Spell Once, Summon Anywhere: A Two-Level Open-Vocabulary Language Model

Authors: Sebastian J. Mielke, Jason Eisner6843-6850

AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Comparing to baselines (including a novel strong baseline), we beat previous work and establish state-of-the-art results on multiple datasets.
Researcher Affiliation Academia Sebastian J. Mielke, Jason Eisner Department of Computer Science, Johns Hopkins University, Baltimore, MD, USA sjmielke@jhu.edu, jason@cs.jhu.edu
Pseudocode No The paper describes the model mathematically and textually but does not contain structured pseudocode or algorithm blocks in the main body.
Open Source Code Yes Code at github.com/sjmielke/spell-once.
Open Datasets Yes We evaluate on two open-vocabulary datasets, Wiki Text-2 (Merity et al. 2017) and the Multilingual Wikipedia Corpus (Kawakami, Dyer, and Blunsom 2017).
Dataset Splits Yes Bits per character (lower is better) on the dev and test set of Wiki Text-2 for our model and baselines...
Hardware Specification No The paper mentions running experiments on 'GPUs' and 'computational resources at the Maryland Advanced Research Computing Center (MARCC)' but does not provide specific hardware details like GPU models, CPU types, or memory amounts.
Software Dependencies No The paper mentions using PyTorch and AWD-LSTM but does not provide specific version numbers for these or other software dependencies.
Experiment Setup No The paper states that 'A detailed discussion of all hyperparameters can be found in Appendix B,' but does not include specific hyperparameter values or detailed training configurations within the main text.