@InProceedings{hao-EtAl:2018:BlackboxNLP,
  author    = {Hao, Yiding  and  Merrill, William  and  Angluin, Dana  and  Frank, Robert  and  Amsel, Noah  and  Benz, Andrew  and  Mendelsohn, Simon},
  title     = {Context-Free Transductions with Neural Stacks},
  booktitle = {Proceedings of the 2018 EMNLP Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP},
  month     = {November},
  year      = {2018},
  address   = {Brussels, Belgium},
  publisher = {Association for Computational Linguistics},
  pages     = {306--315},
  abstract  = {This paper analyzes the behavior of stack-augmented recurrent neural network (RNN) models. Due to the architectural similarity between stack RNNs and pushdown transducers, we train stack RNN models on a number of tasks, including string reversal, context-free language modelling, and cumulative XOR evaluation. Examining the behavior of our networks, we show that stack-augmented RNNs can discover intuitive stack-based strategies for solving our tasks. However, stack RNNs are more difficult to train than classical architectures such as LSTMs. Rather than employ stack-based strategies, more complex stack-augmented networks often find approximate solutions by using the stack as unstructured memory.},
  url       = {http://www.aclweb.org/anthology/W18-5433}
}