Lately, we were busy deploying our Enterprise MT Hub to several large clients and talking at conferences in the US and Europe. Now, it’s a good time to recap the last quarter:
🧭 We published our new MT Landscape — let us know what do you think!
📢 Our first booth at LocWorld — to present Intento Enterprise MT Hub.
🆕 Published two new stock MT engines: Alibaba e-Commerce and Tencent.
📰 Lots of new language pairs and other MT news! Check it out!
🔗 We made some conference talks, see links to the slides below.
1. Machine Translation Landscape
We have published a Machine Translation Landscape prior to the LocWorld41 conference.
You may grab it here: https://docsend.com/view/5gccttz
2. Intento Enterprise MT Hub
We have presented our Enterprise MT Hub at the LocWorld conference in San Jose. Reach us to learn more if you have missed the conference! 🙂
3. Integrations and connectors
- Our Intento Translator extension is approved in Google Chrome Store.
- Intento plugin is officially public in memoQ 9.1.
- Many updates to our plugin for SDL Trados.
4. MT Provider News
Machine translation from Alibaba and Tencent MT is now available in our MT Hub, Google AutoML and ModernMT cut their prices, many MT systems have broadened their language support. Also, we spot new MT engines for Japanese and Arabic (the production integration is coming).
Alibaba
We have published in production Alibaba Translate models for e-Commerce (29 language pairs). There are three different models for translating Product Titles, Product Descriptions, and Product Reviews.
Amazon
Amazon added support for Romanian, Hungarian, Ukrainian, Vietnamese, Thai, Urdu and Greek (already at Intento).
On October 14, Google AutoML Translation moved from Beta to General Availability, with several significant changes:
- training price went down from $76 to $45 per hour
- there’s a training cost cap $300 per training (similar to what Microsoft has)
- three public volume-based tiers for AutoML translation: $80, $60 and $40
- also, models must be re-trained at least once per 18 months (keep that in mind when making a budget!)
IBM Watson
Lots of new language pairs for Eastern Europe! English <> Bulgarian, Estonian, Romanian, Croatian, Slovak, Slovene
Microsoft
Microsoft published a stock model for Irish and enabled NMT customization for Vietnamese, Bahasa, Portuguese, Romanian and Slovene.
ModernMT
ModernMT launched a new website (https://www.modernmt.com/) with lots of changes across the board:
- price for their flagship adaptive NMT dropped about 20x from €4/1K words to $50/1M symbols
- several lower tiers for stock NMT ($15/1M) and slow NMT ($8/1M, 1-hour turnaround)
- upgraded data protection and data retention terms
QCRI
Interesting non-commercial (yet?) MT Engine for Arabic languages. Not integrated yet, coming soon.
Rozetta T-400
We know about Mirai, but recently discovered another Japanese MT engine from Rozetta: T-400.
Systran
New language pairs for Systran: Japanese to French, Portuguese to Dutch, Portuguese to Chinese (both Simplified and Traditional) and back.
Also, Systran has launched SYSTRAN Marketplace for data owners to publish their MT engines for the general audience. There’s no public API for the marketplace yet, just an online demo.
PROMT
New language pairs for PROMT: Japanese to Chinese (both Simplified and Traditional) and back.
Tencent
We have launched a production integration with Tencent TMT (more on that here).
5. New Blog Posts
- RoBERTa: A Robustly Optimized BERT Pretraining Approach (https://blog.inten.to/papers-roberta-a-robustly-optimized-bert-pretraining-approach-7449bc5423e7).
- Speeding up BERT (https://blog.inten.to/speeding-up-bert-5528e18bb4ea)
- Welcome to the Simulation (https://blog.inten.to/welcome-to-the-simulation-dd0d8cb6534d).
6. Our Conference Talks
- [In Russian] “How to Choose and Tame the Machine Translation” (Translation Forum Russia, August 25–26, Saint-Petersburg, Russia)
- [In Russian] “Modern Neural Net Architectures / Year 2019” (UseData conf, September 2019, Moscow, Russia)
- “Cloud Cognitive Services Landscape” (Global Innovation Forum “Transforming Intelligence”, October 16–18, 2019, Yerevan, Armenia)
- “Cloud Artificial Intelligence Landscape” (UC Berkeley Cloud Meetup, October 29, 2019, Berkeley, USA)
- “Progress in Commercial Machine Translation Systems” (Stanford HAI, Workshop “Measurement in AI Policy: Opportunities and Challenges”, October 30, 2019, Stanford, USA)
- “Fantastic Quality MT Engines and Where to Find Them: Evaluating MT in 2020” (LocWorld41, November 6–8, 2019, San Jose, USA)
- “AI Development Trends in The World” (AI Journey, November 9–11, 2019, Moscow, Russia)
