Is there some way a human can be said to propriocept things that they neither saw nor touched?
Proprioception being an abstraction over touch and sight data that generates awareness of how your body is arranged in space is a sense in the same manner. And you can definitely get proprioception over more than your own body - partner dancers will quickly develop a felt sense of how their dance partner's body is moving, as well.
Suppose you were to anesthetize all the nerves leading to your skin. If you define touch in a way that's limited to skin interacting with stuff outside the body, then that would neutralize your sense of touch.
But according to the article, there are "piezo2" receptors embedded in muscle fibers. They can tell you how much the muscle is stretched or contracted. So even without using touch or sight, you could tell whether (say) your elbow is bent or straight based on information coming from your biceps and triceps muscles.
One could imagine a body that doesn't have this and works by some combination of dead reckoning (where the brain knows it pulled on the muscles this much, therefore the limb must have moved that much, and keeps a running total), sight, and cues from how the limbs are interacting with the outside world. But the article says it isn't just that. Instead, the brain is getting direct information from muscles.
You could, though, make a case that touch should be generalized and this should be included in it. The article says piezo2 receptors are used both in touch and in muscles. If you wanted, you could define touch to include sensations within the body, where this part of this muscle is touching this other part of the same muscle and feeling that.
Imagine a robot arm. You have the armature and a beefy electric motor on it. Your task is to make it do some manipulation task. If all you can do is send voltage to the electric motor you will never succeed. You don't know where the arm is at any moment. But even if you would know where it started from you couldn't know how much it moved exactly just by knowing what voltage you sent to it. This is because minute variation in loading and tinny errors would accumulate into huge uncertainty in such an open-loop system. So basically you need something to close the loop. A rotational encoder on your robot arm, or muscle strain measurements on a living creature.
But just having the sensors is not enough. If you want to make the (robot or human) to touch a particular point in space you need to know how to turn the sensor readings into a transformation. I would be lying if I would tell you I know how exactly the human brain does this, but in robotics it's just a few transform matrixes multiplied together.
So it’s more like a combination of sight and touch and forward modeling based on muscle input.