Scaling Hidden Markov Language Models
This work addresses the challenge of making HMMs competitive in modern NLP for language modeling tasks, though it is incremental as it builds on existing ideas.
The paper tackled the problem of scaling hidden Markov models (HMMs) for language modeling, which had poor performance on large datasets, and resulted in models that are more accurate than previous HMM and n-gram methods, making progress toward state-of-the-art neural models.
The hidden Markov model (HMM) is a fundamental tool for sequence modeling that cleanly separates the hidden state from the emission structure. However, this separation makes it difficult to fit HMMs to large datasets in modern NLP, and they have fallen out of use due to very poor performance compared to fully observed models. This work revisits the challenge of scaling HMMs to language modeling datasets, taking ideas from recent approaches to neural modeling. We propose methods for scaling HMMs to massive state spaces while maintaining efficient exact inference, a compact parameterization, and effective regularization. Experiments show that this approach leads to models that are more accurate than previous HMM and n-gram-based methods, making progress towards the performance of state-of-the-art neural models.