Encoding nondeterministic finite-state tree automata in sigmoid recursive neural networks

Abstract

Recently, a number of authors have explored the use of recursive recursive neural nets (RNN) for the adaptive processing of trees or tree-like structures. One of the most important language-theoretical formalizations of the processing of tree-structured data is that of finite-state tree automata (FSTA). In many cases, the number of states of a nondeterministic FSTA (NFSTA) recognizing a tree language may be smaller than that of the corresponding deterministic FSTA (DFSTA) (for example, the language of binary trees in which the label of the leftmost k-th order grandchild of the root node is the same as that on the leftmost leaf). This paper describes a scheme that directly encodes NFSTA in sigmoid RNN. © Springer-Verlag Berlin Heidelberg 2000.