Applications of MT Information retrieval Translations for users who want to find out about the essential content of a document 1-to-1 communication Translations of publishable quality
MT – How does it work? Two main approaches: Direct - Lower-end systems Direct translation via bilingual dictionaries, relatively little analysis of sentence structure Transfer - Higher-end systems Interpretation of meaning with analysis of word and sentence structure and generation of target
MT market Internally developed MT systems for specific environments (mostly large corporations) Low-cost or free MT translation offerings (for non-professional use) MT engines combined with subject- related dictionaries for corporate customers (esp. for larger companies)
Progress in technology How much progress has been achieved in the past years? There has been virtually no progress in MT technology in the past 15 years at all. And what are the prospects for the near future? Not that bright - major players have left the playground
But still it works much better today Processing power has multiplied MT vendors provide MT systems with much more knowledge (terminology databases)
Languages and file formats Major MT solutions focus on a few languages Few file formats supported
Supported target languages xxLinguatech PT xxxxxTranscend xxxxxxPower Translator xxxxLogos xxxxxxSystran PorSpaGerItaFreJapMT system
Basic issues What MT solutions are available for the source and target languages required? Does the amount of documentation justify an investment in MT? What is the type of translatable documentation? What are the quality requirements?
Determine quality requirements Who is the target audience? Is translation provided for free? Will the target audience accept MT like style? What risks could a mistranslation involve?
What is the appropriate quality? Why should expectations for MT be any lower than those for human translations? Goal: MT+Post-editing=quality of human translation Affects attitude of post-editors
How does the source material impact MT output? Well-prepared source material is the pre- condition to successful use of MT MT-specific source issues: typos, semantical ambiguities, improper punctuation, passive voice, proper names Controlled Language may be a solution
What is Controlled Language? Defines authoring Guidelines for technical writers (e.g. style) Leads to simplified language usage (e.g. length of sentences) Restricts language usage (e.g. approved terminology)
Examples Nortel Network: NSE (Nortel Standard English) Caterpillar: CTE (Caterpillar Technical English) GM: CASL (Controlled Automotive Service Language) IBM: Easy English
Controlled Language checker Assists authors in complying with companys controlled language standards Can be customized to text type Examples: LANTMASTER IBM Easy English Analyzer Microsoft Word grammar and spell checker
Terminology Comprehensive terminology database is key to MT use (coverage) Terminology databases for MT have different requirements than for human translation Extract and define terminology before start of translation
Integration of translation memory MT integration available for a variety of MT systems (Trados, Transit, IBM TM) Flagged MT proposals No MT proposal when Human Translation is available (both exact and fuzzy)
Quality assurance Post-editing is NOT the same as QA. It is part of translation process. Quality Assurance is an additional process.
Human resources Train technical writers on using Controlled Language and CL checker Find and train qualified post-editors Decide on payment method (per word, per hour) Consider using internal resources
… not just (human) resources Post-editing requires even higher qualification than translation Prevent post-editors from getting frustrated by just correcting errors Motivate them to maintain and improve the MT system
Conclusion Thoroughly evaluate MT solutions applicable to given scenario Consider all implications from authoring process to final quality assessment Invest in Controlled Language and continous terminology management