This lab will show you how to translate text from one language into another.

• Installation
• Translating
• Next Steps

We'll be using the NLLB (No Language Left Behind) model from Meta which is available under a Creative Commons CC-BY-NC-4.0. To get this model you'll need to run the download script. It's over 500 megabytes in size, so this may take some time.

cd ~/labs/lab5
./download_model.sh

Once the model is downloaded, you'll need to install the Python dependencies for running it.

pip install --break-system-packages ctranslate2 transformers sentencepiece

The translation script takes in a line of text at a time as input, and translates from a given source language to the target language you specify. There are over 200 languages supported and you pass them in with the --source_language and --target_language command-line arguments. Here's an example that will translate from English to Spanish:

python translate.py --source_language=english --target_language=spanish

You should see output like this:

Translating from english to spanish
Enter text to be translated

Type in "Hello, how are you?" and press return.

After a couple of seconds, you should see:

Hola, ¿cómo estás?

You can translate between any pairs of the 200 languages supported, though some of the languages have less data than others and so the model can be less accurate for those. For example, try running Spanish to Arabic:

python translate.py --source_language=spanish --target_language=arabic

Paste in "Hola, ¿cómo estás?" and you should see this result:

مرحبا, كيف حالك؟

You can also read in text files and the script will translate line by line.

python translate.py --source_language=chinese --target_language=english < ../text/chinese_example.txt

The NLLB model from Meta is only licensed for non-commercial use, but the recent MADLAD model from Google offers similar functionality without those restrictions, and is also available in quantized form. It's a bit larger though, so it won't run as quickly.

You could look at translating speech into text using the script from Lab 4, and then feed the results into the translator to convert it into another language.