Internationalizing bots: language detection, message catalogs, plurals, formats and right-to-left text

Decide the user's language

Platforms give hints:

• Telegram: from.language_code (an IETF tag such as es or pt-br), when the user's client shares it.
• Discord: locale on interactions (the user's client language) and guild_locale for community servers.
• Slack: the user's locale is available through users.info with include_locale=true.
• WhatsApp and Messenger: no reliable language field; infer from the phone prefix with care or ask.

Use the hint as a default, offer a /language command, and store the choice. A hint is not a promise: a person's interface language may not be the language they want to chat in.

Message catalogs

Keep strings out of code, keyed by ID:

# en.ftl (Project Fluent syntax)
items-in-cart = { $count ->
    [one] You have one item in your cart.
   *[other] You have { $count } items in your cart.
}

• Use a format with plural rules, such as ICU MessageFormat or Project Fluent. Languages have different plural categories (for example, Arabic has six, Japanese one); Intl.PluralRules exposes them.
• Use named placeholders, not string concatenation, because word order changes between languages.
• Give translators context: where the text appears and its length limit.
• Fall back to a default language when a key is missing, and report missing keys in tests.

Localize the bot's surface

• Telegram: call setMyCommands with language_code for each language; users see the menu in their app language. The command list builder produces the JSON. Bot name, description and short description can also be set per language.
• Discord: application commands support name_localizations and description_localizations. See the Discord command builder.
• Buttons and keyboards need translating too, and button text must still fit.

Formats

• Dates, numbers and currencies: use Intl.DateTimeFormat, Intl.NumberFormat and Intl.RelativeTimeFormat (or Babel in Python) with the user's locale.
• Time: store UTC, display in the user's time zone. Plan schedules with the time zone scheduler.
• Discord renders <t:UNIX:F> timestamps in each reader's local time; use them in shared channels. The timestamp converter builds them.

Length and direction

• Translations are often longer than English, so check limits on buttons, embeds and messages with the message splitter.
• Arabic, Hebrew and Persian are right-to-left. Mixed text with numbers or Latin names can display out of order; Unicode isolates (U+2068 and U+2069) around inserted values help.

LLM bots

Tell the model which language to answer in based on stored preference, not on the last message alone, and evaluate quality per language: models are uneven across languages.