7.12.16 Participants


  • MT
  • Võru FST
  • Papers/articles
  • Next meeting

Estonian FST


Heiki: Consider expest.fst also a documentation for grammar work, try to make the code also grammarian-readable.

Jaak: Continued getting rid of multiple generations. It is now somewhat better, but still uncommitted.

Filter is in src/filters/ Need to include it in the src/filters/Makefile.am to make sure the regex is turned into an fst.

Then, to use the filter when building the Oahpa fst, add a target for analyser-oahpa-desc.%:analyser-oahpa-desc.tmp.%, and specify how to use the filter in that target.

Fst build process:

1: build *-raw-* fst, containing everything with tags in a known order
2: build *.tmp.?fst from the *-raw-* fst (language independent build step)
3: build the final fst from the *.tmp.?fst file (language-specific build step, ie the place to add filters only relevant to estonian); if there are no build steps here, the tmp file is just copied to the final target file

Documentation for namning convetion:



  • Coverage of large corpus
  • Result of generation
    • Number of double generations
  • common lemmatest: Find a lemmalist and test
    • generate the lemma forms of the biggest (or most relevant) est bidix
    • Eventually some other lemmalist
  • different lemmatest: make check
    • In this test we measure each fst against its own lemmalist
    • 70
  • Other?


  • Jaak sets up tests, and Jaak and Heiki (etc) to test


Heiki-Jaan and Sjur have discussed the jspwiki generated documentation issue in Helsinki


est -> fin

Tänases Postimehes langevad viimased maskid: idarahaskandaalist alates on viimased kuus aastat jäetud avalikkusele mulje, et Keskerakonna ja Ühtse Venemaa suhted seisavad külmutatuna.

  • Tänases *Postimehes lankeavat #uusin *maskid:
  • idarahaskandaalist #alkaa ovat #uusin kuusi vuotta *jäetud julkisuudelle *mulje, että #Keskeinen puolueen ja *Ühtse Venäjän maan *suhted seisovat *külmutatuna.


document generation issues Heiki had and Trond corrected: in /doc, make-B; forrest, and get the error message:

X [0]                                     root-morphology.html        BROKEN: Couldn't accept input link ["[ue]"]

this indicates that

    [ ] 

was used in the documentation to be generated (following !! ), and it should not have been.

Võru FST


Not much happened lately. Problem: Identifying A vs N from a 6000 lemma list. Trying to do that via POS info from Estonian. Probably a 2000 residue of those 6000 will remain for manual work.



  • typology papers (typology of MT systems, that is)
  • ordinary MT system presentation papers
    • sme-fin is a good candidate


  • Systems in use papers


  • heiki vs. jaak paper (subtracting or not)
  • presentation FST papers
  • lexc vs twolc: where to put the complexity






  • Push this to a more concrete level to the next meeting (all)
  • Articles as issue next time

Summary of the Helsinki trip

Heiki presented Hki slides, and still has had no lunch.

http: //blogs.helsinki.fi/language-technology/news/bault-2016/ Highlight: Lg tech developers out there meet academia

Next meeting

Monday Dec 19th 13.30 Norwegian time