Cognitive Science
Introduction
Overview
• Aims and learning outcomes• Assessment• Programme
• Cognitive science is interdisciplinary• Cognitive science uses formal models
Beware• This strategy might not succeed (!)• Fashion can influence the perception of research
Aims
• To introduce interdisciplinary approaches to the study of higher cognitive processes
• To familiarise you with computational and other formal modelling
• To illustrate the application of modelling to cognitive processes
Learning outcomes
• To gain direct experience of computational and other formal modelling techniques.
• To integrate material across areas within psychology and across traditional subject disciplines.
• To compare and critically evaluate formal techniques in relation to empirical findings.
• To tackle key theoretical problems in cognitive science, particularly problems linked by the theme of common sense reasoning.
Assessment
• Two hour examination in June, which counts for two thirds of the mark.
• Three pieces of coursework (counting for 4%, 4%, and 25% respectively of the course mark)
• Coursework assesses the first and, to a lesser degree, the third learning objectives.
• The exam will assess learning objectives two, three and four.
Coursework
AW 1 - Connectionist modelling 1 (4%)
AW 2 - Connectionist modelling 2 (4%)
Modelling project (25%)
Programme
1 Introduction - why cognitive science?2 Cognitive modelling3 Cognitive modelling4 Cognitive modelling5 Cognitive modelling6 The development of concepts7 Learning word meanings8 Ambiguous words9 Compositionality and word meaning10 Common-sense reasoning
Cognitive Modelling
Project – construct a model of adjective-noun combination
red apple
fake gun
heavy baby / heavy elephant
Cognitive Modelling
Heavy baby
Heavy Baby
Learns by training over and over
Distributed
3
.2 .3 .7 .2 .4 .6
.2 .3 .5 .8 .4 .6
.2 .3 .6 .4 .2 .2
Distributed
3
.7 .3 .4 .6
.5 .9 .4 .6
.6 .5 .2 .2
NODES:
nodes = 4
inputs = 6
outputs = 2
output nodes are 1-4
CONNECTIONS:
1-4 from i1 – i6
<weight feature>
The development of concepts
What do we mean concept?
Why is concept learning tricky to understand?
Connectionist nets as a simple model of concept learning
Some features of natural concept learning that make the picture less simplee.g. Role of existing background knowledge
Gavagai
Learning word meanings
Ambiguity and vagueness
Complex links between words and concepts
Bank
Newspaper
To paint
Combining concepts
Compositionality is key to language
red apple, red brick, red mist
Watergate, blood gate, Stargate
Commonsense reasoning
Which information is relevant to drawing a conclusion?
Which facts are affected by an event?
• Yale shooting problem
• Property inheritance
Tweety is a bird. So, Tweety can fly?
A little history – the Cognitive Revolution
Skinner (1957) Children learn words (language) through operant
conditioning - stimulus controls response
Chomsky's (1959) review of Verbal Behavior(link on course web pages)
"Dutch" - what stimulus? proliferate "stimuli”
but role of attention etc. mind'Creativity' of language compositionality
Technical concepts of Skinner's behaviorism (stimulus, reinforcement, operant etc.) were used non-technically in "Verbal Behavior“
Eg. the artist is reinforced by the effects his work may have on others
… but the artist's (often) not there when these effects occur. It's not like reinforcement in a Skinner box.
"I now believe that mind is something more than a four letter Anglo-Saxon word - human minds exist and it is our job as psychologists to study them."
Miller (1962) in American Psychologist, 17, p. 761
Nb Piaget, even Freud, were always cognitively oriented
Chomsky (1957; 1965)Transformational Generative Grammar
Account for syntactic facts (linguistics)
e.g. active and passive have same meaning
Judge facts using 'intuitions' (psychology)
the resulting grammars are related to something people know
(linguistic competence)
A small transformational generative grammar S NP, VPNP determiner, nounVP verb, NPdeterminer: {the, a} noun: {boy, dog} verb: {eat, kick, bite, occur}
Passive transformation (simplified):NP1, V, NP2 NP2, BE, V, EN, by, NP1
Captures the fact that selection restrictions match
Congress impeaches Clinton Charlie impeaches a shoeClinton is impeached by Congress A shoe is impeached…
Congress impeaches Clinton
NP1 V NP2
Rule
NP1, V, NP2 NP2, BE, V, EN, by, NP1
Clinton is impeached by Congress
More history – early machine translation
Weaver (1949) memorandum
Georgetown (1954-66)250 words & 6 rules at start
Alpac Commission (1966)speed? cost? quality?
Meteo (1977)English FrenchUse existing materials (style sheets)Translators involved
Fashion and the life cycle of (some) AI projects
Oblivion, fading Rebirth Excitement Claims More excitement Wild claims Unmet expectations
Fading, oblivion.
Cognitive science now
"higher" cognitive functions; processes & representations
InterdisciplinaryPsychology, linguistics, philosophy, computer science, brain sciences, anthropology, ….
Use formal / explicit modelsComputational metaphor
strong v. weak
The original question "Can machines think?“ I believe to be too meaningless to deserve discussion.
Alan Turing
www.warwick.ac.uk/~psrex/cogsci.html
The end