ReMLAV: Relational Machine Learning for Argument Validation
The ReMLAV project aims to extend the state of the art for argument validation, one of the four fundamental scenarios in the area of argumentation machines that the SPP proposal identifies. We define argument validation as the problem of establishing the validity of a single step in an argument chain, within the context of the entire argument chain.
Our project is interdisciplinary: We aim to make a new methodological contribution by building on and uniting work from computational linguistics (embeddings or high-dimensional representations of linguistic objects), machine learning (relational machine learning methods, including tensor factorization and deep learning methods) and data mining (subspace analysis of complex embedding spaces).
The goal is to develop robust methods for real-world scenarios, focusing on the analysis of arguments and their contexts in natural language texts. Furthermore, we aim to develop scalable methods for real-world scenarios by building on methods that are scalable to very large text collections: embedding learning (computational linguistics), tensor factorization (relational machine learning) and subspace clustering (data mining).
The SPP emphasizes that the end user of argumentation technology is a human being. Thus, returning a yes/no or valid/invalid answer and nothing else is not sufficient. An explanation of the decision is also required. Leveraging data mining research, we aim to use subspace models for explanations in computational argumentation.