This was a fun weekend project. The idea is to produce a more rigorous and comprehensive set of Benedict Cumberbatch-sounding names than is available elsewhere.
The makename.py script reads the BENEDICT.txt and CUMBERBATCH.txt files to generate random name pairs:
$ python makename.py
BUGABOO CADILLAC
$ python makename.py
POLITIC COUNTERMAN
$ python makename.py
CANDIDATE CORTISONE
You can also seed it with a string of your own to make it perform as a consistent name generator:
$ python makename.py "brent tubbs"
DEMAGOGUE COUNTERPOINT
$ python makename.py "boyd ocon"
BODYGUARD CONTRABAND
This project was greatly aided by Carnegie Mellon's freely available pronunciation dictionary. It contains lines like this:
BENEDICT B EH1 N AH0 D IH2 K T
CUMBERBATCH K AH1 M B ER0 B AE2 CH
SUPERCALIFRAGILISTICEXPIALIDOCIOUS S UW2 P ER0 K AE2 L AH0 F R AE1 JH AH0 L IH2 S T IH0 K EH2 K S P IY0 AE2 L AH0 D OW1 SH AH0 S
Programmers will mainly be interested in the rank.py file, which creates the BENEDICT.txt and CUMBERBATCH.txt files. It reads the CMU pronunciation dictionary, parses the entries, and scores each word by how much it sounds like "Benedict" or "Cumberbatch". This is done with a custom phonetic distance function that takes breaks each word into syllables and compares those, further breaking each syllable into its beginning consonants (if present), main vowel, and ending consonants (if present). Differences increase the distance score. Differences in the first syllable, last syllable, or number of syllables incur extra penalties.
After ranking all words by similarity, I arbitrarily removed all words below a minimum threshold. I also added in a couple words that the algorithm unfortunately missed (e.g. "bumpercar", "cuttlefish").
TODO:
- Keep fine tuning the distance function. Make "ch" close to "sh" for example.
- Compound words! To really be exhaustive, the script should be able to create things like "numbercrunch".
Inspiration and prior art: