Labor day special: Agreement

This is a short post for labor day I wanted to elevate a concern raised in the comments to a post.

As all agreement (perhaps modulo cases of Suffixaufnahme) is of finitely many finite valued features, we can simply introduce one lexical item for each feature combination, and adjust our formal features accordingly. This is the checking theory of (Chomsky 95), whereby lexical items have fully fleshed out morphological features, and they must be checked by other (agreement) heads with matching features.

While this is formally possible, it feels like we’re missing a generalization; there is just one lexical item, which is unspecified for the features in question, and it inherits values for these features by virtue of its position in a larger syntactic structure. While some have argued that certain features, like case, do not come from a particular other head, but are rather due to more global properties of syntactic structure (such as which other DPs are in the same local domain), the canonical case of feature inheritance is person, number, and gender features, which always seem to come from particular heads.

A particularly influential idea about how to transmit features from one head to another is that there is an operation, Agree, which transmits features between two heads which stand in a c-command relation. Another idea is that heads must be in a Spec-Head relation for feature transmission to occur (and then there is no operation which does it, it just sort of happens). These proposals are related of course; postulating an operation of Agree which makes one head agree with another is sort of impossible to be wrong - no matter what mechanism is ultimately responsible for agreement, it can surely be formulated in this way. The actual work on Agree is attempting to discover and formulate general restrictions on when two heads can agree with one another. The Spec-Head theory just takes these restrictions to be very tightly geometrically constrained.

One aspect of work on Agree that I am not convinced about is that it implicitly assumes that syntactic features are (at least partially) reducible to morphological features. This could end up being true, but I think it is (or would be) an interesting discovery, and should not just be stipulated. Given our current formulation of syntax in terms of establishing dependencies between heads, I want to propose that we reformulate the Agree hypothesis in the following terms:

morphological features are transmitted along syntactic dependencies

In other words, until we start formulating rules about how agreement works, we can be attentive to the fact that in our analyses elements which actually agree with one another should be connected to each other by a sequence of syntactic dependencies.1

  1. But Greg, you say, our structures are connected graphs, which means that every head is connected to every other head. Yes, I answer, but we want systematic connections that obtain in the same way across different structures. ↩︎