I guess it's slightly more (30mb uncompressed for 100k sentences), but text compression ratios are generally really good these days if they have high similarity (which I wager a dialect dataset would)

1 likes 1 replies