Cross2StrA: Unpaired Cross-lingual Image Captioning with Cross-lingual Cross-modal Structure-pivoted Alignment
This addresses the challenge of generating accurate and fluent captions in a target language without paired image-caption data, which is incremental as it builds on existing cross-lingual methods by adding structural alignment.
The paper tackles the problem of unpaired cross-lingual image captioning, which suffers from irrelevancy and disfluency issues, by incorporating scene graph structures and syntactic constituency trees to align semantic and syntactic attributes during transfer, achieving improved relevancy and fluency in English-Chinese transfers.
Unpaired cross-lingual image captioning has long suffered from irrelevancy and disfluency issues, due to the inconsistencies of the semantic scene and syntax attributes during transfer. In this work, we propose to address the above problems by incorporating the scene graph (SG) structures and the syntactic constituency (SC) trees. Our captioner contains the semantic structure-guided image-to-pivot captioning and the syntactic structure-guided pivot-to-target translation, two of which are joined via pivot language. We then take the SG and SC structures as pivoting, performing cross-modal semantic structure alignment and cross-lingual syntactic structure alignment learning. We further introduce cross-lingual&cross-modal back-translation training to fully align the captioning and translation stages. Experiments on English-Chinese transfers show that our model shows great superiority in improving captioning relevancy and fluency.