Advances in Representation. New AI- and XR-Driven Transdisciplinarity

Exploring the intricate bond between text and images has been a topic of contemplation for centuries. The Text-To-Image (TTI) technology has emerged as a powerful tool for training Neural Networks to generate images based on specific descriptions. The contribution presents the research in its preliminary stage, primarily focusing on developing a process to intercept sentences and paradigmatic relationships to aid network formation. The tool used is the AI of neural networks, which emulates the biological functioning of the human brain, building processes based on information interconnections. The algorithms behind this technology can integrate external data (through training) with internal design information (through learning), allowing the structure to adapt and change over time. Text-To-Image (TTI) networks are a valuable tool for research, as they can efficiently convert written language into visual representations. These neural networks are capable of analyzing and comparing textual and visual data, making them a great option for the task at hand. Thorough training is crucial to ensure that these networks function properly and reliably when solving complex problems. Neural networks have a complex structure, making it hard to trace the logical procedure that led to a solution. They learn rules and activities automatically through learning rather than through reasoning deduction (Perlovsky L. I.,2001). For this, they represent innovative support for the definition of a systematic analysis of the relationship between image and text. In the literature, TTI is often used for improving the quality and control of image generation. This is done by describing natural language or defining desired attributes while still keeping content that is not related to the text. (Li et al., 2019; Li et al., 2020). The present study investigates the potential of AI in transforming written texts into visual representations through practical training. The research delves into historical treatises from the Renaissance era and modern texts from the 20th century, encompassing diverse linguistic complexities in the Italian language. The conversion of historical texts, such as treatises, into images is a complex process that requires a deep understanding of linguistics, (M. Dardano, 2017). This task necessitates careful consideration of various factors that are not easily discernible. A key aspect of this effort involves the modulation of language to ensure effective communication with artificial intelligence, thereby facilitating the generation of optimal image reconstructions. As such, a robust understanding of the linguistic nuances of the original text is critical in achieving this goal. The decoding of the variables that contribute to the most congruous definition of the text-image relationship is the goal of future research developments. The neural network will undergo training in various aspects of visual language, including balance, configuration, shape, and development to enhance its ability to effectively interpret and translate the "architecture of the text". The research presented here is centred on the study of lexical semantics as a method of interpreting sentences, as well as understanding the paradigmatic relationships within them, for use in neural networks. The aim of this work is to create visual representations of the written descriptions found in treatises.