Project CIIAM

Contextual Information Inference for Argument Mining

About the Project

Recent research in Natural Language Processing (NLP), particularly in the field of argument mining, has significantly contributed to the systematization of processes underpinning language structuring. Pre-trained language models are now among the most advanced achievements in the field and are central to numerous technologies involved in understanding and manipulating natural language. However, these models do not always capture the full extent of linguistic knowledge necessary for fine-grained contextual understanding and relevant inferences.

In this context, the incorporation of pragmatic analysis aims to address this limitation by facilitating the deduction of information inferred from context in its broadest sense. However, it is worth noting that this dimension of linguistic analysis represents a high-risk scientific domain, which has so far been little explored due to the complexity of its formalization and operationalization. Pragmatic knowledge is inherently multimodal and can be observed at different levels: circumstantial knowledge (e.g., the physical environment of the participants), epistemic knowledge (shared beliefs and values of the participants), linguistic knowledge (preceding utterances), and/or social knowledge (the relationship between the participants).

As part of our research and funding for a postdoctoral position, our goal is to thoroughly explore how the multimodal aspects of communication manifest at the linguistic level. We plan to adopt approaches based on text generation, particularly those leveraging advanced prompting techniques, to develop "contextual frames." These frames will allow us to account for elements such as the situation of enunciation, cultural variability, implicit implicatures, and other factors that influence how arguments are formulated and interpreted in argumentative contexts.

Through this innovative approach, we aim to develop argument-mining techniques that not only capture the fundamental mechanisms underlying debates and discussions but are also robust to the nuances and subtleties of human communication. By synthesizing contextual information, we strive to improve the ability to analyze argumentative discourse with precision and nuance, integrating both linguistic and multimodal dimensions of communication. This approach paves the way for a more comprehensive and enriched understanding of argumentative and persuasive processes.