Unsupervised Learning of A-Morphous Inflection with Graph Clustering
Maciej Janicki · Recent Advances in Natural Language Processing · 2013
This paper presents a new approach to unsupervised learning of inflection. The problem is defined as two clusterings of the input wordlist: into lexemes and into forms. Word-Based Morphology is used to describe inflectional relations between words, which are discovered using string edit distance. A graph of morphological relations is built and clustering algorithms are used to identify lexemes. Paradigms, understood as sets of word formation rules, are extracted from lexemes and words belonging to similar paradigms are assumed to have the same inflectional form. Evaluation was performed for German, Polish and Turkish and the results were compared to conventional morphological analyzers.