Discourse analysis plays a crucial role in Nat- ural Language Processing, with discourse re- lation prediction arguably being the most dif- ficult task in discourse parsing. Previous stud- ies have generally focused on explicit or im- plicit discourse relation classification in mono- logues, leaving dialogue an under-explored do- main. Facing the data scarcity issue, we pro- pose to leverage self-training strategies based on a Transformer backbone. Moreover, we design the first semi-supervised pipeline that sequentially predicts discourse structures and relations. Using 50 examples, our relation pre- diction module achieves 58.4 in accuracy on the STAC corpus, close to supervised state-of- the-art. Full parsing results show notable im- provements compared to the supervised mod- els both in-domain (gaming) and cross-domain (technical chat), with better stability.