The preliminary title of my doctoral dissertation is Hypothetical Reasoning. From Formal Learning Theory to Machine Learning. My research is embedded in a broader project conducted by Peter Verdée entitled Relevance criteria in entailment, epistemic grounding and counterfactuals: semantic and linguistic aspects. The project is funded by Fonds Spécial de Recherche (FSR) and started on March 1, 2017.

The goal of my doctoral thesis is to obtain an interpretation of certain conditional sentences of the natural language in terms of formal learning theory; and more specifically, conditional sentences which correspond to various forms of hypothetical reasoning. This problem relates to an analogous program of interpreting conditional sentences in machine learning, in particular to the issue of opinion mining and feature- or topic-based sentiment analysis (see Narayanan, Liu and Choudhary (2009) for an example of how to use conditional analysis to assign positive, neutral or negative sentiments to topics or product features).

An additional objective of my work is to ensure that the obtained interpretation satisfies certain relevance and explainability criteria, which in machine learning correspond to the problem of black box interpretability. The following two aspects of my dissertation are particularly relevant:

  1. the interpretation of the process of hypothetical reasoning which I propose can be applied uniformly to human and machine learners, and
  2. the probabilistic framework I use allows for producing explanations of all related processes, contributing to increased interpretability of certain machine learning models.