In certain filetypes sometimes you need to not take in account the segmentation rues provided by the TM/language resources and instead keep the original segmentation of the filetype. Some examples include:

Excel: Keep the content of a cell in a single TU

xliff/mqxliff: Keep the original segmentation (this is very annoying and workarounds are really time consuming) 

Custom Grep/txt/csv filetypes: Keep the match/line/csv field in a single TU

Xml: keep a node in a single TU

Having a tm with just "paragraph segmentation" is usually not enough to achieve the same result as the one described above.

Often in workflow you find yourself with different files in the same project that needs this kind of segmentation (i.e.: strings for a software and the manual booklet) and you're forced to either create multiple consequent projects or add files one after the other changing project settings and tm every time.