Recursive Top-Down Production for Sentence Generation with Latent Trees
This work addresses the challenge of learning latent tree structures for natural language processing tasks, offering incremental improvements in specific synthetic and translation benchmarks.
The paper tackled the problem of modeling recursive sentence generation by developing a dynamic programming algorithm that marginalizes over latent binary trees to compute sequence likelihoods, achieving outperformance on the SCAN LENGTH split and comparable results on English question formation.
We model the recursive production property of context-free grammars for natural and synthetic languages. To this end, we present a dynamic programming algorithm that marginalises over latent binary tree structures with $N$ leaves, allowing us to compute the likelihood of a sequence of $N$ tokens under a latent tree model, which we maximise to train a recursive neural function. We demonstrate performance on two synthetic tasks: SCAN (Lake and Baroni, 2017), where it outperforms previous models on the LENGTH split, and English question formation (McCoy et al., 2020), where it performs comparably to decoders with the ground-truth tree structure. We also present experimental results on German-English translation on the Multi30k dataset (Elliott et al., 2016), and qualitatively analyse the induced tree structures our model learns for the SCAN tasks and the German-English translation task.