Charting sensorial style: frontiers in linguistic style analysis

When we communicate, it is not just “what” we say, but also “how” we say it — our linguistic style — that conveys meaning. The way we use language reveals a lot about us - our background, personality, and even how we perceive the world. This thesis explores an understudied aspect of linguistic style: how people talk about sensory experiences like sights, sounds, tastes, and smells, “sensorial style”. This thesis bridges a fundamental gap between stylometry (the study of linguistic style) and sensorial linguistics by demonstrating that communities have identifiable linguistic styles and that sensory language forms a distinct stylistic dimension.

In this thesis, we discovered that sensorial style is a product of meaningful choices in how people communicate. By comparing sensorial language use across cultures, we found both universal tendencies and unique cultural differences in how sensory experiences are described.

We found that people use far more words related to vision than to other senses like smell or touch. This is true regardless of language. However, the balance between senses can shift depending on the context - recipes use more taste-related words, while song lyrics emphasize hearing and emotions. We also found that different cultures vary in how richly they describe each sense. For example, French and German speakers use a wider variety of words to describe tastes and flavors than speakers of other languages especially Arabic and Urdu. Arabic and Urdu speakers tend to use a smaller, more focused set of words when describing sensory experiences, whether they’re talking about sights, sounds, tastes, or smells.

We demonstrate that traditional stylistic features and sensorial style are interconnected. We model this relationship between traditional style and sensorial style using dimensionality-reduced Large Language models that we called SLIM-LLMs. We show that these low-dimensional SLIM-LLMs effectively capture traditional style information relevant to sensorial language prediction.

Our findings not only expand our understanding of language use but also give us a new lens to study how sensory perception is expressed in different cultures and contexts. By developing computational models to represent sensorial style as well as models that integrate traditional stylometric features with sensorial language, we demonstrate the potential for a more comprehensive approach to linguistic analysis. This thesis lays the groundwork for future research exploring the connections between language, cognition, and sensory experience, offering new methods, tools and, perspectives for a wide range of domains.