Learning to Discover, Ground and Use Words with Segmental Neural Language Models
arXiv:1811.09353
Abstract
We propose a segmental neural language model that combines the generalization power of neural networks with the ability to discover word-like units that are latent in unsegmented character sequences. In contrast to previous segmentation models that treat word segmentation as an isolated task, our model unifies word discovery, learning how words fit together to form sentences, and, by conditioning the model on visual context, how words' meanings ground in representations of non-linguistic modalities. Experiments show that the unconditional model learns predictive distributions better than character LSTM models, discovers words competitively with nonparametric Bayesian word segmentation models, and that modeling language conditional on visual context improves performance on both.
References in corpus (7)
- Sequence Transduction with Recurrent Neural Networks
- Representations of language in a model of visually grounded speech signal
- Latent Sequence Decompositions
- Sequence Modeling via Segmentations
- Learning to Create and Reuse Words in Open-Vocabulary Neural Language Modeling
- From phonemes to images: levels of representation in a recurrent neural model of visually-grounded language learning
- Revisiting the Hierarchical Multiscale LSTM