# Catalogue of Artificial Intelligence Techniques

## Tree-Adjoining Grammars

**Aliases:**
TAG

**Keywords:**
LFG, Lexical Functional Grammar

### Categories: Natural Language

Author(s): **Nicolas Nicolov**

Tree-Adjoining Grammar (TAG) is a grammatical (and mathematical)
formalism which constitutes a tree generating system (as opposed to
string generating systems such as Context-free Grammars).
TAG postulates the existence of a finite set of *elementary trees*
out of which bigger syntactic trees can be built. Elementary trees are
divided into *initial trees* and *auxiliary trees*--initial
trees represent minimal syntactic structures while auxiliary trees
correspond to minimal recursive structures. Bigger trees can be built
by means of the operation of *adjoining*. Adjoining takes a
recursive structure (auxiliary tree) and `inserts' it into another
tree (either an initial tree or a tree derived by previous adjoining).
Two important properties of TAGs which enable them to characterise the
strong generative capacity of grammars (that is, their capacity to
characterise the structural descriptions associated with sentences) are:

- Extended Domain of Locality: TAGs have a larger domain of locality than Context-free Grammars and linguistic formalisms based on them (such as Lexical Functional Grammar and Head-Driven Phrase Structure Grammar). Thus, the dependencies between a verb and all of its arguments can be stated (locally) in one construction.
- Factoring Recursion from the Domain of Dependencies: The elementary structures are the domains over which (linguistic) dependencies such as agreement, subcategorisation, and filler gap, for example, are stated. Recursion is factored out from the domain of the dependencies but is reintroduced by the operation of adjoining which embeds recursive structures (auxiliary trees) into other trees. Thus, the long distance behaviour of some dependencies is accounted for.

### References:

No references to display.

### Comments:

No comments.