Multilingual Flexible Neuro-Symbolic Language Generation

The core aim of the M-FleNS (Multilingual Flexible Neuro-Symbolic Language Generation) project is to explore the extent to which combining the strengths of neural and symbolic (grammar-based) Natural Language Generation (NLG) systems is possible.

We will build FleNS generators that (i) exploit grammar-based system aspects to address the vexed problems of poor accuracy (including hallucinations and omissions of content) and data and energy-hungriness in neural generators, and (ii) exploit neural system aspects to address problems with fluency, coverage and robustness in grammar-based generators.

The overall ambition is to find solutions for some of the biggest current challenges in state-of-the-art NLG, including semantic controllability, energy greed and suitability for low-resource languages.

Combining the Applicant's expertise in symbolic NLG systems and data annotation with the Supervisor's expertise in machine learning and evaluation for NLG, and benefiting from the excellent ADAPT research environment, we will develop a new type of NLG system that combines the best of both worlds, symbolic and neural, to create better NLG systems and components for real-world applications.The project is well aligned with the European Green Deal strategy and more particularly with the Digital Europe Programme whose main objectives include bringing digital technology to businesses, citizens and public administrations.

Through a combination of working with the Supervisor and her research group, direct training, international collaboration and self-guided study, the Applicant will expand his spectrum of scientific expertise to Deep Learning methods and human evaluation of NLP systems, strengthen his general knowledge in linguistics, and acquire a comprehensive understanding of the IPR and business related skills needed for further exploitation.

This project will allow the Applicant to establish himself as an internationally recognised research leader in the field of NLG.