All new parents want to pick the perfect, unique name to fit their little ray of baby sunshine. With each generation though, the most unique of baby names tend to fall into quite identifiable trends. If you were a girl born in 1950, you might be called Donna, Barbara or Linda. In 2000, you were more likely to be called Madison, Alexis or Abigail.
Developer and designer Nate Parrott of Brown University has found a way to hand the baby-naming over to the computers, in a bid to generate new and ultra unique names for the babies of the future. Using a list of 7,500 popular American baby names, Parrott trained a neural network to convert each name into a mathematical representation called an embedding. It’s a tactic often used in machine learning.
”Once I had a model that could translate between names and their embeddings, I could generate new names, blend existing names together, do arithmetic on names, and more.”
By analyzing the similarities in names, the algorithm identified which words were more name-like than others. It could tell that words like ‘Winter’ and ‘July’ were more name-like than the word ‘automobile’. The results are pretty interesting. Some of Nate’s new names are plausible, like ‘Seina’, ‘Mannie’, or ‘Aloora’. Others are a little more experimental and may take a little longer to catch on, like ‘P’ or ‘Chhzzu’.
See the list of Nate’s top 30 computer-generated names below:
While some of them do seem a little far fetched now, who’s to say that in 2117 it won’t be cool to be named ‘Hhrsrrrrrr’? See the detailed diagram and explanation of how Parrott’s neural network works at medium.com.