Profiling a Neural Language Model

In-house seminar

Over the last few years, the study and analysis of the internal mechanisms of the most recent Neural Language Models (NLMs) became one of the most productive research lines in the context of Natural Language Processing (NLP). Several methodologies were devised in order to obtain explanations and understand how these models are able to implicitly encode linguistic competence within their representations. The seminar holder will introduce an approach designed to investigate whether, by exploiting the latest NLP methods of text and author profiling and, specifically, of study of the processes of evolution of language skills, it is also possible to learn the behavior of the state-of-the-art NLMs. The seminar will be divided into two parts. In the first part, a stylometric approach will be illustrated to trace the evolution of the language skills of L1 and L2 learners, using a wide set of characteristics linguistically motivated. In the second part, on the other hand, the seminar holder will introduce some studies he carried out in order to investigate the linguistic skills implicitly codified by NLMs, starting from the same set of linguistic characteristics used in the previous experiments.

Speaker(s): Alessio Miaschi

He is a research fellow at the ItaliaNLP Lab of the Institute for Computational Linguistics “A. Zampolli” of the National Research Council of Italy. He received his PhD in Computer Science at the University of Pisa in May 2022 with a thesis focused on the definition of methodologies for the interpretation and understanding of language skills implicitly encoded by the most recent neural language models. Previously he received a Master’s Degree in Digital Humanities at the University of Pisa. His research interests mainly concern the development and analysis of the most recent neural language models, as well as the definition of Natural Language Processing technologies in the field of Education.