The just-announced Pixel Buds, that translates conversations almost in real time, would never have been possible without the recent improvements of Google Translate. What are the different models used behind Google Translate? How to understand and analyze the way Google Translate works? What's Natural Language Processing, and how to integrate it into a product?

For this talk, we'll have the chance to welcome an engineer directly working on Google Translate, Keith Stevens. Keith will talk about how different Neural Network-based Natural Language Processing models can provide solutions to real user needs. While some problems are not yet fully solved with these models, he'll also explain how they provide dramatically better answers for users needing automatic machine translation for text, document summarization, question answering, and even image captioning. Lastly, he'll cover at a high level how these models work, what's needed to train them, and some issues in deploying them within a product.

ApéroTalk is our events series based on a very simple (and very French) concept: Apéro and Talk.

Agenda

• 7:00 pm - Door opens
• 7:30 pm - Talk and Q&A by Keith
• 8:30 pm - Networking session

About Keith Stevens

Keith has been working on Google Translate for the last 5 years. He currently focuses on building tools to understand why Translate makes its decisions and what the neural machine translation model is really learning.

Previously he's worked on crowdsourcing and internal infrastructure for translate after and researched semantic representations methods in a PhD program.