Language Technology at UiT

The Divvun and Giellatekno teams build language technology aimed at minority and indigenous languages

View GiellaLT on GitHub divvungiellatekno/giellalt.uit.no

Meeting setup

Agenda

  1. Opening, agenda review
  2. Reviewing the task list from two weeks ago
  3. Documentation - divvun.no
  4. Corpus gathering
  5. Corpus infrastructure
  6. Linguistics
  7. Speller infrastructure
  8. Other issues
  9. Summary, task lists
  10. Closing

1. Opening, agenda review, participants

Opened at 10:07.

Present: Børre, Maaren, Sjur, Thomas, Tomi, Trond

Absent: Saara

Main secretary: Tomi

Agenda accepted as is.

2. Reviewing the task list from the last meeting

Børre

Maaren

Saara

Sjur

Thomas

Trond

3. Documentation

Documentation tasks:

  1. Add documentation on our corpus infrastructure and our corpus work in general (“To be done by the ones making the corpora”: Børre, Tomi, Trond, Saara).

  2. Now we have 4 documents:

    1. Correct corpus (disamb usage)
    2. Corpus plan (for the disamb corpus cwb)
    3. Corpus conversion, two versions, in infra and in ling. Tomi and Børre have done parallell work ;-(
    4. catxml

For the basic corpora, we need 3 types of documentation, or doc for 3 target groups:

  1. For the users/linguists:
    1. What corpus are found, how do I use them (this info is now scattered)
  2. For the collectors:
    1. How do I add texts, where do I add them, how do I convert them (this is the Corpus conversion doc)
  3. For the programmer
    1. What did I actually do? (this is partly the catxml doc)

For the work on the graphical user interface, we need documentation as well, in principle along the same lines, except that the user is not the same linguist as above.

  1. add/update Aspell documentation (Tomi)
    1. Some documentation has been written, but there still is work to be done.
  2. as always: document what you’re doing:-) (all)

4. Corpus gathering

Tasks:

Contracts

Tasks:

The most problematic issue:

Who has the copyright of extracted material, like single words, collections of words, syntactic structure (potentially with some words filled in)? We need this to be controlled by us, not by the authors. The exact borderline is hard to define.

North Sámi New Testament

Lule Sámi Dictionary

We will invite Anders Kintel to a meeting in Tysfjord on Nov 17th, where we will discuss the possibilities and requirements for cooperation. The overall meeting in Tysfjord is there to prepare the municipality for being included into the legally defined Sámi-speaking area.

Bitte and Børre will participate, as well as Sjur, given that Anders Kintel is available.

5. Corpus infrastructure


For reference: This is what we decided in Helsinki:


admin/depts/  (governmental departments)
      guovda/ (Guovdageaidnu municipality)
      karas/  (Kárášjohka municipality)
      sd/     (Sámi parliament)
      others/ (everything else)
bible/ot/
      nt/
facta/
ficti/
laws/
news/MinAigi
    /Assu
    /NRK
    /YLE
    /other ```
  • Reprocess the old (from new dir.) corpus files

Naming conventions and directory structure

  • The original file should be protected using file and directory permission.
  • The meta information (i.e., the xsl translation files) should be under version control
  • Given that our language detection works well, the intermediate file don’t need to be under version control (the lg identification tool is under gt/script, and it needs to be made part of the coprus processing)

Tasks:

  1. Make a system for file and directory permission (today: we all belong to the cvs group), to only allow people with root user privileges write access to the corpus repository, at least regarding original files
  2. Include the xsl files under version control (cvs? rcs?)
  3. Incorporate language detection as part of the corpus processing.
  4. the dir structure is:
    1. one dir for orig, containing also the meta-info and interm. files
    2. another dir for our ready-to-use xml files after conversion
  5. dir structure for web-posted corpus files:
    1. subdivision according to week or month, we start out with month till we see the amount of traffic (yyyy-mm)
      1. Done
  6. we need a way to deal with hyphenated documents in catxml/preprocess:
    1. in normal cases hyphenation points should be removed
    2. when testing the robustness of our parsers, as well as when testing the hyphenator, the hyphenation points should be retained

Corpus conversion

Pdf to XML

Saara has made a new conversion module, it is almost finished. We’ll return to the issue, evaluation, etc. on the next meeting.

Task: Saara to prepare for this presentation, and to make documentation.

perldoc gt/script/samiChar/Decode.pm

(X)HTML to XML

Tomi has been looking at this, and is making an xsl script for it. The web form developed by Tomi should be augmented to allow posting of URL’s as well as documents from the local file system.

The URL posting need to check whether the same URL has been posted before, and if so, whether the page has changed.

Task: Tomi and Saara to present status quo and suggest routines, merger, etc. on the next meeting.

The documentation for corpus conversion should be added to the gt/doc/ling/corpus_conversion.xml document.

6. Linguistics

Name lexicon

Summary: see the newsgroup

Motivation:

  • Divvun: We want to cross-link different versions of the same locations in different languages
  • Common: We do not want to enter the same names twice. We want a language-independent name lexicon
  • Disamb: Having a richer tag set makes it easier to disambiguate
  • Future: Richer analysis makes new applications possible, within information retrieval, grammar checking, machine translation etc.

Needed: A plan for this project:

a. do the main markup in the present propernoun file b. make a script for converting it to xml (to be done one time) c. make a script for xml2lexc (to be done by the makefile) d. make the tags etc. in the parser

Conversion:

  1. This week
    1. clean up the present infl. lexicons (merge BLIND and BERN, VUOLAB and LONDON) - Trond
    2. Make an emacs mode for markup (Saara). Options: fem, mal, sur, plc, org, obj, none). Combinations: surplc
  2. (end of this week and) Next week:
    1. Mark up as much as possible within a week or so (Maaren to do the Sámi names, and to split CNAME into BERN and LONDON, Trond and Børre to look at the rest)
    2. Then convert to xml
    3. Then mark up the rest with correct semantic tags
    4. This means we would need a seventh option, the unspecified name.
    5. Look into efficient editing of the XML lexicon
    6. Look into synchronisation issues with risten.no - we want the names there as well

Updated status quo:

  • Entries: 20000
  • Converted: 13500
  • Time used: 10 h

Needed tools: An emacs mode doing this (Saara):

  1. Go to next “ NAME ;” ( where NAME is a string of symbols “A-Z-”)
  2. Wait for input, one of these: m f s p o b
  3. Replace “ NAME ;” with “ NAME-mal ;”, “ NAME-fem ;” etc. and go to next “ NAME ;”

Possible refinement: Encode for combined options (both plc and sur, e.g.) already in this phase.

Waiting for emacs mode.

Twol SETS definition issue

The definition of G1, G2, G3 in Lule Sámi is still open. and we would like to have input on this issue.

Update: it is still not working, see [bug 193|http://giellatekno.uit.no/bugzilla/show_bug.cgi?id=193]

SUGGESTION (Trond): Thomas, Trond and Sjur have a new meeting on this issue, on Wednesday.

North Sámi

  • three-part compounds issue still open, as is the number project.
  • The treatment of Sámi place names, we need a contract with “Norge digitalt”, via UFD.
    • Sjur has written an e-mail to the UFD contact person, Øystein Johannessen, who will look into it soon. He has not responded beyond saying he will return to it. Sjur brought this up in the board meeting, and Bjørn Olav Megard will remind Øystein Johannessen about this issue. Sjur will follow up on this one.
  • normativity issues:
    • the Giellalávdegoddi meeting was last Friday, they will have a new meeting in December. They were not able to make any decisions, and there will be a new Giellalávdegoddi beginning next year who won’t make decisions until late spring. This is a serious problem for the Divvun problem.
      • Actions: Sjur will bring this to the Divvun board, write a new letter to the Giellalávdegoddi, emphasizing the needs and timetables of the project

Lule Sámi

Sjur, Thomas and Trond will cont. Lule Sámi issues.

Numerals

  1. An empirical overview
    1. Numeral generation
    2. Numeral inflection
    3. Numerals as parts of compounds
  2. A clear concept of how we want to treat them
    1. Tagging
  3. A treatment

We will return to this issue after the name conversion.

7. Speller infrastructure

Nothing this week either.

8. Other

Technical issues

  • The mac os / perl bug (at least Trond and Sjur has it):
    • utf8 “\xC4” does not map to Unicode at /Users/trond/gt/script/preprocess line
      1. This msg did not show up in 10.3 (perl 5.8.1), but does so in 10.4 (perl 5.8.6). It is probably a perl - OS mismatch. (Trond, Thor Øivind, Tomi)
      • Another example of the same bug:
      • :”\x{00c3}” does not map to utf8 at ../script/preprocess line 113, <> chunk 33.

Bug fixing

10 open bugs (and 24 risten.no bugs)

Buying

  • rucksacks for all

risten.no

  • Organisation: could Tomi be used, in exchange for more linguistic work by (old) GIO members? Yes, it is ok, but how much still needs to be evaluated
  • it is ok to integrate “kvensk” placenames with risten.no
    • this should be integrated with the general proper name work - we want all proper names integrated with risten.no, df above
    • needs further development of risten.no to allow for multiple XML bases to be presented and maintained in parallel. This is to be further worked on by Tomi and Sjur

Meeting with Kvensk revitalisation project

Grammar, dictionary, placename lexicon for Kvensk. They want similar infrastructure as in Sámi language technology. Trond and Sjur will discuss how we can help them without taking too much attention away from our real jobs.

9. Summary, task list

Børre

  • Contact oahpahusossodat and the rest of the SD about texts
    • Doing some digging into WebSak
  • Reorganise the directory structure
  • Put all corpus texts into one place
  • Continue converting text from input format to our xml
  • Have a look at the placenames files.
  • Ask Thor-Øivind to move bugzilla to our new webserver.
  • Gather public texts
  • Work on the name lexicon

Maaren

  • The missing list, both the overall missing list from our xml corpus, and a file-for-file review, in order to get different terminology.
  • continue working with the missing list from risten.no
    • working with the missing list from risten.no this week (today)
  • Start working on Sámi place names
  • Start working at normativity issues (numeral issues with Trond?)

Saara

  • Look at the corpus infrastructure issue
  • Look at the corpus interface issue with Lars
  • Convert texts from .doc to .xml, to get a grasp of our corpus format
  • make an emacs mode for the name project (cf. specs in the memo above)
  • prepare for a presentation of the pdf etc. conversion together with Tomi for the next meeting.

Sjur

  • Lule Sámi twol problems, look again at the sets definition with Thomas and Trond
  • risten.no bugs and fixes
  • follow up on voice group-chat not working to Sámediggi
    • Now awaiting cost evaluation from the IT guys (Geir Kaaby et al)
  • project planning with Trond, continued
    • also look at the development processes - specification and testing
  • Follow up on the meeting with Anders Kintel 17th of November -> ask Berit Karen Paulsen/Bitte
  • Follow up on place names from Norge Digitalt -> remind Bjørn Olav Megard
  • Evaluate SFST as speller (and analyzer) lexicon
    • more thorough analysis than was possible in Guovdageaidnu
  • write a background document on the corpus contracts
  • Discuss the contract issue with Trond, return the new version to the lawyer
  • write to the board about the lack of progress with the Giellalávdegoddi, and the problem it causes for the project
  • write to the Giellalávdegoddi once more, emphasizing timetable and response needs in the Divvun project
  • discuss kvensk project support with Trond
  • write public tender documents

Thomas

  • work on Lule Sami compounding and derivation
  • Look at Linguistic bugs with Trond
  • Meet with Sjur and Trond about the definition of G1, G2, G3

Tomi

  • Aspell: Continue working on the affix file & aspell
    • Contact aspell author (UTF-8 thing)
  • three-part compounding
  • corpus infrastructure: dtd location (both public and internal)
  • corpus infrastructure: file and dir organisation
  • Document aspell and corpus infrastructure
  • Cgi-script for uploading documents to corpus base
  • Specification for new catxml in C++
    • this includes also placing the source and binary
      • clean the script/ catalogue with Trond
  • Common makefile issues

Trond

  • Work on the bug list (7 open).
  • Get the new version of the New Testament
  • project planning with Sjur, continued
    • also look at the development processes - specification and testing
  • Discuss the contract issue with Sjur, return the new version to the lawyer.
  • Work on the name project: Clean up the lexicon file, discuss the emacs mode with Saara and the work with Maaren and Børre.
  • Add docu on the corpus infrastructure
  • clean the script/ dir
  • discuss kvensk project support with Sjur

10. Next meeting, closing

24.10.2005 10:00

Closed at 12:11