99 Designs – Crowdsourced Visual Design


I’m a huge fan of crowdsourcing and have been following and using mechanical turk for years.  Using technology to reach out to workers or experts who can then in turn work when they want.  It is a simple concept but also very powerful and is the way of the future.   It is efficient economically because it is helping suppliers (workers) meet demanders (businesses).  The key to modern crowdsourcing is to define the unit of work property so that it can be executed independently and in parallel.  My fascination with crowdsourcing, design, technology and entrepreneurship made me a fan of 99 designs right from the start.

With 99 Designs has a 200K+ vertical crowd (community) of designers that can work on your tasks.  For example if you need a logo you can create a job then designers can submit their proposals.  One of the cool things is that anyone can “work” on your task often just for the hope of getting select for the next round.  This means that you can get 100+ designers to submit initial proposals then you can start narrowing the field down and do another iteration against the crowd further refining the logo.  At first this does not seem that impressive but it really is quite amazing.

I can’t wait for an opportunity to try it out.

My @nytimes Dialect Quiz Map

Image Source: http://bigkingken.files.wordpress.com/2012/02/total-county.gif

The new york times released a fun dialect quiz map that attempts to pin point where you live based on how you talk.  The basic idea is that a combination of things you say and how you say them can uniquely identify where you grew up.  Most questions are coming from a Harvard Dialect Survey started in 2002.

Typical question

Screen Shot 2013-12-23 at 11.57.50 AM

Working in the NLP space I had high hopes that the quiz would work quite well.  Having grown up in Seattle, which does not have a significant amount of dialectic uniqueness, I also had some doubts on its accuracy.  It turns out that I was right.

My results:


Although it was not totally accurate it does not mean that it isn’t a useful tool.  The accuracy is highly dependent on the individual and their geo-dialectic past.  My wife, for example grew up in an around New York City.  When she took the survey it was able to pin point the exact town that she grew up in.

Real Applications

With this type of dataset there are some really interesting applications that could be used.  For example, marketers could comb social media and simply through the language used in the posts determine demographic information to target adds.  Another interesting idea would be to attempt to uniquely identify a person by the works and phrases that they write.  This could be used as a form of authentication similar to that of examining the canvas of a famous lost work of art.

In order to make this more useful the accuracy has to be better.  I’m especially interested to see if it could be improved by incorporating more mainstream idioms or cultural norms.  For example, if I talk about coffee am I talking about latte’s at Starbucks or a cup of joe at Dunkin.

Take the quiz

Take the quiz and let me know how it went.  Here is the link:


Crawling the Web for Synonyms

Lately I have been interested in writing a web crawler to help me find and exploit information on the web. Two ideas I had that might be feasable would be writing a program to wander the web looking for pronunciation data. Another idea is to crawl domain specific websites for keywords and conversations to help identify and look for aliases and synonyms. For example I went to a popular music site and found the following a conversational thread.

In a single thread here is what I extracted:

The track is caled trinity
music on it
Name that song!
What’s the name of the song from the second half of the trailer
remixed a track
So what is the name of the song at the second half of the trailer?
wait, where can you get “Tempest”?
I’m also interested in the music
what is the second song played
I’m looking for the song
“Born to be Dizzy” by The Starlite Desperation.
It starts with something like “All my heroes turned out to be …”

This process could be automated using NLP and used to collect vast amounts of linguistic information from thousands of websites. The best part is. Its Free!


SmarterChild is a fun and entertaining instant message that your kids are probably more familiar with than you. By adding “SmarterChild” SN (Screen Name) to you Aol IM client you can interact with a program. While hardly a testament to the Turning test it is quite exciting in the knowledge base this chat bot has build up. Combined with a little Elisa magic, natural language processing and back-end content it creates for a very rich experience. Late last year Microsoft purchased the parent company, Colloquis to fit into their Live Services vision. Specifically the purchase allowed Microsoft to leverage the existing technology from Colloquis to build better customer support systems.

See the website here: http://smarterchild.colloquis.com/