Difference between revisions of "User:Popcorndude/GSoC2021Proposal"

From Apertium
Jump to navigation Jump to search
m (→‎Work Plan: typo)
Line 65: Line 65:
Status as of Feb 9: working Python compiler from minimal formalism to trx.
Status as of Feb 9: working Python compiler from minimal formalism to trx.
Feb 23: the coding challenge listed at [[Ideas_for_Google_Summer_of_Code/Robust_tokenisation]] can be found [https://github.com/mr-martian/gsoc2021-coding-challenge here].
Feb 23: the coding challenge listed at [[Ideas_for_Google_Summer_of_Code/Robust_tokenisation]] can be found [https://github.com/mr-martian/gsoc2021-coding-challenge here] and a draft of the file structure for the unit-testing framework can be found at [[User:Popcorndude/Unit-Testing]].
== Work Plan ==
== Work Plan ==

Revision as of 23:24, 23 February 2021

Google Summer of Code 2021 proposal draft


Name: Daniel Swanson

Email: awesomeevildudes@gmail.com

IRC: popcorndude

GitHub: https://github.com/mr-martian

Timezone: UTC-5

Proposal: UNIpertium

This project is really 3 projects UNIfied only by the fact that they all start with UNI:

UNIcode Support

This portion of the project is Ideas for Google Summer of Code/Robust tokenisation. All string processing would be transferred to ICU and all inputs would be normalized. A program would be inserted at the beginning of the pipeline which normalizes incoming text and compilers would emit warnings for non-normalized input.

The normalization will likely be a modified version of NFC. See https://github.com/apertium/organisation/issues/24 for discussion.

See also https://github.com/apertium/lttoolbox/issues/81 and https://github.com/apertium/lttoolbox/issues/85

UNIt-Testing Framework

Our regression tests are few and far between and almost none of them are run automatically. This part of the project would be the creation of a Grand Unified Unit-Testing Framework™, probably mainly derived from https://github.com/TinoDidriksen/regtest (though I have included time in the schedule for bikeshedding the exact feature set).

I would also plan to enable this testing on all existing repositories unless the maintainers really didn't want it. This will entail converting any existing tests from whatever state they are currently in.

See also https://github.com/apertium/apertium-init/issues/51 User:Ilnar.salimzyan/On_testing

UNIversal Dependencies-Based Structural Transfer

I have worked out in broad outlines how to get apertium-recursive's rtx-proc to parse a dependency grammar. This third phase would be implementing that concept and seeing if it's actually useful.

If it works, we'll have a way of creating UD parsers (though I suppose we already have that with CG) though I think the more useful thing is that I have some ideas for a UD generator uses the rules for the parser and uses them to reorder a set of relations. The result of that would be a semi-monolingualized syntactic transfer system, where translation pairs could immediately start using the UD parsers in the monolingual repos and would only need rules for particular problematic constructions.

There's probably also cool stuff we could do with treebanks.

See also [1] [2] [3] [4] [5]


I am a senior at Swarthmore College studying math and linguistics. I have a lot of experience with Python and a decent knowledge of C++. I am a native speaker of English and can read Spanish and Biblical Hebrew.

Repositories I maintain:

Repositories I have been very involved in:

Coding Challenge

Preliminary UD work: https://github.com/mr-martian/ud-experiments/tree/master

Status as of Feb 9: working Python compiler from minimal formalism to trx.

Feb 23: the coding challenge listed at Ideas_for_Google_Summer_of_Code/Robust_tokenisation can be found here and a draft of the file structure for the unit-testing framework can be found at User:Popcorndude/Unit-Testing.

Work Plan

The following work plan is written assuming the number of hours that GSoC recommends (17 or so per week), but it is very likely that having fun or not having much else to do will result in me putting in substantially more time than that. Either this will balance out my inevitable underestimate of the complexity of the task, or I'll end up way ahead and start inventing new tasks for myself.

Time Period Goal Details Deliverable
Community Bonding Period

May 17-June 5

Finalize implementation plan for Unicode support
  • Decide on normalization scheme
  • Figure out how ICU custom normalizers work
  • Determine which ICU types to make default
Draft of string-handling guidelines (= updated version of Code style)
Week 1

June 6-12

Convert Lttoolbox
  • Apply plan from Community Bonding Period
PR on lttoolbox repository
Week 2

June 13-19

Convert Apertium
  • Apply plan from Community Bonding Period
PR on apertium repository
Week 3

June 20-26

Convert remaining tools
  • Apply plan from Community Bonding Period
PRs on apertium-separable, apertium-anaphora, apertium-recursive, and apertium-lex-tools
Week 4

June 27-July 3

Transition week
  • Deal with anything that comes up in testing
  • Finalize string-handling coding guidelines
  • Start documenting existing unit testing frameworks
Unicode transition branches ready for merging
Week 5

July 4-10

Propose Grand Unified Unit-Testing Framework™
  • Present proposed framework on the mailing list
  • Begin bikeshedding
  • Begin prototyping UD compiler for apertium-recursive
Nothing terribly concrete
midterm evaluation Full Unicode support Full Unicode support with proper tokenization and normalization in all native pipeline modules and coding guidelines for future contributors
Week 6

July 11-17

Finish initial UD compiler
  • Get parser working
  • Converter to CONLL-U
  • Figure out how to describe and implement transfer
  • More bikeshedding
Preliminary UD-based transfer system
Week 7

July 18-24

Implement Grand Unified Unit-Testing Framework™
  • Incorporate mailing list bikeshedding
  • Finish implementing/porting/whatevering the system
  • Begin applying to repos
Working unit-testing framework
Week 8

July 25-31

Framework rollout
  • Enable unit-testing in all languages and pairs
  • Convert existing tests to new format
Test reports everywhere
Week 9

August 1-7

Further UD experimentation
  • Try some actual UD grammars
    • See how well it translates with no pair-specific rules, just the reorderings specified by the monolingual grammar
    • See how fast or slow it ends up being
  • Deal with the inevitable issues that arise from rolling out changes to several hundred repositories
Preliminary evaluation of the effectiveness of UD transfer
Week 10

August 8-14

Unknown What I write here will have little bearing on what actually happens and also by not making a plan for this week, I ensure that my overall plan can be at most 90% wrong. Unknown
final evaluation Project done Unicode support with accompanying developer documentation, unit-testing framework for all language and pair repositories, and some potentially useful messing around with UD-based transfer

I have no other commitments this summer and would be able to work on this project full-time.