Smelting Gold and Silver for Improved Multilingual AMR-to-Text Generation

Recent work on multilingual AMR-to-text generation has exclusively focused on\ndata augmentation strategies that utilize silver AMR. However, this assumes a\nhigh quality of generated AMRs, potentially limiting the transferability to the\ntarget task. In this paper, we investigate different techniques for\nautomatically generating AMR annotations, where we aim to study which source of\ninformation yields better multilingual results. Our models trained on gold AMR\nwith silver (machine translated) sentences outperform approaches which leverage\ngenerated silver AMR. We find that combining both complementary sources of\ninformation further improves multilingual AMR-to-text generation. Our models\nsurpass the previous state of the art for German, Italian, Spanish, and Chinese\nby a large margin.\n