The only thing we really
know about phrase structure is that, just like everything on the analytical side of music, any discussion about it will eventually turn into a fight!
This one's just simmering so far though.
Singing the lines of what you play helps get a good sense of the phrasing. Listen for the places where, when you take a breath, the musical line makes the most sense. Mark the print music with a comma at the breath points. You might discover later that you were hearing phrase sections rather than full phrases, but it's still an important step in developing a clear sense of what's really happening in the musical line. 4 and 8 bar lengths are common, but this varies a lot. This approach doesn't work with all music but it does with most, IME, even if the only singable line is a bass line below a torrent of arpeggios. The formal/traditional approach to phrase structure defines phrases as ending in certain harmonic ways (cadence) and that approach fits well with some of the repertoire.
Edit: I see that this is in the Composer's Workshop. What I already said is obviously from a performer's perspective but some applies to the composition process also. From a composer's perspective phrase lengths depend completely on your own sense of structure, balance, etc. I think music that's intended for focused listening - hard listening - is conducive to less predictable phrase lengths, whereas music intended for casual listening by a less sophisticated listening audience will likely be heard as simply sounding "weird" if the phrases are too surprising - similar to listener reactions to polytonality, atonality, even chromaticism.