Enriched In-Order Linearization for Faster Sequence-to-Sequence Constituent Parsing

Sequence-to-sequence constituent parsing requires a linearization to\nrepresent trees as sequences. Top-down tree linearizations, which can be based\non brackets or shift-reduce actions, have achieved the best accuracy to date.\nIn this paper, we show that these results can be improved by using an in-order\nlinearization instead. Based on this observation, we implement an enriched\nin-order shift-reduce linearization inspired by Vinyals et al. (2015)'s\napproach, achieving the best accuracy to date on the English PTB dataset among\nfully-supervised single-model sequence-to-sequence constituent parsers.\nFinally, we apply deterministic attention mechanisms to match the speed of\nstate-of-the-art transition-based parsers, thus showing that\nsequence-to-sequence models can match them, not only in accuracy, but also in\nspeed.\n