Development of relational cognition

Development of relational cognition

1. human thinking qualitively diff from animal. Difference= we make inductive inferences orders of magnitude more complex than those of other species

   • but non human see like we do, remember like we do, and even learn simple languages

2. Human inductive inferences: weird situation but valid inference. how? size, relationship between size and hiding behind. Can still use inductive inference

3. human vs non-human induction

   • match to sample task: honeybees can do this. What do you need to represent in order to do this task?: same-colour (x,y)-untrue doesn't need to know why it's choosing. nope just blue. Realtional match to simple task: choosing B requires you to represent same-colour as an explict relation. Complex, animals can't do it. wait, pigeons can. Entropy is a measure of disorder, lots of disorder in diff pics. babies and pigeons are sensitive to this. But still plently of problems humans solve routinely that animals can't.
   • Follow the leader task: train animal to mimic trainer, cant train animal ball is too big but basket fits, however four year olds can do this task
     • Hiding task Ravens matrices: hard problem definitely off limits for animals
   • why can't animals do this task? hint from another problem
     • Call and tomasello (1998) TOM task: a reciever and a signaler box switch. Chimps learn quickly good at it.  if boxes get switched not by signaller will reciever know they have info signaler does not? chimps can't but fiver year olds can. Requires TOM: ability to attribute mental states to others and to appreciate that others have mental states different from ones own
     • ToM = understanding relation between person and mental state. Requires: an appreciation that thought and the think are independent, the ability to bind thoughts to individuals dynamically
       • seems processing a relations is what humans do that animals can't
       • About a person and a belief using the roles having or not having to drive inference

4. Relation is a structure that takes some arguments and then tells you something about about those arguments without changing any other aspects of those arguments or being changed by the arguments

   • A relation: understanding between animals of species and how they care for their young, not colour or fur eg extraneous features, attends soles to relations! Realtional roles not just features
     • Many inferences made to be able to make relations based on roles. as a result can permit generalisation to any new objects that fill same roles
   • Use relations to make an analogy between those two situations
     • Analogy: the process of finding correspondences between one thing or situation and another. Can be based on shared relations with no shared object similarities
       • finding analogies between domains allows us to make inferences about one of the corresponding things of situations based on info about the other

5. Component process of analogy. During retrieval a target activates potential sources (driven by surface similarity)

   • Memory retrieval: Novel target problem retrieve potentially applicable sources(s) from memory Mapping: Discover and represent the structural (relational) corresponences between elements of target and source Analogical Inference: source drives inferences about target Schema Induction: Induce a more general schema that retains what source and target have in common
   • Generalization occurs via schema induction, abstract a general schema that retains what the two situations have in common
     • use info to make inferences about something else similar. Realtionally relevant contexts. Can use schema retrieval to solve future problems, generalise these schemas. (schema as a genral knowledge structure used to categorize future situations and allow you to generalise info about them) 

6. Realtional thinking: power a function of the representations of relations, only humans do this with flexiblity, represent in a certain way

   • Requirements: cat doing the same thing looks the same but in two situatuions chased and chaser. To understand this you need consistent system that are coding for what it means to be a chaser independent of what its chasing. Its unaffected by other features and feature similarity. 'chaser' as some entity in own right that can be mapped across those situations. chase means the same thing no matter its chasing (independent rep of roles and fillers) what it means to be a chaser applies to one item in one instances and a diff item in a diff instances. Bindings can flip: change who chases who, same elements involved but didd correspondences. Break old ones make new ones = dynamic bonding .and chased flips.  binding tag (is dynamic-creats and destroys on the fly) is explicit-specifies what is bound to what, doesnt change the individual meaning of the elements ie cat still cat, it only links them
   • In the limit allows universal generalisation-or generalisation over completely variablaised argumnts. Gen to any possible input. Any possible X. Can extend this rep

7. meaning is a function of: the stated relations and their arrangement (ie role filler bindings). relational representations are structured.

8. Relations are hard to represent (but very powerful thinking). Other problem is they are hard to learn

   • Why? relations not properties of single objects always tied to specific examples, but we learn representations that are independent of examples ie. same shape relation- truthvalue is completely uncorrelated with any specific features of any specific shapes. BUT we always see the same-shape relation in the context of two specific shapes ie we never see disembodied 'same shapeness'
   • Are we born with relational thought?
     • Goswami: children are born making analogies. Dont make judgement based on overall similaries appreciate relations and analgies are the way to go. Casual relations fundamental to understanding. part of cognitive dev, born with it.
     • Smith (1984) not inborn but leaned through development
       • Holistic (identical object) matching, property matching, relation matching. Chhilds task is to choose two objects (from 3) that match by the same rule. Kids appreciation of similarity starts off holistic, becomes able to focus on individual object properties and eventually becomes able to focus on relations. 
       • The relational shift: Genter 1988, 2003: children initially represent and reso about the world in terms of object features gradually turn to relational concepts. Can ignore featural similarity and use relational infomation to drive inference
         • 3 main theories to why this happens
           • 1. Increased relational ability is the result of learning more and more complicated relations (Genter) 2.The result of increased processing capacity (Halford) 3.Result of increased inhibitory control which looks like increased processing (Richland et al)
         • Kotovsky & Gentner (1996)
           • If relational thinking develops via learning relational concepts then we should see increased relational processing in very young children with the right training (more experience = more relational) "progressive alignment" (systematic ordering of examples) helps kids acquire relational concepts, even without feedback The task: Relational match to sample. Starts easy within-dimesion match :size maps size, then even easier features are simialr too. Another rleation: symmetry. Harder cross-dimesion match increase in size = increase in darkness the 4-6 years were at chance when trials were presented i n random order, relations too complex. BUT when tirals in systematic ordereasy (within) then harder (cross) they did much better and success on first 8 within predicted success on second 8 cross
           • conclusion: strong evidence for learning, refute theory relational learning is inborn
             • Progessive alignment 1. gave kids the chance to solve early problems featurally 2. gave kids the chance to compare later problems to earlier (solved) problems, bootstrapping the solution to the harder problems thereby 3.helped kids to: acquire the relational concept and thereby apply it to cross-dimensional trials. experience drive development of relational thought ie learning
         • Halford: analogy develops as changes in WM capacity affect our ability to make good analogies (less wm =worse analogical ability)
           • v young kids no matter how you train them won't solve problems. not neurologically developed enough
         • Richland, Morrison & Holyoak (2006)
           • Childrens ability to solve analogies require the capacity to intergrate multiple relations and ignore featural/semantic distractors develops between ages 3-13  Ability yo have inhibitions responsible for limited relational performance in children. trials where you need to do two things (harder to relate) The task: a picture analogy task. view 2 pics asked which object in the target pic goes with a designated object in source pic Relational match (correct) the boy like the cat is chasing something. featural is matching cat to cat. other non match (one of many) 2 relations is harder and semantic distractor is harder than none 3-4 years highly susceptible to distractions when intergrating relations is hard around 30% drop in relational responses
           • Conculsions: 1. reduced inhibitory control on younger children A) makes it harder to ignore a semantic distractor b) makes it harder to intergrate multiple relations(while you are thinking about one inhibit the other) 2. development of realtional knowledge in older kids likely plays a role as well
   • problem: Really learning structure from an unstructured world is hard
     • There are domains where it looks as if associations not sufficient for children to concieve world, must be relational learning. But lack of explanation for how this happens is detrimental to structural view of cognition
     • Extreme views tend to be wrong, little bit of both. Innate aspects susceptible to learning, end up with a bit of both.
       • Gelmans position: how does learning occurs? characterise one state from another, need to find mechanism. Can't find it. either its not found or mis characterised in the second state. Linda says its all association just features of objects drives observations, shes say relational stuff small mark of mechanism. It seems comlex but not. No innateness. Basically shes not answering Q. Needs to provide mechanism.
     • Not told us how to solve this problem: puts a damper on theories of relational cognition (McClelland)
     • where does this structure come from is is worth thinking about?