I'm thinking of an app the pulls full song lyrics from an API and runs them through a text "summarizer". This means instead of, say, 50 lines of lyrics for a song, it would be reduced to 10, but the remaining words would not have been changed from the original.
Is this legal?
I am in the United States, but it is possible I would publish this in other countries as well.
It is unlikely to be legal. It is basically plainly illegal since you need permission to copy somebody's work, and copying the work is the first step. The possible escape would be the "fair use" defense, where you can really only be sure after you try it and successfully defend an infringement lawsuit using that argument. There are 4 factors that are "balanced": Purpose and character of use, The nature of the copyrighted work, Amount and substantiality and Effect on market, plus the wild card of Transformativeness. The ideal situation is where you quote a brief passage from a work, in order to comment on e.g. the author's style, do this in a educational setting, and do this in a way that transforms the work into a new creative work.
The Copyright office has a fair use index where you can get a summary of fair use findings (at least those that underwent appeal), but there are no bright lines that tell you what the threshhold is. We can assume that you intend to profit from this and this does not create a "commentary", which counts against you. Since music lyrics are typically licensed for use, there is a non-negligible effect on market, another problem. The amount of copying is very high (technically, 100%, but if you only count the part that is redistributed it's still well above the 1% number that is "out there" – again, there are no official rules especially regarding how much you can copy). It is unlikely that the product would fare well in terms of transformativeness, if you're just mechanically re-phrasing text (through a phone program). An individual might look at song lyrics and creatively construct a parody by discerning the "ideas" behind the words (an idea is not copyrightable), then writing their own text expression of the idea, but AI programming hasn't yet progressed that far.
