Publication at Faculty of Mathematics and Physics |
2021
Abstract
In this article, we present a proof-of-concept method for creating word-formation networks by transferring information from another language. The proposed algorithm utilizes an existing word-formation network and parallel texts and creates a low-precision and moderate-recall network in a language, for which no manual annotations need to be available.
We then extend the coverage of the resulting network by using it to train a machine-learning method and applying the resulting model to a larger lexicon, obtaining a moderate-precision and high-recall result. The approach is evaluated on French, German and Czech against existing word-formation networks in those languages.