Song translation is a really interesting "intelligence-complete" [constrained] optimization problem. You have an objective function that needs to take into acco

Song translation is a really interesting "intelligence-complete" [constrained] optimization problem. You have an objective function that needs to take into account: - closeness of meaning to the source language for each line or group of lines - rhyme scheme in the target language (often completely absent in the source language) - meteor/scansion in the target language (i.e. strong syllables on downbeats, see: http://www.patpattison.com/perfectmatch/ http://www.patpattison.com/perfectmatch/ - and, as is often the case when the source language is a language like Japanese where syllable stresses are ignored, the melody is not written to make this easy) - if the target language is tonal or semi-tonal, the degree to which words match the melody increasing/decreasing in pitch - overall natural feel of the lyrics as dialogue So the song re-writing is an iterative process as indicated here: over time, you explore the space of possible translations, evaluating against these metrics, and recognizing that changing one line could have higher-order effects, i.e. you need to change the rhyme scheme and therefore require a reordering of two other previous lines, leaping out of a local optimum and throwing out a lot of progress to make the overall song even better! Skilled songwriters, of course, can instantly discard huge portions of the space, and know intuitively where to begin. But even the best still need to iterate. And I find translation a great exercise for songwriting, since it lets me practice navigating this subspace of the much more complex general problem: how do you write a song starting with a blank page?