Duration 16:41
16+
Play
Video

Engage and translate: Text and audio chat in 100+ languages

Dale Markowitz
Applied AI Engineer at Google
+ 1 speaker
  • Video
  • Table of contents
  • Video
Google Cloud Next 2020
July 14, 2020, Online, San Francisco, CA, USA
Google Cloud Next 2020
Request Q&A
Video
Engage and translate: Text and audio chat in 100+ languages
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free
Add to favorites
3.21 K
I like 0
I dislike 0
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free
  • Description
  • Transcript
  • Discussion

About speakers

Dale Markowitz
Applied AI Engineer at Google
Sarah Weldon
Product Manager at Google Cloud AI

Experienced leader and self-starter with diverse background spanning from public policy to engineering, with a passion for collaboration and transferring capabilities and ideas between sectors. I bring deep passion for understanding the needs of the customer and continuous enhancement of the end to end customer experience, validated by qualitative and quantitative evidence.

View the profile

About the talk

As users worldwide connect to the net and travel globally, it’s more important than ever to make interactive applications chat in many languages. But when users speak 100s of languages, this quickly can get challenging. Where to start? If you are new to translation, we’ll touch on a couple of simple starters, then provide examples of how you could advance your translations once you have gone further along the learning curve.

Sarah Weldon, the Product Manager for Cloud Translation, and Dale Markowitz, an Applied AI Engineer and Developer Advocate, shares how the Translation API can quickly globalize an app, with no multilingual expertise required. Increasing language coverage can drastically increase engagement, even for internal applications. When Mercy Corps integrated Translation API in their internal hub, traffic volume increased 70%.

Learn about integrating the Google Cloud Translation API Advanced with a chatbot client, using the machine translation glossary feature to control a set of terms for more relevant translation. Hear about new Translation features.

Speakers: Dale Markowitz, Sarah Weldon

Watch more:

Google Cloud Next ’20: OnAir → https://goo.gle/next2020

Subscribe to the GCP Channel → https://goo.gle/GCP

#GoogleCloudNext

AI213

product: Cloud Translation, Cloud Media Translation API, Custom Translation; fullname: Dale Markowitz, Sarah Weldon;

event: Google Cloud Next 2020; re_ty: Publish;

Share

Hello and welcome to engagement. Translate text and audio chat in over a hundred languages. I'm Sarah, Weldon, product manager for Google Cloud. Translation services, which include apis for free train and customizable machine translation models and a new beta API streaming audio translation in the talk with Dale Markowitz applied AI engineer at Google cloud. And so, bring some of the features that we're discussing to life in a couple of dinners. Today's companies their employees and their customers are truly global. And

everyday, they're also more digital, keeping Pace with the exchange of information in multiple languages at scale is a grind challenge especially so when more work is moving to Virtual. So companies are turning to machine learning to augment their workloads and make content, more accessible faster than would be possible without automation. Local dentist, in machine translation, continue to drive up, quality and new features on our products. Give customers more control over how their content is translated, more on that later. Let's take a quick look at where we started. Pre

2016. The worlds are relying on statistical machine translation for this paragraph from a Japanese language, translated into into English for readability, 7710 of the mountain, covered with snow, and it is said to be the highest mountain in Africa is a little awkward and how it's written, as well as there's some words missing with your flu every year. But if we started today, we've done a lot of work to improve machine translation using machine learning. And so we introduced Google introduced neural machine translation

after 2016 and into 2017 and newer model, the same translation would come through as Kilimanjaro is a mountain of 19710 Ft covered with snow and is said to be the highest mountain in Africa. It's a much more fluid translation. It doesn't feel awkward and it's a good enough, fer words We keep improving even Beyond us and summary the last major drop in quality. We saw I was in 2016 to 2017, but was continuously improved over the years as well. And in the most recent, what do you see what this crap and green? We saw an increased jump over

multiple languages through a number of different technology Investments driven by our research team lose or jump over a hundred plus languages. And this Improvement is comprable to the gain observed years ago. When translating from Seussical jr. Machine translation as they spoke about a few minutes ago, So, our products include translation API, basic translation, you had danced and auto Amo translation. And then the new one that I was talking about a little bit later for basic stations on it really intended for his employees cases,

it's intended for user-generated Content, where you're not trying to control a lot of the specific translations. So we wanted to make it very simple to use and so translation, API basic does the important thing still with language detection? And it also continues to use the advancements of the new machine learning models as we release them and develop new languages. Can you translate native tribe basic for chops Mario's and user-generated content? Cuz it's easy to integrate with and you're not trying to change the translations after the facts for a number. But I'm in the

area content. It still is really easy to use and will enable you to easily integrate with many languages at a very simple platformer. I'll tell you a little bit about a scenario where we have for our customers and in this case has games and mobile David, and there are international unsettled, when the cameras are wanting to talk to each other and collaborate, in order to achieve certain goals or should just chat line, they don't need to be able to cross between multiple languages and so by integrating with the translation a VIP say

Ulysses, able to improve engagement experience for their end users and improve the overall game experience, just like an idiot. This is just a trick to get attack scenarios as well. So let's talk about another customer scenario. This time, public sector within the US, many languages are spoken in government required to make information accessible to their constituents. Play Sarah Google and I'm talking to you from Austin, Texas. Texas is a really interesting State because it's actually the second most

multilingual say to the US after California and something like one in three people that I actually speak English as their first language in the house. So it's really important especially in times like these your own languages. So I recently heard about this project was actually done by the state of Hawaii could understand document, so I kind of lost it so I decided I would go to the Austin website. I found a bunch of different questions that people ask, you know,

how do I report a leaky fire hydrant and I decided to build This is a chat app that I built answer questions about the City of Austin. It recognizes lots of languages, so free sample. If I ask her to check, if I'm registered to vote in Spanish, it will detect my language and use the translation API to respond in the same language. The translation API supports over a hundred different languages, and lots of different character sets so you can write your app code once and have it

work all over the world. The demo that I just showed you, it was working with a chatbot that has been preset answers, but actually the art exactly the same for like four different languages. So the texts sent from the chat client to a server translation a language translator to translate back to the descenders language. And I know the translation were there was some really specific like Austin terminology like the Metro System are some really specific words in Austin

so you can prove quality Fordham Institute cases. Thanks Dale. That was great, talk about what we can do to modify contents to better represent different possible translations within a language depending on the industry contact store location. Mario, I'm thinking of public-sector again and it's pretty common thing to want to be able to provide comment to a public agency attorney general. If you were talking to two users that are in Spanish speaking Spanish,

there's a pretty common translation for attorney general which is fiscal enter all do in this case our machine translation what automatically selects that's for the British translated response countries, I say in Latin America that might have a different version of what they call their attorney general's office. In this case, ministerial public attorney general's office in Spanish, So, what if you could customize your translations to your domain? Is translation HDI Advance. You can actually do this and if you

have a terminology base, which a lot of public organizations and companies do as well, as you can use this, as a starting point with an easy integration with transitioning, Advanced English terms and your terms and other Target languages. If you just create this ESP file, an imported into your Google Cloud project, then you can use. This was future cases for machine translation, just by referencing your glossary in your translation, API call and get the response back. It'll actually preserve your preferred term over with a creepy machine translation might have

otherwise predicted and translation. API Advanced also has other features that can help with inflammation as well as more advanced customization capabilities with our custom machine translation. The Great Debaters act project. What say you will have even more than just terminology and you want to go further your company or organization that has a lot of past human translations and use those translations to prove the machine translation prediction for what a translation would likely be for your content,

your contacts, through your industry. It's really important to understand, conceptually that Google has those pre-trained models, which is what you typically use a machine translation API, our translation Epi basic model, that we've created using a lot of training data, but if you have your own training data, you can actually use those as a starting point and then build and improve on them with a custom Training on auto Amo translation. So, I take it, you're patched human translations bringing that to our platform and running

trainings on it, then you're improving the statistical. Likelihood. That our model is going to provide back to you a translation that students to your contacts and to your industry. So it doesn't for transfer learning platform, where you can go and learn and manage more at the model of your platform. What's the factor attack scenario has the right language? But it doesn't speak, it actually just using text translation, do it since users both right and talk

there is an area is where we might want to have audio as well as text translation support at 4 or are you cases? So can you translate speech in real time? Organizations are seeing value in delivering their products and content globally in text. It's in video, it's an audio, it's in call. And so for these scenarios, you want to be able to have an easier way to implement translations for those few cases and media entertainment, technology. So what's so complicated about

translating dontics data with audio and video? We have our species guy is in her video apis that allow for interacting with those types of content. Translation. So we can just wish we can all three of them. But it's kind of hard to do. And it comes with other challenges that are nearly impossible to solve on the developer client-side. So we're looking at how to solve that. On the couch side. One thing late in the day, can be a lot higher. If you're in a grating across multiple apis especially want to set up for streaming and the other one is set up. For more synchronous case is

worth waiting for the content to finish before it starts to figuring out what a translation would be additional. You have to do multiple API calls for developers to work with, especially if you're trying to figure out how to improve accuracy, etcetera across these two different API. And then you're also dealing with a streaming client which creates and complexities with integration and usability and it's it's a challenge in itself. Show me the translation Yuki I just process it supports directly audio files or a microphone input through mobile phone or through your browser

and so it sounds. Content directly if they're not having to run through my purse and then a translation model in order to get a translation pack, you can go dressed in Translation, media translation API with your audio translation in the language which you thought of your target language, translation outcome. In addition to the distracting away, the fractions of multiple API calls, we have some other features that overall improved experience. So I may as well as models for Revere elephantine, youth cases allows you to

be able to make the bows to get the best quality optimized for your audio translation, outfits for support for the translation model. So we separate friends. German Italian, Hindi Portuguese, Russian Spanish, English and others for you to be able to translate languages for your surveys. I'll take it back to Dale and should we can see the transit media translation API in action. Thank you, Sarah. So it's Robert. I can confirm that it is tricky to do anything in real time. Especially real-time translation

translation API and protector. Hi, my name is Dale. I'm using the media translation API and a Raspberry Pi to translate what I'm saying in real time. My talk next time I get to talk somewhere, a translation customers for media translation, API. When they integrated with our video translation API, they were able to see James and Lily and see as well as your overall quality performance. And so they're able to provide real-time streaming translation, video chat, OnePlus is a phone provider and they've created a

beta application that allows for their customers to be able to opt in and be able to see the translations from the person that they're speaking a different language while they're still engaged them in in in video chat. And so we were able to demonstrate how text translation audio translation can work seamlessly together to create a picture and user experience. I wanted to fight you to our documentation on our Cloud website where you can get more information on Automall translation, translation of guide, pants and translation. Please start them out, thanks for

joining us.

Cackle comments for the website

Buy this talk

Access to the talk “Engage and translate: Text and audio chat in 100+ languages”
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free

Ticket

Get access to all videos “Google Cloud Next 2020”
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free
Ticket

Similar talks

Abdul Kinadiyil
Solutions Architect at Google Cloud
+ 1 speaker
Chuck Rhoades
Director of Engineering at Yum
+ 1 speaker
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free
Peter Dykstra
No-Code Apps & Automation at Google
+ 1 speaker
Tony Fader
Software Engineer at Google
+ 1 speaker
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free
Kaz Sato
Developer Advocate at Google
+ 1 speaker
Rakesh Talanki
Google Cloud Platform Principal Architect Services at Google
+ 1 speaker
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free

Buy this video

Video

Access to the talk “Engage and translate: Text and audio chat in 100+ languages”
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free

Conference Cast

With ConferenceCast.tv, you get access to our library of the world's best conference talks.

Conference Cast
635 conferences
26170 speakers
9693 hours of content
Dale Markowitz
Sarah Weldon