Proceedings of the Annual Meeting of the Cognitive Science Society

This paper proposes a recurrent neural network model thatlearns to perform morphological affixation, a fundamental op-eration of linguistic cognition, and has interpretable relationsto descriptions of morphology at the computational and algo-rithmic levels. The model represents morphological sequences(stems and affixes) with distributed representations that sup-port binding of symbols to ordinal positions and position-basedunbinding. Construction of an affixed form is controlled at theimplementation level by shifting attention between morphemesand across positions within each morpheme. The model suc-cessfully learns patterns of prefixation, suffixation, and infixa-tion, unifying these at all levels of description around the theo-retical notion of a pivot. Connections of the present proposal toneural coding of ordinal position, and to computational modelsof serial recall, are noted.