In time, alternative approaches to syntax arose in response to the challenges in early generative syntax. Some of these were highly formal in their representation and prioritizes computational precision. One such approach was Lexical Functional Grammar.
In the 1970s, Joan Bresnan and Ronald Kaplan took a hard look at where Chomsky’s ideas were headed and did not like what they saw. As a mechanism, movement was powerful. It did too much and was problematic for generating all the grammatical sentences and none of the ungrammatical ones. They suggested that lexical rules would function as a better alternative to movement and they suggested that it would better to separate grammatical relations out from the constituent structure. The model they developed, they called Lexical Function Grammar (LFG). It made significant changes for the constituent structure. No empty nodes in a tree were allowed. No redundant layers that had now function should appear. No movement allowed; use lexical rules instead. What you see is what you get. And then they went a step further. If grammatical relations cannot be defined by constituency in every language, then they should simply always be treated as basic categories linguistic privatives. This has significant implications for representing syntax. The complete lack of movement and derived grammatical relations then also requires a new way of presenting grammatical relations in relation to constituency.
The early issues that Bresnan and Kaplan noted were less about information flow and more about the status of movement in relation to grammatical relations. While perhaps the most effective arguments Bresnan and Kaplan put forward came from languages like Warlpiri with so-called “free word order”, they also gave evidence from English. A sentence like:
That issue with transformations] we argued about ___ for days
We argued about [that issue with transformations] for days
At the time, these sorts of sentences were viewed a excellent evidence for movement in English. The basic argument in early transformational grammar was that if movement was necessary for English, then sure, on the basis of considerations of learnability and simplicity, movement should be required feature of all languages.
Bresnan and Kaplan, however, encountered what they called movement paradoxes. The above clauses are examples of object fronting, but that initial position at the front of the clause is more complicated than merely movement from one location in the clause to another.
[That movement is compelling] we argued about ___ for days.
*We argued about [that movement is compelling] for days.
They noted that even though the basic structure trees were the same, headless relative clauses functioning in that initial NP position do not arrive in that position as a result of movement from the prepositional phrase “about [noun phrase]”. The asterisk (*) in the second sentence above denotes the ungrammaticality of the sentence.
The relative clause, that movement is compelling did not move from the prepositional phrase. This is a movement paradox. It’s structurally parallel to the second of sentences above, the one that supposedly involved movement that left a gap after the preposition about. This sentence here has the same gap following about, but no movement from that position. These issues are compounded when we consider languages whose word order is “free”.
Our original Greek sentence, from the top of the page, would likely look something like the following in early LFG:
Here the tree is completely flat at the phrase level (the NP within the PP notwithstanding). No claims are being made as to a hierarchy between phrasal constituents. Even the discontinuous phrase is treated as flat, directly subordinate to the top level.
Grammatical relations are specified in a separate, parallel structure hence the name of this section Parallel and constraint-based syntax. A larger syntactic representation for this sentence in early LFG with the functional structure (‘function’ because, LFG calls grammatical relations grammatical functions):
There is a certain logic to this approach of separating this information out from constituent structure, especially for a language like Greek that has well-established inflectional morphology for things like case that in turn specify the grammatical relations in the sentence. If the grammatical relation subject is determined not by syntactic position in the tree, what is the point of either defining it or representing it within a tree. LFG uses a set of syntactic rules between the constituent structure on the left and the functional structure on the right to constrain each other. The excessively simplified explanation is that:
- Lexical rules rules for what kinds of elements a verb requires form the basis for the functional structure.
- Phrase structure rules similar to the ones for early Chomskyan syntax guide the formation of the constituent structure.
LFG’s phrase structure rules, however, are always language specific. They are not a universal grammar. The expectation is that Greek has a different set of phrase structure rules than English does. These language specific rules also provide information for mapping a constituent structure to a functional structure and vise versa. This results in a framework that perhaps lacks the sort of intuitive elegance of early Chomskyan syntax, while still being extremely workmanlike and useful for computational approaches to grammar.
When questions of information flow and discourse became a more and more pressing issue in syntacitc theory, Lexical Functional Grammar proposed its own solutions distinct from the proposal of mainstream generative theory that we saw previously. But before we examine those ideas, it is necessary to take a look at one of the other major responses to early Chomsky in formal syntax.
At the very same time that Bresnan and Kaplan were doing their work, Gerald Gazdarwas coming to similar conclusions, together with Ewan Klein, Ivan Sag, and Geoffrey Pullum. We will examine their proposal next time.