Each factor captures a shared dynamic pattern across thousands of words. Select a factor to see its trajectory and the words that load most strongly onto it. Click any word chip to explore it in the word explorer.
Words positioned by their Factor 1 × Factor 2 loading — revealing clusters of shared linguistic dynamics, not semantic similarity. Colored by dominant regime. Click any word to explore it.
Using dynamics visible up to year 1995, the model predicts which words are about to surge over the next 5 years. Slide to any year and reveal what actually happened.
Cultural Shift Detector
Detects periods of cultural change purely from linguistic signals — without being told anything about history.
ShiftScore = z(LII) + z(changepoint density). High values = the language system is under stress.
ShiftScore timeline
Shaded regions = known historical events. Peaks show where our system detects cultural upheaval.
LII z-score
changepoint density z-score
Shift type classification
Each year classified by the pattern of its LII and changepoint signals.
Event detection results
Did the system detect each known historical event?
Feature importances (supervised model)
Detected cultural shift peaks
Years where ShiftScore > 1.5 (strong signal)