Computational Complexity in Neurally-inspired Morphology processing

Pre-compilation of memory "chunks" in the mental lexicon is beneficial for on-line processing because it enhances the capacity to predict the completion of an upcoming input word and to maintain longer and more complex word sequences in the STM store: the more we memorize, the less material we need to integrate on-line. Morpheme-based representations are like memory chunks that come into the picture when memory of whole words fails, due to either novel, degenerate and noisy input, or to limitations in perception/memory spans. Due to the combinatorial nature of language and our memory limitations, less "chunking" and more on-line integration is expected at processing levels beyond morphology.