User:Khannatanmai/GSoC2020Progress

Work Plan: http://wiki.apertium.org/wiki/User:Khannatanmai/GSoC2020Proposal_Trimming#Work_Plan

Ongoing

Phase 3 (July 31 - August 24)

• Proper error handling of wordbound blanks
• Use transfuse in apy, wikimedia translations

Completed

Application Review Period (March 31 - May 3)

• Compile all the discussion about the modification to the stream format (in talk pages)
• Create dedicated page for the development of the new stream format: User:Khannatanmai/New_Apertium_stream_format
• Going through the documentation again and reading the wikis for each module just to ensure I haven't missed anything in the overall working of Apertium as I've never really made a language pair.
• http://wiki.apertium.org/wiki/User:Khannatanmai/New_Apertium_stream_format : Document modification to Apertium stream format (see talk pages for relevant discussion)
• Document how much change is needed in which parsers and what the change is
• Proof of Concept for the new format
• Document the change needed in tokeniser, bidix lookup, and generation to include surface form: User:Khannatanmai/Eliminating_Dictionary_Trimming
• Document all the proposed benefits with including secondary information

Community Bonding Period (May 4 - June 1)

• Create a suitable development and debugging environment for the pipe (Xcode)
• Modifying transfer to pass secondary tags ahead. Updates can be found here.
• Modify generator to ignore secondary tags while matching
• Deal with MLUs in generator, and special characters in sectags, etc.
• Analyse the code of the parsers of the modules
• Fix transfer behaviour with LUs with invariable parts and MLUs
• Need to deal with sec tags appearing before lemq if lemq comes from variable
• Wiki for all features being implemented for secondary tags here.
• Testcase: lemq comes from variable
• Create test t1x file which covers all test cases.
• Run thorough regression tests on eng-spa (multi stage transfer) and spa-cat(single stage transfer)
• Manually insert secondary tags in the stream and test if they reach the generator
• Prepare an alternate proposal to secondary tags: User:Khannatanmai/Alternate_stream_modification

Phase 1 (June 1 - July 4)

• Deal with the community's objections to secondary tags.
• Come up with a method everyone is happy with
• Analyse the needs of WikiMedia's markup handling.
• New page for the development of word bound blanks: User:Khannatanmai/Wordbound_blanks
• Add tests an examples which have merging, splitting, deletions, insertions, etc.
• Changed formalism so that wordbound blanks are now before an LU
• Modify chunker to deal with wordbound blanks
• Write tests for the chunker

Phase 2 (July 3 - July 27)

• Make sure regression tests show no regression
• Modify interchunk and postchunk to deal with wordbound blanks
• Write tests for chunker, interchunk, postchunk blank handling
• Modify pretransfer to split wordbound blanks
• Write tests for pretransfer blank handling
• Deal with separable and merge blanks when multiwords are formed
• Add tests to -separable for wordbound blank handling
• Make lt-proc parse wordbound blanks as normal blanks correctly for analyser, generator, biltrans, and post generator
• Add tests for lt-proc analysis of wordbound blanks
• Add test for anaphora handling wordbound blanks
• Handle wordbound blanks in apertium streamparser.
• Add feature in transfer and postchunk so that it outputs the wordbound blank automatically if there's only one LU in the matching pattern.
• Wordbound blank handling in postgeneration as it has many-many rules
• Tests for wordbound blank handling in postgeneration
• Changes in separable for wblank handling to work with revautoseq as well.

Phase 3 (July 31 - August 24)

• Parse wblanks and store with LU in -recursive
• Get wblanks in recursive output using parallel wblank stack that mimics main stack operations
• Add tests in recursive
• Fix wblank handling with XML TRX rule files
• Wordbound blank handling with variables in recursive
• Test wblanks from variables and MLU wblanks in recursive
• Modify apertium-tagger to parse wblanks as normal blanks
• Modify hfst-proc to parse wblanks as normal blanks (analysis, generation)
• Fix this error: http://codepad.org/yU4uaSNX (transfer error in afr-nld) - used old transfer, convert to new
• Fix wblank printing error in pairs that use t4x
• Test if wordbound blanks go through the pipe properly in all pairs