One of them is particularly interesting for me as a native speaker of German, and perhaps for those of you who speak German, too:
Klaus Kohler [sic] [demonstrated] among other things that German listeners needed no more than the palatalization of a single segment n to hear kann Ihnen rather than just kann, deeply buried in the middle of a rapidly spoken colloquial sentence.
And I tried it myself and I think I can nearly reproduce that, at least for the speaking bit (harder to test the comprehension bit): kann is [kʰan] while kann Ihnen is [kʰanʲː]. (Not quite [kʰaɲː], I don’t think.) I can imagine that in rapid speech, the final length would get lost, leaving only the palatalisation.
...and here I thought German had no palatalisation! (True, it's phonetic only, not phonemic, but still: very interesting. To me, at least :D)
It also provides a lovely synchronic example of how segments can get lost while their ghost remains in the effect they have on the surrounding segments: similar thing occur in all sorts of areas such as umlaut, tone, or Greenlandic uvularisation. And also how this can cause phoneme splits if segments get lost, where the previously allophonic distinction (caused by the presence of the affecting segment) becomes phonemic when the segment drops entirely (as with Greenlandic vowels, where three phonemic vowels [six, if you count vowel length] split into six [twelve], once the uvular consonant got assimilated completely to a following consonant, forming a non-uvular geminate, while the vowel remained uvularised).