Sequence Modeling via Segmentations
This addresses sequence modeling challenges in domains like natural language processing and speech recognition, but it is incremental as it builds on existing tools like RNNs.
The paper tackles the problem of modeling sequences with unknown segmental structure by proposing a probabilistic model that sums over all valid segmentations, using dynamic programming for efficient computation. It demonstrates the approach on text segmentation and speech recognition tasks, showing it can discover meaningful segments.
Segmental structure is a common pattern in many types of sequences such as phrases in human languages. In this paper, we present a probabilistic model for sequences via their segmentations. The probability of a segmented sequence is calculated as the product of the probabilities of all its segments, where each segment is modeled using existing tools such as recurrent neural networks. Since the segmentation of a sequence is usually unknown in advance, we sum over all valid segmentations to obtain the final probability for the sequence. An efficient dynamic programming algorithm is developed for forward and backward computations without resorting to any approximation. We demonstrate our approach on text segmentation and speech recognition tasks. In addition to quantitative results, we also show that our approach can discover meaningful segments in their respective application contexts.