Morphological segmentation without a lexicon

This article concentrates on the question To what extent can automatic morphological segmentation be achieved without the use of a lexicon? This question seems to us important in two respects, first, because of new insights into language structure, and secondly, because there are applications (for example, in information retrieval) which require nothing more than a morphological segmentation and where it turns out to be very costly to construct a lexicon We describe here two non-lexical segmentation algorithms which we developed for Spanish and German, respectively The approach adopted for Spanish is qualitative in that it uses rules on short grapheme sequences which carry information of morpheme boundaries The one adopted for German is quantitative, based on a study of frequencies of grapheme pairs having a morpheme boundary between, directly before or directly behind them 1