In this article, I'll show you how to translate content programmatically with PHP using the TransformersPHP library.
Translating text is essential for reaching a global audience and ensuring your content is accessible to speakers of different languages.

Step 1: Set up the project

To get started, please make sure you have the TransformersPHP library installed. You can install it via Composer by running:

composer require codewithkyrian/transformers

During the installation, you have to answer a question:

Do you trust "codewithkyrian/transformers-libsloader" to execute code and wish to enable it now? (writes "allow-plugins" to composer.json) [y,n,d,?]

You'll need to answer yes to enable the Composer plugin to download all shared libraries necessary for TransformersPHP.

Once installed, require the autoload file to load all necessary classes and dependencies:

<?php
require "./vendor/autoload.php";

Step 2: Import the necessary classes

Next, you’ll need to import the relevant classes and functions that handle translation:

use Codewithkyrian\Transformers\Transformers;
use function Codewithkyrian\Transformers\Pipelines\pipeline;

• Transformers: This class manages the setup and configuration for translation models.
• pipeline: This function initializes your specific translation pipeline.

Step 3: Initialize the Transformers class

Before translating content, you must configure the Transformers class:

Transformers::setup()->setCacheDir("./models")->apply();

• setCacheDir(): This method defines the directory for caching models, which speeds up the process by avoiding repeated downloads.
• apply(): Finalizes the setup and applies the configuration.

Step 4: Set Up the Translation Pipeline

The next step is to create a pipeline for translation using a pre-trained model:

$translationPipeline = pipeline("translation", 'Xenova/nllb-200-distilled-600M');

• pipeline("translation", 'Xenova/nllb-200-distilled-600M'): This function sets up a translation pipeline using the specified model, Xenova/nllb-200-distilled-600M, which is capable of handling multiple languages efficiently.

The model used for translations in this example is https://huggingface.co/Xenova/nllb-200-distilled-600M

Step 5: Provide content for translation

Define the sentences you want to translate:

$inputs = [
    "The quality of tools in the PHP ecosystem has greatly improved in recent years",
    "Some developers don't like PHP as a programming language",
    "I appreciate Laravel as a development tool",
    "Laravel is a framework that improves my productivity",
    "Using an outdated version of Laravel is not a good practice",
    "I love Laravel",
];

This array contains English sentences that will be translated into Italian.

Step 6: Translate the content

Loop through each sentence and translate it:

foreach ($inputs as $input) {
    $output = $translationPipeline(
        $input,
        maxNewTokens: 256,
        tgtLang: 'ita_Latn'
    );
    echo "🇬🇧 " . $input . PHP_EOL;
    echo "🇮🇹 " . trim($output[0]["translation_text"]) . PHP_EOL;
    echo PHP_EOL;
}

• $translationPipeline($input, maxNewTokens: 256, tgtLang: 'ita_Latn'): This function call translates each English sentence into Italian, with maxNewTokens limiting the length of the translation and tgtLang specifying the target language as Italian (ita_Latn).
• trim($output[0]["translation_text"]): Cleans up the translated text by removing any leading or trailing whitespace.

The model supports a lot of languages. To define the target language with the tgtLang parameter, you must use the language code FLORES-200. Here there is a list: https://github.com/facebookresearch/flores/blob/main/flores200/README.md#languages-in-flores-200

In the first execution of the script, the pipeline() function will download all the model files into the directory: models/Xenova/nllb-200-distilled-600M. Be patient, the model is huge, more than 800 MB.

Conclusion

With TransformersPHP, translating content programmatically is a streamlined process. By setting up the environment, initializing the necessary classes, and defining a translation pipeline, you can easily convert text from one language to another. This is particularly useful for creating multilingual websites, applications, or content, allowing you to reach a broader audience effectively.

References

• TransformersPHP website: https://codewithkyrian.github.io/transformers-php/
• TransformersPHP source code: https://github.com/CodeWithKyrian/transformers-php
• Intro article about TransformersPHP: https://dev.to/robertobutti/machine-learning-with-php-5gb
• How to generate Alt Text with TransformersPHP https://dev.to/robertobutti/how-to-auto-generate-the-image-alt-text-using-ai-and-transformers-php-3onc
• TransformersPHP official documentation: https://codewithkyrian.github.io/transformers-php/introduction
• The author, the amazing Kyrian https://x.com/CodeWithKyrian, thank you for all your effort in building this open source PHP project ✨