Pragmatics as the secret ingredient: a multilingual study of reference tracking and evolving anaphors in culinary narratives

Anaphora resolution, namely the correct interpretation of anaphoric means, poses numerous challenges for natural language processing (NLP) technologies, particularly when approached from a multilingual perspective. Reference tracking mechanisms exhibit considerable cross- linguistic variation, with some languages heavily relying on inference and common sense. Ad- ditional complexities are introduced by phenomena such as null pronouns, partial coreference, and evolving anaphora, namely an expression that refers to an entity which, in the interim, has undergone changes in its properties or attributes. Crucially, these phenomena are frequent in procedural texts such as recipes, where precise coreference disambiguation is essential for the successful execution of tasks in a technologically driven culinary landscape. The present paper delves into the complexity and variation of reference tracking across languages NLP systems need to face. After an overview of coreference encoding and resolution from a cross-linguistic perspective, with a focus on zero, bridging, and evolving anaphors, it seeks to unravel the in- tricacies of anaphoric encoding in culinary narratives through a study of the type, frequency, and nature of anaphoric devices used in authentic recipe texts. It examines texts in English, Italian, and Chinese, which differ substantially in terms of coreference tracking systems and in the degree of reliance on inference for disambiguation.