merge

a fun thought I've had on linguistic history and one that I haven't seen expressed elsewhere:

So in Syntactic Structures (/LSLT) Chomsky looks into ways to formally capture some linguistic facts and starts off with some simple finite state automata. He shows that there are some facts about language that these simple machines are too weak to handle. Like, they are formally not powerful enough to capture the facts. In light of this failure, he says that we need something more powerful to capture language AND we should scrap finite state automata. This is fine, though he just as easily could have said: we need something more powerful to capture language IN ADDITION TO finite state automata. This is not nutty as there are some phenomena that might be sensibly captured by these (I'm thinking 'flat' stuff like iterated adjuncts, coordination, etc.)

Ok, so he goes on and looks into phrase structure grammars. As you might expect he finds that these do not succumb to the same failures as finite state machines, but he also uncovers some shortcomings for these more powerful devices. The difference here is that the failures are not those of formal power, but of explanatory adequacy: they can capture the facts, but only in ungainly and unenlightening ways. In response to this he says: we need something more powerful to capture language IN ADDITION TO phrase structure grammars. the more powerful thing is transformations, things that that in this case take the structures created by phrase structure grammars and alter them (they move things around, delete them, substitute in things, etc). Echoing the above paragraph, he could have just as easily said: we need something more powerful to capture language AND we should scrap phrase structure grammars.

But he doesn't, and for the longest time I just assumed that the reason he didn't say we should scrap them is because their failures were of a qualitatively different, weaker type. Not formal expressive powerful, but explanatory adequacy. These things coincide here, but there's no real logic tying them together from what I see. So it's kinda odd, or at least not immediately justified, that he does not scrap phrase structure grammars here. Though he's a smart guy and I trust him and so did a lot of other people and I'm sure he had his reasons.

Fast forward to, um, 93/95, and we are met with merge, which is a real departure from phrase structure grammars (though x-bar technically is the first real diversion from them, it's not in ways that matter really here). With merge as we know, the same process that builds structure also is responsible for moving things around. That is, structure building and transformations are collapsed. Or, more interestingly put: there is no more phrase structure grammar, only transformations, only putting two things together.

This is basically the step that Chomsky seemingly arbitrarily avoided way back in 55/57, just delayed by 40 years.