220 likes | 224 Views
This research explores how cognition prevents old items from interfering with the current item in a serial attention task. It investigates the role of functional decay and strengthening through use in preventing interference. The findings have implications for understanding within-run slowing, error increase, and switch costs in cognitive performance.
E N D
Functional decay in serial attention ACT-R Workshop August 1999 Erik M. Altmann (altmann@gmu.edu) Wayne D. Gray (gray@gmu.edu) hfac.gmu.edu
B more active than A Item A attended Item B attended When is decay functional? • When Item B is current Activation Time
Research question • How does cognition prevent old items from interfering with the current item? • Especially when there are many items • And when items are used while current • Use increases strength • The serial attention task...
Instruction Even Odd
Target 7 4 3 ...
Instruction High Low
Target 8 9 ... 2
High Low EO EO EO HL EO HL EO Potential interference Task structure Even Odd ... ... 8 7 4 Perception Cognition Memory Time
Analytical framework: Activation • What happens to an instruction’s activation over time? • Assumptions: • Episodic representation of instructions • One instruction per chunk • Retrieval increases instruction activation • Memory delivers the most active instruction
Assume 1 instruction retrieval per trial (every ~500 msec) New instruction less active Strengthening through use Activation
New instruction more active 1 use every 500 msec 1 use every 100 msec Decay through use Activation
Predictions of decay model • A performance decline as instruction decays • Within-run slowing • Within-run error increase • Encoding time • Encoding takes 100 msec per cycle • How many cycles needed to ensure decay?
Within-run slowing Response time (msec) Replicated in many conditions
Encoding time • Ho — Instruction is a detection trial • Takes less cognition than a classification trial • Expect 200 to 500 msec RT • H1 — Instruction is an encoding trial • Duration predicted by number of 100 msec encoding cycles required to achieve decay
Assume minimum activation at end of run How much encoding?
How much encoding? • Find the slope of the activation curve • R = number of retrievals (10) • E = number of encoding cycles (?) • Slope = d/dR [ln(2(R+E)/R-0.5)] • When slope is zero, E = R • N encodings ensure decay for N retrievals • Predicted encoding time: • 10 cycles x 100 msec per cycle = 1000 msec
E1 E2 Empirical encoding time Encoding time = E1 + E2 Response time (msec)
H1 RMSE: 6.5% H0 Empirical encoding times Response time (msec)
Implications • Model predicts repetition effect • One trial has to prime the next(Altmann & Gray, 1999a) • Optimized learning is true • Non-optimized learning discounts encoding • Don’t need PAS parameter • Permanent noise comes from transient noise during encoding
Conclusions • Decay is crucial – need to forget old items • Forgetting by deletion impossible in wetware • Forgetting by decay predicts within-run slowing • Decay demands up-front strengthening • Predicts switch costs (e.g., Rogers & Monsell, 1995) • An integrated theory of serial attention • Switching and maintenance functionally related
References Altmann & Gray (1999a). Preparing to forget: Memory and functional decay in serial attention. Manuscript submitted for publication. Altmann & Gray (1999b). Serial attention as strategic memory. Proc. Cog. Sci. 21. Altmann & Gray (1998). Pervasive episodic memory: Evidence from a control-of-attention paradigm. Proc. Cog. Sci. 20.