Neural Semantic Parsing for Dutch
Rik van Noord, Antonio Toral and Johan Bos

Neural networks have had many successes in semantic parsing over the last few years. Recently, neural parsers have been developed that can parse sentences to a (formal) semantic representation, such as Abstract Meaning Representation or Discourse Representation Structure (DRS). These parsers often only work for English, since that's the only language with (enough) gold standard data available. However, recently the Parallel Meaning Bank released a data set of manually annotated DRSs for four languages (English, German, Italian and Dutch). Since these DRSs themselves are language neutral, we can leverage gold standard English instances to improve parsing of the non-English sentences, focusing mostly on Dutch. We do this by using a sequence-to-sequence neural network with a multi-encoder setup, in which we employ the translations of the English input sentences. Our goals are threefold: (a) design a DRS parser specifically tailored to Dutch, (b) design a multi-lingual parser that can parse DRSs for all four languages, and (c) improve our English DRS parser by including the other languages. We will discuss our results during the presentation.