Introduction to Cognitive Science
Q270: Experiments and Models in Cognitive Science Lecture 2
What is Cognitive Science? •
Not that new…at least 30 years old now
•
“Scientific study of intelligence and intelligent systems” (Goldstone)
•
“Scientific study of the mind” (Thagard)
•
“Cognitive science is concerned with understanding the processes that the brain uses to accomplish complex tasks including perceiving, learning, remembering, thinking, predicting, inference, problem solving, decision making, planning, and moving around the environment. The goal of a cognitive model is to scientifically explain one or more of these basic cognitive processes, or explain how these processes interact.” (Busemeyer)
What is Cognitive Science? •
Hardware/Software analogy If neuroscience studies hardware (wetware), cognitive science studies software Can’t touch software, must operate using hardware constraints Our software comes prepackaged by nature/learning, so we have to reverse engineer to determine analytically how it works
•
How? Experimentation and model building/testing
•
For example: Homer cognitive simulator [Homer.exe]
Interdisciplinary Nature
We borrow approaches from many areas to study the mind/intelligence
•
Subfields of CogSci: neuroscience, psychology, artificial intelligence, linguistics, philosophy, anthropology, physics, biology, education, economics
•
Focus on: scientific method of induction, as well as simulation/modeling of theory…compare to human behavior
Marr/Poggio’s Levels of Analysis •
Computational Level (Behavior): Describes the directly observable output (behavior) of a system
•
Algorithmic Level (Functional): Describes how input is processed to produce the behavioral output
•
Hardware Implementation Level (Physical): Describes the physical hardware/wetware of which the system is composed
E.g., Free recall? Cognitive science aims at multiple levels of analysis to fully understand the system
magazine keyboard fountain camera ******** giraffe gravel paper angel door +
Methods of Inference •
Behavioral Experiments: Reaction time, accuracy, psychophysical judgments, eye tracking, dynamics
•
Computational Modeling: symbolic systems, connectionist (artificial neural) networks, matrix modeling, production systems
•
Brain Measurement: Non-invasive scanning: PET, EEG, fMRI, TCM; Invasive techniques: single-cell recording, animal models, electrode stimulation
Typical Research Areas •
Attention: How do humans filter information and allocate processing resources? Dichotic listening, exogenous/endogenous cueing, executive control, overt/covert attention
•
Memory: How are memories formed, organized, retained, and retrieved? Short-term/working, episodic, semantic, serial (temporal), procedural, recognition/recall, retroactive/proactive interference
Typical Research Areas •
Perception: How do we process and interpret information from the sensory systems? (vision, audition, tacticion, ollafaction, gustation) Object/pattern recognition, visual search, motion perception, navigation, word perception, face processing
•
Learning: How do we use information in our environment to adapt behavior? Language acquisition, error-driven learning, dynamic systems, categorization, identification, associative learning, induction
Typical Research Areas •
Language Processing: How do we process and represent written and spoken (or signed) languages to interpret as meaningful content? Sentence processing, word recognition, semantics, syntax, pragmatics, orthography/phonology/phonetics, discourse processing
•
Decision Making: How do we integrate information and consider consequences to choose between alternatives? Heuristics, speeded decision-making, uncertainty, Tversky and Kahneman, utility theory, risk, judgments
Human vs. Machine intelligence •
Why study humans? AI in 1960s vs. today: babies vs. supercomputers/robots Kasparov vs. Deep Blue
•
Learning: All info vs. adaptive heuristics Generalizability / induction / insight Humans don’t require negative instances to learn
•
Memory: Distributed vs. localized (Lashley’s engram) Reconstructive vs. veridical Flexible to adaptation; emergence/swarm intelligence
•
Decision making: Tree search vs. resonance and wisdom Lenat/Kurtzweil: speed of neural processing But…we can be easily fooled…availability heuristic (k)
Reconstructive Memory:
Loftus and Palmer (1974) “How fast were the cars going when they ____ each other?” (hit, bumped, smashed)
How do we categorize perceptual stimuli? (Lets use novel stimuli so we have control)
Tasio Galli Galli
Internal Representation
Radok Tasio
Radok Samar
Internal Representation
“Samar”
Prototype model vs. exemplar model; ordinal/magnitude diffs
Typicality Effects Rosch (1973, 1975): Typical members of a semantic category can be processed more efficiently than atypical ones
1.30 1200
1.25
RT (ms)
1150
0.62
Euclidean Distance
Rosch Prototype Label
0.64
0.60 1.20
1100
0.58 0.56
1050
1.15 0.54
1000
1.10 High
Med
0.52
Low
Typicality Jones & Kintsch (2006)
Schema Abstraction Task Posner & Keele (1968)
Prototype
Low-level Distortion
High-level Distortion
Exemplars are all distortions of their category prototype Subjects learn to classify exemplars into A, B, C categories, but never see the prototype (manipulate # exemps per category) After learning, subjects classify the protoype better than other exemplars, and better than exemplars used in training
Schema Abstraction Findings: 1) Old exemplars are a bit better than the prototype on an immediate test, but classification for the prototype tends to be much better than the old exemplars with a greater delay 2) Old exemplars are classified better than new exemplars 3) New low-level distortions are classified better than high-level distortions on both immediate and delayed tests 4) Classification of new exemplars is better for large categories (categories w/ more exemplars) than small categories 5) Random patterns are more likely to be assigned to a large category than to a small category This (especially #1) has been taken as evidence that we create a central “abstract” representation rather than storing exemplars (think of how you learned about dogs or birds)
For Category B (6 exemplars) Interacts w/ category size
Strength and Memory Retrieval
LETTER +
MUFFLER +
Frequency? # of senses? Contextual diversity?
…the letter that Microsoft sent to Yahoo!’s Board of Directors... …an open letter to environmentalists… … the discovery of the 27th letter of the English Alphabet …
If you've ever heard a car engine running without a muffler... … had a new muffler and pipe installed on my old car... …took the car in for repair and found out the muffler bracket is shot…
Strength and Memory Retrieval Strength Theories: Principle of Repetition: Repeating a word strengthens its trace in memory, or adds additional traces…the effect is due to frequency Rational Theories: Principle of Likely Need: A word that has been seen in more contexts strengthens its representation b/c it is more likely to be needed in an unknown future context…the effect is due to contextual diversity Think of babies learning about words or objects Frequency and diversity are confounded. What source is actually used by humans when forming memories?
“Imagine that you are an explorer charged with the task of learning the alien language of Xaelon. You will see several Xaelon sentences. Each sentence will be paired with a scene that is described by the sentence…” Recchia, G., Johns, B. T., & Jones, M. N. (2008). Context repetition benefits are dependent on context redundancy. In V. Sloutsky, B. Love, and K. McRae (Eds.) Proceedings of the 30th Cognitive Science Society Meeting, 267-272.
flurn twops plorm
shech bloup thrig
shech twops thrig
leuts bloup dralp
Experiment 2
• Xaelon contains • four words for each of four blue Fribbles (from Michael Tarr’s stimuli repository, 2006)
• four words for each of four possible locatives (up, down, left, right) • four words for each of four possible gray background objects
SUBJECT
LOCATIVE
OBJECT
leuts bloup dralp SUBJECT
OBJECT
Experiment 2
• Participants view 450 such images, broken into 10 blocks • Afterwards, a surprise pseudo-lexical decision task: participants distinguish Xaelon “words” from non-Xaelon foils as quickly and accurately as possible
Experiment 2
• Allows us to manipulate variables of interest: • High-WF words occurred 180 times during training; low-WF words occurred only 45 times • High-SD words occurred in all eight possible semantic contexts; low-SD words occurred in only one • Maps directly onto variables of Study 1
Experiment 2
Discussion
• Both our studies point to the same conclusion: repetition is most beneficial when it occurs in distinct contexts • Consistent with distributional accounts of child language learning (Mintz, Newport, & Bever, 2002) • Corroborates evidence that contextual diversity has fundamental effects on memory access (Adelman et al., 2006; McDonald & Shillcock, 2001; Pexman et al., 2008)
An example of a large-scale cognitive model BEAGLE: • Jones, M.N., & Mewhort, D.J.K. (2007). Representing word meaning and order information in a composite holographic lexicon. Psychological Review, 114, 1-37. • Jones, M.N., Kintsch, W., & Mewhort, D.J.K. (2006). High-dimensional semantic space accounts of priming. Journal of Memory and Language, 55, 534552.
“Pitcher”
Context + Order from Wikipedia Baseball is a sport played between two teams usually of nine players each. It is a bat-and-ball game in which a pitcher throws (pitches) a hard, fist-sized, leather-covered ball toward a batter on the opposing team. The batter attempts to hit the baseball… Beer is the world's oldest and most popular alcoholic beverage. It is produced by the fermentation of sugars derived from starch-based material — the most common being malted barley; however, wheat, corn, and rice are also widely used, usually in conjunction with the barley. Less widely used starch sources… An automobile (from Greek auto, self and Latin mobile moving, a vehicle that moves itself rather than being moved by another vehicle or animal) or motor car (usually shortened to just car) is a wheeled passenger vehicle that carries its own motor. Most definitions of the term specify that automobiles are designed to run primarily on roads… A book is a set or collection of written, printed, illustrated, or blank sheets, made of paper, parchment, or other material, usually fastened together to hinge at one side. A single sheet within a book is called a leaf, and each side of a sheet is called a page. A book produced in electronic format is known as an e-book.
1
.
4
1
.
3
1
.
2
1
.
1
1
.
0
.
9
.
8
.
7
.
6
.
5
.
4
.
3
.
2
.
1
.
0
astronomy physics chemistry psycholog biologyy scientific mathematics technolog y scientists science
financial
research
saving s finance pay invested lend loaned borrow invest bank investments spend save
-
.
1
-
.
2
-
.
3
-
.
4
-
.
5
-
.
6
-
.
7
-
.
8
-
.
9
sports -
1
.
0
-
1
.
1
-
1
.
2
-
1
.
3
-
1
.
4
-
1
.
5
team teams football coach sport players baseball soccer tennis basketball -
1
.
7
-
1
.
6
-
1
.
5
-
1
.
4
-
1
.
3
-
1
.
2
-
1
.
1
-
1
.
0
-
.
9
-
.
8
-
.
7
-
.
6
-
.
5
-
.
4
-
.
3
-
.
2
-
.
1
.
0
.
1
.
2
.
3
.
4
.
5
.
6
.
7
.
8
.
9
1
.
0
1
.
1
1
.
2
1
.
3
Neighbors in context space
Animals
2.0
1.5
purple
france tuna canada
1.0
.5
blue
catfish
0.0
red
green -.5
swordfish australia
germany yellow pickerel
-1.0
mexico -1.5 -2
-1
0
1
2
3
1.5
green red blue yellow purple 1.0
.5
france canada australia germany
0.0
mexico -.5
pickerel tuna
-1.0
swordfish mackerel catfish
-1.5
-2.0 -2.0
-1.5
-1.0
-.5
0.0
.5
1.0
1.5
2.0
Perceivability Hypothesis Nouns are learned faster than verbs (Fleischman & Roy, 2005; Gentner, 1982; Snedeker & Gleitman, 2004) Dominant hypothesis is perceivability Gillette, et al. 1999; Gleitman, 2004 What is the statistical behavior of these types?
Jones & Kintsch (2005)
Learning Lexical Classes
Our Projects Will Contain: •
Area (memory, percepion, etc.)
•
Background on what has been done, what is known, and current theories/models
•
Prediction: Hypothesis for one or more of the competing theories
•
Experimentation: manipulation, control, causal link between manipulated variable (input) and change in behavior
Our first study: A shark researcher believes that he has created an effective shark deterrent. A underwater beacon gives off a sonic pulse that humans can’t hear, but that drives sharks away. To test his device, he places one at a beach in Boston. Over the next two months, he records the number of shark attacks at his beach. He then sends his assistant to a beach in Miami to counts the number of shark attacks there for two months. There were only 12 shark attacks on his Boston beach that had the beacon, compared with 38 attacks at the Miami beach. The researcher concludes that his device is effective, and starts to market it. Would you feel safer at a beach with his beacon? What other alternative explanations can you come up with to account for this result (fewer shark attacks at the test beach) ?
Alternative Explanations for Shark Attack Differences: 1. There were more shark attacks in Miami than Boston because there are more people in the water in Miami than Boston 2. There were more shark attacks in Miami than Boston because there are more fishing trawlers in Miami, which draws the sharks closer to shore. 3. There were more shark attacks in Miami than Boston because….