Saturday, May 11, 2013

A tale of two Google searches

1.  I had a song stuck in my head, but I didn't know the words.  It dated back to childhood, most likely from Sharon, Lois & Bram, and the lyrics as I remembered them were "My mother need to tell me that you omungowah."

Clearly, I had misheard it or was misremembering it, and was jamming a bunch of phonemes together to make "omungowah".  And whatever the omungowah really was, it was probably the crucial word in googling up this song.

Expecting nothing, I started typing my mother need to tell me that you omungowah into Google, and before I even got to the omungowah, the suggestion feature gave me "My mother didn't tell me that you go mango walk".  Which is exactly the song I was looking for!   Well done Google!

 (Here's an example of the song, although I have no idea what the source is.)

2.  In 2009, they had a public art project where people could stand on an empty plinth in London's Trafalgar Square and do whatever they wanted for the audience of whoever happened to be in Trafalgar Square and as a live real-time webcast audience.  In the middle of one of these plinth performances, Eddie Izzard finished his marathons, also in Trafalgar Square, and the crowd and cheering of his marathon finish interrupted one of the performances and distracted the camera operator.

I was looking for this video, so I googled eddie izzard plinth. Not only did Google not find the video, but it gave me one of those despised "Results for similar searches."  And the "similar search" that it proposed was eddy izzard!

Yes, they not only eliminated the key search term, they introduced a spelling error!  (Interestingly, the results for eddy izzard were Eddie Izzard's website, IMDB page and wikipedia entry, all of which spelled his name correctly.)

It seems like Google's algorithms missed a few crucial points. First of all, "Eddie" is a far more common spelling than "Eddy". How do they end up "correcting" away from the more common (and correct) spelling?

Second, Eddie is a celebrity with an unusual surname, which means that a disproportionate number of instances of the word "Izzard" on the recorded internet will have the word "Eddie" next to them.  Surely their concordance function should have figured this out - at least enough not to change what I entered!

And third, if your search contains something general (a celebrity's name) and something specific (the word "plinth"), the specific thing is probably there for a reason.  It is in no way helpful to completely eliminate the specific and give the user only general information about the celebrity!  If Google is going to insist on using this "Results for similar searches" function, they should use synonyms of the most specific search term, or use words that correlate with the specific search term ("Trafalgar" might have been helpful, for example.)

How is it possible that Google could fuck up this badly while still being capable of finding my "omungowah"?

(The video of Eddie Izzard finishing this marathons in Trafalgar Square and interrupting the plinth performance can be found at 31 minutes here.)

No comments: