Syntactically Motivated Task Definition for Temporal Relation Identification

Georgiana Marsic*
*Research Institute in Information and Language Processing; University of Wolverhampton; MB Building, Stafford Street; Wolverhampton, WV1 1LY, UK;
The annotation of temporal relations remains a challenge, being a very difficult task for humans, not to mention machines, to reliably and consistently annotate temporal relations in natural language texts. This paper advocates a change in the definition of the problem itself, by proposing a staged divide-and-conquer approach guided by syntax, that offers a more principled way of selecting temporal entities involved in a temporal relation. The decomposition of the problem into smaller syntactically motivated tasks, and the identification of accurate and linguistically grounded solutions to solve them, promote a sound understanding of the phenomena involved in establishing temporal relations. We illustrate the potential of linguistically informed solutions in the area of temporal relation identification by proposing and evaluating an initial set of syntactically motivated tasks.