Translate Languages with MarianMT and Hugging Face Transformers

Picture by Writer | Canva

Language translation has develop into a necessary device in our more and more globalized world. Whether or not you are a developer, researcher, or traveler, you’ll all the time discover the necessity to talk with individuals from completely different cultures. Therefore, the flexibility to translate textual content shortly and precisely may be very useful for you. One highly effective useful resource for attaining that is the MarianMT mannequin, part of the Hugging Face Transformers library.

On this information, we’ll stroll you thru the method of utilizing MarianMT to translate textual content between a number of languages, making it accessible even for these with minimal technical background.

What’s MarianMT?

MarianMT is a machine translation framework based mostly on the Transformer structure, which is widely known for its effectiveness in pure language processing duties. Developed utilizing the Marian C++ library, the MarianMT fashions have an enormous benefit of being quick. Hugging Face has integrated MarianMT into their Transformers library, making it simpler to entry and use via Python.

Step-by-Step Information to Use MarianMT

1. Set up

To start, it’s worthwhile to set up the mandatory libraries. Guarantee you might have Python put in in your system, then run the next command to put in the Hugging Face Transformers library:

You’ll additionally want the torch library for dealing with the mannequin’s computations:

2. Selecting a Mannequin

MarianMT fashions are pre-trained on numerous language pairs. The fashions observe a naming conference of Helsinki-NLP/opus-mt-{src}-{tgt} in hugging face, the place {src} and {tgt} are the supply and goal language codes, respectively. For instance, for those who search Helsinki-NLP/opus-mt-en-fr in hugging face, the corresponding mannequin would translate from English to French.

3. Loading the Mannequin and Tokenizer

Let’s say you resolve to translate English to a particular language, i.e., French. Then you definately would want to load the best mannequin and its corresponding tokenizer. Right here’s the way you load the mannequin and tokenizer:

from transformers import MarianMTModel, MarianTokenizer

# Specify the mannequin identify
model_name = "Helsinki-NLP/opus-mt-en-fr"

# Load the tokenizer and mannequin
tokenizer = MarianTokenizer.from_pretrained(model_name)
mannequin = MarianMTModel.from_pretrained(model_name)

4. Translating Textual content

Now that you’ve got your mannequin and tokenizer prepared, you may translate textual content in simply 4 easy steps! Right here’s a primary instance.Initially, you’d specify the supply textual content in a variable that you just need to translate.

# Outline the supply textual content
src_text = ["this is a sentence in English that we want to translate to French"]

Since transformers (or any machine studying mannequin) doesn’t perceive textual content, we need to convert the supply textual content into numeric kind. For that, we might tokenize our textual content. For an intensive understanding of how you can do tokenization, you may consult with my Tokenization article.

# Tokenize the supply textual content
inputs = tokenizer(src_text, return_tensors="pt", padding=True)

Then we’ll move the tokenized sentence to the mannequin and it’ll output some numbers.

# Generate the interpretation
translated = mannequin.generate(**inputs)

Discover that mannequin outputs tokens, and never textual content immediately. We must decode these tokens again to textual content so people can perceive the translated output of the mannequin.

# Decode the translated textual content
tgt_text = [tokenizer.decode(t, skip_special_tokens=True) for t in translated]
print(tgt_text)

Within the above code, the output would be the translated textual content in French:

c'est une phrase en anglais que nous voulons traduire en français

5. Translating to A number of Languages

If you wish to translate English textual content into a number of languages, you should utilize multilingual fashions. For instance, the mannequin Helsinki-NLP/opus-mt-en-ROMANCE can translate english to a number of Romance languages (French, Portuguese, Spanish, and so forth.). Specify the goal language by prepending the supply textual content with the goal language code:

src_text = [
    ">>fr<< this is a sentence in English that we want to translate to French",
    ">>pt<< This should go to Portuguese",
    ">>es<< And this to Spanish",
]

# Specify the multilingual mannequin
model_name = "Helsinki-NLP/opus-mt-en-ROMANCE"
tokenizer = MarianTokenizer.from_pretrained(model_name)
mannequin = MarianMTModel.from_pretrained(model_name)

# Tokenize the supply textual content
inputs = tokenizer(src_text, return_tensors="pt", padding=True)

# Generate the translations
translated = mannequin.generate(**inputs)

# Decode the translated textual content
tgt_text = [tokenizer.decode(t, skip_special_tokens=True) for t in translated]
print(tgt_text)

Output would appear like this:

["c'est une phrase en anglais que nous voulons traduire en français",
 'Isto deve ir para o português.',
 'Y esto al español']

With this setup, you may simply translate your English textual content into French, Portuguese, and Spanish. There are some teams of languages aside from ROMANCE languages as effectively. Here’s a checklist of them:

GROUP_MEMBERS = {
 'ZH': ['cmn', 'cn', 'yue', 'ze_zh', 'zh_cn', 'zh_CN', 'zh_HK', 'zh_tw', 'zh_TW', 'zh_yue', 'zhs', 'zht', 'zh'],
 'ROMANCE': ['fr', 'fr_BE', 'fr_CA', 'fr_FR', 'wa', 'frp', 'oc', 'ca', 'rm', 'lld', 'fur', 'lij', 'lmo', 'es', 'es_AR', 'es_CL', 'es_CO', 'es_CR', 'es_DO', 'es_EC', 'es_ES', 'es_GT', 'es_HN', 'es_MX', 'es_NI', 'es_PA', 'es_PE', 'es_PR', 'es_SV', 'es_UY', 'es_VE', 'pt', 'pt_br', 'pt_BR', 'pt_PT', 'gl', 'lad', 'an', 'mwl', 'it', 'it_IT', 'co', 'nap', 'scn', 'vec', 'sc', 'ro', 'la'],
 'NORTH_EU': ['de', 'nl', 'fy', 'af', 'da', 'fo', 'is', 'no', 'nb', 'nn', 'sv'],
 'SCANDINAVIA': ['da', 'fo', 'is', 'no', 'nb', 'nn', 'sv'],
 'SAMI': ['se', 'sma', 'smj', 'smn', 'sms'],
 'NORWAY': ['nb_NO', 'nb', 'nn_NO', 'nn', 'nog', 'no_nb', 'no'],
 'CELTIC': ['ga', 'cy', 'br', 'gd', 'kw', 'gv']
}

Wrapping Up

Utilizing MarianMT fashions with the Hugging Face Transformers library offers a robust and versatile technique to carry out language translations. Whether or not you’re translating textual content for private use, analysis, or integrating translation capabilities into your purposes, MarianMT provides a dependable and easy-to-use answer. With the steps outlined on this information, you may get began with translating languages effectively and successfully.
 
 

Kanwal Mehreen Kanwal is a machine studying engineer and a technical author with a profound ardour for information science and the intersection of AI with drugs. She co-authored the book “Maximizing Productiveness with ChatGPT”. As a Google Era Scholar 2022 for APAC, she champions range and tutorial excellence. She’s additionally acknowledged as a Teradata Variety in Tech Scholar, Mitacs Globalink Analysis Scholar, and Harvard WeCode Scholar. Kanwal is an ardent advocate for change, having based FEMCodes to empower girls in STEM fields.