Home

Since 2007 the field of natural language generation has organised shared-task evaluation events every year, under the Generation Challenges umbrella (in some years with funding from EPSRC, UK).

In the course of these shared tasks, a wealth of data has been created, along with associated task definitions and evaluation regimes.

This repository brings together legacy Generation Challenges resources as well as some other standardised data resources that have the characteristics of a shared-task but haven't been run as a competition (e.g. Prodigy-METEO). For each task, the repository aims to provide:

1. The task definition;

2. Input and output data;

3. Evaluation software and descriptions of any human evaluations;

4. Documentation;

5. A list of publications reporting previous results for the same task.

Tasks and resources are grouped together according to NLG subfield, as shown in the navigation panel to the left. The task language is English unless stated otherwise.

Credits

Site created by Anya Belz and Albert Gatt, see Belz & Gatt (2012) A repository of data and evaluation resources for natural language generation, LREC'12 Conference, European Language Resources Association.

Site currently maintained by Anya Belz, Simon Mille, and Anastasia Shimorina.