TRACING SENSORY EVENTS: A FRAME-BASED APPROACH TO AUTOMATICALLY CAPTURE AND DIACHRONICALLY ANALYSE OLFACTORY AND GUSTATORY INFORMATION FROM TEXTS

The study of sensory language has long been influenced by Europe’s visual-centric perspective, which relegated smell and taste to the status of “minor” human senses. Investigating sensory language generally requires identifying the words associated with these particular domains, raising the challenge of distinguishing between the senses and isolating the specific features of their linguistic expressions. In this dissertation, we focus on these two often overlooked senses, smell and taste, by proposing a novel and comprehensive framework to capture not only their specific terms but also their contextual information, covering English and Italian for smell and English for taste. Using this framework, we create two manually annotated benchmarks, which we then use to implement two systems for the automatic extraction of olfactory and gustatory information from texts. Subsequently, we employ these systems to collect large-scale olfactory and gustatory data, addressing the underexplored topic of the diachronic evolution of sensory language. To this end, we conduct four studies that address this topic from different angles: a taxonomic categorisation of smell-related terms, two studies on the automatic detection of shifts in the perception of specific smell-related objects, and a final study exploring and comparing the evolution of taste and smell vocabularies in terms of lexical diversification and evaluative content. This work not only provides practical tools for large-scale data collection but also introduces a series of methodologies that combine historical, linguistic, and computational approaches to the study of smell and taste.