Caveat EmptorIntroductionClass # 3 (1/24/06)Embryonic developmentClass #4 (1/26/06)Neurobiological resultsIntroduction to McCulloch-Pitts NeuronsClass #5 (1/31/06)The McCulloch-Pitts Neuron and ``triangle nodes''Psycholinguistic evidence for a global constraint theoryClass #6 (2/2/06)Techniques for electrophysiological investigationsNeurological experimentsClass #7 (2/7/06)Class #8 (2/9/06)BackpropagationClass #9 (2/14/06)Artificial neural networks: wrap-upRecruitment learningClass #10 (2/16/06)The physiology of visionColor NamingClass #11 (2/21/06)Categories & ConceptsBasic-level categoriesInternal Structure of categoriesBrain StudiesClass #12 (2/23/06)Image SchemataCross-linguistic variationSpatial Image SchemataClass #13 (2/28/06)Class #14 (3/2/06)Regier model: denouementForce dynamicsMotor areas in the brainClass #15 (3/7/06)Class #16 (3/9/06)X-SchemataClass #17 (3/19/2006)Intermezzo: fitting models to dataClass #18 (3/16/06)The Bailey model resumedClass #19 (3/21/06)AspectX-schemata and aspectClass #20 (3/23/06)Class #21 (4/4/06)Metaphors as mapsBoroditsky's psycholinguistic experimentsClass #22 (4/6/06)A computational model for metaphor?Bayes NetsKARMA: A computational model for processing metaphoric sentencesClass #23 (4/11/06)Class NotesJ.G. MakinApril 11, 20061 Caveat EmptorIf a lion could talk, we could not understand him.Ludwig Wittgenstein, Philosophical InvestigationsMedium rotation,The shock of the new,And a memo from Feldman saying,‘Everything is true.’Dan Bejar, “The Sublimation Hour”2 IntroductionThese notes are intended as a supplement to the powerpoint slides, and not a replacement.There are many references to figures, none of which appear in this document, and mostof which are extremely helpful if not absolutely nec essary in understanding the material.So the reader is strongly encouraged to keep a copy of the slides at hand while readingthrough these notes; and to follow the external links, although they’re usually of lesserimportance. If there’s a conflict between the content of these notes and the content of theslides, you should go with the latter—and you should also send an e-mail saying as much [email protected]. At the present moment, these notes are not complete.3 Class # 3 (1/24/06)3.1 Embryonic developmentNervous-system cells develop from the “ectoderm,” one of the layers of the original embryo(the one from which skin cells also develop). There are two main types of these cells: neuronsand glial cells.1Differentiation:(a) Where the cell is born, with respect to chemical concentrations (e.g. SHH), determineswhat type they become: glial; motor and control; and interneurons, as the concentrationdecreases.(b) When the cell is born determines where it ends up (“inside-out” arichitecture). Inci-dentally, radiation appears to (adversely) affect the migration of neural cells.Axon guidance: Filopodia (little “feet” on the axon) follow the chemical gradients, bring-ing the axons to their proper targets. There are short-range as well as long range gradients.Chemical gradients can be repulsive as well as attractive.Connexions: Many axons connect to dendrites of “downstream” neurons. A winner-take-all mechanism prunes all but one axon. There can, however, be multiple growth cones fromthe single axon.Activity dependent tuning: Chemical gradients lead neurons to their proper connectingsites, but there is “over-connection” (130%). These are then “pruned” in accord with neuralactivity.4 Class #4 (1/26/06)4.1 Neurobiological resultsSperry’s exp eriment involved cutting the optic nerve in frogs and rotating the eye 180 degrees.The behavioral result: the frogs moved their heads away from food (flies). Conclusion: theoptic nerve re-wired itself so that each area of the eye was connected to the same area ofthe optic tectum that it was prior to the surgery. Of course, now that the eye was rotatedthis was maladaptive. This re-growth indicated that the optic nerve wiring was geneticallyhard-wired.On the other hand, Hubel and Wiesel’s experiments demonstrate the importance of earlypost-natal development. The LGN in the thalamus is composed of layers, each of whichreceives input from a single eye. These layers are in turn connected to layer IV of the visualcortex. Now, cells that are connected to similar layers in the LGN tend to get “stackedtogether” in the visual cortex, forming columns of cells which are dominated by a single eye.However, if one eye is deprived of stimulus for a long period, esp. during weeks 1-9, thenthe ocular dominance columns never develop, no matter how long the eye is open later.The fine-tuning need not be post-natal. In the womb, the infant moves its limbs, providingfeedback (proprioceptive, etc.), and visual feedback is provided by systematic patterns ofretinal stimulation. –Presumably this stimulation comes from the infant’s own brain, so it isnot a solution to the problem of limited genetic information: in some sense, this informationhad to be encoded, presumably in the genetic code. Rather, it is a solution to the problemof an infant’s needing to see very soon after its birth. Similarly, colts are born able to walk.Another example of pre-natal tuning is the auditory stimulation provided by the mother’svoice, which includes language information: newborns have been shown to be respond pref-erentially to the sounds of their native language (this presumably is true for white people2born in China as well as etc....)There is of course an information-theoretic issue (cf. the hw ass’t): our genes can onlyencode so much information; the rest of the constraints must be supplied by environment.That’s one reason why the stunting of the formation of ocular dominance columns shouldnot be surprising.Having stressed genetic pre-wiring and pre-natal/post-natal fine tuning, we should men-tion that the brain is nevertheless plastic in many ways later in life. Recovery from braininjury (esp. stroke) and even the elimination of phantom limb are examples of this phenom-enon. No one knows exactly how this happens, but it may be that redundant connectionswhich had previously been subjected to inhibition are now “released” to carry inf ormationin ways they didn’t before.4.2 Introduction to McCulloch-Pitts NeuronsWe now move on to some computational abstractions of the neuron, and the computationalmodels that are based on it. What features do we want to capture?• conductivity delays??•