Joseph Pentheroudakis and Lucy Vanderwende
We describe an automated method for identifying classes of morphologically related words in an on-line dictionary, and for linking individual senses in the derived form to one or more senses in the base form by means of morphological relation attributes. We also present an algorithm for computing a score reflecting the system's certainty in these derivational links; this computation relies on the content of semantic relations associated with each sense, which are extracted automatically by parsing each sense definition and subjecting the parse structure to automated semantic analysis. By processing the entire set of headwords in the dictionary in this fashion we create a large set of directed derivational graphs, which can then be accessed by other compo nents in our broad-coverage NLP system. Spurious or unlikely derivations are not discarded, but are rather added to the dictionary and assigned a negative score; this allows the system to handle non-standard uses of these forms.