Section: Overall Objectives

Highlights

Generating Instructions in a Virtual Environment.

Talaris participated in the preparation of the international GIVE 2.5 challenge on the Generation of Instructions in a Virtual Environment. This challenge brought together researchers from six universities and evaluated the participating systems on their ability to generate instructions in a dynamic 3D setting [40] . One of the two systems developed by TALARIS [26] won the second place both in terms of objective and of subjective metrics.

Generating from Knowledge Bases.

Talaris' work on data-to-text generation has attracted increasing interest this year. The GenI sentence generator developed by Talaris was licensed to be used by Stanford Research International (SRI) in the large scale AURA (Automated User-centered Reasoning and Acquisition System) project whose aim is to provide an Intelligent Electronic Textbook that could be used by teachers and students; negociations are currently underway with SeeReason Partners LLC for another commercial license; and a two day invited visit to Bolzano lead to the launch of a new collaboration which will focus on integrating GenI in the Quelo system developed by the KRDB group to verbalise queries on knowledge bases. In parallel, Claire Gardent gave an invited talk on data-to-text generation at the French NLP conference (TALN, Traitement Automatique des Langues Naturelles) and an invited tutorial on Generation for the Semantic Web at the K-CAP (Knowledge Catpure) conference in Banff, Canada.