[07:52:05] <wm-bb>	 <marijke00> Where you want or ought to store the constructors also depends on what you envision or assume them to be. If they’re for step 1/content determination in the reference NLG pipeline (which I assume it to be), then it makes sense to go for wikidata, since it will facilitate the content selection process.
[07:52:05] <wm-bb>	 <marijke00> If they’re aimed to be for steps 3-6,  i.e., more on the language generation  up to realisation side, which requires computation, then on/closer to wikifunctions may be more effective. The example in the update letter is a chimera of all sorts of things that suggests trying to do the declarative parts of all 6 steps at once in one place, which I’d advise against. (There are m <clipped message>
[07:52:07] <wm-bb>	 <marijke00> any good reasons for why all (structured content -to-text) NLG systems use a pipeline approach rather than anything else.)