Developmental Minimalist Syntax

A long-term project that I’ve been working on has been to find ways to translate the findings of generative  (Chomskyan) syntax to theory-external phenomena within human cognition. This was mainly motivated by my cognitive science students here in Claremont, who have always struggled to understand how to relate the models of grammar I taught them with what else they knew about cognition. This is especially so when it came to the Minimalist model: building structure derivationally in a bottom-up fashion clearly matches neither perception nor production of language, so it is hard for them to fathom why our model would work like that. From my perspective, I’ve never had good answers, except to say that it works to capture the properties of adult grammars. So I starting wondering why it worked so well, when it clearly didn’t correlate with language processing in any obvious way. Surely the burden of responsibility is on generative syntacticians to communicate the findings of our field to cognitive scientists more broadly; I figured my students deserved better answers, so I tried to find them.

The hypothesis that I arrived at was that the sequence of structure-building in a Minimalist derivation works so well because those are the same sequences that language is acquired in by a child, and children retain their earlier stages of knowledge, building additional structures on top of their foundational structures. In work together with Madeline Bossi PO ’17, she explored a large range of acquisition literature and was able to teach me that 1) ideas like this had been around generative syntax for a long time, because 2) broadly-speaking, language is acquired in this fashion. We articulated the proposal in this way:

In many ways, this claim is affirmed in the data on child language acquisition: in general, structurally lower material is acquired before structurally higher material.

As was pointed out to us by a NELS reviewer years ago, however, there are a number of complications in which acquisition doesn’t clearly match bottom-up structure building, which is part of why the initial forays into these ideas in the mid-90’s appeared to have failed. To give just one example, children at very early stages of acquisition will be asking wh-questions quite frequently: what doing? where going? etc. This is just one example of many.  And while some recent work has attempted to revive this broad instinct, the empirical problems persisted.

This led us to focus particularly on those exceptions, at which point we found some quite striking correlations. Specifically, in those areas where there appears to be a breakdown of bottom-up acquisition of grammar, there are corresponding breakdowns of bottom-up derivation of adult grammatical structures. It is a well known fact that (on Minimalist analyses)  there are grammatical constructions where operations take place before they should have based on the canonical sequence of structure building (these are known as look-ahead problems). Likewise, there are many instances where operations take place after they ought to have on the canonical sequence of structure building: these include instances of Late Merger and of delaying Probing, among other things. Broadly speaking, these are counter-cyclic operations, i.e. grammatical properties that appear to operate against the normal cycles of bottom-up structure building in a Minimalist model.

The claim that we advance in our manuscript (currently being revised for resubmission, and which is now also coauthored with Katherine Johnson and Galia Bar-Sever) is that this correlation holds quite broadly: look-ahead problems in adult grammars correspond to `early’ acquisition (i.e. acquiring a pattern before the complete requisite grammar is acquired for that pattern), and patterns like Late Merger correspond to children learning a structure after subsequent, higher structures have been acquired.  For example, in adult grammars, in long-distance wh-movement wh-phrases move to the embedded CP before the ultimate target of their movement (matrix CP) has been merged into the structure; this correlates to children learning the linear position of wh-words (sentence-initial in English) long before learning about tense and complementizers. Likewise, relative clauses have been proposed to be merged late in a structure, explaining why they lack reconstruction effects as evidenced by binding (Takahashi and Hulsey 2009 summarizes this succinctly). We think it is no accident that relative clauses are one of the last structures that children grammaticalize in fully adult-like ways.

On this approach, many of the Minimalist analytical constructs that make up proposals about `UG’ (operations like Merge, Agree, etc) are are simply devices that children use to arrive at grammatical abstractions in language acquisition. Children acquire a pattern, and then they build additional knowledge on top of the knowledge that they already grammaticalized. This results in hierarchical syntactic knowledge, with structurally lower (and often unpronounced/invisible) representations of syntactic material occurring because those were a child’s first generalizations about that material, which were later revised, resulting in `movement’ of that material elsewhere in a sentence (i.e. a reanalysis of the position of an element, as the child has acquired more grammar). Therefore the lower positions, and even the sequence of operations itself that can model adult knowledge, can be considered ontogenetic fossils: remnants of earlier stages of knowledge that are retained in adult grammatical knowledge.