Hacker News new | past | comments | ask | show | jobs | submit login
New, more powerful Google Translate app (googletranslate.blogspot.com)
313 points by caiobegotti on Jan 14, 2015 | hide | past | favorite | 91 comments



The conversation feature is pretty phenomenal, and comes close to a true Babelfish device. I just tried it with English<->Mandarin with my wife (we both speak fluently), and works pretty accurately and is fast.

I will definitely use this when traveling.


I just tried English<->Mandarin with my wife...though not always grammatically correct, it gets the point across.

For those who would use it while travelling abroad, keep it mind this requires an Internet connection. In China, you would need VPN since Google services are blocked.


You can download the language for offline usage, at least in android


That only works for the (amazing) WordLens functionality. Live audio translation still requires Google's cloud for processing.


WordLens works offline without any additional downloads, at least for the languages I tried (Spanish, French, English).

Audio translation and TTS works offline if you have the proper languages downloaded, but being able to detect what language is being spoken seems to be an online-only feature.


Baidu translate works great in China and will translate anything you copy to the clipboard automatically. Requires the usual insane Chinese app permissions so Xprivacy or something similar is a must.


But Baidu doesn't have the Babelfish.


It is really perfectible for French<->Mandarin at least.

You actually have to translate both sentences into English to understand how they somewhat mean the same thing when taken literally (but that's just the same problem as on translate.google.com)


> It is really perfectible for French<->Mandarin at least.

You might prefer to say "imperfect" instead of "perfectible". While "perfectible" is a, well, perfectly good word, it is unusually rare enough that I thought you had meant to say "perfect" and made an error. "imperfect" is much more common and less likely to be misinterpreted. Apologies for the unsolicited criticism but I really was confused by this at first.


it was the perfect word choice to illustrate translation difficulty


True, but "really perfectible" sounds strange.


It sounds strange in english. Saying "c'est perfectible" in french is correct and usual. It is translated by google translate into "This is perfectible". As usual, google translate is very poor.


The irony being the poster above used that exact translation themselves?

Translation is difficult, and at least in this case the translation is an accurate one (albeit one many english speakers would need to look up in a dictionary).


Funny thing is, English is not my first language.

I guess it sounds a bit weird even in my mother tongue.


That's his point


I've laboured under the misapprehension that it was spelt "perfectable" for my entire life.


Yeah, the similar English sentiment would be "it could be better" or "leaves a lot to be desired"


I don't know whether they have a specific French <--> Chinese model. They might, they might not.

It's hard to train for all n^2 language pairs, so MT systems usually back off to English as a pivot language. i.e., they'll translate French --> English --> Chinese.


New neural machine translation architectures are experimenting with pairs of neural encoders / decoders, one pair for each language and a shared language independent vector space for the meaning of all words:

http://arxiv.org/abs/1406.1078

http://arxiv.org/abs/1409.3215

http://arxiv.org/abs/1410.8206

So the total number of models is still linear with the number of languages.

I do not know whether this new generation of translation models is leveraged by the google translation app though.

Also pairs of languages for which their are big amount of parallel training data will still be favored.


> Also pairs of languages for which their are big amount of parallel training data will still be favored.

wouldn't bible translations help?


It might but:

- the vocabulary and topics covered in the bible is quite different from today's written and spoken text, especially phone discussions or social network messages.

- other aligned corpora such as http://www.statmt.org/europarl/ are much larger than the bible (several millions of tokens for most pairs vs less than 1 million for the Bible)

Agreed that http://www.statmt.org/europarl/ does not cover non-European languages.


> so MT systems usually back off to English as a pivot language

That's an interesting choice, because English lacks features some other languages might have, and thus you end up distorting through English. I remember considerable work from different sources a ways back toward a constructing artificial languages for this purpose so to mitigate the introduction of ambiguity by using an existing natural language as a pivot language, I'm surprised that natural language as the pivot is the state of the art (though I'm not surprised that English is the pivot language given that.)


It's just weight of research hours, and weight of data. Our English numbers are almost always better in NLP.


Am I correct in assuming it requires a good fast internet connection to work at all? For example does it even work on 2G or is it WiFi only?


It works offline.


Google put it's translation engine into the client? I doubt that.


Yeah it looks like they did:

• Word Lens: Just point your camera to a sign or text and the Translate app will instantly translate the text, even without Internet/data connection. Currently available in: English ↔ French, German, Italian, Portuguese, Russian, Spanish.

(From https://play.google.com/store/apps/details?id=com.google.and... )


That's amazing. Major kudos to the Google Translate team if it really does work fully offline.


The live video translation was a startup that they acquired ("Word Lens"). It already worked offline so it would be strange if they had removed that feature.

I doubt this implies that the Google translation engine has been ported to work offline.


It does and has for some time[1]. Just switch to airplane mode and try it before asserting this? You have to download the language packs first, but it works.

[1] http://googletranslate.blogspot.com/2013/03/the-worlds-langu...


Does anyone know of any equivalent open source systems for the word lens or audio translate apps? Or even what kind of algs they are using?


The translation quality won't be (anywhere near) as good offline as online, but it works well enough for word lens to be useful without a data connection!


They've been able to do offline voice-to-text, for English at least, for the last 1-2 years, so it's possible. But I would tend to agree...seems doubtful that they could have best-in-class real-time translations....offline, right?


and yet that never works for me. the offline bit


The "Google Translate" app has supported offline translation for "pinned" languages for a while now.


> The conversation feature is pretty phenomenal,

Yeah, it is. I had pretty much the same idea a few years back, and called it "SpaceChat". I ended up dropping the project because I was too busy with other things. Smh.


Word Lens was the first iPhone app that I saw that really felt like Magic, glad to see it was bought up and put infront of a mass audience (where I assume it will gather data and learn)


> (where I assume it will gather data and learn)

Yes, if you look in the settings the following option appears:

☑ Improve camera input (Allow Google to retain your images related to translation)

It is set on by default.


Precisely my thinking. I was wondering what'd happened to Word Lens and so to see them baked in to the Google engine is really exciting. Great work and congratulations!


Totally agree. It's the first time in many years where I was genuinely surprised at the capabilities of a computer.


Congrats to Otavio and crew for bringing Word Lens all the way to such a spectacular success.


Word Lens was the first iPhone app that I saw that really felt like Magic

When Augmented Reality works correctly, that's how it should feel.


Congratulations to Otavio and the Word Lens team, I'm excited to see this in the Google Translate app.


Yeah, I noticed they finally pulled in Otavio's Word Lens into google translate after they acquired him.


Really curious about the tech / arch. behind its ASR. Must have a lot of nodes and tricks in place to support high throughput and low latency.

If I'm not mistaken, this paper [1] is the last time Google published about the related architecture. We know very little about improvements over the last 7 years.

[1] http://www.aclweb.org/anthology/D07-1090


AFAIK there are several new advancements for both ASR and Translation front. DBN and Special RNN called LSTM based systems surpassed classical approaches. A recent Translation mechanism based on LSTM from Google: http://papers.nips.cc/paper/5346-sequence-to-sequence-learni...


The original word lens app was much closer to a hashmap of words in language src to dest.

http://stackoverflow.com/questions/15656331/real-time-image-...


I speak English, Spanish and Italian. I've been talking to this app for the past hour in those languages and translating to whatever languages have speech output. This is so amazing.


Works pretty well, but so does Microsoft's Translation app for Windows Phone. Having used Microsoft's app for 2 years now, I am really wondering what took Google so long. Google's app offers coloring of the translation and a way more languages. But in the end, it's as fast and reliable as Microsoft's solution on my devices..


For those wondering, the app is called "Translator". I've also found the translation to be pretty good, though it tends to trip when it comes to handwriting.

https://www.windowsphone.com/en-us/store/app/translator/2cb7...


The word lens is impressive. Tried it out on my university ID against some bright light, and it was able to detect even the low-contrast regions. Even my Mac keyboard. It works on almost anything.

It doesn't work on Chinese quite yet.


It would be really cool if they could recognise the font (or a close match) and display the translated words in almost the same style.


Cool, but pretty useless.


Google has published a lot of work recently about using deep learning for language translation with really impressive results. Anyone know if Google Translate is now neural net based? Or would a many-layered net not scale to the amount of API calls that Google translate receives?


Having just spent a month travelling through Japan, I can attest to the brilliance of the Translate app even before this update! That said, the Android version seems more feature rich than the iOS version (maybe I just don't know where to look for the features)


What are the differences between the versions ?

Also, is Google legally allowed to only release the app for android ?


Companies are free to only release wherever they want. The legal aspect is when your phone depends on an appstore that refuses to allow competition in (I.E, if Android only allowed Google Maps and not any others). That's the anti-trust illegal aspect.


> Also, is Google legally allowed to only release the app for android ?

Are they "legally allowed"? Why wouldn't it be legal?

Ask the same question with regards to Apple and iOS, and the answer is patently clear.


> Why wouldn't it be legal?

don't waste your brain capacity on stupid people

there are too many of them


I'm just amazed this runs on my iPhone 4. Brilliant upgrade.


If they haven't already started doing this, it seems it would be inevitable, to turn this into a full-fledged API for use in real-time communications, like WebRTC apps, etc...

I see that they have an API, but not sure how feasible it would be to use in real-time apps.


I work on software that, among other things, runs a call center and I've been getting requests to do something like this in real time so that we can communicate with our users that don't speak english (We have agents who speak other languages but they wanted a fallback for when they weren't there). I told them real-time was still a little bit out (even with this as an API it would take time to integrate). This is really exciting though and I look forward to the day that we can do this will little to no lag and talk to people who don't share a language in common.


Some friends and I actually did something like this at a hackathon. We built a Google Hangouts app that used Google's translate API for real-time video conferencing in different languages.


English <-> Portuguese needs some work. I had an hilarious conversation with myself.


Do we know the Word Lens (Quest Visual) acquisition price in the end ?


I wonder when they'll integrate it with google glass to allow e.g. realtime subtitles for when people are talking to you in a foreign language.


This is Amazing. Very helpful here in Japan. I'll be glad when the word lens feature supports Japanese.


Before this they kept the Android app a lot more powerful. Are the iOS and Android apps finally on par now?


I doubt that was an intentional strategy. Google seems to not push worse apps to iOS in general. The new Maps UI launched on iOS first, for example.


Meanwhile on Windows Phone, we have YouTube blocked. Anyway, I have to admit Google Translate is phenomenal now, even though it's thanks to the guys behind Word Lens. Kudos to everyone involved. This is probably going to be one of my most used apps from now on.


Unfortunately crashes on my Nexus 5 when trying to take a picture :(


So what languages are actually supported in conversation mode?

Few months ago spoken Thai input was not supported.

Also, World Lens works nothing like in the animation. The app scans for words and provides translation in a traditional UI, it does not overlay translation over the image.


The Google Translate app (and Word Lens, I believe - it's been a few years since I played with it) do use augmented reality to overlay directly on the image.


Not on my phone. Shows square [align text] which does not disappear.



It could be it behaves differently on older iPhone.


The photo translation doesn't work offline for me. I downloaded the languages but when I take a photo it just says "no network". I can mark the text but nothing happens.


Know what's funny? "This is fucking funny!"

(try it)


OCR is really fast and accurate. Impressive.


It doesn't translate to indian english and to japanese english...


As cool as the new features are, the app is now difficult and slow to use the way I used to. Little things, like focusing the textbox when you tap clear, are just missing. I'm willing to take that for the automatic OCR features, but it's just worse for everything else (probably due to their new "Material Design" strategy which doesn't belong on iOS).


Unfortunately, all of their text-recognition is still done on the cloud. So If you're traveling, you most likely won't have a data plan in the country you're visiting, and word lens won't work. This seems like a huge missing piece to me.

It's strange because they let you download language packs so that all the translation (and voice recognition) can be done locally. I have no knowledge in this area, but honestly: is there something fundamental about text recognition that it can't be done client-side?


> is there something fundamental about text recognition that it can't be done client-side?

Since the blog post talks about how Google Translate now does client-side text recognition, I suspect the answer to this question is "no".


Nowadays Pocket Wifi's or data SIMs for your phone/tablet are fairly cheap and ubiquitous, if you're travelling to the parts of the world supported by the feature, you'll likely have a decent connection.


Google Translate came pre-installed on my handy, but since data is expensive- at least here in Germany- I removed it and defaulted to an offline German <=> English dictionary. After reading this post and watching the video I decided I'd give it another go and I'm glad I did.

In order to do translation w/out using data each and every time you have to:

1. Open the Translate app

2. Tap the options icon; top right

3. Tap "Settings"

4. Tap "Manage Offline Languages"

5. Select the language you would like available offline and how you'd like to download it( wifi only, wifi 3G, etc )

The German language pack is about 300mb, but that's nothing compared to the cost over time of using the app; data cost. Let's just hope they don't pull a SketchUp with this.



Why poor news? Google bought Quest Visual (the company behind Word Lens) some time ago.


In fact, the blog post subheading for that section is "Instant translation with Word Lens"

I'm curious if that part has improved with integration with Google's Translate. The video-based translated word substitution was amazing, but my understanding was that it was doing a very simple translation in the original app, essentially a word at a time.


Word-at-a-time use was perfect for me travelling around europe, as basically all I wanted it for was translating signs


Yeah, it works for most of the cases where you'd want to use something like Word Lens. I was more just wondering how deep the new integration goes.


the new google translate stuff is world lens. like, that is world lens code inside.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: