In a revolutionary step forward in the field of Artificial Intelligence, a team of Italian researchers has introduced Minerva, a family of Large Language Models (LLMs) trained entirely in the Italian language. Baptized in honor of the Roman goddess of wisdom and loyalty, Minerva represents a significant leap forward in AI technology specifically developed for the Italian language.

Minerva, created by the Sapienza NLP research group led by Professor Roberto Navigli of the Department of Computer, Control, and Management Engineering “Antonio Ruberti” at Sapienza University of Rome, marks a milestone as the first set of LLMs trained from scratch on Italian language content. This innovative approach sets Minerva apart from existing models, which rely on adaptation from other languages.

The training process for Minerva was meticulous and comprehensive. The models were trained on a wide range of Italian and English sources, totaling over 500 billion words, equivalent to more than 5 million novels. This massive corpus aims to provide a solid foundation for AI systems applicable in various fields, from natural language understanding to text generation and automated customer support.

“We have worked tirelessly to create Minerva, the first set of Large Language Models entirely in Italian,” “This milestone represents a crucial step towards the technological autonomy of our country and the affirmation of Italian excellence in Artificial Intelligence.”

Prof. Navigli.

Minerva stands out for several innovative features. Firstly, its modular architecture allows for the development of models of different sizes and computational capabilities, to meet the needs of a wide range of applications. This flexibility is essential to ensure accessibility and adaptability of the technology to different contexts and use cases.

Furthermore, Minerva is based on a “from scratch” training approach, without relying on pre-existing models in other languages. This approach allows for a more faithful capture of the peculiarities and nuances of the Italian language, offering a solid foundation for the development of AI systems capable of understanding and interacting naturally with Italian users.

“One of our main objectives was to create language models that faithfully reflected the richness and complexity of the Italian language,” “Too often, Artificial Intelligence solutions have been developed based on languages like English, overlooking the specificities of other languages. With Minerva, we aim to bridge this gap and offer Italian users an authentically Italian AI experience.”

Prof. Navigli.

In addition to linguistic fidelity, Minerva distinguishes itself for its transparency and interpretability. Instead of operating as “black boxes,” Minerva models have been designed to be as explanatory and understandable as possible, allowing experts to understand the decision-making process behind their predictions and outputs.

This feature is crucial for promoting trust and acceptance of Artificial Intelligence, especially in sensitive areas such as healthcare, finance, and education, where transparency and accountability are essential.

“Transparency is a crucial aspect for the widespread adoption of Artificial Intelligence,” “With Minerva, we want to demonstrate that it is possible to develop advanced AI models while maintaining a high degree of interpretability and explainability. This is a fundamental requirement for gaining the public’s trust and promoting responsible adoption of the technology.”

Prof. Navigli.

Currently, Minerva is only available to the scientific community as part of the FAIR (Foundation for Artificial Intelligence Research) project. However, it is expected that the advanced version of the model will be released to the public shortly, allowing for direct interactions in the Italian language.

This step towards open-source availability represents a significant milestone in democratizing access to cutting-edge AI technologies developed in Italy. Making Minerva accessible to a wider audience means promoting innovation and adoption of AI solutions made in Italy, providing Italian companies, institutions, and citizens with the tools to fully leverage the potential of Artificial Intelligence.

“We are excited to share Minerva with a broader audience,” “Our goal is to make this technology accessible and to stimulate innovation and creativity in the Italian AI ecosystem. We want Minerva to become a reference point for the development of authentically Italian AI applications.”

Prof. Navigli.

Impact and Applications

The impact of Minerva extends far beyond the field of academic research. This family of Large Language Models has the potential to revolutionize numerous sectors, from natural language understanding to text processing, content generation, and automatic translation.

In the customer service field, for example, Minerva could enable the development of chatbots and virtual assistants capable of interacting naturally and contextually with Italian users, offering a smoother and more personalized support experience.

In education, Minerva could be employed to create adaptive learning tools, capable of understanding students’ needs and providing personalized content and feedback. Additionally, Minerva models could be integrated into sentiment analysis and emotion detection systems to enhance user experience and communication effectiveness.

In the healthcare sector, Minerva could support the development of medical language processing systems, facilitating the understanding of medical reports, generation of diagnoses, and communication between healthcare professionals and patients. Similarly, in the financial sector, Minerva could be used for automatic document analysis, report generation, and personalized consultancy.

“The potential of Minerva is truly limitless,” “This technology represents a turning point for Artificial Intelligence in Italy, paving the way for a myriad of innovative applications that can improve citizens’ lives and promote our country’s competitiveness on the global AI stage.”

Prof. Navigli.

The development of Minerva has not been a solitary journey. The Sapienza NLP research team has actively collaborated with academic institutions, research centers, and technology companies, both nationally and internationally, to ensure a holistic and multidisciplinary approach.

“We firmly believe in the importance of collaborations to promote innovation,” “We have established synergies with excellent partners, both in Italy and abroad, to enrich our work with complementary skills and diverse perspectives.”

Prof. Navigli.

Among the main partners of the Minerva project are the Institute of Information Science and Technologies (ISTI) of the National Research Council (CNR), the Italian Institute of Technology (IIT), and the National Institute of Nuclear Physics (INFN). These collaborations have allowed for the integration of advanced expertise in areas such as natural language processing, machine learning, and high-performance computing.

Additionally, the Minerva team has entered into agreements with leading companies in the technology sector, such as Microsoft, Google, and Amazon, to explore possible synergies and technology transfer opportunities. These partnerships accelerate the adoption of Minerva in real-world application contexts and ensure a tangible impact on society.

“Working in synergy with excellent partners allows us to amplify the impact of Minerva and accelerate its path to industrialization,” “Together, we can transform this technological innovation into concrete solutions that improve citizens’ lives and strengthen Italy’s competitiveness in the global AI landscape.”

Prof. Navigli.

The advent of Minerva marks a crucial moment in the development of Artificial Intelligence in Italy. This family of Large Language Models, entirely trained on the Italian language, represents a significant step towards the technological autonomy of our country and the affirmation of Italian excellence in AI.

Thanks to its innovative features, such as linguistic fidelity, transparency, and interpretability, Minerva has the potential to revolutionize numerous sectors, from natural language understanding to content generation, offering authentically Italian AI solutions.

Minerva’s journey does not stop here. With the upcoming release of the advanced version to the public, this technology will have the opportunity to reach a broader audience, stimulating innovation and creativity in the Italian AI ecosystem.

“Minerva represents an extraordinary milestone for Italy,” “We are proud to have contributed to creating a technology that can have a significant impact on our society and project Italian excellence into the world of Artificial Intelligence. This is just the beginning of an exciting journey that will take us to new horizons of innovation and progress.”

Prof. Navigli.

The sources used by Rivista.AI Navigli for his quotations include a series of articles and journalistic sources, ANSA, Il Sole 24 Ore, RaiNews.