Translate and Label! An Encoder-Decoder Approach for Cross-lingual Semantic Role Labeling

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 603–615, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 603 Translate and Label! An Encoder-Decoder Approach for Cross-lingual Semantic Role Labeling Angel Daza and Anette Frank Leibniz ScienceCampus “Empirical Linguistics and Computational Language Modeling” Department of Computational Linguistics Heidelberg University 69120 Heidelberg, Germany {daza,frank}@cl.uni-heidelberg.de Abstract We propose a Cross-lingual Encoder-Decoder model that simultaneously translates and gen- erates sentences with Semantic Role Labeling annotations in a resource-poor target language. Unlike annotation projection techniques, our model does not need parallel data during in- ference time. Our approach can be applied in monolingual, multilingual and cross-lingual settings and is able to produce dependency- based and span-based SRL annotations. We benchmark the labeling performance of our model in different monolingual and multilin- gual settings using well-known SRL datasets. We then train our model in a cross-lingual set- ting to generate new SRL labeled data. Finally, we measure the effectiveness of our method by using the generated data to augment the train- ing basis for resource-poor languages and per- form manual evaluation to show that it pro- duces high-quality sentences and assigns accu- rate semantic role annotations. Our proposed architecture offers a flexible method for lever- aging SRL data in multiple languages. 1 Introduction Semantic Role Labeling (SRL) extracts semantic predicate-argument structure from sentences. This has proven to be useful in Neural Machine Trans- lation (NMT) (Marcheggiani et al., 2018), Multi- document-summarization (Khan et al., 2015), AMR parsing (Wang et al., 2015) and Read- ing Comprehension (Mihaylov and Frank, 2019). SRL consists of three steps: i) predicate detection, ii) argument identification and iii) role classifica- tion. In this work we focus on PropBank SRL (Palmer et al., 2005), which has proven its validity across languages (van der Plas et al., 2010). While former SRL systems rely on syntactic features (Punyakanok et al., 2008; T¨ackstr¨om et al., 2015), recent neural approaches learn to model both ar- gument detection and role classification given a Figure 1: We propose an Encoder-Decoder model that translates a sentence into a target language and applies SRL labeling to the translated words. In this example we translate from English to German and label roles for the predicate have. predicate (Marcheggiani et al., 2017; He et al., 2017), and even jointly predict predicates inside sentences (He et al., 2018; Cai et al., 2018). While these approaches alleviate the need for pipeline models, they require sufficient amounts of training data to perform adequately. To date, such models have been tested primarily for English, which of- fers a considerable amount of high-quality train- ing data compared to other languages. The lack of sufficiently large SRL datasets makes it hard to straightforwardly apply the same architectures to other languages and calls for methods to augment the training data in lower-resource languages. There is significant prior work on SRL data aug- mentation (Hartmann et al., 2017), annotation pro- jection for monolingual …

Content truncated to 500 words