Cisco is being a bit literal. 10/100 "T" (twisted pair) does have independent TX and RX conductors. In a point-to-point situation, the link is fundamentally full-duplex. However, when a hub is involved, everyone's TX is connected to everyone else's RX. As a result, it's impossible for more than one node to transmit at a time, thus _half-duplex_ , but it's the same cable. A switch returns us to the p-t-p model; a node isn't connected to any other nodes, just the switch, and the switch has buffers to store frames so they don't collide.
In the end, it's a combination of things. The cable is part of the equation, but does not inherently make something full-duplex. It can rule it out, though. 10base-2, for example, is literally a shared wire, so full-duplex cannot be done here. Two nodes _could_ , in theory, talk at the same time and make sense of _each other_ \-- using echo cancellation -- but all others on the line will hear nonsense.