What is Surprisal?
Surprisal measures how unexpected a word is given its context. It's calculated as -logâ‚‚(probability). High surprisal (red) means the word was unexpected; low surprisal (green) means it was predictable. This metric correlates with reading time, brain activity (N400 ERP), and comprehension difficulty.
Try these examples:
Simple sentence
Unexpected ending
Complex sentence
Technical text
Calculating surprisal values with GPT-2...
This may take 10-20 seconds
Results
Low surprisal (predictable)
Medium surprisal
High surprisal (unexpected)
--
Average Surprisal
--
Max Surprisal
--
Words Analyzed