2

A little background: this paper concerns forgetting in neural networks (a computer science concept). The word "activation" might mean the activation (=output) of neuron(s), but since I cannot parse the sentence, I'm not sure. Another one, "at test" - it likely means "at test time", that is, after the network has been trained, when we "test it" and no longer change any weights etc.

The model was applied by McClelland and Rumelhart (1985) to data from the time course of growth of activation at test as a function of variables such as familiarity and priming condition.

I cannot parse the passage in bold specifically. To me, the sentence is illegible and possibly incorrect, but my English skills are limited. Please help.

7
  • 2
    Oh, it's grammatical all right, but too much has been left out. Most likely this resulted in well-established formulas that serve to identify the authors' attempt to meet anticipated questions or criticisms; idioms are thickest in the most tightly-focussed speech communities, and this is a lot of what makes technical literature so hard for ordinary mortals to read -- we don't have the presuppositions and speech habit to follow it. Like an American listening to cricket talk. Mar 1, 2015 at 15:32
  • To a native English speaker, it also makes no sense. One might have to know the jargon used by neural network researchers in 1990 to understand it (or it might be incomprehensible due to typos). Mar 1, 2015 at 15:33
  • 2
    McClelland's papers appear to be online, by the way. psych.stanford.edu/~jlm/papers
    – anemone
    Mar 1, 2015 at 15:35
  • I wasn't familiar with this use of the time course, but presumably it's by extension from timecourse - (medicine) The varying activity of a medicine over time following administration. Mar 1, 2015 at 15:38
  • 1
    I think it's "at test" which is the technical collocation. Compare "decision making at test": researchgate.net/publication/…
    – TimR
    Mar 1, 2015 at 15:45

1 Answer 1

4

Thanks to anemone's comment, I went onfollowingto check the referenced paper (should've done that earlier, silly me!). The below image (from McClelland and Rumelhart 1985 paper) helped a lot. Looking at this image, I'd say:

  • The time course would be consecutive processing cycles
  • We can see there's a "growth of activation" (with time, that is, processing cycles)
  • at test = during testing as opposed to training the network
  • It's a "function of familiarity" because there are plots for familiar and unfamiliar data
  • As for "function of priming condition", there are some other plots for "primed" and "not primed" performance

It's still a mind-boggling sentence, but I think now I have a feel of what Ratcliff was referring to. Luckily this part of the paper is not my main focus point, so I can safely move on. enter image description here

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.