What legacy system could possibly prevent one from simply recording the position of every train every few seconds? This is not a particularly difficult problem for a billion-dollar business to solve.
The control systems of the MTA are purely electromechanical. It goes way beyond what one would normally call a "legacy system".
You can't record the position of trains because there is no way to snapshot the whole system. Only the signal tower local to that section of the system is aware of what's going on in its zone. This is a vestige of the design where there was no reasonable way to centralize that amount of signaling over such great distances.
To get a snapshot of the whole system you'd have to simultaneously snapshot the state of every signal tower in the system and find some way to collect this information centrally. This sounds reasonable until you realize that the control system, being 50 years pre-computer, is unaware of individual trains. It doesn't detect trains, only presence of at least one car on a section of track (detected by the third rail making contact with the train).
To ascertain which train is where you'd have to either have dispatchers identify the train by communicating with operators (hard to poll), or infer the identity of a train by monitoring state. Both are quite difficult when all of your data is spread all over the entirety of NYC in isolated distributed control centers.
The IRT trains spent billions centralizing all command and control into a single building in Manhattan, and thus snapshotting state is trivially easy. The BMT/IND segments of the system are substantially bigger, with more stations, than even that. A full modernization to electronic control is necessary, which will cost many billions.
In cases where delays are more substantial and you need to hold trains further upstream, this is communicated manually between towers.
Much of the control here is not based on semantic data, but rather verbal. The actual control system itself isn't capable of much more than indicating which piece of track is occupied.
It is fairly easy to hook up a "'low resistance between the rails aka 'wheels on the rails' aka 'there is a train here'" detector to a 'do not enter' signal a few hundred meters up, and to a 'drive slowly' signal some distance before that (http://en.wikipedia.org/wiki/Automatic_Block_Signal)
I mean, are you telling me for a 100 million dollars this couldn't be done?