Current Bio May 2018

George Kachergis discusses a recent publication by Berens, Horst, and Bird (2018) investigating whether word learning is primarily based on associations or hypothesis testing using an fMRI cross-situational learning task. In a hypothesis testing model, one would first hypothesize the meaning of a word, but correct that meaning as needed/necessary with more experience with that word. In an associative model, for example, associations between words are learned. New words are learned based on their association with already known-words or their novelty/lack of association with a new word. Therefore, associative learning models are thought to occur more gradually. Berens et al found that learning word-object pairs occurred somewhat rapidly, and hippocampal activity was consistent with a rapid-learning/hypothesis-testing model. Just over half the subjects also reported using hypothesis-testing strategies during the task as well. The scan data looks convincing, but subjects were only learning 9 novel word associations, and they were “pre-familiarized” words and objects – reducing the novelty of these “new/to-be-learned” words. Kachergis also points out that more word-object pairs should be included to make it more like how actual language/word learning may occur.

Here is a link to the dispatch article by Kachergis: https://www.sciencedirect.com/science/article/pii/S0960982218302999

And a link to the article by Berens et al from March: https://www.cell.com/action/showMethods?pii=S0960-9822%2818%2930227-6

For those interested in memory, a really interesting paper just came out from Indiana University (and they also had a talk this year at CO3!). Rats that were able to discriminate 12 different odors were placed into arenas that had covered food holes. A list of odors would be presented through these holes to the rats, and list length was inconsistent so there was no way to predict when any given trial would end. Rats were then moved into two different “contexts” (distinct arenas that differed in size, shape, and color). For one, they would have to discriminate between a foil odor and an odor that was presented second-to-last. For the other, they would discriminate between a foil odor and the odor that was presented fourth-to-last. For this first experiment the rats performed above chance, with an accuracy of about 86% correct.

To control for the possibility the rats were using relative familiarity (memory traces for the odors) for the discriminations rather than memory replay, they doubled the time between list items for the last four odors. The idea here was to present the second to last item (that now had the “typical” memory trace for the fourth-to-last item from Experiment 1) with the last presented item (that now had the “typical” memory trace for the second-to-last item from Experiment 1). Again, rats performed above chance. My concern here is that the rats could still potentially be using length of time for these discriminations, rather than relying on the “typical” memory trace carried over from the first experiment? I’m not sure though, I spent a lot of time trying to sort out this manipulation and whether or not I liked it when I was reading over the paper, so maybe someone can help me out here?

Several other experiments were run to a) show that replay memory survived a 60 minute retention interval between odor presentation and memory assessment (rats again performed above chance), b) show that replay memory is resistant to interference by including a new-old odor discrimination between list presentation and memory assessment (rats again performed above chance on the memory assessment – even when the foils for the new-old discrimination task was either the second- or fourth-to-last odor from the list), and to c) show that this task, as episodic memory replay should be, is hippocampal dependent, by suppressing hippocampal activity in the rats and observing deficits in the rats performance on these (However the rats still performed at about 75% accuracy here), but not hippocampal-independent memory tasks.

I think this paper is really cool, and the data seems convincing. The whole time I was reading it I was worried about experimenter cueing. I found tucked in at the end of the methods that to control for this they would replace experimenters and look for changes in accuracy after a switch, which they did not find, but I’m still a little skeptical and wished they had included some trials without an experimenter present. I also think a study like this is a perfect example of a study in which replication is important. Overall, still a really interesting read (and talk at CO3).

Link to article: https://www.sciencedirect.com/science/article/pii/S0960982218304342