Results 1 to 3 of 3

Thread: Neural networks for name generators - anyone tried them?

Hybrid View

Previous Post Previous Post   Next Post Next Post
  1. #1
    Administrator ChickPea's Avatar
    Join Date
    Jan 2015
    Location
    Sunny Scotland
    Posts
    6,884

    Discuss Neural networks for name generators - anyone tried them?

    Been meaning to ask about this for a while, but recent comments in the software discussion reminded me.

    The tl;dr summary... has anyone ever used neural networks to generate fantasy names, and if so, any tips/advice/how to's to share?

    Slightly longer version...
    I love the idea of neural networks, though my programming knowledge is extremely rudimentary. However I know just enough python to be dangerous...

    I used textgenrnn from here...

    https://github.com/minimaxir/textgenrnn

    I used a dataset of Scottish place names from a government website, as I wanted to start with something I was fairly familiar with, just to see how it worked. In some ways, it worked almost too well, as I was getting actual place names even when I bumped the temperature up to over 1 (I think I ran 15 epochs to train it, maybe should have done less.)

    Out of every 10 results, I'd get one or two actual place names, then maybe 2 - 4 that were usable (or usable if I substituted a letter or two). About half were pretty much garbage.

    Anyone else tried this? Not going to lie, I had a massive amount of fun with it, and I'm going to be hunting down other datasets to try out. Just curious to know if anyone else has dabbled with this, and how it worked for you.
    "We are the music makers, and we are the dreamers of dreams"

  2. #2
    Guild Journeyer eepjr24's Avatar
    Join Date
    Nov 2020
    Location
    East Coast, USA
    Posts
    150

    Default

    I have not played with them, although I am familiar with the concepts. They are partially an ML implementation of techniques developed for matching names based on degrees of similarities used for decades in name and address match routines. The problem you have with real place name generation is hard to overcome without either very large sample sets, an exclusion list or significantly increasing the temperature as you noted.

    If you like rabbit holes, check out some of the more advanced routines that attempt to generate hypocoristic alternates. I am not sure where to go to find a public implementation of that, but I know several of the larger IM players (SAP, Informatica, etc.) have attempted to develop something that has a reasonable success in that arena.

    - E
    Content available for use under CC BY-NC-SA 4.0.

  3. #3
    Administrator ChickPea's Avatar
    Join Date
    Jan 2015
    Location
    Sunny Scotland
    Posts
    6,884

    Default

    Quote Originally Posted by eepjr24 View Post
    The problem you have with real place name generation is hard to overcome without either very large sample sets, an exclusion list or significantly increasing the temperature as you noted.
    Yeah, my dataset had nearly 6000 entries, which I thought would be plenty, but I suppose more is always better.

    My main issue with including real place names is that I wanted everything to be fictitious. I can usually pick out the real names from my Scottish list, but if I wanted to use different datasets from around the world, I'd have no clue what was real and what wasn't. Your idea of an exclusion list would work though. I can get my python script to check if the name generated is in the original list, and if so, delete it. That won't be too hard, even for me!


    If you like rabbit holes, check out some of the more advanced routines that attempt to generate hypocoristic alternates.
    Hehe I think I will very quickly get out of my depth with this. Sounds very interesting, though.
    "We are the music makers, and we are the dreamers of dreams"

Tags for this Thread

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •