The headlines have been ablaze the past few days with praise for Microsoft and Google’s new translation technologies. Microsoft is working on their real-time interpretation for Skype calls. Google, on the other hand is working on their Google Translate app, adding the ability to overlay translations in real-time. When I saw the news about the new app version, I was genuinely curious about its capabilities. Since ICD is a LSP, I decided to use older translation projects we have to test the new Google app to see if it is a threat to the translation business.

Google Translate Brief History

Since 2007, Google has developed its own translation engine. In 2010 Google released a Translate app for Android devices and then in 2011 it released Translate for iOS devices. Translation originally was for user typed text, then incorporated voice recognition and the app could speak the translations using the Google Text-to-Speech engine. Prior to the current release of the app, you could use your device’s camera to snap a photo of the untranslated text, highlight the text you want, and then the app would give you its translation in a text box.

The Test Documents

To test the Google Translate app, I chose translated pages from a German document, a French document, and a Spanish document. I chose each page specifically to test different situations you might want to use the app for. The German page has text which isn’t high contrast (in this case the text is white on gray) and could simulate posters or brochures. The French page I chose because it has large, easy to read text like a street sign. The Spanish page has very small text and I thought this would be similar to restaurant menus. Let’s see how the app performs for these three tests.

Test 1: German into English

Low Contrast Text

Screenshot_2015-01-16-08-33-09For this first test, I wanted to test to see how Google Translate works when text is not black on white. I will admit, I started with a tough one but I think it’s a good place to start because we can find the apps limitations sooner. If you look to the graphic on the left, you will see the results.

The app didn’t do too well but I expected this. You can see the text isn’t being overlaid properly and most of the text is not being recognized. I held my phone over this bit of text for a while, in good lighting, hoping it might need to scan the text to find the right words and the right translation. However, the whole time it was scanning and replacing words here and there but never all the words at the same time.

Another issue seems to be for German text, the app is recognizing parts of words, but not the word as a whole. This can be frustrating knowing part of the meaning of a word, but not all of it.

General German Block Text

Screenshot_2015-01-16-09-30-33

Here we can see a section title, followed by a block header, followed by a block of text.

In the header, we can see the DSM5 is something digital, but it misses Schallpegelmesser. No worries Google, I don’t know what Schallpegelmesser is either.

Let’s jump over the block header for a second to look at the block text. We can see a lot of the words are being picked up, but not enough of them to really understand the idea of the passage. Some of the translations are questionable, such as “THE SICKLE” doesn’t seem to fit. This could be a result of German having complex ideas in single words.

Look back up to the block header, which reads “FUNCTIONS”. Text like this is where the app shines at the moment. It is high contrast and it conveys a simple idea. Not once  while scanning this section of text did Google change “FUNCTIONS” to anything else.

List Items

Screenshot_2015-01-16-08-32-49List items also seem to be a place where the app is shining at the moment. Nearly all the items are clearly labeled and the translations are spot on. Again, this is an example that has simple ideas as opposed to complex sentences.

The app defaults to text in all caps and removes any formatting. If you’re looking at a text that has a word or two capitalized, I’m guessing those words were formatted that way to draw emphasis to the word. Losing the formatting could mean you don’t know what you need to focus on anymore.

Test 2: French into English

Cutoff Text

Screenshot_2015-01-16-08-32-24In the French into English tests, the Google Translate app seemed to have an easier job translating the text than the German. This could be the complexity of German words coupled with the app constantly scanning the text makes it hard to lock down the whole meaning of the words.

As shown by the title of this block of text, we see the section is about “PRECAUTIONS AND PROCEDURES OF SECURITY”. Checking this against the source text, the only mistranslation is “SECURITY” should be “SAFETY”.

Block Headers

Screenshot_2015-01-16-08-31-04It is clear these sections are about Dangers and Warnings  you might encounter in the manual.

To be fair, “Danger” is the same in English and French. However, looking closely at the text, you can see Google Translate did pick up the text and then overlay it (compare it to the images above and you’ll see the font is different.

 

Test 3: Spanish into English

Spanish Lists and Small Text

Screenshot_2015-01-16-08-29-11Finally we come to Spanish, and its small text. The samples I have are set up in list form which should be easy for Google Translate to handle.

However, the app seems to be struggling. My guess is the smaller text is responsible; not all the words are being recognized. I had to hold my phone very still so the app could focus on the text, which might not be possible with average use.

Conclusion

While the app has the potential to be impressive, I see it more as a novelty and I don’t see it as being a threat to traditional translation. There are too many bugs, like how the German words are translated in parts and not as a whole and mistranslations of which there were many. Also, the app is constantly scanning the text and forgets the text it found before, making its job harder. My guess is this is why we see some words being translated but not whole sentences and is why the meaning isn’t always quite there even when the sentence is recognized.

I see this app being a great tool for traveling to recognize signs and menus and bits of text here and there. However, I wouldn’t rely on it to replace human translated and localized documents or anything long or important. There is too much room for error with the app that it would probably end up being more of a liability than a way to save on translation costs. I’m looking forward to testing future iterations of the app and to see how it improves over time.