Hi, my name is David.  Welcome to the Utah Baby Namer, which came into being when I made a machine learning algorithm read 200 thousand Utah baby names to see if it could learn the essence of a Utah name and how to make up new ones.  I think we can all agree that this is the noblest use for artificial intelligence yet invented by man.

Follow on Twitter at @utahbabynamer and share the best of the best.

 

How does the baby namer work?

Every name delivered by the baby namer is 100% generated by a particular kind of artificial intelligence algorithm called a recurrent neural network. (LSTM, for the nerds).  If you’ve heard recently of A.I algorithms coming up with absurd recipes, hilarious paint colors, Candy heart messages, Shakespeare, Harry Potter fanfic, and Scrubs episodes: this is basically the same kind of technology.

To get Utah names, what I did was feed the algorithm a large collection of current Utah baby names — every baby name since 2005 that was at least twice as popular in Utah as in the rest of the U.S.  The more popular the name was in Utah relative to the U.S., the more times the name appeared in the data.

The A.I. starts by not knowing anything about what constitutes a name (or even a word), so at the beginning it tries nonsense names like “dcrksxii” and “tAunxp” (actual attempts).  But after several rounds of training, it starts learning how to put consonants and vowels together, then the names start sounding reasonably like names, then before you know it, downright Utahn.

One cool thing about recurrent neural networks is you can tune their creativity level when they’re generating stuff.  So when you use the namer on the “low creativity” setting, the algorithm regurgitates names close to—or even exactly the same as—names it’s seen before.  On Medium Creativity it gets a little more adventurous, suggesting some actual names and some names that aren’t actually real Utah names, but maybe sound like they could be real.  On the high creativity setting the namer will create entirely new names as best it can in the style of Utah.

Give it a try!

So why Utah?

Utahns are well-known for their creativity in baby naming (see here, here, here, and here for examples).  In fact, Utah is home to the most different names of any state in the continental U.S. [2]

So Utah has ended up being kind of an innovation hub of baby naming for the rest of the country.  A lot of the names that become popular in Utah get picked up by the rest of the country later.   Names like Brooke, Natalie, Paisley, Braxton, Lacey, Charlotte, Ruby, Alice, and Nora all became popular in Utah well before they came into vogue in the U.S. at large. (Read more here, here, and especially here). So who knows, if this little baby naming bot can learn to name like a Utah parent, maybe it’ll come up with the next national naming fad!

 

About me

I’m a data scientist who lives in Salt Lake City.  For my day job I try to convince artificial intelligence to solve biology and cure all diseases.  I live with my lovely wife Taixy and two adorable little boys, Riber and Brude.

(ps their names are not really Taixy and Riber and Brude, but they would have been if the Namer had been available when they were born, that’s my point.)

 

Notes

[1] Data for the Namer and the articles comes from the U.S. social security administration (ssa.gov).

[2] By “most different names per capita”, what I mean specifically is Utah has the highest percent of children that are given names that are least 10 times more common in their home state than in the nation. Except Hawaii.  Hawaii is in a class all its own.

[3] No discussion of recurrent neural networks could be complete without mentioning Andrej Karpathy’s blog post, which unleashed the flood of A.I. text generation on the world.  See also Janelle Shane, the godmother of making RNNs come up with stuff.