Case studies

Pioneering Adaptive Machine Translation with Large Language Models

Discover how Pi School leads the way in Adaptive Machine Translation using Large Language Models. Explore our innovative approach to overcoming language barriers in the digital era.

In the dynamic world of Language Translation, the recent advancements in Neural Machine Translation (NMT) and Large Language Models (LLMs) are setting new benchmarks. At Pi School, we’re at the forefront of this technological revolution, applying these advancements to Adaptive Machine Translation.


Our latest project, in collaboration with Translated, a global language services company and Pi School founder, focuses on leveraging LLMs to enhance the quality and flexibility of translation systems. Translated’s unique approach combines machine translation with artificial intelligence, improving efficiency and accuracy in the translation process.

One significant challenge in this domain is the efficient fine-tuning of these models and identifying the optimal prompt to maximise performance. Our data-driven approach involves analysing datasets in Italian, Spanish, and English, ensuring a comprehensive understanding of language nuances.


We’ve employed a variety of metrics such as BLEU (BiLingual Evaluation Understudy), chrF (CHaRacter-level F-score), and COMET (Crosslingual Optimised Metric for Evaluation of Translation) to evaluate the performance of our translation models.
These metrics provide insights into the quality of machine-generated text compared to high-quality reference translations.


The study revealed promising results, especially in Italian-to-English and Italian-to-Spanish translations. We found that the GPT3.5 model with fuzzy prompting outperformed other models across all metrics, showcasing the potential of LLMs to enhance translation accuracy.
In conclusion, our work at Pi School demonstrates the significant potential of LLMs in breaking down language barriers. We aim to contribute significantly to adaptive machine translation by continuously refining our models and techniques.

Do you have a similar challenge to solve? We can help you!