The Advantages of 3D-Trees in Modeling Human Sentence Processing

A 3D-Tree is an efficient representation of a set of regular syntactic trees (2D-trees) that have the same tree skeleton but which may have different labels for the nodes. A node of a 3D-tree may have one or more sets of children as well as one or more parent nodes as long as the 2Dtrees represented have the same tree skeleton. A 3D-tree is therefore able to efficiently represent the type of ambiguous structures which never seem to cause the Human Sentence Processor (HSP) to garden path. The advantages of 3D-trees over other parallel structures and the computational properties of building, manipulating, and maintaining, 3D-trees are presented in this paper. In addition, when the number and accessibility of 3D-trees are limited by the memory architecture of a parser in order to model certain psycholinguistic phenomena, the parser has the desirable computational properties of linear space requirements for parsing any sentence, and linear time requirements for parsing non-garden path sentences.