Failure-Driven Reminding for Incremental Learning

Much of adul t learn ing is gradual , almost impercept ib le. Our model for t h i s knowledge-based, incremental learning is to augment normal story comprehension processing w i th a f a i l u r e t rack ing mechanism. When a comprehension ru le f a i l s , the f a i l u r e and i t s cor rec t ion are stored in an exception episode attached to the f a i l i n g r u l e . The ru le is otherwise unchanged. Subsequent f a i l u res of that ru le t r i gge r the r e t r i e v a l of these exception episodes (failure-driven reminding). Rule mod i f i ca t ion occurs when classes can be found for the known exceptions. The ALFRED program is a pre l iminary implementation that c l a s s i f i e s and remembers f a i l u r e s of "everyday knowledge" in the domain of p o l i t i c a l economics.