In this paper, we detail our approach as part of the runs submitted on behalf of the University of Glasgow Terrier Team (uogTr) for the 2021-A/B edition of the Incident Streams track. Our approach employs the use of transfer learning between component labels of the dataset; more specifically, we decompose the traditional multilabel approach and investigate the relationship between each label as a binary classification task. We submit a total of three official runs to the 2021-A/B edition of the track, namely: uogTr-01-pw, uogTr-02-pwcoocc, and uogTr-04-coocc. Our results show that there exists potential for performance increase through transfer learning.