You seem to be speculating that a solution to parsing is going to pop out of understanding compression. Your argument is that parsing is essentially a problem of modelling P(s), and compression is just that. But I’m skeptical that the compression angle is going to give you special insight. People have been keenly interested in language models since approximately forever, since they’re critical for translation and other applications. They’re pretty much the bedrock of the field.
I’d note though that text compression would be a useful evaluation technique for language technologies. There are lots of competing encoding schemes for grammars, and they’re difficult to compare them, which makes tools difficult to compare. “Which one compresses text best” seems like a very good solution. I don’t know whether it’s been proposed before. It sounds familiar, but that might just be because good ideas have that sort of resonance sometimes.
About the NLP stuff:
You seem to be speculating that a solution to parsing is going to pop out of understanding compression. Your argument is that parsing is essentially a problem of modelling P(s), and compression is just that. But I’m skeptical that the compression angle is going to give you special insight. People have been keenly interested in language models since approximately forever, since they’re critical for translation and other applications. They’re pretty much the bedrock of the field.
I’d note though that text compression would be a useful evaluation technique for language technologies. There are lots of competing encoding schemes for grammars, and they’re difficult to compare them, which makes tools difficult to compare. “Which one compresses text best” seems like a very good solution. I don’t know whether it’s been proposed before. It sounds familiar, but that might just be because good ideas have that sort of resonance sometimes.