Sentiment Analysis
------------------
Original: ['why', 'make', 'a', 'documentary', 'about', 'these', 'marginal', 'historical', 'figures', '?']
Base reduction: ['documentary', 'marginal', 'figures']
Combined reduction: ['why', 'a']
Stop words: {'why', 'a'}

Original: ['nothing', 'is', 'sacred', 'in', 'this', 'gut', '-', 'buster', '.']
Base reduction: ['nothing', 'is', 'sacred', 'buster']
Combined reduction: ['nothing', 'is', 'in']
Stop words: {'in', 'is'}

why make a documentary about these marginal historical figures?

Gradient Baseline: [0.02353, 0.34230, 0.02116, 0.04192, 0.11603, 0.10397, 0.00020, 0.12693, 0.01987, 0.03363, 0.07968, 0.09080]
Gradient Merged (First Token): [0.01066, 0.70125, 0.00809, 0.01908, 0.05369, 0.04803, 0.00016, 0.05715, 0.00958, 0.01500, 0.03585, 0.04144]
Gradient Merged (Stop Token): [0.00232, 0.04769, 0.00280, 0.86601, 0.01532, 0.01522, 0.00006, 0.01761, 0.00318, 0.00520, 0.01198, 0.01261]
