Discussion Panel: Convergences and divergences in generation from data, text, and conceptual input.

It has often been observed that the generation of natural language, broadly conceived, can start from many different types of input, varying from text to numerical data to logically structured data (e.g., from a Knowledge Base). We believe the time is ripe to discuss what these different approaches might have in common, and whether one should think of NLG as a unified field. We have asked a number of panelists to consider these issues during a special session at ENLG 2013. Panelists will be giving brief presentations, followed by an open discussion.

Our panelists were asked to consider these issues in the light of the following questions:

  1. Do you believe there is essentially one kind of NLG? Or are there just a bunch of largely separate endeavours, all of which happen to involve the computational generation of natural language?
  2. To what extent do you think your approach to the generation of language could be combined with others in a single computational architecture?
  3. Do you believe that the different types of NLG should use the same evaluation methods? Or are there important differences between them that dictate an entirely different approach to evaluation?
  4. Is it a good idea to have dedicated NLG conferences/workshops (as we largely do)? Or would it be more productive to divide the intellectual cake differently? For instance, NLG events could team up with summarisation, MT, psycholinguistics, etc. Would that be a good idea? Have we got it right, or is it time for a change in we way we organise ourselves as an academic discipline?


  • Pablo Gervás, Universidad Complutense de Madrid
  • Guy Lapalme, Université de Montréal
  • Frank Schilder, Thomson Reuters
  • Yaji Sripada, University of Aberdeen


  • Kees van Deemter, University of Aberdeen

Generation Challenges 2013

This year's edition of the Generation Challenges will be presented during a special session of the ENLG workshop. The session will feature presentations of results of two challenges that have been running over the past year:

  • KBGen: Generating from Knowledge Bases, organised by Eva Banik (Computational Linguistics Ltd, UK), Claire Gardent (CNRS/LORIA, France) and Eric Kow (Computational Linguistics Ltd, UK).

  • The First Content Selection Challenge from Open Semantic Web Data, organised by Nadjet Bouayad-Agha (Universitat Pompeu Fabra, Spain), Gerard Casamayor (Universitat Pompeu Fabra, Spain), Chris Mellish (University of Aberdeen, UK) and Leo Wanner (Universitat Pompeu Fabra, Spain)