Andy, do you understand duty cycle? all PWM does is change the duty cycle of a square wave, and typically the duty cycle will be made to be proportional to an input voltage. the stuff others are talking about with triangle waves and comparators are convenient methods that you can use to GENERATE a PWM signal. there are no sine waves involved...
you need to stop thinking about it in terms of communications, because that's not the only meaning of modulation; in fact you'll probably understand it much better if you totally forget about what you learned in communications for a minute. In communications, yes, you are combining a couple of sine waves in some way or another to produce the modulation on a carrier wave which can be more easily transmitted, which can then be decoded. in PWM, the signal of interest is the PWM signal itself, because the object is generally to use it to drive things like LEDs and motors, and the "input signal" so to speak, is only important as a control signal for producing this output; you generally don't use it as a communication scheme where the input signal is decoded at some output.
with PWM driving an LED, for example, 0% duty cycle leaves the LED all the way off, and 100% leaves it fully on, and duty cycles in between 0% and 100% vary the brightness from full off to full on. the reason it's used instead of simply driving the same device with a varying analog voltage, is that it avoids many of the problems such as non-linearity of the devices, and in the case of motors, gives improved torque, and it lets you use a digital output driver (such as an inverter/buffer, or some transistors) for high-power output, instead of trying to make an analog amplifier that can handle the power.