In our research, we want to understand how people produce language. How they manage to quickly convert abstract ideas into more or less coherent linguistic expressions, possibly supported with adequate non-verbal cues such as facial expressions and gestures. How, in brief, they go from “intention to articulation”, or, briefer still, from “mind to mouth”. To find out, we conduct controlled experiments and study texts in corpora.
We also build computational models which automatically generate text from data. We believe that such natural language generation (NLG) models offer an excellent way to increase understanding. Some of the models are inspired by what we know of human language production, others rely on state-of-the-art statistical natural language processing (NLP) and machine learning (ML) techniques.
An added value of building computational models is that they can have practical applications, in domains such as education, health, media and journalism. Examples of application areas we work on include:
- developing ‘robo-journalists’ that can automatically report on sports and other events,
- generating spoken output for social educational robots,
- summarising online discussion forums to make them more accessible on mobile devcies, and
- generating individualised treatment reports on medical data.
Much of our research is funded by external funding agencies, in projects where we collaborate with academic partners such as Malta University, Twente University, Radboud University, Aberdeen University, and the Fontys Future Media Lab. Commercial partners include Sanoma Media BV, SoftBank and Philips Research.
Flow.ai is a commercial spin-off of research conducted in the language production group. They develop conversational AI interfaces which enable people to communicate with smart machines through natural language.
The Language Production Group proudly hosts the NWO-funded Tilburg Table Soccer Lab, used to study how sports and emotions influence language production.