From Phoneme to Morpheme: A Computational Model

Zellig Harris proposed a method for grouping phonemes in an utterance into morphemes by simply using counts of each of the phonemes in a corpus relative to their position in sequences contained in the data set. Thus, using an n-gram model, one can model this process and see whether a computational model can actually group representations of phonemes into segments which correspond to morphemes. Here, we use a general n-gram modelling tool created for melodic grouping in music corpora and apply it to a natural language data set. We show that this method which approximates Harris’s can indeed find morphemes in a given language corpus by calculating the distributions of phonemes across a corpus.