Localize Bn

Translate to Bengali (bn → বাংলা) — High-Fidelity Skill

Scope & Variants

Bengali (বাংলা / Bāṅlā) is the official language of Bangladesh and the second-most-spoken language in India (West Bengal, Tripura, parts of Assam). It is written in the Eastern Nagari script (also called Bengali-Assamese script), an LTR abugida derived from Brahmi.

~270 million speakers worldwide (~170M Bangladesh + ~100M India + diaspora). The 7th most-spoken language in the world.

| Locale | Name | Speakers | Notes | |--------|------|----------|-------| | bn-BD | Bangladesh Standard | ~170M | Official language of Bangladesh. Heavier Arabic/Persian loanword influence. Currency: BDT (৳ / Tk). | | bn-IN | Indian Bengali (West Bengal / Tripura) | ~100M | One of India's 22 scheduled languages. Heavier Sanskrit/English loanword influence. Currency: INR (₹). |

Both write THE SAME Bengali language in THE SAME script — fully mutually intelligible — but have systematic differences in:

| Axis | bn-BD (Bangladesh) | bn-IN (India / West Bengal) | |------|--------------------|------------------------------| | Loanword bias | Arabic / Persian / Urdu (esp. for everyday concepts) | Sanskrit (তৎসম) / English | | "Water" | পানি (pani — Persian-origin, dominant in BD) | জল (jol — Sanskrit-origin, dominant in WB) | | "Salt" | নুন / লবণ | নুন (more common) | | "Invitation" | দাওয়াত (Arabic-origin) | নিমন্ত্রণ (Sanskrit) | | Greeting | আসসালামু আলাইকুম (Muslim) / নমস্কার (rare) | নমস্কার (default) / সালাম (Muslim minority) | | "Bath / shower" | গোসল (gosol — Arabic) | স্নান (snan — Sanskrit) | | "Help" | মদদ (madad — Persian) / সাহায্য | সাহায্য (default) | | Spelling reforms | More phonetic, fewer silent letters retained | More etymological / Sanskritic spellings | | Currency | ৳ / Tk / টাকা (BDT taka) | ₹ / টাকা (INR rupee) | | Date | DD/MM/YYYY (both) | DD/MM/YYYY (both) | | Religious calendar overlay | Bengali calendar (বঙ্গাব্দ) + Islamic Hijri sometimes | Bengali calendar (বঙ্গাব্দ) + Hindu Saka/Vikram | | Phone | +880 country code | +91 country code | | Time-of-day greetings | শুভ সকাল / শুভ দুপুর | Same |

On first use, ASK the user:

> Which Bengali variant should I target? > - bn-BD (Bangladesh — Muslim-majority context, BDT taka, পানি/দাওয়াত vocabulary) > - bn-IN (West Bengal / Tripura, India — secular/Hindu-majority context, INR rupee, জল/নিমন্ত্রণ vocabulary) > - Pan-Bengali neutral (technical content for both audiences — minimize region-specific lexical choices, use English loanwords as common ground)

For TECH PRODUCT UI (the most common case): the lexical drift between bn-BD and bn-IN is much smaller because both share English loanwords for technical concepts (ক্লিক, সেভ, ডাউনলোড, লগইন, ফাইল). English tech loans serve as common ground. The differences mostly surface in marketing copy, idioms, religious framing, and currency. If the surface is just buttons/labels/errors, pan-Bengali neutral is usually safe.

---

Identity Guardrail — Bengali Is NOT Hindi (and NOT Assamese either, though close)

Bengali is Indo-Aryan (a daughter branch of Indo-European), descended from Magadhi Prakrit through Eastern Apabhraṃśa. Its closest living sister language is Assamese (অসমীয়া) — they share a writing system (Eastern Nagari) with only two letter differences and are roughly 70% mutually intelligible. Bishnupriya Manipuri, Chittagonian, Sylheti, Rohingya, and Rangpuri/Rajbanshi are close relatives often classified as dialects or sister languages.

| Often-confused with | Why it's NOT Bengali | |---------------------|------------------------| | Hindi (हिन्दी) | DIFFERENT script (Devanagari देवनागरी vs Bengali বাংলা), DIFFERENT grammar (Hindi HAS grammatical gender — masculine/feminine — Bengali does NOT), different verb conjugation patterns, different number system look. Sister branch within Indo-Aryan but not mutually intelligible in writing. Many Bengalis read/understand spoken Hindi via Bollywood exposure but they are not the same language. | | Urdu (اردو) | Different script entirely (Perso-Arabic, RTL). Linguistically sister to Hindi. Bengali has zero structural overlap with Urdu beyond shared Persian/Arabic loans in bn-BD. | | Assamese (অসমীয়া) | THE closest relative. Same script family (Eastern Nagari), ~70% mutual intelligibility. Distinguishing letters: Assamese has ৰ (different from Bengali র) and ৱ (Bengali uses ব). Different verb agreement patterns. Do NOT confuse a Bengali translation with an Assamese one — they look almost identical but are distinct standards. | | Sanskrit (संस्कृत) | Bengali borrows heavily from Sanskrit (তৎসম "same as that" loanwords), but Sanskrit is the ancient liturgical ancestor, not modern Bengali. Sanskritic spelling preserved in formal Bengali but pronunciation drifted significantly. | | Odia / Oriya (ଓଡ଼ିଆ) | Different script (Odia uses curvilinear forms). Geographic neighbor (Odisha state, India). Related but distinct Indo-Aryan language. |

Critical identity markers — Bengali has these, Hindi/Urdu do not:

1. NO grammatical gender — ছেলেটি লম্বা (the boy is tall) and মেয়েটি লম্বা (the girl is tall) use the SAME adjective form লম্বা. Hindi would require लंबा (m.) vs लंबी (f.).
2. Honorific verb conjugation, NOT gender conjugation — verbs change for তুই / তুমি / আপনি honorific level, not for subject's gender.
3. Single 3rd-person pronoun for all humans — সে (she/he, no gender) and তিনি (formal s/he, no gender). Hindi distinguishes वह (with verb agreement gendered).
4. Bengali script (Eastern Nagari) — distinct from Devanagari. The Bengali letter অ resembles a different glyph from Devanagari अ.
5. Inherent vowel ô (অ) — pronounced as /ɔ/ in Bengali, NOT /ə/ as in Hindi.
6. 3 sibilants merge to one — Bengali script preserves শ, ষ, স but all pronounced /ʃ/ (sh-sound), unlike Hindi which distinguishes them.
7. No retroflex /ʈ/-vs-/ʈʰ/-vs-/ɖ/-vs-/ɖʱ/ contrast in pronunciation for many speakers (script preserves the contrast).

When the source mentions "translate to Indian languages" or "South Asian languages," confirm with user which language — DO NOT assume Bengali = Hindi.

---

Register — Three-Tier Honorific System

Bengali has THREE honorific levels baked into verb conjugation. Unlike European T/V distinctions (just 2), Bengali distinguishes:

| Pronoun | Register | When used | Example verb "to do" (present) | Example verb "to do" (imperative) | |---------|----------|-----------|--------------------------------|-----------------------------------| | তুই (tui) | Intimate / very informal | Close childhood friends, siblings (in some families), children, God in prayer, animals | তুই করিস (kôris) | তুই কর (kôr) — abrupt | | তুমি (tumi) | Familiar / peer | Friends, peers, younger relatives, spouse, children (in some families) | তুমি কর (kôro) | তুমি কর / করো (kôro) | | আপনি (apni) | Respectful / polite | Strangers, elders, professionals, ALL SOFTWARE UI | আপনি করেন (kôren) | আপনি করুন (kôrun) |

Possessives match the register:

| English | তুই (intimate) | তুমি (peer) | আপনি (polite — DEFAULT) | |---------|----------------|--------------|---------------------------| | your | তোর (tor) | তোমার (tomar) | আপনার (apnar) |

Default for ALL product UI, marketing, docs: আপনি (apni) — respectful register. Using তুমি or তুই in software UI is rude, condescending, or alien — sometimes intentionally for hyper-casual brands targeting Gen Z, but generally avoided.

Imperative endings — the most common UI form

| English | আপনি form (DEFAULT for UI) | তুমি form (avoid in UI) | তুই form (NEVER in UI) | |---------|------------------------------|--------------------------|--------------------------| | Click | ক্লিক করুন (klik korun) | ক্লিক কর / করো | ক্লিক কর | | Save | সেভ করুন | সেভ কর / করো | সেভ কর | | Delete | মুছুন (muchhun) / মুছে ফেলুন | মোছ / মুছে ফেল | মোছ | | Open | খুলুন (khulun) | খোল / খুলো | খোল | | Close | বন্ধ করুন (bondho korun) | বন্ধ কর | বন্ধ কর | | Submit | জমা দিন (jôma din) | জমা দাও | জমা দে | | Send | পাঠান (pathan) | পাঠাও | পাঠা | | Cancel | বাতিল করুন (batil korun) | বাতিল কর | বাতিল কর | | Continue | চালিয়ে যান (chaliye jan) | চালিয়ে যাও | চালিয়ে যা | | Try again | আবার চেষ্টা করুন | আবার চেষ্টা কর | আবার চেষ্টা কর | | Sign in | সাইন ইন করুন / লগইন করুন | সাইন ইন কর | সাইন ইন কর |

Hard rule: NEVER mix registers in the same UI

Mixing আপনি verbs with তুমি possessives (or vice versa) is the #1 Bengali UI failure.

| ✗ Wrong (mixed) | ✓ Correct (consistent আপনি) | Issue | |------------------|------------------------------|-------| | আপনি ক্লিক কর | আপনি ক্লিক করুন | আপনি pronoun + তুমি verb | | তোমার পাসওয়ার্ড দিন | আপনার পাসওয়ার্ড দিন | তুমি possessive + আপনি verb | | তোমার অ্যাকাউন্ট সেভ করুন | আপনার অ্যাকাউন্ট সেভ করুন | Mixed possessive | | আপনার ফাইল মুছে ফেল | আপনার ফাইল মুছে ফেলুন | আপনি possessive + তুমি verb |

Check EVERY verb, EVERY possessive, EVERY pronoun for consistency.

Register auto-detect from source

| Source signal | Target register | |---------------|-----------------| | Banking, legal, government, healthcare, enterprise, default product UI | আপনি (apni) — respectful, imperative -ুন/-ান | | Marketing aimed at Gen Z, casual social apps, gaming, friend-to-friend chat | তুমি (tumi) — explicit user decision, document it | | Religious texts (addressing the Divine), poetry, literary translation | তুই (intimate) — only when source matches |

When in doubt: pick আপনি. It is NEVER wrong for UI.

---

Critical Hard Rules (Severity ≥ 2.0 — Auto-Fail If Wrong)

1. Script integrity — Bengali script (Eastern Nagari) throughout (severity 2.5)

• All Bengali text in Bengali script, NOT romanization, NOT Devanagari (Hindi), NOT Assamese (which differs by two letters).
• Latin tech identifiers (Git, API, JSON, URLs, code, file names, placeholders {var}, {{count}}) MUST stay in Latin script, embedded LTR within LTR Bengali (no bidi complexity since Bengali is also LTR).
• Conjunct consonants (যুক্তাক্ষর) must be correct:
• ক + ষ = ক্ষ (kkh / kṣ) — as in পরীক্ষা (exam), লক্ষ্য (goal)
• জ + ঞ = জ্ঞ (gg / jñ) — as in জ্ঞান (knowledge), বিজ্ঞান (science)
• ন + ত = ন্ত — as in অনন্ত (eternal)
• ত + র = ত্র — as in ত্রিভুজ (triangle)
• ষ + ণ = ষ্ণ — as in কৃষ্ণ (Krishna)
• হ + ম = হ্ম — as in ব্রহ্ম (Brahma)
• শ + চ = শ্চ — as in নিশ্চয় (certainly)
• শ + র = শ্র — as in শ্রী, আশ্রয়
• Common typing errors: writing the conjunct as two separate letters with halant (e.g., ক্‌ষ instead of ক্ষ). Always render as the proper conjunct glyph.
• Halant / hôshonto (্) suppresses the inherent vowel — used when a consonant is final or part of a conjunct.
• Inherent vowel অ (ô) — every Bengali consonant carries an inherent /ɔ/ sound unless suppressed. So ক alone = "kô," not just "k."
• Vowel signs (কার চিহ্ন) attach to consonants in specific positions (some before: ি, কি; some after: া, কা; some above: ে, কে; some below: ু, কু; some around: ো, কো, ৌ, কৌ).

2. Chandrabindu (ঁ) vs Anusvara (ং) — different sounds (severity 2.0)

These two nasal markers are NOT interchangeable.

| Marker | Name | Sound | Example | |--------|------|-------|---------| | ঁ | চন্দ্রবিন্দু (chandrabindu, "moon-dot") | Nasalizes the preceding vowel (~), no /n/ or /ŋ/ consonant | চাঁদ (chãd — moon), কাঁচা (kãcha — raw) | | ং | অনুস্বার (anusvara) | /ŋ/ velar nasal consonant | বাংলা (Baŋla — Bengali), সংগীত (sôŋgit — music) | | ঃ | বিসর্গ (bisarga) | /h/ aspiration, rare in modern Bengali | দুঃখ (duḥkho — sorrow) |

| ✗ Wrong | ✓ Correct | Reason | |---------|-----------|--------| | বাংলা spelled বাঁলা | বাংলা | "Bengali" uses anusvara (ŋ-sound), not chandrabindu | | চাঁদ spelled চাংদ | চাঁদ | "Moon" has nasalized vowel, not /ŋ/ | | ইংরেজি spelled ইঁরেজি | ইংরেজি | "English (language)" uses anusvara |

3. SOV word order — verb at the end (severity 2.0)

Bengali is Subject-Object-Verb, unlike English's SVO.

| English (SVO) | ✗ SVO calque | ✓ Bengali SOV | |---------------|--------------|------------------| | The user saves the file | ইউজার সেভ করে ফাইল | ইউজার ফাইল সেভ করে | | Click the button to continue | ক্লিক করুন বোতামে চালিয়ে যেতে | চালিয়ে যেতে বোতামে ক্লিক করুন OR চালিয়ে যেতে বোতামটি ক্লিক করুন | | I download the file from the server | আমি ডাউনলোড করি ফাইল সার্ভার থেকে | আমি সার্ভার থেকে ফাইল ডাউনলোড করি | | Open settings to change your password | খুলুন সেটিংস পাসওয়ার্ড পরিবর্তন করতে | পাসওয়ার্ড পরিবর্তন করতে সেটিংস খুলুন |

Modifiers and adjuncts (time, place, manner) come BEFORE the verb. Adjectives come BEFORE the noun (নতুন ফাইল — new file).

4. Case-marking postpositions — attached WITHOUT space or hyphen (severity 2.5)

Bengali expresses case relations via postpositional suffixes attached directly to nouns. NO space, NO hyphen between noun and postposition.

| Case | Postposition | Example | |------|--------------|---------| | Nominative | (none) | ফাইল (file) — subject | | Accusative / Dative | -কে (-ke) | ইউজারকে (to the user), ফাইলটিকে (to the file) | | Genitive | -র / -এর (-r / -er) | ইউজারের (user's), ফাইলের (file's), আমার (my) | | Locative | -তে / -এ (-te / -e) | ফাইলে (in the file), ঘরে (in the room), টেবিলে (on the table) | | Ablative | থেকে (theke — separate word) | সার্ভার থেকে (from the server) | | Instrumental | দিয়ে / দ্বারা (diye / dwara) | চাবি দিয়ে (with the key) |

| ✗ Wrong (space/hyphen) | ✓ Correct (attached) | Issue | |------------------------|----------------------|-------| | ফাইল এ | ফাইলে | Locative -এ attaches directly | | ইউজার এর | ইউজারের | Genitive -এর attaches directly | | সিস্টেম-এর | সিস্টেমের | NO hyphen for native postposition | | ফাইল কে | ফাইলকে | Accusative -কে attaches directly | | ফোল্ডার -তে | ফোল্ডারে OR ফোল্ডারেতে | -তে / -এ attaches without hyphen | | অ্যাকাউন্ট এ লগইন করুন | অ্যাকাউন্টে লগইন করুন | Sign in to the account |

Exception: When the noun ends in an English-script word (e.g., Git, API), insert a hyphen for readability:

| ✓ Acceptable | |--------------| | Git-এ পুশ করুন (push to Git) | | API-তে অনুরোধ পাঠান (send a request to the API) | | GitHub-এর মাধ্যমে (via GitHub) |

5. Classifiers REQUIRED with numbers (severity 2.0)

Bengali requires a classifier (পরিমাণবাচক / শ্রেণীবাচক) between numbers and the noun being counted. This is similar to Chinese/Japanese measure words. Numbers without a classifier sound broken.

| Classifier | Used for | Example | |------------|----------|---------| | -টি / -টা | General objects (most common; -টি slightly more formal, -টা more colloquial) | একটি ফাইল (one file), পাঁচটি ফোল্ডার (five folders), দুটো বই (two books — colloquial) | | -জন | Humans (only for people) | একজন ইউজার (one user), পাঁচজন সদস্য (five members) | | -খানা / -খানি | Flat objects, certain household items, sheets of paper, books | একখানা চিঠি (one letter), দুখানা বই (two books — slightly old-fashioned) | | -গাছা / -গাছি | Long thin objects (sticks, ropes, hairs) | একগাছা চুল (one strand of hair) | | -টুকু / -টুকুনি | Small amount, abstract quantity | একটুকু জল (a little water) | | -পাটি | Pairs | একপাটি জুতা (a pair of shoes) |

| ✗ Wrong (missing classifier) | ✓ Correct (with classifier) | |-------------------------------|------------------------------| | তিন ফাইল | তিনটি ফাইল | | পাঁচ ইউজার | পাঁচজন ইউজার | | ১০ ভাষা | ১০টি ভাষা OR দশটি ভাষা | | দুই ফোল্ডার | দুটি ফোল্ডার OR দুটো ফোল্ডার | | ৭ সদস্য | ৭ জন সদস্য |

Number 1 has a special classifier-fused form: এক + টি = একটি (e-k-ti) is the standard "one [thing]."

**Spe

…

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: i18n-agent
• Source: i18n-agent/i18nstack
• License: MIT
• Homepage: https://i18nagent.ai

Install and usage instructions live in the source repository linked above.