nigbot's model is REALLY insistent on making certain sequences show up repeatedly. Not surprising due to it's repetitive post history, but it's causing repetitive substrings to appear in owl's output. I'm wondering what would happen if I just toned his importance way down in the final coherent model.
I did try filtering out textual repeats but it actually didn't help that much. The issue is lack of variation.
@BowsacNoodle that might be a good idea. I think rap lyrics may have a similar problem since they also tend to be really repetitive.
on top of that I am actually not sure where I could find rap lyrics with phonetic spellings since I believe many of them are translated into white people spellings.
@BowsacNoodle I have the typical generation case narrowed down to a pool like this now, which is great, but the remaining problem is 100% nigbot. You're right that I need to figure out somewhere else to scrape from and add to his data.
@RustyCrab@PurpCat@BowsacNoodle me and blink used to have a twitter bot called princess that had a fat black woman pfp and used that as the source of tweets
it's sourced from a bunch of accounts lol
@Kerosene yes it might be a good idea to capture samples of ebonics from real nigga(ers) on the federated network. Any particular accounts with strong niggaspeak would be appreciated.
@RustyCrab Sorry, despite being La Luz Extinguido I don't post nor see much of that. I wonder if there's anything like "black Twitter" but here on the fediverse.
Add comment