COCA is a high-quality corpus (although the part of speech tagging isn't always perfect). You can verify that you're getting good results by looking at the context.
Looking at the hits, they all look genuine to me.
Most of the academic hits come from "AmerIndianQ", and specifically a Native American story about a coyote that was also printed here.
Some of what is labeled fiction is from movies, such as Fargo. There's also written fiction, which is often informal. Even if it's not as extreme as it was in other times (e.g. "git fo’ dollars mo’ at de en’ er de year" from Huckleberry Finn), it's still common to see dialect (a good, popular — albeit British — example would be characters like Hagrid in Harry Potter).