Transfer Learning for Multilingual Natural Language Understanding


Multilingual SaaS

In the scope of project TRAINER, transfer learning methods applied to deep learning technologies for natural language processing tasks will be researched. The main goal of the project is the creation of a multilingual SaaS (software as a service) offer of natural language processing services that enables the internationalization of this Priberam’s business area.

Growth of Contents

The exponential growth of the volume of text content available, both inside and outside the organizations, is one of the main consequences of the digitization not only of the economy but also of an important part of the human activity, even not professional.

Value Extraction

To extract value from all this text information available to the organizations, several software services are available to identify people, organizations, places and events mentioned in a text, link these entities to external knowledge bases, do sentiment analysis on posts in social media or summarize documents, only to give a few examples.

The lack of enough data to train robust models for certain tasks and/or languages and the high cost of creation of those resources, involves a large investment of any company who wants to broaden its offer to a new language or even to a specific knowledge or technical domain.

Clear Vantage among competition

By integrating transfer learning methods in its SaaS offer of natural language processing services, Priberam will gain a clear advantage in regard to its competitors, as it will be able to reuse resources available in any specific language, to create services for languages where those resources are nonexistent.

Integration in Priberam’s solutions

Additionally, the project results will, in the future, be integrated in Priberam’s solutions for Media & Publishing, Media Intelligence, Legal Research and, in particular, Health, also contributing to the growth and internationalization of these business areas of the company.

Project sheet

Project name | TRAINER: Transfer Learning for Multilingual Natural Language Understanding

Project code | LISBOA-01-0247-FEDER-045347

Main goal | Reinforce the research, technological development and the innovation

Region | Lisbon

Beneficiary | Priberam Informática, S.A.


Approval date | 19-Dec-2019

Start date | 1-Jan-2020

End date | 31-Dec-2021


Total eligible cost | 568.169,57 EUR

European Union financial support | ERDF – 284.084,79 EUR