kantanfest: mindaugas kazlauskas
TRANSCRIPT
SMT or NMT,
it is HT
by SLV via MLV for €€€
Mindaugas Kazlauskas, 2017-06-30, Dublin
500 fastest growing technology companies in EMEA
Our LanguagesALL EUROPEAN LANGUAGES:
Finnish Swedish Danish Norwegian
French Portuguese Spanish German Italian Spanish English Dutch Polish Czech Portuguese French Greek Romanian Slovak Bulgarian Croatian Slovenian SerbianHungarianLithuanianLatvianEstonian
ASIAN:
Japanese
Chinese
Korean
Indonesian
Thai
Hindi
Malay
Vietnamese
RUSSIA, UKRAINE,
TURKEY, MIDDLE EAST:
Turkish
Arabic
Hebrew
Russian
Ukrainian
Facts
Established in 2005. www.synergium.se
Operates in Sweden, with production offices in Estonia,
Latvia, Lithuania, Ukraine
80 people in 5 countries
ISO 14971:2012: Life Science Services Risk
Management
ISO 9001:2015: Total Quality Management Systems
ISO 27001:2013: Information Systems Security
ISO 13485:2012: Life Science Services Quality
ISO 17100:2015: Translation Quality
Customer Focus Category, “For putting clients at the heart of its business“
Customers: MLVs (several verticals) + life science manufacturers
MLVs …..- on demand service, value the speed of taking orders- pressurize on price every year- need capacities- insert MT themselves and ask discount in-front, if you
are not productive you just sacrifice your profit margin- use MT and have no fears about it- multi-CAT- multi-domain- change providers daily (auctions)
(we love our MLV customers )
…end customers in life-sciences- less “on demand“, more long time contracts and better
forecasting- price is stable for that level- need competence- need solutions for unique scenarios- one CAT/TMS per company- one domain per company- change providers rarely, every 2 year
(we love our life-science customers too )
MT for MLVs- multi CAT fragmentation -> solution is to convert- content fragmentation -> solution: MT for all content- problem: domain fragmentation (ask for many engines)- translators fragmentation -> works with MT or not, is productive or not- MT is a must, if gives no return, then SLV finances - MT is only a driver for a cost reduction (speed does not matter as onboarding a freelancers/SLVs is faster)- “MT productive“ and “MT not productive“languages (MT only for some languages)
MT for life-science end customers
- MT for pharmacovigilance (value – speed and cost for cheap-and-dirty analysis)- MT for reducing costs is not relevant as seen too risky
Life-science thrives with innovation, but MT is only a productivity driver for them, albeit with bigger risk (legal issues if there is a legal case because translation error)
(in comparison TM is both productivity AND language quality)
Brain of Human Translator is the real mystery, not NeuralMT or other tech- what is translator`s real productivity? Where does it happen? In the brain
or on the surface of the keyboard?- is he properly paid, or not?- is he MT averse or not- is he MT productive or not (productivity can be w/o MT 1300w/h, with MT 900 w/h- does he dislike a specific CAT or not- does he report false hours or not- be emphatic: how translators see MT? No excitement: one more tool, new skills, instant pressure for productivity, demoralization (MT vs HT)
Speed
How the translator [can be] more productive with MT:
- [can] grasp the meaning faster- [can] avoid typing simple words- [can] find some difficult terms guessed, no need to research- [can] avoid inserting tags and numbers (tags do not go along with NeuralMT)- [can] be comfortable being influenced by a machine
How the translator [can be] more productive with MT:
- is every translator capable reading MT text written by machine and being similar to text of The absent minded professor with funny accent? - can every translator be productive by manipulating the text:
a. copying and pasting longer fragments and removing endings?b. fast enough and does not get tired when speeding?
- speed always get you more tired, and not everybody wants to participate in Dakar auto marathon EVERY DAY, from 9 to 5.
How SLV [can be] more productive:.
We can get some productivity by inserting MT only:
- if file format has less tags- if files are not small, at last 1000 words- if tool is not complicated and MT can be inserted- if that particular translator is productive with MT- if language and content allows be productive with MT- if we have time
Goal of MT: faster, better [quality], cheaper, bigger quantity:.
- start as a risky research project- fund properly as there will be no quick gains- measure constantly your KPIs- talk to your translators, they ride your MT vehicle, you are only The support personnel
SMT advantages
- controllable terminology- safe tags- not translatable product names- no missed words
bad endings in morphology rich languages, English-like word order
NeuralMT advantages
- fluency is much better - word order is much better - endings are much better - significant risk : not 100% controllable, needs testing after building new engine- an absent minded professor :needs to be verified and helped by a down-to-the-earth
spouse- translator`s attention wears-off more quickly
What matters in the end
- did translators saved time, money and passed PART of it to you? (->time and costs savings)- did translators save their energy? (->productivity)- did translators make less mistakes
because of less mental workload? (->better quality)Does NeuralMT help to produce more during the same time? It depends. On the particular translator, on the resourceful MT provider, on the disciplined PM Team and on the translators-loving Vendor Management Team (as they need to motivate freelancers and support)
Summary:
- TM is better investment, MT is not guaranteed to make translator more productive- SMT is much more researched and predictable, NMT research potential is limitless (as it replicates human mind)- MT is more usable when you control all processes in value chain, from the customer to the supplier
Summary:
- MT is much more usable with bigger budgets and bigger volume,
as this technology needs expensive setup than tools.- MT is a cab, translator is the cab driver.
Love your translator, support extensively and pay properly (should not receive less money for time spent on MT projects than for projects without MT, should not take the all the risks of uneven productivity gains because of different content types).
Any questions – write to me mindaugas(at)synergium.se