Character-Aware Neural Language Models
This work addresses efficient and effective language modeling for multiple languages, particularly benefiting those with rich morphology, though it is incremental in applying neural architectures to character inputs.
The authors tackled language modeling by using only character-level inputs, achieving state-of-the-art performance on the English Penn Treebank with 60% fewer parameters and outperforming baselines on morphologically rich languages.
We describe a simple neural language model that relies only on character-level inputs. Predictions are still made at the word-level. Our model employs a convolutional neural network (CNN) and a highway network over characters, whose output is given to a long short-term memory (LSTM) recurrent neural network language model (RNN-LM). On the English Penn Treebank the model is on par with the existing state-of-the-art despite having 60% fewer parameters. On languages with rich morphology (Arabic, Czech, French, German, Spanish, Russian), the model outperforms word-level/morpheme-level LSTM baselines, again with fewer parameters. The results suggest that on many languages, character inputs are sufficient for language modeling. Analysis of word representations obtained from the character composition part of the model reveals that the model is able to encode, from characters only, both semantic and orthographic information.