Abstract
Zero-shot constituency parsing is a recent methodology in unsupervisedparsing that aims to extract parse trees from pre-trained language models(PLMs) with no extra training. This paper improves upon the existing paradigmby introducing the combination of a novel chart-based method and an effectiveensemble technique, attaining performance competitive to other unsupervisedparsers on English PTB. Furthermore, we broaden the range of zero-shot parsingapplication by examining languages other than English. Specifically, we firstdemonstrate that the approach is applicable to the languages that are equippedwith their respective monolingual PLMs. Finally, we propose to introducemultilingual PLMs into the zero-shot parsing framework, confirming that it ispossible to generate reasonable parses for sentences in nine languages in anintegrated and language-agnostic manner.