Rendering i18n messages

Optionally, you can structure your messages as nested objects.

{
  "auth": {
    "SignUp": {
      "title": "Sign up",
      "form": {
        "placeholder": "Please enter your name",
        "submit": "Submit"
      }
    }
  }
}

import {useTranslations} from 'next-intl';
 
function SignUp() {
  // Provide the lowest common denominator that contains
  // all messages this component needs to consume.
  const t = useTranslations('auth.SignUp');
 
  return (
    <>
      <h1>{t('title')}</h1>
      <form>
        <input
          // The remaining hierarchy can be resolved by
          // using `.` to access nested messages.
          placeholder={t('form.placeholder')}
        />
        <button type="submit">{t('form.submit')}</button>
      </form>
    </>
  );
}

As your app grows, you’ll want to reuse messages among your components. If you use component names as namespaces to structure your messages, you’ll automatically benefit from reusable messages by reusing your components.

Examples:

• You’re displaying products in your app and often need to resolve a category identifier to a human readable label (e.g. new → “Just in”). To ensure consistency, you can add a ProductCategory component that turns the category into a styled badge.
• You commonly need a dialog that displays a “confirm” and “cancel” button. In this case, consider adding a ConfirmDialog component to reuse the messages along with the functionality.

There might be cases where you want to use the same message in different components, but there’s no reasonable opportunity for sharing the message via a component. This might be symptom that these components should use separate messages, even if they happen to be the same for the time being. By using separate labels, you gain the flexibility to use more specific labels (e.g. “not now” instead of “cancel”).

For edge cases where the reuse of messages among different components is hard to achieve with shared components, you can either use messages from different namespaces via useTranslations(), or you can consider adding a hook that shares the translation behavior:

export default function useLocaleLabel() {
  const t = useTranslations('useLocaleLabel');
 
  function getLocaleLabel(locale: 'en' | 'de') {
    return t('label', {locale});
  }
 
  return getLocaleLabel;
}

{
  "useLocaleLabel": {
    "label": "{locale, select, en {English} de {German} other {Unknown}}"
  }
}

next-intl is heavily based on the useTranslations API which is intended to consume translations from within React components. This may seem like a limitation, but this is intentional and aims to encourage the use of proven patterns that avoid potential issues—especially if they are easy to overlook.

If you’re interested to dive deeper into this topic, there’s a blog post that discusses the background of this design decision: How (not) to use translations outside of React components.

There’s one exception however: Using next-intl with Server Actions, Metadata and Route Handlers.

Namespace keys can not contain the character ”.” as this is used to express nesting—all other characters are fine to use.

If you’re migrating from a flat structure that uses ”.” in message keys, you can convert your messages to a nested structure as follows:

import {set} from 'lodash';
 
const input = {
  'one.one': '1.1',
  'one.two': '1.2',
  'two.one.one': '2.1.1'
};
 
const output = Object.entries(input).reduce(
  (acc, [key, value]) => set(acc, key, value),
  {}
);

Output:

{
  "one": {
    "one": "1.1",
    "two": "1.2"
  },
  "two": {
    "one": {
      "one": "2.1.1"
    }
  }
}

This keeps the hierarchy while removing the redundancy of repeated parent keys. You can either execute this is in a one-off script or before passing your messages to next-intl.

If you’ve previously used human readable sentences for keys, you can theoretically map the . character to a different one (e.g. _) before passing the messages to next-intl, but it’s generally recommended to use IDs as keys. You can still view human readable labels from your messages in your editor by using a VSCode integration in case that’s your primary motivation for using human readable keys.

Value names are required to be alphanumeric and can contain underscores. All other characters, including dashes, are not supported.

Languages have different rules for pluralization and your messages should reflect these rules.

For example, English has two forms: One for the singular (e.g. “1 follower”) and one for everything else (e.g. “0 followers” and “2 followers”). In contrast, Chinese only has one form, while Arabic has six.

The other aspect to consider is that from a usability perspective, it can be helpful to consider additional cases on top—most commonly the special case for zero (e.g. “No followers yet” instead of “0 followers”).

While you can always use the =value syntax to match a specific number (e.g. =0), you can choose from the following tags depending on what grammar rules apply to a given language:

• zero: For languages with zero-item grammar (e.g., Latvian, Welsh).
• one: For languages with singular-item grammar (e.g., English, German).
• two: For languages with dual-item grammar (e.g., Arabic, Welsh).
• few: For languages with grammar specific to a small number of items (e.g., Arabic, Croatian).
• many: For languages with grammar specific to a larger number of items (e.g., Arabic, Croatian).
• other: Used when the value doesn’t match other plural categories.

next-intl uses Intl.PluralRules to detect the correct tag that should be used for a given number. The exact range that few and many applies to varies depending on the locale (see the language plural rules table in the Unicode CLDR).

Depending on the language, different forms of ordinal pluralization are supported.

For example, English has four forms: “th”, “st”, “nd” and “rd” (e.g. 1st, 2nd, 3rd, 4th … 11th, 12th, 13th … 21st, 22nd, 23rd, 24th, etc.). In contrast, both Chinese and Arabic use only one form for ordinal numbers.

next-intl uses Intl.PluralRules to detect the correct tag that should be used for a given number.

These tags are supported:

• zero: For languages with zero-item grammar (e.g., Latvian, Welsh).
• one: For languages with singular-item grammar (e.g., English, German).
• two: For languages with dual-item grammar (e.g., Arabic, Welsh).
• few: For languages with grammar specific to a small number of items (e.g., Arabic, Polish, Croatian).
• many: For languages with grammar specific to a larger number of items (e.g., Arabic, Polish, Croatian).
• other: Used when the value doesn’t match other plural categories.

The exact range that few and many applies to varies depending on the locale (see the language plural rules table in the Unicode CLDR).

To match a specific number, next-intl additionally supports the special =value syntax (e.g. =3) that always takes precedence.

Values are required to be alphanumeric and can contain underscores. All other characters, including dashes, are not supported.

Therefore, e.g. when you’re mapping a locale to a human readable string, you should map the dash to an underscore first:

"label": "{locale, select, en_GB {British English} en_US {American English} other {Unknown}}"

const locale = 'en-GB';
t('message', {locale: locale.replaceAll('-', '_')});

Common tags for rich text that you want to share across your app can be defined in a shared module and imported where relevant for usage with t.rich.

A convenient pattern is to use a component that provides common tags via a render prop:

import {useTranslations} from 'next-intl';
import RichText from '@/components/RichText';
 
function AboutPage() {
  const t = useTranslations('AboutPage');
  return <RichText>{(tags) => t.rich('description', tags)}</RichText>;
}

In this case, the RichText component can provide styled tags and also a general layout for the text:

import {ReactNode} from 'react';
 
// These tags are available
type Tag = 'p' | 'b' | 'i';
 
type Props = {
  children(tags: Record<Tag, (chunks: ReactNode) => ReactNode>): ReactNode
};
 
export default function RichText({children}: Props) {
  return (
    <div className="prose">
      {children({
        p: (chunks: ReactNode) => <p>{chunks}</p>,
        b: (chunks: ReactNode) => <b className="font-semibold">{chunks}</b>,
        i: (chunks: ReactNode) => <i className="italic">{chunks}</i>
      })}
    </div>
  );
}

If you need to combine shared tags with values from your component, you can merge them accordingly by using the spread operator:

function UserPage({username}) {
  const t = useTranslations('UserPage');
  return (
    <RichText>{(tags) => t.rich('description', {...tags, username})}</RichText>
  );
}

Messages can use tags without any chunks as children, but syntax-wise, a closing tag is required by the ICU parser:

{
  "message": "Hello,<br></br>how are you?"
}

t.rich('message', {
  br: () => <br />
});

Attributes can only be set on the call site, not within messages:

{
  "message": "Go to <profile>my profile</profile>"
}

t.rich('message', {
  profile: (chunks) => <Link href="/profile">{chunks}</Link>
});

In case you have attribute values that are required to be configured as part of your messages, you can retrieve them from a separate message and pass them to another one as an attribute:

{
  "message": "See this <partner>partner website</partner>.",
  "partnerHref": "https://partner.example.com"
}

t.rich('message', {
  partner: (chunks) => <a href={t('partnerHref')}>{chunks}</a>
});

For the use case of localizing pathnames, consider using pathnames.

To dynamically iterate over all keys of a namespace, you can use the useMessages hook to retrieve all messages of a given namespace and extract the keys from there:

import {useTranslations, useMessages} from 'next-intl';
 
function CompanyStats() {
  const t = useTranslations('CompanyStats');
 
  const messages = useMessages();
  const keys = Object.keys(messages.CompanyStats);
 
  return (
    <ul>
      {keys.map((key) => (
        <li key={key}>
          <h2>{t(`${key}.title`)}</h2>
          <p>{t(`${key}.value`)}</p>
        </li>
      ))}
    </ul>
  );
}