I have been a little surprised to learn over the past couple of years that there are still generative syntacticians who think a system without Late Insertion is viable. Most notably, Chris Collins and Richie Kayne have recently put forth a radical proposal that aims to dispense with sui generis morphological operations,For whatever my opinion is worth, the goal of eliminating sui generis morphological operations seems like a laudable one to me. Whether it is in fact a viable goal is a different matter, one which I … Continue reading but also, crucially, rejects Late Insertion.
With that in mind, I would like to present you with what I honestly think is the shortest argument ever presented in favor of Late Insertion (in this case, on the PF side, i.e., Late Insertion of phonological content).
· · · · · · · · · · · · · · · · · · · ·
Preminger 2014 shows that checking-based theories of agreement are untenable, and only valuation-based theories are adequate. In valuation-based theories, the featural content of an agreement node cannot be determined prior to the syntactic derivation, and ipso facto, its form cannot, either.
· · · · · · · · · · · · · · · · · · · ·
That’s it. That’s the argument.
ETA: I’m not expecting anyone to buy the conclusions of Preminger 2014 simply because I assert them to be true – later in this post, I link to a short paper that provides a highly abridged version of the relevant part of the book; please do inspect both the paper and the book for yourself! But note: there are no “short” arguments if every new argument needs to include the already-published arguments it builds upon. 😁
I’ll give a slightly elaborated rendition of the same argument below, but the basic contours remain the same: a result concerning agreement theory which, to my knowledge, has never been seriously challenged, entails that morpho-phonological form cannot be determined prior to the syntactic derivation.
(This is far from the only argument in favor of Late Insertion. Nor is Late Insertion restricted to morpho-phonological content; it applies equally to semantic content, as well. But the current argument is of a different kind than what is usually given, certainly for Late Insertion on the PF side.)
Okay, so, here’s the argument in slightly more detail. Broadly speaking, there are two main approaches out there to the syntax of agreement. The first, checking, is a “generate-and-filter” architecture, wherein syntax is in principle able to generate agreement nodes and noun phrases with absolutely any features; and the role of syntactic agreement is to check (i.e., verify, validate) that the respective sets of features stand in the right kind of correspondence (e.g. that the features that Infl was generated with match the features of the closest c-commanded DP, if there is one).
The second approach is valuation, wherein agreement nodes are base-generated with no features at all (and instead, simply bear a diacritic that says “Find me a feature value!”, or more simply stated, “Probe!”), and they acquire their feature values derivationally. On this approach, a node like Infl would be born with no phi-features at all, and would acquire them in the course of the derivation by probing for the closest DP in its c-command domain.
Now, with respect to the question at hand, it should be obvious that Late Insertion is literally unavoidable on a valuation-based view. A theory without Late Insertion holds that the form of individual terminals is fixed from the start of the derivation. But agreement nodes bearing different features often (or always) differ in their forms. (E.g., an Infl node bearing “2nd person plural” features will have a different form than an Infl node bearing “3rd person singular” features.) The only way to accomplish this without Late Insertion is to adopt a checking-based theory, where agreement nodes are base-generated with both their formal features and their (corresponding) forms, and syntactic agreement serves to merely validate that the featural content of the agreement nodes chosen in a particular derivation matches the feature content of the corresponding noun phrase(s).
So far so good; but as it turns out, only valuation-based theories of agreement are viable. Checking-based theories simply don’t work. This is argued in detail in Preminger 2014 (but don’t worry, my 2018 paper provides a handy cheat sheet for the argument).
And since there is no adequate checking-based theory of agreement, there is also no adequate theory of grammar without Late Insertion of phonological content. Full stop.
This post was prompted by a recent discussion with Stefan Keine, and I honestly doubt I would have realized any of this without that discussion. Stefan should not be held responsible for what I say here, though.
|1.↩||For whatever my opinion is worth, the goal of eliminating sui generis morphological operations seems like a laudable one to me. Whether it is in fact a viable goal is a different matter, one which I will leave unexamined here.|