Generalized Transformations and the Theory of Grammar

Frank, Kroch 1995

Frank and Kroch show how TAG can be used as an alternative to Chomsky's minimalist framework. They argue that TAG provides a much cleaner formalism, especially for capturing locality constraints. Movement is broken into two separate operations: form-chain within individual elementary trees, and adjunction of X' auxilliary trees.

Contraints on Elementary Trees.

Frank and Kroch assume that the set of elementary trees available defines the language; and that this set is generated by some set of elementary principles. They assume that at least the following restrictions hold for elementary trees:

  • CETM: each elementary tree is the extended projection of a single lexical head.
  • Projection principle: all non-terminals in an elementary tree are licensed.
  • equation0.png Criterion: all of a predicates equation0.png-roles are assigned within its elementary tree.

Cyclic movement.

Using TAG, cyclic movement is replaced by adjunction. In particular, we can account for repeated raising of a SPEC/XP by saying that we adjoin in X' auxilliary trees. Under this theory, the reason that we can't do "cyclic movement" past XPs with filled specifiers is that that would require adjoining in XPs. But that would mean adjoining above the SPEC. Alternatively, it would mean adjoining in an X' auxilliary tree with an internal XP, but that kind of tree would violate CETM. Note that under this analysis, there is not actually any movement occuring, and there are no traces left behind.

Form Chain.

Non-cyclic movement is captured by applying form-chain (=move-equation1.png) within individual elementary trees. This operation coindexes a SPEC with a c-commanded equation0.png position (the trace is substituted in later). It is used for wh-movement, head-movement, etc. Note that this type of movement can interact with "cyclic movement" (which is not actually movement under this analysis). Note also that "crosssing" movement is allowed, as long as it's within a single elementary tree.

Long Movement and Multicomponent TAG.

In some types of form-chain type movement, the trace and the landing site are in two different elementary trees. Thus, basic form-chain can't account for these examples. Frank and Kroch propose a special type of form-chain operation which produces a multicomponent tree set, consisting of the tree with the landing site, and a degenerate tree for the trace. This tree set can then be adjoined in, as long as both parts are adjoined to the same elementary tree. TAG features are used to ensure that traces and wh-elements are actually inserted, and a version of the ECP is used to ensure that the trace is properly governed.

No intermediate traces.

Finally, Frank and Kroch discuss 2 phenomena that were traditionally explained with intermediate traces: binding of fronted elements (e.g., "Which picture of herself did Marsha like?") is explained by saying that the subject binds anaphors in spec/CP, and by using multi-component derivations. Scope effects are explained by saying that quantifiers recieve scope in the elementary tree where they're gnerated; and using both adjoining & multi-component derivations.


author =       {Robert Frank and Anthony Kroch},
title =        {Generalized Transformations and the Theory of Grammar},
journal =      {Studia Linguistica},
year =         1995,
volume =       49,
pages =        {103--151},
url =          {citeseer.nj.nec.com/frank95generalized.html}