VF4TA

Development Roadmap

vaytt
vf4ta
ruleMngr

Rules for writing rules

• Faster
  • Multi-thread:
    • node.js?
    • ajax calls?
    • reactphp?
• Offline?

Give feedback

• Minimum: false positive
  
• Maximum: full explanation
  

QA functions reworked

Unique rule-based function

• if srcExpr found in srcSeg
  • if tgtExpr (!) found in srcSeg
    • new error

• Already existing options:
  • Same mumber (bool)
  • Regex or Wildcard or Literal
  • Full segment (bool)

• Become options:
  • Disallowed: (!)
  • BackRef: () -> \1
  • ForRef: \1 <- ()
  • Back/For: dir = s2t/t2s, or f/b

Profiles

• To store and reuse preferences/settings
• Central maintenance by manager
• Different scopes:
  • Default (all)
  • Language preferences
  • Container preferences
  • User preferences

Database

• Profiles
• Rules (previously as CSV, now as SQL)
• Checks
  • language/container-specific (rule)
  • generic (no rule)
• Reports
  • errors
• Errors
  • rule_id, seg_id, edit_id
• Logs (stats)

Input types

• Existing (file or zip batch):
  • XLIFF
  • Excel (2 columns)
  • TMX
• New:
  • OMT packages
    • Analyses target XLIFF files
    • Analyses project_save.tmx
  • RFF: Run Omt to create target files
    • if latest date in project_save > latest date in target files

Redefinition of some checks

Untranslated vs. Empty

• in bilingual files
  • empty: tgt = 0
  • unaltered (xlat identical to src): src = tgt
• in OMT packages
  • empty: ><
  • unaltered: src = tgt
  • untranslated: src = tgt + !E seg in proj_save

New check: Fuzzy (VF4TA)

• Find similar in
  • File
  • Batch
  • Container
  • Global (all ML)

New check: Consistency+

• Find inconsistent translations in
  • the file
  • the batch/project
  • the container
  • globally

RFF: Edit translation

• Editable target version
• Saved in the report
• Merged back in file

• Run on:
  • input files in Vaytt
  • TMX files in MemoryLn
    • btn on the right for each row

ruleMngr

• CRUD approach, ruleset database
• Keep track of modifications to a rule, see history
  • modified by upload
  • modified by editing
• New value: version -> allows to compare with previous version
  • table.edits
• Bullet Two
• Bullet Three

ruleMngr

get latest version from rules table +
user who last updated the rule +

history of the rule (row per version w/ user) +

list of checks where it was applied

User feedback onboarded

Client sends request with files and prefs

Server analyses batch and returns report

Client sends back feedback on the report

1

2

3

Frontend rendering (1)

• User feedback (votes)
  • used to fine-tune results:
    • Error (OK)
    • Ignore (FP)
      Never again
  • saved at the end
    • button Save
    • on closing tab
    • inactivity

Frontend rendering (2)

• Display of results staggered by scope:
  • Segment
  • Batch/project
  • Container
  • Global (all ML)
• New column in QA report: Scope
•  

API

• Requests from another server
• Bullet Two
• Bullet Three

R&D

• Exploit language assets (ML = corpora)
  • Train MT engines
  • Extract language information
• Exploit statistics
  • Who did what when?
  • What checks are more/less useful?
  • What checks produce more/less false positives?