[19:29:52] <wm-bb>	 <Hogü-456> What do you think how many different structured sentences are needed to write well readable texts. I have looked to the proposal for a Natural Language Genaration System and at the demonstrations of Ninai at https://gitlab.com/mahir256/ninai/-/blob/main/demonstrations.md I currently want to understand how complex it will be to generate texts so how much needs to be done before that is possible.
[19:32:17] <wm-bb>	 <Jan_ainali> That depends on what you want to write about and how long you want it to be. For a lead section of a biography, 3-4 sentences should be plenty. (re @Hogü-456: What do you think how many different structured sentences are needed to write well readable texts. I have looked to the proposal...)
[19:42:26] <wm-bb>	 <Hogü-456> So my question is, how many different examples are needed, that people can write based on these examples well readable texts through using the same structure and only replacing the Q-IDs and the Lexemes. If 3-4 different sentences are enough I think that would be great and not much. Are there statistics what kind of sentence is used how often within a specific Wikipedia Language Version.
[19:45:23] <wm-bb>	 <Jan_ainali> _Needed_ is ambiguous. Do you want to be able to generate a text for all Wikidata items? Well, then it's probably going to be millions of different sentences. But to get started in one corner, I think just a couple of well curated ones will be good enough. (re @Hogü-456: So my question is, how many different examples are needed, that people can write based on these examples well readable texts thr...)
[19:47:39] <wm-bb>	 <Hogü-456> Yes the question is about being able to generate text for all Wikidata items. Thank you for your answer and it is good to know that there are many different sentences necessary to cover all.
[19:48:01] <wm-bb>	 <Jan_ainali> Also, hopefully people will _not_ be manually replacing neither Q-ids or lexemes for each article. If we can write the functions smart enough, you'll probably be able to cover entire fields (and in any language) without further intervention..
[19:50:02] <wm-bb>	 <Jan_ainali> My sense is that we will never achieve that (since new Wikidata items are continuously created). Just getting to a couple of percent would be a huge success. (re @Hogü-456: Yes the question is about being able to generate text for all Wikidata items. Thank you for your answer and it is good to know t...)
[19:54:08] <wm-bb>	 <Jan_ainali> (This assumes the renderers for languages are good enough for the purpose, that all lexemes needed already are well defined and that the necessary items have labels in the right language. Which is a huge assumption. 🙂 ) (re @Jan_ainali: Also, hopefully people will not be manually replacing neither Q-ids or lexemes for each article. If we can write the functions s...)
[19:58:03] <wm-bb>	 <mahir256> It's possible to some extent with Ninai to write "renderers" that can generate a sentence given only a Qid as input; there's definitely some streamlining needed, though (re @mahir256: I have begun modularizing person lead sentences a bit (haven't yet decided how to add birth/death date to the returned construct...)