Extracting relational triples from texts is a fundamental task in knowledge graph construction. The popular way of existing methods is to jointly extract entities and relations using a single model, which often suffers from the overlapping triple problem. That is, there are multiple relational triples that share the same entities within one sentence. In this work, we propose an effective cascade dual-decoder approach to extract overlapping relational triples, which includes a text-specific relation decoder and a relation-corresponded entity decoder. Our approach is straightforward: the text-specific relation decoder detects relations from a sentence according to its text semantics and treats them as extra features to guide the entity extraction; for each extracted relation, which is with trainable embedding, the relation-corresponded entity decoder detects the corresponding head and tail entities using a span-based tagging scheme. In this way, the overlapping triple problem is tackled naturally. Experiments on two public datasets demonstrate that our proposed approach outperforms state-of-the-art methods and achieves better F1 scores under the strict evaluation metric. Our implementation is available at https://github.com/prastunlp/DualDec.