Acquisition and enrichment of morphological and morphosemantic knowledge from the French Wiktionary

International audienceWe present two approaches to automatically acquire morphologically related words from Wiktionary. Starting with related words explicitly mentioned in the dictionary, we propose a method based on orthographic similarity to detect new derived words from the entries' definitions with an overall accuracy of 93.5%. Using word pairs from the initial lexicon as patterns of formal analogies to filter new derived words enables us to rise the accuracy up to 99%, while extending the lexicon's size by 56%. In a last experiment, we show that it is possible to semantically type the morphological definitions, focusing on the detection of process nominals