Other authors have studied the ability of DTRNN-like architectures to learn context-free grammars from examples. In some of the cases, the DTRNN is augmented with an external stack which it can manipulate, in much the same way as a pushdown automaton (the recognizer for context-free grammars) is a finite-state machine with an external stack (Mozer and Das, 1993; Giles et al., 1990). Others rely on the ability of the DTRNN to organize their state space in a stack-like fashion.