Valuazione automatica della complessità di frasi in Italiano e Inglese tramite Deep Learning.

In modern society, the capability of comprehending a text is crucial to promote processes such as the social integration of individuals, the improvement of the personal capital, and the personal independence. Although the importance of literacy skills, the current panorama of linguistic proficiency is wide, and it extends from highly literates to people who have difficulties understanding daily-life texts. Society underestimates the problem, while suitable means should be put in place to address it. A simple but effective solution is to accompany the original complex to understand text with a simplified version drafted by exploiting a Text Complexity Evaluation (TCE) process. Such a process is usually done manually, and it is entrusted to expert human annotators who combine the numerous features which affect the text comprehension to assess the text complexity. The process is made harder in contexts characterized by a large production of documents where the difficulty of tackling the TCE process and many documents to be analyzed makes the task impracticable. This results in the need for tools capable of making the assessing procedure more affordable. The aim of the doctoral project has been the development of innovative approaches for supporting the evaluation of text complexity in English and Italian languages via Deep Learning methodologies. The developed systems have shown learning abilities capable of inferring and relating in a non-trivial way features which affect the text comprehension to assess the text complexity, and versatility for tackling the problem in different languages. Moreover, a new resource suitable to tackle the problem for the Italian through machine and deep learning has been provided.