I am working in the field of referring expression generation (REG), a subfield of Natural Language Generation. In particular, I work on content selection for noun phrases in both initial and subsequent reference. I am mainly interested in uncovering psycholinguistic factors that have an influence on human language production and modelling these factors in a cognitively plausible way.
|One focus of my thesis work was on the use of spatial relations in referring expressions. I have collected two corpora of descriptions of objects in simple 3D scenes to study how humans use spatial relations: the GRE3D3 Corpus and the GRE3D7 Corpus.|
|I try to maintain an extensive bibliography on referring expression generation.|
|To get a feel for existing referring expressions algorithms I conducted an initial experiment comparing their performance to a data set of human-produced descriptions.|
As a result of my first experiment, I got interested in the issues involved in evaluation of referring expression generation and NLG more general.
|I have done some work on an overgeneration and ranking approach to REG leveraging similarities between domains to be able to produce more than just one description for an object.|
|In collabortion with Mariët Theune and Emiel Krahmer, I have done some work on optimising the use of redundancy in referring expressions.|