Obligatory Post of the Month

Harro. I've got some experiment scripts running, and thought I'd post while I wait for the first few results. I'm home for thanksgiving, w00t. My niece seems to be growing in ability but not in size... :p She can kind of walk. Or actually I think she's just managed to fall over a bit more horizontally. Maybe she's about to start flying in an Adamsian manner?

These dang ol' simple recurrent networks are really good at finding statistical predictors of what they are being trained on. Too good! I want them to learn something about boundaries in speech by showing them utterance boundaries (utterance boundaries we think we can get for free because they are followed by lots of silence). Instead, they are learning a whole lot about utterance boundaries in speech, which is not helpful because we assume kids can get them for free, or at least pretty close to free.

I piloted the study I want to do for my M.A. on Tuesday night and it went astonishingly well. We only have 9 people total, but it looks like I might be able to show that people treat sounds completely differently if they think they are hearing language as opposed to hearing other stuff. This has been shown for one aspect of language, but I'm looking at statistical learning of words. Since the sorts of statistics that the word-learning task demands are not all that complicated, everyone has sort of assumed that statistical learning means that the mechanisms for word learning are not specific to language. It'll be neat (and unexpected!) if this holds up over 30 kids to a condition. I want to make 5 year olds do this too.... we think they can probably do the task since it's so short, even though it's harder to cast what actually happens as a "game."

oh yeah! happy thanksgiving! ^_^

hmm, thinkin' that's it. bai bai

-=-raptur-=-