Learning Intentions Following Evolution

Understanding why a certain utterance was said by a participant in a conversation is fundamental for successful communication to take place.

Notwithstanding the central role of understanding intentions in communication, this and other forms of pragmatic reasoning remain beyond the current state-of-the-art in Natural Language Processing (NLP).

For this reason, critical applications of these technologies, such as personal assistants, remain limited to few hand-coded use-cases such as adding events to a calendar and fail to serve as general-purpose communicative assistants. This proposal aims at making significant progress towards genuine Natural Language Understanding (NLU).

It introduces a system that can reason about the pragmatic meaning of a sentence by searching for useful explanations in a space that represents all possible concepts (often called a ""Language of Thought"").

Because such a search process is generally intractable, this project proposes to let the system discover the more efficient way to search.

To achieve this goal, it will model an open-ended evolutionary process that is here hypothesized to lead to emergent exploration strategies.This project bridges Linguistics and Complex Systems with connections to Cognitive Science and Evolutionary Theory.

To maximize its success probability, I will conduct it at the COLT Lab in Universitat Pompeu Fabra (UPF), a leading group in Computational Linguistics, with a secondment at Santa Fe Institute (SFI), a global leader in Complexity Science.