Greenland2017

This document contains an overview of the topics and training given in Nuuk in June, 2017. It contains links to the relevant parts of the existing documentation. The idea is that this page can be used as an overview document also in other cases.

Basic organisation: presentation of topic, followed by excercises. Roughly one topic + excercise before lunch and one topic after.

Topics (with a roughly schedule with time left at the end - to be adjusted as needed):

  • Day 1:
    • presentation of all tools
    • introduction to the infrastructure
    • unix crash course
  • Day 2:
    • dependencies on external tools
    • in-source documentation
    • tag conventions in the Giella infrastructure
    • morpheme boundaries
  • Day 3:
    • testing
    • debugging
    • developer tools
  • Day 4:
    • refactoring the code

Presentation Of All Tools

Introduction To The Infrastructure

Overall goals

  • language independent, while still adaptable to the needs of each language
  • separation of concerns - build structure vs linguistic work
  • scalable:
    • add as many languages as you want
    • add as many tools as you want
    • with a new language, you get whatever all the other languages have
    • with a new tool, all languages get a basic version of the tool
  • predictable - same thing is called the same in different languages
  • understandable - names should be understandable as is
  • modular
  • technology neutral (but rule-based by default - the only thing that works with the languages we work with)
    • it supports both Xerox, Hfst and Foma
    • it supports Apertium, but could also support other (rule-based) MT systems
  • should give strong support to reuse of resources
    • shared lexical data
    • one fst starting point for every tool

Means for achieving the goals

Unix Crash Course

See Unix for Linguists

Dependencies On External Tools

Language technology tools

  • Hfst or Foma or Xerox
  • VislCG3

Optional:

  • Apertium (MT)

Infrastructure support tools

  • Autotools
  • Forrest (documentation verification and publishing)
  • Python
  • Subversion
  • a number of other tools listed on a separate page

External dependencies for final products

This varies from product to product. In most cases (like spellers, keyboards and more) all these dependencies are precompiled by programmers and added as library resources to be included automatically so that the linguists do not have to have everything installed and correctly configured for building the end user products (s)he wants.

The exact details for each product is listed separately in each case, on individual pages.

In-Source Documentation

See a separate document.

Tips for in-surce documentation:

  • open two terminal widows
  • run forrest in one
  • run make in the other

To run forrest: 

forrest run -Dforrest.jvmargs="-Dfile.encoding=utf-8 -Djava.awt.headless=true"

To debug, edit the generated jspwiki file till the error is found, then correct it in lexc.

Tag conventions in the Giella infrastructure

Morpheme boundaries

Clean code

Guidelines for clean code:

  • all forms should have an analysis
  • all analyses should be consistent - same tags for the same feaure, in the same order
  • consistent use of whitespace for increased readability

Testing

See this document

Debugging

Developer Tools

Short intro to the tools in the devtools/ directory: 

check_analysis_regressions.sh
generate-*-wordforms.sh
test_ospell-office_suggestions.sh

Refactoring the code

When the Yaml files are covering the relevant parts of lexc, one can rewrite the lexc code and at the same time ensure that the analysis doesn't change. One can also use one of the devtools/ tools to help in that process.

This can also be seen the other way: when you know what area of the grammar you want to change, you write yaml tests to specify the intended output, and work on the grammar until you get there.

Excercises and practical work

  • correct bugs identified
  • search for new bugs
  • change tags
  • run tests