Mark Granroth-Wilding

Current Research

Since moving to Silo.AI in August 2021, I am no longer actively doing academic research, though I hope to maintain to some extent some of my ongoing collaborations in the areas below.

My research interests lie in natural language processing (NLP), computational creativity and music processing/cognition.

Particular areas of research that I have recently focused on are:

  • multilingual NLP and language transfer for low-resource languages;
  • application of NLP to noisy, historical data;
  • cross-lingual representation learning;
  • combination of Bayesian modelling and neural representation learning.

Before moving to Silo.AI, I was working on the following projects in the Discovery Group at the University of Helsinki.

  • NewsEye
    Building a digital investigator for historical newspapers. I lead work package 4 on topic modelling for analysis of multilingual, noisy output from automated text recognition on historical newspapers.
  • Embeddia
    Within the EU, access to fundamental resources such as local news and government services is limited by the great diversity of the EU’s 37 languages. Embeddia seeks to address these challenges by leveraging innovations in cross-lingual embeddings to allow existing monolingual resources to be used across languages. The project will develop novel multilingual techniques for news analysis and media production.

Previous project: Digital Language Typology

Digital Language Typology:
Producing a platform to assess the structurally manifested family relationships within a set of languages with large digital textual and speech material. We focussed in particular on low-resourced Uralic languages.

Previous project: Immersive Automation

Immersive Automation was a project about news automation in which our team in Helsinki worked on Natural Language Generation. The aim was to build a demonstration of a future news ecosystem, using automatic data analysis and NLG.

The project ended in May 2018.

Previous project: WHIM

My previous work, with Stephen Clark at the Cambridge Computer Laboratory, was part of the What-If Machine (WHIM) project.

Computational Creativity studies how to engineer software which can take on some of the creative responsibility in arts and science projects. The aim of WHIM was to derive, implement and test novel formalisms and processes which enable software to not only invent, but assess, explore and present novel, creative ideas.

For an introduction to Computational Creativity and this project, with a slight leaning towards Cognitive Science, take a look at my talk from KVIT 2015.

Previous project: Jazz Parsing

In my PhD, I investigated the use of linguistic grammars to model and automatically analyze the structure of jazz harmonic sequences.

More about my PhD project