Salmon tool: I don't understand the principle

Hello,
I am trying to understand how the Salmon tool works (for doing transcripts quantification in RNAseq). I have read the Salmon page, the article of its creation and also the article about the work process of Minimap2… but I cannot understand a point. In fact, I do not understand how the software can decide if a fragment transcript can be associated to a specific position on the original transcript. For being more clear, if I have a fragment that can map on various splicing variants of the same transcript, how Salmon decides to which one is it associated? I don’t need a mathematical/ theoretical super answer, just a simple one + possibly the reference where this is explained…
I thank you very much for your help.
Have a nice day
Marta