@inproceedings{liu2023compositional, author = {Liu, Fangyu and Liu, Qianchu and Bannur, Shruthi and Pérez-García, Fernando and Usuyama, Naoto and Zhang, Sheng and Naumann, Tristan and Nori, Aditya and Poon, Hoifung and Alvarez-Valle, Javier and Oktay, Ozan and Hyland, Stephanie}, title = {Compositional Zero-Shot Domain Transfer with Text-to-Text Models}, booktitle = {TACL 2023}, year = {2023}, month = {March}, abstract = {Label scarcity is a bottleneck for improving task performance in specialised domains. We propose a novel compositional transfer learning framework (DoT5 - domain compositional zero-shot T5) for zero-shot domain transfer. Without access to in-domain labels, DoT5 jointly learns domain knowledge (from MLM of unlabelled in-domain free text) and task knowledge (from task training on more readily available general-domain data) in a multi-task manner. To improve the transferability of task training, we design a strategy named NLGU: we simultaneously train NLG for in-domain label-to-data generation which enables data augmentation for self-finetuning and NLU for label prediction. We evaluate DoT5 on the biomedical domain and the resource-lean subdomain of radiology, focusing on NLI, text summarisation and embedding learning. DoT5 demonstrates the effectiveness of compositional transfer learning through multi-task learning. In particular, DoT5 outperforms the current SOTA in zero-shot transfer by over 7 absolute points in accuracy on RadNLI. We validate DoT5 with ablations and a case study demonstrating its ability to solve challenging NLI examples requiring in-domain expertise.}, url = {http://approjects.co.za/?big=en-us/research/publication/compositional-zero-shot-domain-transfer-with-text-to-text-models/}, }