Services to Weblicht

[stranak]

I couldn't find a better place to make this issue. If there is one, feel free to move it and close it here.

Currently we have only one service, Cesilko, in Weblicht.

Identify all our suitable web and add them to Weblicht:

• [ ] Morphodita (CZ and EN)
• [ ] NameTag (CZ and EN)
• [ ] Korektor (CZ)
• [ ] ElixirFM (Resolve method?)
• [ ] Moses translations (EN-CZ, CZ-EN, at the very least)
• [ ] UDPipe
  • selectable language
  • selectable level of processing.
  • Maybe also output format
• ???

Do NOT add Treex::Web scenarios as Weblicht services for now, it is too slow.

Look into which option to choose (or talk to Weblicht devs):

• [ ] simple services with some "sensible default" settings. Probably what is the default of the service web page. E.g. Morphodita: Tag plain text, use guesser, original tags.
• [ ] configurable services (see SfS: Tokenizer/Sentences in Weblicht) that allow to set things similar to our web pages of the services.

[stranak]

One issue in case of UDPipe but probably not only: TCF I/O.

Milan wrote:

Bylo by potřeba přidělat načítání (do https://github.com/ufal/udpipe/blob/master/src/sentence/input_format.cpp) a ukládání (do https://github.com/ufal/udpipe/blob/master/src/sentence/output_format.cpp).

Bude to spíš technické -- hlavně kvůli přidání nějaké XML knihovny (která musí být celá v UDPipe, kompilovatelná na všech systémech, zabalitená do bindingů, a ne moc velká :-) Kdysi jsem na to měl vybranou pugixml, mám jí dokonce v MicroRestD https://github.com/ufal/microrestd/tree/master/src/pugixml s nějakými potřebnými úpravami (když jsem uvažoval, že bychom MorphoDiTu zapojili do Weblichtu, přidal jsem si tu XML knihovnu do našeho REST serveru).

[stranak]

I would advocate for Weblicht to add CoNLL-U ➡️ TCF converter as a module. There is a lot of SW outputting (and consuming) CoNLL format and it seems sub-optimal that all of that should implement TCF I/O independently.

[kosarko]

Closing for now; if there is anything urgent let's reopen the issue.