In our models of grammatical knowledge, words play an important role. They are phonological atoms, influence the surrounding syntactic structure, and provide the criteria for evaluating that sentence's truth conditions. However, as linguistic theory has advanced, particularly in investigating the structures of polysynthetic/agglutinative languages, the primacy of words as the atoms of grammar has been scrutinized. Theories of grammar must have some notion of morphosyntax -- building complex words out of sublexical units. However, the character of morphosyntax is far from well understood.

In the modern instatiation of Chomsky's generative grammar, the "timing" of word formation has been left unclear. Historically, there has been reason to think that at least some kinds of word formation must occur in the syntax, such as verbal morphology in Chomsky's (1957) "affix hopping" analysis. Alternatively, other aspects of morphosyntax, such as the derivational relationship between 'destroy' and 'destruction', were claimed to occur in a presyntactic generative lexicon (Chomsky 1970). However, even these kinds of analyses have been called into question. With the advent of Halle's (1990) and Halle and Marantz's (1993, 1994) theory of Distributive Morphology, the notion of a "generative lexicon" has been shown to be unnecessary, and Chomsky (2000) and Boeckx and Stjepanović (2001)have shown that the key syntactic operation in forming words -- head movement -- may also need to be abolished for technical reasons. Thus, the timing of word formation has become a complicated question with many theoretical ramifications, albeit based on very unclear and indirect empirical considerations.

This talk will present an offshoot of ongoing work looking at the structure of noun phrases in Bengali. I will claim that word formation rules must be interspersed with syntactic rules, and that they must be able to occur post-syntactically as well. I will address a system of feeding and bleeding relationships between a noun-to-classifier (N-to-Cl) substitution operation and a classifier-to-determiner (Cl-to-D) operation with semantic reflexes (suggesting it occurs before the syntactic derivation is complete). Specifically, I will show that for some classes of nouns, the N-to-Cl substitution operation feeds the Cl-to-D adjunction operation, and for another class of nouns, it is counterfed by Cl-to-D adjunction. This suggests an ordering between two distinct N-to-Cl relationships, one of which must occur before the derivation is complete. Given general concerns of cyclicity, the latter must be post-syntactic. I will then show that a series of diagnostics from Embick and Noyer's (2001) thoery of pre- and post-syntactic operations confirm the suspicion that N-to-Cl substitution has a particular pre-syntactic footprint when it feeds Cl-to-D adjunction, and it has a particular post-syntactic footprint when it is counterfed by Cl-to-D adjunction. This demonstrates that (1) there is empirical reason for thinking some word formation must occur before the syntax is complete (by hypothesis, in the syntactic derivation) and word formation after syntax is complete, and that (2) Embick and Noyer's diagnostics are confirmed by independent rule ordering evidence.