Frustratingly Simple but Surprisingly Strong: Using Language-Independent Features for Zero-shot Cross-lingual Semantic Parsing

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 5848–5856 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 5848 Frustratingly Simple but Surprisingly Strong: Using Language-Independent Features for Zero-shot Cross-lingual Semantic Parsing Jingfeng Yang‡ Federico Fancellu† Bonnie Webber⋄ Diyi Yang‡ ‡ Georgia Institute of Technology † Samsung AI Center, Toronto ⋄ ILCC, School of Informatics, University of Edinburgh yjfllpyym@gmail.com bonnie@inf.ed.ac.ukfederico.f@samsung.com dyang888@gatech.edu Abstract The availability of corpora has led to signif- icant advances in training semantic parsers in English. Unfortunately, for languages other than English, annotated data is lim- ited and so is the performance of the de- veloped parsers. Recently, pretrained multi- lingual models have been proven useful for zero-shot cross-lingual transfer in many NLP tasks. What else does it require to apply a parser trained in English to other languages for zero-shot cross-lingual semantic parsing? Will simple language-independent features help? To this end, we experiment with six Dis- course Representation Structure (DRS) seman- tic parsers in English, and generalize them to Italian, German and Dutch, where there are only a small number of manually anno- tated parses available. Extensive experiments show that despite its simplicity, adding Univer- sal Dependency (UD) relations and Universal POS tags (UPOS) as model-agnostic features achieves surprisingly strong improvement on all parsers. We have publicly released our code at https://github.com/GT-SALT/ Multilingual-DRS-Semantic-Parsing . 1 Introduction Semantic parsing is the task of transducing natu- ral language to meaning representations, which in turn can be expressed through many different se- mantic formalisms including Discourse Represen- tation Theory (DRT) (Kamp and Reyle, 2013), Ab- stract Meaning Representation (AMR) (Banarescu et al., 2013), and so on. However, manually anno- tating meaning representations in a new language is a painstaking process which explains why there are only a few datasets available for different for- malisms in languages other than English. For in- stance, in the case of DRT, even in the latest re- lease of the Parallel Meaning Bank (PMB) (v3.0, †Work done while Federico Fancellu was at the University of Edinburgh. Abzianidze et al., 2017), there is no gold training data available for Italian and Dutch and the few annotated sentences are used as development and test data. How can one train a parser when train- ing data is unavailable? This work answers this question by looking at what it is required for zero- shot cross-lingual semantic parsing – learning a semantic parser for English and testing it on other languages. Prior research on cross-lingual semantic pars- ing leveraged machine translation techniques to map the semantics from a language to another (Da- monte and Cohen, 2018). However, these methods require parallel corpora to extract automatic align- ments which are often noisy or not available at all. Other work exploited parameter-shared models, which are based on language-independent represen- tations. In particular, cross-lingual word embed- dings and pretrained multilingual models (Hu et al., 2020) have been used in various cross-lingual NLP tasks, including semantic parsing (Oepen et al., 2020; Sherborne et al., 2020). However, pretrained multilingual models are …

Content truncated to 500 words