That doesn't make much sense.
Still, it doesn't make a lot of sense that they need to do that many location requests. Why should where I'm standing right now affect what words are in my dictionary? It's a poor proxy at best, the words that I use in my language are a product of both where I grew up and where I spend my time. Where I happen to be standing does not always matter that much.
I wonder if bucketing users into "these people use the same words" would give better results?
Since Swype has do to a lot of inference from your gestures, it seems reasonable that it may perform better when inferring proper names with some location awareness.
For example, if I am in a particular city, and wanted to type the name of a street where I am.
I am not saying that's a feature I find compelling; it does sound reasonable though. Letting the user chose this behavior (and the permission), would be best.
If my keyboard thinks, just because I'm in New York City, that I meant to say, "Stand on line" instead of "Stand in line" it's insane.
If not, then ok it's terrible programming, but nothing worth getting paranoid about.
If it is checking location data for location specific words it there should me a minimum time between checks as location can can't really change enough to matter in the 20-30 sec it currently uses.
Not that I think fine location data is necessary (I don't, and would like clear permissions options), but not sure IP geolocation is a user-friendly substitute if needed.