Unsupervised grammar induction and similarity retrieval in medical language processing using the Deterministic Dynamic Associative Memory (DDAM) model

This paper is an overview of unsupervised grammar induction and similarity retrieval, two fundamental information processing functions of importance to medical language processing applications and to the construction of intelligent medical information systems. Existing literature with a focus on text segmentation tasks is reviewed. The review includes a comparison of existing approaches and reveals the longstanding interest in these traditionally distinct topics despite the significant computational challenges that characterizes them.Further, a unifying approach to unsupervised representation and processing of sequential data, the Deterministic Dynamic Associative Memory (DDAM) model, is introduced and described theoretically from both structural and functional perspectives. The theoretical descriptions of the model are complemented by a selection and discussion of interesting experimental results in the tasks of unsupervised grammar induction and similarity retrieval with applications to medical language processing. Notwithstanding the challenges associated with the evaluation of unsupervised information-processing models, it is concluded that the DDAM model demonstrates interesting properties that encourage further investigations in both theoretical and applied contexts.