Toward a Totally Unsupervised, Language-Independent Method for the Syllabification of Written Texts

Unsupervised algorithms for the induction of linguistic knowledge should at best require as few basic assumptions as possible and at the same time in principle yield good results for any language. However, most of the time such algorithms are only tested on a few (closely related) languages. In this paper, an approach is presented that takes into account typological knowledge in order to induce syllabic divisions in a fully automatic manner based on reasonably-sized written texts. Our approach is able to account for syllable structures of languages where other approaches would fail, thereby raising the question whether computational methods can really be claimed to be language-universal when they are not tested on the variety of structures that are found in the languages of the world.