This document discusses stacked convolutional neural networks as a model for imagined speech in EEG data. It outlines several challenges in studying imagined speech, including the difficulty of obtaining clean EEG data without external stimuli interfering, and the brain's complex and diffuse representation of imagination. The document argues that progress requires a robust language model that accounts for the brain's organizational structure, and that understanding imagined speech could have applications like restoring communication for paralyzed patients.