Course program
Human Languages and Large Language Model: Theoretical Models in Comparison
The course aims to provide students with essential theoretical tools to critically compare natural-historical languages and Large Language Models (LLMs), which currently represent the most advanced paradigm of Artificial Intelligence. It will begin by illustrating the shift from symbolic AI models to sub-symbolic models based on Deep Neural Network architectures and designed for Deep Learning. This transition - enabled by technologies that have been known for years - has taken place over the past two decades and has brought about a transformation in theoretical models, as well as a renewed debate concerning the similarities and differences between human and artificial intelligence.
In the context of Natural Language Processing (NLP), symbolic models require that every level of language be analyzed and explicitly represented. However, the linguistic behavior of machines developed under this framework is neither qualitatively nor quantitatively comparable to that of human language users or even to that of current LLM-based systems. The latter rely instead on techniques for extracting patterns and structures based on statistical regularities from large quantities of raw linguistic and non-linguistic data. By harnessing enormous computational power, these systems revitalize a conception of language previously theorized by structuralism—particularly the Hjelmslevian and Jakobsonian variants and Distributionalism. While the “autonomist” structuralist model now occupies a marginal position within contemporary linguistic and philosophical debates, a critical examination of its limitations—also in light of internal developments within linguistic structuralism and insights from cognitive semantic and cognitive semiotic theories—will serve to highlight the fundamental gap between LLMs and natural-historical languages.
Prerequisites
No specific prerequisites are required.
Books
1. Jakobson, R. (1966), Saggi di linguistica generale, Milano, Feltrinelli (o successive edizioni).
2. Tavosanis, M. (2018), Lingue e intelligenza artificiale, Roma, Carocci.
3. Gastaldi, Juan Luis & Pellissier, Luc, “The calculus of language: explicit representation of emergent linguistic structure through type theoretical paradigms”, Interdisciplinary Science Reviews, 46:4, 569-590, DOI: 10.1080/03080188.2021.1890484 - https://doi.org/10.1080/03080188.2021.1890484
Additional readings and bibliographic references will be provided during the course.
Frequency
Attending the course is highly recommended.
Exam mode
The exam consists of an oral interview. Depending on the number of attending students, it will be possible to organize group presentations (comprising a minimum of 3 and a maximum of 5 students) on a topic covered in the syllabus. The group presentation will account for 30% of the final grade, while the remaining 70% will be based on the individual oral examination.
Lesson mode
The first part of the course will mainly consist of frontal lessons, while the second part will stimulate open discussion, also increasing direct involvement. The modalities of direct involvement will be specified during the course, depending on the number of students and their willingness to cooperate.
