If I was going to scrape this data and re-purpose it, I would've absolutely cleaned up those apostrophes. The pivoting between straight and curly would certainly be a pet peeve. Unless there's a semantic difference between the two I'm unaware of.
The song was released almost a year before UTF-8 was first presented at USENIX, so I think it is reasonable for the lyrics to be expressed using the commonly available technology available of the time.
I queried for this song's lyrics, and I found only a single quoting symbol (') used entirely throughout.
Taking a layperson's reading of the lyrics, I didn't find anything off due to quoting issues. If there was any disruption, it was minimal and unnoticed.
Yeah, makes sense, but this is still a pretty good approach. Inserting invisible or unusual Unicode symbols would prompt the scraper to carefully cleanup the read files (maybe even fixing these apostrophes as a result). Unusual whitespace is also likely to be removed and cleaned up.
On the other hand, these alternating apostrophes have a chance to stay unnoticed (or neglected), falling through the cracks.
There is a semantic difference between the two. The straight quote is a superset of the curly one.
So "rock 'n' roll" is correct. And "rock ’n’ roll" is correct. But "rock ‘n’ roll" is not correct, since the wrong apostrophe is used. We're not quoting the letter n, we're showing that the letter a was removed.
On this quantity of data, you wouldn't be able to do this manually.
If you hope to avoid being caught this way, I'm going to assume you noticed this without the benefit of hindsight and plan to correct all out-of-place Unicode characters automatically. How will you avoid over-correcting?
There's also no reason to believe this is the only fingerprinting Genius has done (they only need to publish the most obvious fail). For example, I can use the same fingerprinting technique but switch between American and British spellings.
I think if you had multiple corpuses of lyrics, you could cross-check for anomalies of any variety (odd quoting, switching between american/british english), etc.
The fingerprinting isn't likely applied to every song, to prevent obvious detection. If you went through multiple databases, you might see N prevailing copies of a song's lyrics, and 1 that seemed different. The one that's different has the anomaly.
