Latent Tree Learning with Differentiable Parsers: Shift-Reduce Parsing and Chart Parsing
This work addresses the challenge of improving sentence representation models for natural language processing by learning latent parse trees without external syntax, though it appears incremental as it builds on existing differentiable parsing methods.
The paper tackles the problem of latent tree learning for sentence representation by introducing a new model based on shift-reduce parsing, which achieves competitive downstream performance and induces non-trivial parse trees, while also analyzing trees from shift-reduce and chart-based models.
Latent tree learning models represent sentences by composing their words according to an induced parse tree, all based on a downstream task. These models often outperform baselines which use (externally provided) syntax trees to drive the composition order. This work contributes (a) a new latent tree learning model based on shift-reduce parsing, with competitive downstream performance and non-trivial induced trees, and (b) an analysis of the trees learned by our shift-reduce model and by a chart-based model.