Thursday, March 11, 2010
Google Street View Hong Kong
Google released Street View imagery for Hong Kong. Hong Kong is a fun and busy place where it’s easy to get lost, perhaps less so with these photo maps. [Thanks Keith!]
If you’re looking for a Leberwurstbrot in Hong Kong...
Reader Play, Google’s Slideshow of Interesting Stuff
OK, I can see myself waste some time with this. Google Reader Play is a casual, push style way to feed yourself distracting/ enlightening/ inspiring snippets, imagery and videos. You can star a specific page, Like it or share it, but just jumping from page to page using the bottom navigation works too.
Note Reader Play is personalized. Google writes, “We use the technology behind Recommended Items in Reader to populate Reader Play with the most interesting content on the web. While you don’t need a Google account to use Reader Play, your experience will be personalized if you sign in.” Google says that “Reader Play adapts to your tastes” (click Like, and more stuff like that should appear, Google suggests).
You can also set this app to auto-play, which sort of clashes with specific YouTube videos though... it would have been smarter to wait until a video ended playing before moving to the next bit. Now, I noticed that the actual source or author of a particular piece ends up as a kind of by-the-way footnote in this stream of stuff – even clicking on the “from” link will merely load that blog into Reader Play, and not open the source site – but I guess that might be the way of RSS and/ or the future.
[Thanks MZaza!]
Wednesday, March 10, 2010
No, we can’t translate "Yes we can"
By Roger Browne
Ten years ago, the best-available translation software analysed the source text to determine its structure: subject, object, nouns, verbs, phrases, etc. From the structure tree, a new text could be generated in the target language.
The precise details of Google’s translation algorithms are not published, but the structure tree is not the main mechanism. Instead, there is a corpus – an enormous database of parallel works. These are works available in more than one language as a result of a previous human translation.
Based on equivalents found in the corpus, Google obtains translations for various multi-word fragments from the source text, then blends those together into what is usually a coherent sentence in the target language.
The system doesn’t work so well on fragments that weren’t translated in the corpus. For example, the phrase “Yes we can” was used prominently in Barack Obama’s election campaign, and was therefore included untranslated in many foreign language news reports. You can see this in a search for [obama “yes we can”] on google.de.
As a result, Google Translate is not always able to translate that phrase, even when used in a context unrelated to Barack Obama.
In a test I performed today, I found that the phrase “Yes we can” was not translated into these languages:
Catalan, Czech, Dutch, Finnish, French, German, Hungarian, Italian, Polish, Portugese, Slovak, Spanish, Turkish.
It was translated into these languages:
Afrikaans, Albanian, Arabic, Belarusian, Bulgarian, Chinese, Croatian, Danish, Estonian, Filipino, Galician, Greek, Haitian, Hebrew, Hindi, Icelandic, Indonesian, Irish, Japanese, Korean, Latvian, Lithuanian, Macedonian, Malay, Maltese, Norwegian, Persian, Romanian, Russian, Serbian, Slovenian, Swahili, Thai, Ukrainian, Vietnamese, Welsh, Yiddish.
What can we conclude from this? Probably that the corpus for each language on the first list includes a higher proportion of Obama campaign reports than the corpus for any language on the second list.
[Thanks Ilan and Philipp!]
Monday, March 8, 2010
Variable Speed for YouTube Videos
By Roger Browne
If you’ve opted in to YouTube’s HTML5 experiment, the videos will play in your own browser’s player, instead of in an Adobe Flash player. This was discussed in the Forum a few weeks ago.
YouTube can serve HTML5 videos if you have an HTML5-capable browser with the h.264 video codec (i.e. Chrome but not Chromium, Safari v4+, or IE with Google Chrome Frame). Videos containing ads continue to use the Flash player.
YouTube has implemented a variable speed control for their HTML5 videos. Next to the play/pause button is a small arrow. When clicked, it opens up a panel containing a slider and two icons: a hare and a tortoise.
As you can probably guess, moving the slider towards the hare makes videos play faster, and moving the slider towards the tortoise makes them play slower.
The pitch of the sound is compensated, so that the audio doesn’t become squeaky at high-speed and deep at low-speed. However, the audio adjustment is not quite smooth enough for listening comfortably to accelerated speech, even at the minimum speed-up.
The “slow speed” feature works well for action replays. The “fast speed” feature is great for quickly finding the part of a move that you want to watch. But the audio processing needs to be just a little better before it will be practical to save time by watching long talks at an accelerated pace.
I used Safari 4.0.4 on OS/X Leopard to view the HTML5 videos and to take the screenshot, because Firefox on my Linux computer doesn’t have the h.264 codec.
[Thanks Qrczak!]
Criticism of Google at Wikipedia
Wikipedia has a whole page dedicated to criticism of Google.
Do Some People Confuse Public and Private Buzz Messages?
Do some people confuse public Buzz messages with private email conversations? I don’t know, but some of the following conversations I found using Gmail’s Buzz search sounded a bit odd. One of the conversation, for instance, was by a person who only follows one person and only has one follower. Then again, how would this misunderstanding come about, if that’s the case, as Buzz says “Public on the web” next to the Post button, and doesn’t offer a field to address only a single person? Gmail puts private emails and public buzz messages in one app, one click away in the interface... is that enough to trigger misunderstandings? I’ve obfuscated some parts in these screenshots.
Sunday, March 7, 2010
The Ask-Our-Computer Service (from 1967)
This snippet is from Popular Science, May 1967, page 94:
[Hat tip to George R!]
Friday, March 5, 2010
15 YouTube Killer Features of 2025
In the year 2025, perhaps we’ll be seeing:
1. YouTube Super Zoom
What it is: When watching a video, you can press the pause button and zoom in on smaller stuff you see in the background. For example, we see a mayor walking through a museum, and there’s a painting by Vermeer in the background. Hit pause and the zoom icon, and click on the Vermeer painting, which is highlighted with a “zoomable” color (as are some other objects in the scene). It will be enlarged up to the point where you can see individual paint strokes.
How it works: Google’s image comparison engine will run in the background as soon as you select the zoom tool. For instance, the comparison engine will notice that images of the region containing Vermeer’s Milkmaid are available in many other places all over the web, like museum sites which have super large imagery. Google will merge those images to create an authoritative larger image, making user zooms feasible.
2. YouTube Speech Dubbing
What it is: When you watch a foreign language short film on YouTube, the voices will automatically be dubbed into the language of your current content preference. You won’t even notice people aren’t speaking your mother tongue.
How it works: Behind the scenes, Google’s YouTube runs a speech-to-text program, followed by machine translation, followed by text-to-speech. To make the outcome more seamless, face recognition understands who is speaking and slightly adjusts the lip movements of the speaker so that it looks like the person really says the translated tone.
3. YouTube Music Composer
What it is: You’re an amateur film maker. You created a thrilling, low-budget short film, but you don’t have any money (or the needed talent) to add a great music score. You click “Compose music”, and YouTube suggests a complete, copyright-free soundtrack to you. You check the score and possibly tweak the results where needed, and go live with your short film.
How it works: By analyzing light, pace, speech patterns – does the speaker sound nervous, happy, stressed, relaxed? – and more, YouTube’s AI music composer engine automatically creates a score. Most of the “smartness” of the algorithm is actually brute force: YouTube analyzes millions of movies and their respective scores, and compares what the visuals of your clip at a specific time most likely resembles. It then merges and distorts music from the scores found, adjust harmonies and rythm, and attaches it to your clip. The music distortion is just substantial enough so that no music composer can sue Google for ripping off their piece.
4. YouTube 3D Rotation
What it is: You’re viewing a clip with an interesting building in the background. You can hit the pause button and then rotate the building, showing it from different sides.
How it works: Google’s all-encompassing Street View cameras collect 3D data from all kinds of scenery. They also match objects from photos snapped from different angles found all over the web. When a matching image is found in the video, it’s connected to its 3D counterpart to allow you to rotate it.
5. YouTube’s I’m A Star
What it is: You send a film clip of a famous Hitchcock movie scene to a friend. Your friend sees you being attacked by a bunch of scary birds, with Tippi Hedren screaming for your life!
How it works: You can pick any movie available at YouTube. Many full-length movies, like all of Hitchcock’s work, have passed into the public domain by now, allowing creative remixing and more. You click the “I’m A Star” button, and let your webcam snap a few portraits of you. Google’s face recognition will then show you a selection of actors of the movie. You click on Rod Taylor, and instantly have his face be replaced with yours across all of the movie.
6. YouTube’s Clipteller
What it is: Atop every video there’s a “Clipteller” tab. Click it, and the contents of the clip will be explained to you in story form – both in plain text, as well as a voice reading the text – complete with descriptions of what’s happening in every scene, dialogue and so on. This creates very accessible videos, and it’s also content which in turn gets indexed by Google, ready for full-text searching.
How it works: Google’s film analysis algorithms have become so smart, they know what’s happening. If a train is driving by in a scene, the algos will understand the object “train”, the action “passing by”, they will be able to categorize the sound into “train sounds” and so on. From that information, an AI creates a textual description, which is then “beautified” by comparing it with typical story writing patterns, found in the millions of books which Google scanned.
7. YouTube’s Filmmaker
What it is: Filmmaker lets you edit a video, apply filters, crop, tweak and mash videos at your pleasure. You can add titles and special effects, like an explosion. You can even choose from a variety of famous actors to play along in your video.
How it works: Filmmaker is an app sitting in the browser, with no further downloads necessary, thanks to the power of HTML9. It integrates with mobile phone cameras and so on so that making short movies becomes very casual. Based on old films, which Google all digitized, Filmmaker allows you to pick from a vast range of classic actors, which are all available as smart 3D avatars. This way you can tell Arnold Schwarzenegger to jump off a balcony and then let the house behind him explode. (Arnold will get a share of your ad revenues from movie displays.)
8. YouTube’s GeoFilter
What it is: Depending on the location of the video viewer, you’ll only see scenes which are legal per your government’s rules. For instance, if violent movies are illegal in your country, then the scene which would contain a sword fight will be replaced with a ballet scene. Similarly, if demonstrating for free speech rights is illegal in your country, then a news report showing a demonstration will turn into coverage of a baseball game.
How it works: Google’s YouTube offers all governments an API to feed them their legalese and religious preferences in simple terms, like “May contain violence”, “May not contain demonstrations” and so on. When a user visits YouTube, their location is known by their IP, and YouTube – the AI of which understands the actual content of a clip – compares with the legalese API data to understand which scenes should be replaced. Replaced scenes are either taken from other, copyright-free videos, or re-enacted using YouTube’s smart 3D objects and avatars.
9. YouTube Change the Cam
What it is: When watching an archived news report on YouTube – say, a press conference by the president – a camera symbol shows up in the bottom of the video. Click it, and the same press conference will be displayed from another camera angle.
How it works: YouTube checks, rather fuzzily, if the specific audio track of a clip has matches in other clips. If it finds other clips, it checks if the video differs substantially. If that’s the case then YouTube figures it’s the same scene shown from a different angle.
10. YouTube Shut Out the Kids
What it is: YouTube allows you to switch on a setting so that kids won’t be able to see certain content, like brutal kung fu movies.
How it works: It doesn’t really work, because kids end up outsmarting their parents and blocking them from seeing anything. YouTube cancels this feature after a two week trial and creates a support hotline for confused parents.
11. YouTube AutoRatings
What it is: Every clip is assigned a rating expressing how funny it is, how dramatic, how brutal, how sad, how romantic and so on. Nobody manually sets these ratings, though. You can then search, for instance, for a movie that’s funny but thrilling, and get back a comedy action buddy movie.
How it works: A great number of voluntarily participating YouTube users have their webcam turned on whenever they surf the web, including YouTube. A small program implemented in Google’s browser recognizes the face and parses the facial emotion, and also listens to the user voice. If you smile, the cam will know, and attach that emotion (happiness/ funny/ good) to the individual video clip or other page you’re on. This in turn helps them identify the emotions a particular clip or web page triggers, which then helps them deliver better web search and YouTube clip search results. (SEO spammers quickly learn to abuse this feature by paying large crowds of people to visit their websites with a forced smile.)
12. YouTube Social Network
What it is: Google turns YouTube into a giant social network, based on videos you viewed.
How it works: Similar to the Gmail Buzz program, Google connects everybody to everybody else, ignoring certain privacy details for a while. People will be shocked and complain about how their viewing habits are suddenly connected to their web search history and email address book, so Google will refine their opt-out setting, but at that point millions of users have already been connected, and Facebook is dead.
13. YouTube FindYourProduct
What it is: A company, like the Coca Cola Company, can enter their product – “bottle of coke” – and find videos of people happily consuming their product. They can then add their banners to the side of that video, or create lotteries where everyone consuming their product has a chance to win big money, or they can simply analyze consumer behavior.
How it works: A straightforward image recognition is the basis of FindYourProduct. YouTube content creators later understand that they can make more money by strategically placing products inside their videos – because this will get them better ads, and a chance to win in the product lotteries – so YouTube is flooded with lots of grassroots product placement.
14. YouTube 3D
What it is: You can view any film on YouTube in 3D. Even without special glasses, and without the respective film being shot in 3D.
How it works: Based on a patent by Matt Cutts, Google’s algos analyze depth layers in clips by checking movements and generating a pseudo-3D scenery out of it, extrapolating existing visuals with a lot of guess work. Using a very complicated Subpixel Rendering Blinking Effect built into the Google Chrome OS for that purpose – lucky Google, as their major OS market share allows them to roll out whatever browser feature they need for their web sites – the movie transform into 3D.
15. YouTube Smell the Vid
What it is: Once activated, a scene in a movie will smell just like it would if you were there. If the movie protagonist is on a fishing hunt on his boat then you’ll smell the salt water and the fish. If the protagonist of a romantic movie puts on perfume, you will know its brand without seeing.
How it works: The Google Chrome Pad computer will have an odor emitter. Just like speakers emit sounds, this piece of hardware emits smell, by mixing different base fragrances to create e.g. cinnamon, strawberry, salt water and so on. Websites can define these fragrances in their stylesheets. YouTube on the other hand automatically analyzes video clip content semantics, and emits the needed fragrances, outputting them in real-time.
16. YouTube’s Just Let Me Watch the Movie, OK?
What it is: The “Just Let Me Watch the Movie, OK?” setting (JLMWMovie-OK), once activated, will show a movie as it was actually intended to be seen by its director.
How it works: JLMWMovie-OK is rather simple: activate it, and the killer features 1 to 15 as mentioned above will be deactivated. The artistic integrity of the movie is preserved, and you can lean back with a pack of popcorn and actually just enjoy the video.
[Photo source CC licensed by Scott Eric.]
Thursday, March 4, 2010
YouTube With Auto-Subtitles Option for All English Videos
TechCrunch reports that YouTube announced they will roll out auto-captions to all English videos now, for content creators who want this (before, only a portion of videos received this option). As a disclaimer, Google noted that (quote TechCrunch) ...
... Just like any speech recognition application, auto-captions require a clearly spoken audio track. Videos with background noise or a muffled voice can’t be auto-captioned. President Obama’s speech on the recent Chilean Earthquake is a good example of the kind of audio that works for auto-captions.
When video creators now choose a platform to upload their video, a feature like this – well in sync with Google’s stated mission – could make all the difference.
Ads Can Provide The "Best Answer" For a Search Query, according to Google
By Roger Browne
SiteLinks are now appearing in Google’s Sponsored Links. An AdWords SiteLinks unit consists of a regular text ad with four additional links. These ad units were sighted in August 2009, and trialled in November 2009 in conjunction with some of Google’s Agency Partners, including Gillissa in the UK.
Each link can be up to 35 characters. Up to 10 links can be entered by the advertiser, although only 4 will show at a time. The links are not always displayed, and none will appear unless the ad is the only ad being shown for the search term. The advertiser pays the same for each click, whether it is on the main part of the ad or on one of the SiteLinks. If a viewer clicks more than one of the links, the advertiser is not charged for the subsequent clicks.
Ads Can Provide The “Best Answer” For a Search Query
Here’s what Google AdWords Help has to say about the display of AdWords SiteLinks:
Ad Sitelinks are designed to trigger in situations where an ad provides the “best answer" for a search query, and Ad Sitelinks are most likely to trigger on unique brand terms.
The screenshot above was the result of a search for [bing]. However, I got the same ad when I searched for the placename ["Bing, Montana"] and the ad is certainly not the “best answer" for that search query.
Wednesday, March 3, 2010
Onion: "Google Responds To Privacy Concerns With Unsettlingly Specific Apology"
Fake news site Onion has a piece titled Google Responds To Privacy Concerns With Unsettlingly Specific Apology. [Thanks Scjm]
Topeka, Kansas Is Now Google, Kansas
By Roger Browne
Bill Bunten, the Mayor of Topeka (the capital city of Kansas, USA) has signed a proclamation declaring that during March the city will be known as Google, Kansas.
This is being spun by the press as a name change, but that’s not the case. The city’s official name remains Topeka, even though during March it will be (also) known as Google.
The wording of the proclamation is interesting. It starts by asserting that Google “is a corporation who maintains a small company feel”, then lists ten core principles of Google’s philosophy. (Somehow, the list still includes “It’s best to do one thing really well”.)
Then we get to the purpose of the proclamation. The City, and it’s 11,000-strong “Think Big Topeka” organization, are begging Google to bring its Fiber for Communities pilot project to Topeka.
It’s just a publicity stunt, and I wonder whether Google will bite.
By the way, Kansas is the Center of the World According to Google
[Thanks Juha-Matti!]
Knol: The State of Play
By Roger Browne
Knol is a Google service that allows anyone to publish an article about almost anything.
When Google’s VP of Engineering, Udi Manber, announced Knol in December 2007 it was widely regarded as a “Wikipedia killer”. If Knol’s users could replicate the success of Wikipedia, the pundits proclaimed, then Google would gain an enormous commercial benefit. No longer would users leave the Google’s search results to find their answers at Wikipedia; they would go to Knol instead.
Google, meanwhile, was promoting Knol as a place for named authors to write authoritative articles to share “units of knowledge”, or knols.
Google’s announcement was accompanied by a sample Knol page, an article about insomnia by Rachel Manber (Udi Manber’s wife) who is an insomnia specialist from Stanford University. The article was comprehensive, carefully-written, well-illustrated, neatly formatted and very readable.
Knol conducted a private invitation-only test until the gates were opened to the public in July 2008. There was an immediate flurry of interest, encouraged by the potential to earn AdSense income by allowing Google to place ads on knols.
Unlike many of Google’s other ventures, where Google’s team can be quite secretive, the Knol team was remarkably open, engaging directly with the authors. They solicited improvements, delivered them, discussed problems, helped those who showed potential, and even submitted collaborative edits to many articles (including one of mine).
The first hundred thousand knols
Less than six months later, Google proudly announced that the 100,000th knol had been published. The announcement was met with cries of “So What?”. The media has a short attention span, and those who remembered the launch of Knol didn’t think much of where it hand ended up. Slate magazine hadn’t even waited that long before writing off Knol.
Matt Cutts thought it was weird that the same people who had hailed Knol as a dominant player were now the ones now dismissing it. Matt blogged that, in his opinion, Knol was doing just fine, but the comments posted to Matt’s blog were overwhelmingly negative.
Platform development
Meanwhile the Knol team slogged away, improving and promoting Knol. It seems to me that they’re taking a long-term view of this project. A steady stream of new features, improvements, and marketing activities was delivered:
- Knol debates, with opposing knols pitted head-to-head (in contrast with Wikipedia’s “neutral point of view” policy)
- articles permitted in any language, and eight languages for the Knol interface
- version control with rollback
- various awards and badges for “top viewed” and “top pick” knols
- a very comprehensive advanced search system
- display of related knols in the sidebar
- table and caption editing tools
- email notifications and activity feeds
- integration with Google Analytics
- facilities to help embed knol content into other sites
- a competition in conjunction with dummies.com with a $1000 prize
- a new kind of knol called a collection, which works like a directory
- expanded name verification options (but still for the US only)
- the ability to embed images, equations, slideshows, video, calendars, maps, documents, presentations, spreadsheets, gadgets and thousands of special characters and symbols
- audio playback of knols
- “dofollow” links for trusted contributors
- pageview statistics
- comment rating
- a promotion for knols about holiday crafts
- crowdsourced categorization of knols
- a ten-point review scorecard for each knol
- a standard citation format
- a twitter interface
- ten million professional stock photos licensed for use at Knol via PicApp
- and many, many more enhancements
20 months later
Now, almost 20 months after its launch, the service still sports the beta tag. Google hasn’t announced an updated count of knols, but a search for [site:knol.google.com] shows 163,000 results, although there may be more. Those results are a real mixed bag.
There are lots of spam pages which have no purpose other than to drive traffic to dubious commercial websites. If you’re interested in acai berries, you’ll find plenty of knols about them.
Next are the crank pages. Knol provides a soapbox for any author to pontificate about whatever they like, so it’s not surprising that you can find frauds, lies and delusions. Some of the more egregious of these, such as the “run your car on water” knols, are being flagged and removed. Other knols just reflect the obsessions of their authors. If you want to read articles like “Horniness is a spreading epidemic that must be stopped”, knol is the place to go.
There’s still much copying of content from Wikipedia, despite great efforts to stamp it out (helped by the “similar content” links generated in the sidebar of the knol).
There are thousands of pages about Knol itself. Many are full of navel-gazing and pontificating about the service, rather than real content. Around these pages have formed several “organizations”, informal affiliations of knol authors who give out awards, place their stamps of approval on articles, or conduct lobbying campaigns. It tends to be a small clique of knol authors who get involved with this, and the whole process seems to me to be somewhat self-congratulatory.
A “currency” of sorts has evolved, called “knol translation points”. The idea is that you earn these points by translating knols into other languages. You can then spend the points to have your own knols translated, or you can “award” your points to others in recognition of their good work.
There are some interesting pockets of light. A number of eHow authors have contributed over six hundred knols under the joint username eHowKnol. The quality is variable and unexceptional, but nevertheless above average.
More significant is the use of Knol by the Public Library of Science, an open access publisher, as a platform for “open communication and discussion of new scientific data, analyses, and ideas about influenza research”. Under the title “Influenza: Currents” new articles are drafted and collaboratively edited. They are then subjected to stringent vetting by team of qualified moderators, after which the successful articles are included in PubMed.
Knols can be sorted by rating or by page views. Looking at the top knols, it’s clear to me that the most-viewed knols are, on average, much higher quality than the top-rated knols.
Only ten knols have had more than 100,000 page views over the past 20 months, and some of those are knols written by Googlers back in 2008. The majority of knols have had far fewer page views, mostly below a thousand, and sometimes only several dozen. It seems that Knol is a great platform for writers, but not so much for readers.
A killer?
How has Knol fared as a Wikipedia killer? Google Search Trends provides a dramatic answer:
(I couldn’t compare the sites using Google Website Trends, because that service does not accept queries for google.com and its subdomains.)
The Knol team no longer seems to be using the phrase “authoritative content”, and I can’t help thinking of something else Matt Cutts wrote after the 100,000th knol was posted:
My personal conception of Knol is that when you want to write a quick article or put some information on the web, Knol is a great place to do it.
Is that it then? Is Knol destined to be nothing more than the next Geocities?
Weirdchina With Thousands of Followers on Buzz
At Gmail Buzz I’ve been subscribing to Robert “Weirdchina” Kong Hai and noticed that some of his buzz posts received tons of comments. I’ve asked Robert about this, and here’s his reply:
Your right, my posts do get a lot of attention, I’m having a fun time on buzz! I think the social media community is very surprised that my 19 posts have received almost 3000 comments and 1500 likes. This might even be the most on Buzz. I also have more then 13,000 followers right behind Mashable, so I could be the 2nd most followed person on Buzz at the moment.
I’m in a unique situation, I have built up a world community based around the Chinese netizens. For a long time I have been sort of a bridge between the Chinese social media users and the rest of the world. I often speak about this. When people talk about social media, they never include the Chinese in the discussion, although there are 384 million China internet users, (more then the population of the USA) and by 2015 that number grows to 500 million, so instead of ignoring them I embrace them and give them a much needed voice. So this has attracted a lot of attention, sort of a viral type of thing. I expect these numbers to grow much higher on Buzz as this is just the beginning of something very special that is happening.
....and It also helps that I have 180,000 followers on twitter as @weirdchina & @weirdchina2
This site unofficially covers Google™ and more with some rights reserved. You can subscribe to the feed, email your tips and join our forum!