব্যাখ্যা আছে।
Progressive-specificity ladder-এর Step 2-এ কী যোগ করি?
ভুল হলেও সমস্যা নেই — প্রতিটা option-এর সাথে ব্যাখ্যা আছে।
"No preamble" instruction কেন save token করে?
ভুল হলেও সমস্যা নেই — প্রতিটা option-এর সাথে ব্যাখ্যা আছে।
Bengali speaker-দের জন্য সবচেয়ে common politeness trap কোনটা?
ভুল হলেও সমস্যা নেই — প্রতিটা option-এর সাথে ব্যাখ্যা আছে।
একটা LLM-কে "act as a senior code reviewer" বলা — এটা কোন category-র phrase?
ভুল হলেও সমস্যা নেই — প্রতিটা option-এর সাথে ব্যাখ্যা আছে।
Prove It To Yourself
চারটা ধারণা মনে রাখো। নিজের ভাষায় লেখো। Day 6 শেষ।
চারটা ধারণা মনে রাখো — Imperative verb, প্রথম word: "Summarize" not "Can you summarize"। ৮ family-তে ~৪০টা verb: Analysis, Extraction, Transformation, Reasoning, Generation, Classification, Editing, Critique — মুখস্থ থাকলে "কী verb use করব" আর প্রশ্ন থাকে না। Specificity ladder ৫ ধাপে climb করো: verb → scope → length → focus → audience + format। আর phrasebook + "No preamble" — ~২৫টা reusable phrase পকেটে রাখো, প্রতি prompt-এ lego block-এর মতো combine করো।
আজকের শব্দগুলো — action verb, imperative, specificity ladder, structural marker, persona, negation, uncertainty handling, no preamble, phrasebook, politeness trap — এখন থেকে English-ই থাকবে; team meeting, PR review, নিজের prompt লেখার সময় এভাবেই দেখবে। কাল কেউ "no preamble" বললে তুমি জানো সেটা কী। পুরো তালিকা নিচের glossary-তে।
শব্দার্থ — পরে ফিরে দেখার জন্য
একনজরে সব key term। কোনো শব্দ ভুলে গেলে এখানে এসে খুঁজে নিও।
- Act as
- Persona declaration-এর common phrasing — "Act as a senior engineer।" System prompt-এ frequent।
- Action verb
- Prompt-এর প্রথম word — model কী করবে সেটা specify করে। ~৪০টা production verb ৮ family-তে organized।
- Analysis verbs
- analyze, examine, evaluate, assess, review — existing জিনিসের structure/quality বোঝার জন্য।
- Audience marker
- "for an executive audience", "for a 6th grader" — output-এর vocabulary + complexity calibrate করে।
- Classification verbs
- classify, categorize, label, tag, sort — items-কে buckets-এ assign।
- Critique verbs
- critique, audit, check, validate, verify — correctness বা quality judge।
- Editing verbs
- edit, refactor, fix, polish, rewrite — existing content improve।
- Extraction verbs
- extract, identify, list, find, pull out — বড় text থেকে specific data বের করা।
- Focus filter
- "focusing on financial implications" — Generic output-এর বদলে purposeful angle।
- Format specifier
- "Output as JSON", "Return in markdown" — output shape define।
- Generation verbs
- write, draft, create, generate, compose — নতুন content production।
- Hedging language
- "maybe", "perhaps", "try to", "could", "possibly" — uncertainty signals যা prompts-এ weakening। Production-এ avoid।
- Imperative
- Direct command form — "Summarize." not "Can you summarize?" Production prompt-এর default mood।
- Length constraint
- "in 3 bullet points", "under 100 words" — output size bound করে।
- Negation
- "Do not include speculation" — explicit boundary, কী model করবে না।
- No preamble
- "Output directly. No preamble." — model-এর default introduction ("Sure! Here's...") strip করে। ~30-80 token save।
- Persona
- Role assignment — "You are a senior reviewer।" System prompt-এ best fit।
- Phrasebook
- Reusable English phrase fragment-এর categorized collection — output format, length, tone, audience, persona, negation, uncertainty, no-preamble। ~২৫টা phrase।
- Politeness trap
- Bengali-speaker-specific anti-pattern: "Could you please possibly try to..." — hedging stack যা prompt weaken করে।
- Progressive specificity
- Vague প্রম্পট-কে production-ready বানানোর ৫ ধাপ — verb → scope → length → focus → audience/format।
- Reasoning verbs
- explain, justify, argue, compare, contrast — show the why or the how।
- Tone calibrator
- "in a professional tone", "warm but neutral" — output voice/style adjust।
- Transformation verbs
- summarize, paraphrase, translate, convert, rephrase — meaning preserve করে form change।
- Uncertainty handling
- "If unsure, say 'I don't know'" — confabulation prevent, refusal permission।
Sources consulted to author this lesson. Citation style is informal — follow the links if you want to dig deeper.
- Prompting best practices — Be clear and direct — Anthropic (2025)
- Best practices for LLM prompt engineering — Palantir Foundry team (2024)
- 10 Best Practices for Production-Grade LLM Prompt Engineering — Latitude (2024)
- Prompt engineering — Write clear instructions — OpenAI (2024)
- 12 Prompt Engineering Best Practices for Reliable LLM Output — Brilworks (2024)
দুই বাক্যে নিজের ভাষায় লেখো
নিজের ভাষায় দুই বাক্যে লেখো — ৮টা verb family-র নাম কী কী, আর progressive-specificity ladder-এর ৫টা ধাপ কী কী।