Memory compression strips content
Huang & Awh (2018) demonstrate through 4 experiments that memories are chunked in a manner that removes their content. This phenomenon explains why we are better at remembering a group of things (objects or letters for example) that make sense together (like a word, in the case of letters), even though it might take a little longer. A good example of this is presented in the paper.
"... Someone remembering the string “internationalizationcongratulationmisinterpretation” would probably agree that they are not holding online individuated representations for each letter in this string, even if every letter could be recalled perfectly given sufficient time. ... [but] it is not obvious whether the memory of a “Belgium flag” contains the active representation of the three colors or not, or the memory of a word “dog” contains active representations of the three letters or not. (Huang & Awn 2018).
My hot take is that this is an interesting paper that explains a phenomenon of human behavior that we take for granted. Indeed, it also makes sense in the sequence learning literature, where once a movement (like a dance) has been mastered, it may be hard to execute a particular aspect of the movement in isolation. The movement is stored as a chunk, and the individual components, like the letters in the example above, become separate from the whole.