webinar automotive and engineering content 16.06.16
TRANSCRIPT
KantanMT Partner Webinar “Machine Translation: Managing Translation Memories
for Engineering and Automotive Translation”
16 June 2016
Christian TaubeMatrix Communications GmbH
Technical Lead
About me and my company
I have worked in the translation industry since 1997 as a project manager, technology lead, and consultant. I have co-founded two translation companies, including Matrix Communications GmbH
Matrix is a mid-size language service provider (LSP) with approx. 50 employees, founded in 2006. We are based in Munich and service a wide range of industry and service sector clients throughout Europe.
We are actively using MT since 2011 and have been partners of since 2012.
Machine translation today
Many people are interested in MT.
Buyers realize that MT has limits butoften do not know how to push them.
Few enterprises are ready to invest intoa complex solution.
What the market expects is a tangible and transparent MT service at fair price.
What the market is lacking are sustainedrole models for everyday experience.
Everybody has used MT!
What to look out for?
MT in Industry: Business case context
Disrupt the cost structure!
Where do we see potential for MT?
Productivity aid when translating texts in organizations that have...
• ... some level of control over authoring guidelines• ... a high degree of in-domain specificity• ... well-documented terminology• ... a well-established translation memory process
We find that this potential can be realized in, for example...
• Technical documentation (of many sorts)• User documentation• Financial reporting• Catalogue texts
MT case study
Customer: Engineering company with world-wide presence and a wide variety of high-tech products and service solutions
Content: Service documentation
Language pairs: German English, Spanish, French, Italian, Dutch, Danish, Norwegian, Chinese (Simplified) – further pairs currently in planning
Translation platform: Trados Studio
Implementation: TM + output, post-editors can work in their usual environment
Productivity: Productivity improves significantly per language
Case study: Building MT engines in KantanMT We use customer-specific translation memories as input
(no other, less customer-specific corpora, although that is an option that Kantan has).
We use customer-specific terminology and make sure this is actually in the TMs
We edit our „raw“ TM input for the MT purpose.
Preparatory steps in clean-up before training may include:
Elimination of non-aligned segments, empty segments, segments that are too short/too long on either side, etc.
Normalization of capitalized words, acronyms, etc.
Certain linguistic checks (consistency), terminological checks.
TMX input size ranges from 100,000 to > 700,000 segments.
Our smallest engine was built on 150,000 words (!) – Specificity works!
Automated post-editing rules (PEX) are constantly optimized and extended.
Case study: Success factors and benefits
Success factors we have identified:
• Manage expectations on all sides.
• Identify suitable text types.
• Work with authors to optimize source.
• Invest in your TMs. Use suitable tools to maintain them.
• Train post-editors. Translators can be post-editors!
• Don‘t be afraid of ending up in dead ends. Reverse, and try again.
Benefits gained: • Turnaround times significantly reduced (translator throughput)
• Significant cost reductions
• Final workflow retains the same high quality standard
Summary: Benefits of cooperation
Fully automated system training: One-click system customization
Automatic data pre-processing and cleansing (Kantan Gentry)
Fully automated translation step: Automated pre- and post-processing of files we upload
PEX rules to aid translation
Quality assessment of engines: KantanWatch
Gap Analysis
Reject Report
No maintenance and infrastructure
Q & A
Thank you!
@KantanMT [email protected]
Brian Coyle
Brian Coyle
Brian Coyle is the Chief Commercial Officer at KantanMT. He has 25+ years’ experience in senior sales and marketing management roles in a range of industries and across a number of markets including the US, the UK and Europe. Over the last number of years, Brian has worked on the commercialisation of products in the SaaS market with Kefron Document and Information Management Solutions and also AccountsIQ, the specialist cloud accounting software provider. Brian holds an MBA from the Smurfit Business School University College Dublin, and is a member of Tekom, European Association of Technical Communication.
What is KantanMT.com?Statistical MT Platform
Cloud-based Highly scalable Inexpensive to operate Fusion of TM & MT & rules High speed, high quality translations
Our VisionTo put Machine Translation
Customization Improvement Deployment
into your hands
Active KantanMT Engines
9,109Training Words Uploaded
380,371,660,160Member Words Translated
6,546,558,712
The KantanMT Community.
Case Study #1
@KantanMT [email protected]
Technical Manuals
Technical Manuals
ObjectivesLanguages: English->DutchPOC to validate rapid ROIFocus on translation productivitySolution must scale to 50 manuals pa. Integrate to existing L10N workflow
ChallengesClient had already built in-house MTOutput of in-house system very poor
Technical Manuals
ResultsSeamless integration into existing L10N
workflow (and MemoQ)Translation productivity: + 25% Reduced Post-Editing effort using PEXReduced project turnaround timeSignificantly higher quality translations
achievedProcess will scale to meet future
requirements
Case Study available for download.
Technical Manuals
Case Study #2
@KantanMT [email protected]
Vehicle Specifications
Vehicle Specifications
ObjectivesVehicle descriptions service6,000 websites & 17,000 dealershipsLanguages: English->Spanish &
Canadian FrenchFocus on translation productivity &
speed
ChallengesClient had own internal L1ON platformSelf-developed, customised CAT tool
Vehicle Specifications
ResultsSeamless integration into existing
L10N workflow using KantanAPITranslation productivity: + 70% Reduced Post-Editing effort using
PEXReduced project turnaround timeReduced project costs
Conclusions.
Technology enabled localization workflows are becoming the norm
Introducing automated translation delivers productivity benefits
Those that ignore this change will struggle!
@KantanMT [email protected]