Supervised Morphological Segmentation in a Low-Resource Learning Setting using Conditional Random Fields

We discuss data-driven morphological segmentation, in which word forms are segmented into morphs, the surface forms of morphemes. Our focus is on a lowresource learning setting, in which only a small amount of annotated word forms are available for model training, while unannotated word forms are available in abundance. The current state-of-art methods 1) exploit both the annotated and unannotated data in a semi-supervised manner, and 2) learn morph lexicons and subsequently uncover segmentations by generating the most likely morph sequences. In contrast, we discuss 1) employing only the annotated data in a supervised manner, while entirely ignoring the unannotated data, and 2) directly learning to predict morph boundaries given their local sub-string contexts instead of learning the morph lexicons. Specifically, we employ conditional random fields, a popular discriminative log-linear model for segmentation. We present experiments on two data sets comprising five diverse languages. We show that the fully supervised boundary prediction approach outperforms the state-of-art semi-supervised morph lexicon approaches on all languages when using the same annotated data sets.