Research and development on multilingual AI

Tilde is a pivotal language technology research hub in the Baltic region with 30+ years of experience in European research projects and numerous local projects in Estonia, Latvia and Lithuania. Our team of in-house researchers cooperate with leading European research centres to advance the state-of-the-art in areas of language technologies such as machine translation, conversational AI and dialogue systems, speech recognition and synthesis, and foundational language models.

In-depth research expertise areas

Machine translation

We innovate in domain-adapted, adaptive, and term-aware neural machine translation, developing methods for robustness, mitigation of biases, and large language models.

We focus on multilingual natural language understanding, semantic indexing, LLM-based retrieval-augmented generation, and personalisation.

Conversational AI

We research cascaded and end-to-end speech recognition and speech translation, multi-speaker and multilingual speech synthesis, real-time speech recognition, automatic subtitling and dubbing.

Speech technologies

We develop multilingual named entity recognition, anonymisation and pseudonymisation, term recognition and extraction, morphological analysis, lemmatization, part-of-speech tagging and other tools.

Text analysis

We maintain the largest termbank in Europe – EuroTermBank, and develop term management and electronic dictionary tools.

Knowledge management

TildeOpen LLM is now live on Hugging Face 🎉

Meet TildeOpen – our new open-source foundational large language model for European languages.

Supported by:

Current research projects

AI language technologies and their implementation support to promote Ukraine’s integration into the European Union

This project addresses the challenge of translating Ukraine’s national legislation and aligning it with the EU acquis. It requires precision, legal accuracy, and coordination across institutions. Tilde is tailoring its secure, AI-powered translation platform for Ukraine’s legal and administrative texts. The platform will deliver high-quality translations at scale, reduce manual work, and ensure consistent use of EU legal terminology. It combines machine translation, translation memory, and a terminology portal, all adapted to Ukraine’s public sector. The project is being implemented with funding from the Latvian state budget and support from the Ministry of Foreign Affairs of Latvia and the Central Finance and Contracting Agency (CFLA).

European Language Data Space

Through the Language Data Space (LDS) relevant stakeholders will be able to share and also monetise their language data and other language resources through a single platform, taking EU values and compliance with EU rules fully into account.

FORTISSIMO PLUS subproject: Locally Deployable Enterprise Search and Q&A solution

The current state-of-the-art open LLMs inadequately support most European languages, posing a significant challenge for organisations operating in multilingual environments. While these models demonstrate reasonable performance for several major European languages, they often lack sufficient coverage and accuracy for many other languages, including Eastern European languages. This project aims to address this disparity by developing a locally deployable AI-based enterprise search and question-answering solution for underrepresented European languages, particularly those in the Balto-Slavic family. The project will create a secure, scalable and customizable AI system that integrates seamlessly with an organization’s infrastructure, ensuring robust data privacy and compliance with stringent governance requirements.

All projects