2019/08/29 by Angel Daza, Daza, Angel, Anette Frank +1
Computer Science · #Computation and Language (cs.CL) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Natural Language Processing Techniques #Text Readability and Simplification #Topic Modeling
paper · pdf · doi:10.48550/arxiv.1908.11326
openalex publication_date 2019/08/29 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
We propose a Cross-lingual Encoder-Decoder model that simultaneously\ntranslates and generates sentences with Semantic Role Labeling annotations in a\nresource-poor target language. Unlike annotation projection techniques, our\nmodel does not need parallel data during inference time. Our approach can be\napplied in monolingual, multilingual and cross-lingual settings and is able to\nproduce dependency-based and span-based SRL annotations. We benchmark the\nlabeling performance of our model in different monolingual and multilingual\nsettings using well-known SRL datasets. We then train our model in a\ncross-lingual setting to generate new SRL labeled data. Finally, we measure the\neffectiveness of our method by using the generated data to augment the training\nbasis for resource-poor languages and perform manual evaluation to show that it\nproduces high-quality sentences and assigns accurate semantic role annotations.\nOur proposed architecture offers a flexible method for leveraging SRL data in\nmultiple languages.\n