Building Multilingual Web Apps with Web-Translate

Building Multilingual Web Apps with Web Translate

By Mark Volkmann, OCI Partner and Principal Software Engineer

September 2018

Introduction

Creating web applications that display text in a variety of written languages is a common need.

One might think this is a solved problem and that simple solutions exist. I have not found any that fit the bill, though, so I created web-translate, an open source library that can be obtained via npm at https://www.npmjs.com/package/web-translate.

The web-translate library provides a set of JavaScript functions and a command that make translating English to another language in web applications easy! It is not specific to any framework and should work with all the popular choices, including React, Vue, and Angular.

The translations are performed as a build step, so no cost for translation services is incurred during web application usage.

Goals

The goals for the web-translate library are as follows:

  • Determine required translations by parsing source files for calls to a certain function
  • Determine additional required translations by parsing a JSON file containing English translations
  • Support selection between the two most popular translation services, Google Cloud Translate API and Yandex Translate Service
  • Support build-time translation to avoid incurring run-time translation costs during each user session
  • Support easy generation of translations for new languages to be supported
  • Allow generated translations to be overridden by manually creating language-specific JSON files that describe the desired translations
  • Support run-time translations for the rare cases where dynamically generated text must be translated

No Free Lunch

The most highly recommended language translation services are the Google Cloud Translation API and the Yandex Translate Service. Both require an API key.

The Google Cloud Translation API requires setup of a Google Cloud Platform (GCP) project. The cost is $20 (USD) per million characters translated.

To enable use of the Cloud Translation API and obtain an API Key, follow these instructions:

  1. Browse https://console.cloud.google.com/, choose an account, and log in.
  2. Select the project for which you want to enable the Translation API from the dropdown near the top.
  3. Scroll down and click "API Enable APIs and get credentials like keys."
  4. Click "Enable APIs and Service" at the top.
  5. Enter "translation" in the search input.
  6. Click "Cloud Translation API" which is not enabled by default.
  7. Press the "Enable" button.
  8. Click "Google Cloud Platform" in the upper-left to return to the project page.
  9. Scroll down and click "API Enable APIs and get credentials like keys" again.
  10. Click "Credentials" in the left nav.
  11. Click the "Create Credentials" dropdown and select "API key."
  12. Copy this API Key to a secure location.

The Yandex Translate Service has free and commercial tiers.

The free tier currently allows translating up to 1,000,000 characters per day, but not more than 10,000,000 per month.

Commercial tier pricing is documented at https://translate.yandex.com/developers/offer/prices. Currently the cost is $15 (USD) per million characters translated up to 50 million. The rates go down slowly above that.

Information on obtaining a Yandex API key can be found at https://tech.yandex.com/translate/.

Setup

To use the web-translate package, follow these instructions:

  1. npm install web-translate
  2. Set the environment variable TRANSLATE_ENGINE to "google" or "yandex." When not set, this defaults to "yandex" because it has a free tier.
  3. Set the environment variable API_KEY to the API key for the desired service.
  4. Add the following npm script in the package.json file for your application.
    "gentran": "generate-translations",

Supported Written Languages

Create the file languages.json to describe the languages to be supported.

For example,

{
  "Chinese": "zh",
  "English": "en",
  "French": "fr",
  "German": "de",
  "Russian": "ru",
  "Spanish": "es"
}

It is important to get the language codes (like "en") correct because those are used to request translations. You can find a list of valid language codes at https://www.wikipedia.org/wiki/List_of_ISO_639-1_codes/.

This file must be in a directory that is accessible at the domain of the web app. For example, if your web application is running on localhost:3000, then an HTTP GET request to localhost:3000/languages.json must return the content of this file.

For a React application created with create-react-app, placing this file in the public directory will achieve this.

Getting Translations

Use the i18n function to get translations. For example, when the language is Spanish, calling i18n('Hello') might return "Hola."

i18n can be imported with import {i18m} from 'web-translate';.

The string passed to the i18n function can be English text or a key used to lookup the translation in a language-specific JSON file.

Keys are useful for long phrases, sentences, and even paragraphs. For example, i18n('greet') might return "Welcome to my wonderful web application!".

Translations for keys must be defined in language-specific JSON files. For example, the file en.json could contain the following:

{
  "contact": "For more information, contact Mark Volkmann.",
  "greet": "Welcome to my wonderful web application!"
}

Specifying Translations

Translations of all the text for all supported languages can be manually entered in language-specific JSON files like es.json (for Spanish) and fr.json (for French). However as we will see next, these files can be generated using one of the supported translation services.

In some cases, it may be desirable to use different translations. Those provided by the translation service can be overridden by creating a JSON file whose name starts with the language code, followed by -override.json. For example, the file es-overrides.json could contain the following to change the translation for "Hello" from "Hola" to "Oh La."

{
  "Hello": "Oh La"
}

Generating Translations

Manually looking up and specifying translations can be very tedious! Translations files for all the languages to be supported can be generated by simply running npm run gentran.

Here is a description of what this does.

  1. Get all the languages to be supported from the file languages.json
  2. Get all the literal strings passed to the i18n function in all the source files under the src directory where source files are any with an extension of "js", "jsx", "ts", or "tsx".
  3. Get all the English translations from the file en.json.
  4. For each language to be supported except English ...
    1. Set translations to a map of all the translations found in an overrides file for the current language (ex. fr-overrides.json). If none exist, then translations begins empty.
    2. For each key/value pair in the English translation file en.json ...
      • If there is not already a translation for this key, get the translation for the value from the selected translation service and save it in translations.
    3. For each key found in source files ...
      • If there is not already a translation for this key, get the translation for the value from the selected translation service and save it in translations.
  5. Write translations to a new translation file for the current language.

If you choose to generate the translation files, and you should, remember not to manually edit them because they will be overwritten the next time npm run gentran is run.

If the i18n function is passed a variable instead of literal string, no translations will be provided. This is because tracing the flow of the code to find all possible values is a very hard problem.

In these cases, manually enter the desired translations in the en.json file. This will enable translations for all the other supported languages to be generated.

Allowing User to Select Language

web-translate makes it easy to allow the user to change the current language. Here is an example of how this can be done in a React application. 

(It should be somewhat similar for other frameworks. I welcome contribution of code for other frameworks to share here.)

  1. import React, {Component} from 'react';
  2. import {
  3. getLanguageCode,
  4. getSupportedLanguages,
  5. i18n,
  6. setLanguage
  7. } from 'web-translate';
  8. import './App.css';
  9.  
  10. class App extends Component {
  11. state = {languages: {}};
  12.  
  13. async componentDidMount() {
  14. const languageCode = getLanguageCode();
  15. const languages = await getSupportedLanguages();
  16. this.setState({languages});
  17. }
  18.  
  19. changeLanguage = async event => {
  20. const languageCode = event.target.value;
  21. await setLanguage(languageCode);
  22. this.setState({languageCode});
  23. };
  24.  
  25. render() {
  26. const {languageCode, languages} = this.state;
  27. const languageNames = Object.keys(languages);
  28.  
  29. return (
  30. <div className="App">
  31. <div>
  32. <label>Language:</label>
  33. <select onChange={this.changeLanguage} value={language}>
  34. {languageNames.map(name => (
  35. <option key={name} value={languages[name]}>
  36. {name}
  37. </option>
  38. ))}
  39. </select>
  40. </div>
  41. <div>{i18n('Hello')}</div>
  42. <div>{i18n('some-key')}</div>
  43. </div>
  44. );
  45. }
  46. }
  47.  
  48. export default App;

With watch and live reload (as supported by create-react-app), changes to the list of supported languages and their translations are made available in the running application seconds after they are saved.

Run npm run gentran again whenever any of the following occur:

  • New languages are added to the list of supported languages in languages.json.
  • New translations are added in the English file en.json.
  • A change is made in a language-specific -overrides.json file.
  • New calls to i18n with literal strings are added in any source file or the literal strings passed to existing calls are modified, and translations for the new values are not already present in all the language .json files.

Dynamic Translation

Some web applications may need to dynamically generate text that requires translation. This can be accomplished using the translate function. It takes a "from" language code, a "to" language code, and the text to be translated.

For example, to translate "I like strawberry pie!" from English to French,

const text = 'I like strawberry pie!';
const translatedText = await translate('en', 'fr', text);

This requires ensuring that the TRANSLATE_ENGINE and API_KEY environment variables are set in the environment where the web app is running.

Bear in mind that run-time translation will incur per-user session charges from the selected translation service. Avoid using this approach when possible.

Example Application

In our example application, we begin with the following files:

languages.json

{
  "English": "en",
  "French": "fr",
  "Spanish": "es"
}

en.json

{
  "some-key": "My English key"
}

es.json

{
  "Hello": "Hola"
}

App.js

Inside the render method we have:

<div>{i18n('Hello')}</div>
<div>{i18n('some-key')}</div>

Before running npm run gentran, when the application is rendered, we see the the languages "English," "French," and "Spanish" in the language dropdown with "English" selected. We also see "Hello" and "My English Key" as the results of the i18n calls.

Selecting "French" from the dropdown displays "Hello" and "some-key." This happens because the file fr.json does not yet exist.

Selecting "Spanish" from the dropdown displays "Hola" and "some-key." We see "Hola" because that is the translation for "Hello" in the es.json file. We see "some-key" because es.json does not yet provide a translation for that key.

Run npm run gentran. This generates the files es.json and fr.json, which will now contain translations for both "Hello" and "some-key."

Depending on your build process, the application may automatically re-render in the browser.

When "French" is selected from the dropdown, we see "Bonjour" and "Ma clé anglaise" which are the French translations for "Hello" and "My English Key."

When "Spanish" is selected, we see still "Hola," but we now also see the Spanish translation for "My English Key," which is "Mi clave en inglés."

Add the following lines to languages.json:

"German": "de",
"Russian": "ru"

Run npm run gentran again. The language dropdown will now contain "German" and "Russian." Selecting these languages will display their translations.

Add the following in App.js:

<div>{i18n('Where is your pencil?')}</div>

Run npm run gentran again. We now see that text for English.

Select other languages from the dropdown to display the translation for this text.

Sometimes the translations provided by Google Translate will not be ideal for the application. To override translations, create a language-specific -override.json file. For example, to display "Oh La" instead of "Hola" for the Spanish translation of "Hello," create the file es-override.json with the following content:

{
  "Hello": "Oh La"
}

Run npm run gentran again. Select "Spanish" from the dropdown. The new translation for "Hello" in Spanish will be displayed.

Placeholders

Translation text can contain placeholders for inserting dynamic text. For example, en.jsoncould contain the following line:

"greet": "Hello ${name}, today is ${dayOfWeek}.",

To use this, pass a second argument to i18n that is an object where the keys are the placeholder names, and the values are the values to be inserted. For example,

i18n('greet', {name: 'Mark', dayOfWeek: 'Tuesday'});

Acknowledgements

web-translate uses code from the npm package "translate" from Francisco Presencia (franciscop on GitHub). See https://www.npmjs.com/package/translate. Thank you Francisco!

Thanks to AJ Levinson for details on getting a Google Cloud Platform API key!

Summary

web-translate provides the simplest approach to language translations in web applications that I have seen!

Please post suggestions for improvements and any issues at https://github.com/mvolkmann/web-translate/issues.

Software Engineering Tech Trends (SETT) is a monthly publication featuring emerging trends in software engineering.

More from Mark Volkmann

* You may unsubscribe from our mailing list at any time using the 'unsubscribe' link at the bottom of every email. You can customize your email subscriptions here. To see how we keep your personal information secure, please visit our privacy policy.

secret