Reproduce “On the Practical Ability of Recurrent Neural Networks to Recognize Hierarchical Languages” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Satwik Bhattamishra, Kabir Ahuja, Navin Goyal
The paper tests whether LSTMs fail on long sequences because of length or nesting depth. It evaluates Dyck languages, where valid brackets require tracking unmatched openings, and gives a theoretical RNN construction using a numeric stack encoding. On Dyck-2, an LSTM scores 75.1% unrestricted versus 99.6% with depth capped at 10, but performance declines as tested depth rises.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.