Talk:Psycholinguistics/Connectionist Models

From Wikiversity
Jump to navigation Jump to search

Here are some comments on your chapter (Note: still in progress):

  • it needs proofreading. Lots of typos and grammatical errors. You are producing "live" content so you should read it over right away and clean it up
  • The chapter is very disorganized. I suggest starting with a general overview of connectionist modelling, then general properties of connectionist models, then a few specific examples.
  • for the most part, keep it about connectionist models and the general philosophy of connectionism, rather than connectionists (people who do connectionist modelling and/or subscribe to the tenets of the philosophy)
  • "Connectionism as a cognitive theory" provides a good overview of this topic. Given that this is a chapter in a Psycholinguistics texbook/course, it would be good to make this section a little more language-focused. You do this nicely in the last paragraph but this could be expanded on.
  • please be attentive to your choice of wording. For example, in the last paragraph of "Connectionism as a cognitive theory" you write "...simple genetic rules..." but you are referring to learning rules, not genetic ones. Do you mean that simple learning rules are genetically programmed? Similarly I don't know what "bioprogrammed logically programmed areas of the brain" means.
  • the core concepts of the simple learning rules used by connectionist networks should be clearly described prior to launching into examples such as a network for word learning. At the very least, Hebbian learning and backpropogation should be described. BTW, at the very start of the chapter you state taht weights are adjusted by backpropogation but this is not actually a universal feature of connectionist models.
  • this chapter would benefit a lot from illustrative images
  • "Layers in a model for word recognition" should start with a general description of the purpose of the model, before going into the details of individual layers
  • rather than starting with the model of word recognition, start with a more general description of the architecture of connectionist models (input, hidden, and output layers)
  • when you do start to describe a model of word recognition, pick one model, make it clear which one you're talking about, and describe it. Your description gets a bit confusing in that you mention aspects of different models. Better to fully describe one and then in a subsequent section describe how other models have improved on that one
  • avoid statements like "of course" -things are not necessarily obvious to learners.
  • Tabula Rasa should not be capitalized, but since it's Latin it should be italicized
  • You should elaborate on the Roe et al paper you cite - you should provide enough information that the reader can understand your chapter without going to the papers you cite.
  • The localization discussion seems very out of place. Since most connectionist models (certainly those for language) are so far removed from real neurons or "wet" neural networks, they are not suited for addressing the question of localization vs. distributed representation. There are numerous other topics in Psycholinguistics that are much more relevant, such as development (including word learning and inflectional morphology), speech production, reading, and aphasia

AaronJNewman 21:47, 7 March 2011 (UTC)Reply[reply]