Robsut Wrod Reocginiton via Semi-Character Recurrent Neural Network
Authors: Keisuke Sakaguchi, Kevin Duh, Matt Post, Benjamin Van Durme
AAAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In our experiments, we demonstrate that sc RNN has significantly more robust performance in word spelling correction (i.e. word recognition) compared to existing spelling checkers and character-based convolutional neural network. We conducted spelling correction experiments to judge how well sc RNN can recognize noisy word sentences. We used Penn Treebank for training, tuning, and testing. |
| Researcher Affiliation | Academia | Keisuke Sakaguchi, Kevin Duh, Matt Post, Benjamin Van Durme Center for Language and Speech Processing, Johns Hopkins University Human Language Technology Center of Excellence, Johns Hopkins University {keisuke,kevinduh,post,vandurme}@cs.jhu.edu |
| Pseudocode | No | The paper provides mathematical equations for LSTM and Char CNN, but no structured pseudocode or algorithm blocks are present. |
| Open Source Code | No | The paper provides links to third-party codebases for comparison models (Char CNN and Enchant) but does not state that the code for their proposed sc RNN model is openly available. |
| Open Datasets | Yes | We used Penn Treebank for training, tuning, and testing. Section 2-21 for training, 22 for tuning, and 23 for test https: //catalog.ldc.upenn.edu/ldc99t42. |
| Dataset Splits | Yes | We used Penn Treebank for training, tuning, and testing. Section 2-21 for training, 22 for tuning, and 23 for test... The data includes 39,832 sentences in training set... 1,700 sentences in the tuning set... and 2,416 sentences in test set. |
| Hardware Specification | No | The paper states training 'can be made within a hour with a CPU machine' but does not provide specific hardware details such as CPU model, GPU type, or memory specifications. |
| Software Dependencies | No | The paper describes the model architecture and refers to external tools like Char CNN and Enchant, but it does not specify the software libraries or their version numbers (e.g., Python, TensorFlow, PyTorch) used for implementing and running the sc RNN experiments. |
| Experiment Setup | Yes | The input layer of sc RNN consists of a vector with length of 76 (A-Z, a-z and 24 symbol characters). The hidden layer units had size 650, and total vocabulary size was set to 10k. We trained the model by running 5 epochs with (mini) batch size 20. We set the backpropagation through time (BPTT) parameter to 3. |