You could see what you were typing, as others have said, because the print head (which was cylindrical) rose up and rotated until the right character was in place, then slammed forward to smack the (simultaneously raised) tape and make a mark on the paper, and then dropped back down. The raise-spin-slam-unspin-and-lower sequence was rather noisy -- and not just the "slam" part.
On a distantly-related note, the data-transmission mechanism for directly-connected model 33s was the "20 mA current loop" -- a loop of wire that went from A to B and back. Machine A would send pulses into this loop by raising the voltage on one end until a current of 20mA flowed, hold it there for a short time, and then release. That was a "1" bit, I believe. A similar stretch of time with no current was a "0" bit.
The cool thing about the 20mA design was that it really didn't matter how long the wire was, or how resistive (up to a point!): if 20mA is flowing HERE, then 20mA must be flowing THERE, and be enough to activate a solenoid or whatever at the remote location. For long wires and higher speeds, things look less and less like switched-DC, and more and more like high-frequency analog signals, and things like inductance and capacitance make the whole plan less workable. But it sure was a cute first draft!