Nunc Est Tempus. Jaap van der Meer, TAUS Amsterdam, The Netherlands

Transcription

1 Nunc Est Tempus Jaap van der Meer, TAUS Amsterdam, The Netherlands

2 Headlines 2017 Nearly Indistinguishable From Human Translation Google Claims Breakthrough (September 26, 2016) Systran Goes Neural, Lets You Try It (October 18, 2016) Baidu Sells Human Translations at Rock-Bottom Rates to Gain Edge in AI (March 16, 2017) A Convolutional Encoder Model for Neural Machine Translation Facebook (July 25, 2017) How SDL s Secure Neural Machine Translation Helps Enterprises Reduce Risk (August 14, 2017) Sogou Invests USD 6-7m in Shanghai-Based UTH in Scramble for Translation Data (August 22, 2017) Linguee s Founder Launches DeepL in Attempt to Challenge Google Translate (August 30, 2017) Amazon Pits Neural Machine Translation Framework Against Google, Facebook, Others (December 20, 2017)

3 January 1954: In Five Years... FAHQT This slide may not be used or copied without permission from TAUS

4 TAUS Vision 2005: Translation out of the Wall

5 Hello World! How do we communicate? It s Time for a Big Idea!

6

7 TAUS Vision December 2013 Human Language Project Collaboration between business, government and academia worldwide Think of: the Human Genome Project: a $3.8 Billion investment in sharing data about the human genome drove $796 Billion in economic impact, and spurred growth in the life sciences industry.

8 TAUS Forum in Dublin June 2014 English or Machine Translation as the New Lingua Franca 53.5% against 46.5% for MT

9 1954: Electric Brain In five years we will translate all of the Soviet Union into the Queen s English in just one week. 2017: Neural MT The time has come..

10 The S-Curve is stronger than all of us

11 Law of Accelerating Returns

12 Evolution of the Translation Technology Landscape

13 The Neural Wave Neural MT Five years of NMT = twenty years of SMT 5,000 lines of code 200,000 lines of code Zero shot translation Thousand(s) decisions taken by machines Every week we see improvements in quality Enables ambient intelligence The progress in what we can call the consumer translation tech industry is of much larger relevance than the classic corporate, outbound targeted, translation business. Chris Wendt: Going Vocal with Ambient Translation, Nunc Est Tempus, December, 2017

14 Democratization of Algorithms Open source algorithms From SPSS and SAS Institute to TensorFlow, Open NMT, MXNet, Nematus, ConvNet 4 million R and Python users vs. 600 million Excel users The democratization of AI has only just begun Once you have found your best service, you are less likely to switch. That s how the circle of open algorithms and customer lock-in gets closed. Nunc Est Tempus, December, 2017

15 Follow the Data Our language data travel freely across all borders to economies driven by free enterprises and long-term big data strategies. Nunc Est Tempus, December, 2017 We are moving from a peopledriven to a data-driven industry. Adolfo Hernandez, Nunc Est Tempus, December, 2017 Cross-language big data inspires our future. Eric Yu, CEO of GTCOM, Nunc Est Tempus, December, 2017

16 Not Business as Usual Parallels to advertising, travel and banking industries: Standardization and automation lead to mass market adoption Ubiquitous, always-on translation triggers new business and pricing models The ease and omnipresence of translation will be the foundation for a plethora of new business opportunities. Razor-sharp localization Fifty percent of the services we produce today, we were not doing five years ago, and they are not translation. Smith Yewell: In the Business of Global Talent Search, Nunc Est Tempus, December, 2017

17 Going Vocal Ambient intelligence Our environment automatically adjusts to the language and best means of communication. Spoken translation in wearable technology, robots, smart home devices Confluence of technologies, business models and markets Machine Translation will become more acceptable not because it miraculously becomes perfect, but because it can act as a utility for people with reasonable expectations. Chris Wendt: Going Vocal with Ambient Translation, Nunc Est Tempus, December, 2017

18 Going Vocal

19 Going Vertical Bringing relevance to verticalization Companies that specialize in the nuts and bolts: managing infrastructure Service companies that fan out into many verticals and cannibalize sturdy domestic industries Alternatives to translations that nobody needs 70% of what we do will be the same from one industry to another. But 30% of the work will be industry-specific and this is where we need to be fully relevant. Christophe Djaouani: Bringing Relevance to Verticalization. Nunc Est Tempus, December, 2017

20 The Long Tail of Languages Geolinguistic shifts We are in a semi-state of globalization: 30% international trade China One-Belt One-Road Initiative (OBORI) China Next Generation AI Development Plan Zero shot translations adds MT for more languages Stimulates interest in new cultures and languages Chinese Government Going All in to Back Language Industry (Slator, January 10, 2018) The web, in the mentality of the Chinese, has always been a key tool to discover the world outside. Eric Liu: Translation World Shifting from English-Centric to Chinese-Centric, Nunc Est Tempus, December, 2017

21 Quality Matters Boutique style translations: Don t be blinded by technology Dumb translations are not enough Stories need to be told and imagination needs to be triggered

22 Reading the Charts Dashboards become commonplace: Can t manage, if you can t measure ROI on a per segment basis Data help to decide where to invest Localization on the whole has been more of a scattergun approach until now for most companies. Nunc Est Tempus, December, 2017 Using our dashboards we will be able to switch to a razor-sharp approach to localization. Nunc Est Tempus, December, 2017

23 Making Translation Invisible Towards Self-Driving Autonomous Workflows: Automate dozens of human touchpoints in a multi-layer multicompany supply chain Microservices and SOA have helped solve the challenges of interoperability Resolve diverging issues in localization: quality levels, nomenclature, metrics, how we measure In the past, the keyword was first cost, then quality, then efficiency. Today it s agility. Aiman Copty: Making Translation Invisible. Nunc Est Tempus, December, 2017

24 The Modern Translation Pipeline FAUT: Fully Automatic Useful Translation Content Good Enough: Review Content Content Content Algorithmic management Content profiling UTS scoring Matching resources Confidence scores Quantum Leap Datafication Dashboards MT TM CT MT Training: Data loopback DQF Machine Learning: Data loopback Content High Quality: Transcreation

25 The Modern Translation Pipeline 1. made great progress in interoperability 2. is adopting DQF as a quality standard Data-Driven Invisible Self-Learning Autonomous

26 and where are the Jobs?! A search for global multilingual talents to undertake a plethora of new value-added services: Brand Validation The Engineers The Reviewers The Data Analysts Multilingual Digital Marketing ediscovery Usability Testing Campaign Management Content Curation The Transcreators Community Development

27 Quality Management Becomes So Vital 1. Jobs of post-editor/translator, quality evaluator, editor and reviewer are blurring 2. They feed the data that train the translation pipeline, to fork content in three levels: FAUT Good enough High Quality 3. Their annotation and corrections train the machine translation engines.

28 Key Messages of Nunc Est Tempus: 1. The consumer is driving the change 2. Democratization of algorithms has only just begun 3. Follow the data 4. Not business as usual 5. Going vocal 6. Going vertical 7. Geolinguistic shifts 8. Quality DOES matter 9. Reading the charts 10. Making translation invisible And now is the time!

29 Key Take-Aways 1. Change jobs, processes, technology 2. Control measure, dashboard 3. Analyze use your data

30 TAUS Customer Experience Journey We guide adoption of technologies and data-driven processes, normalized quality metrics and future-proof business models. IMPROV E

31 This slide may not be used or copied without permission from TAUS