Games-With-A-Purpose for Data Labelling

Supervisor: Prof Massimo Poesio

Research group(s): Cognitive Science, Game AI

The development of Natural Language Processing systems (or, indeed, any system developed using Machine Learning methods) still depends for the most part on the availability of large collections of annotated data. One of the most promising approaches to tackle this problem has been the development of Games-With-A-Purpose such as FoldIt or GalaxyZoo - attractive online games that label data as a byproduct of playing them. In our lab we have developed a number of such games, in particular for NLP applications, and we are carrying out research both in developing better and more entertaining games (including research on the HCI of such games), and in using the data thus generated for NLP applications (using e.g., Bayesian inference methods for aggregating the labels, and ML techniques to train systems over noisy labels). The candidate could be involved in any of these activities, in collaboration with researchers from both the Cognitive Science / Computational Linguistics areas and from the Games and AI research group.