We examine phrase alignment in a hybrid connectionist framework. We describe the architecture and the learning algorithm of our approach. Simulations have been carried out to demonstrate the feasibility of this approach using a real-world title phrase corpus. Although the results of our approach are still at an early stage, we found that a hybrid approach to phrase alignment has the potential to provide good results using relatively little training data. While there have been a lot of statistical approaches to alignment, to the best of our knowledge this is the rst hybrid connectionist approach for phrase alignment.