In this paper we approach the segmentation of on-line cursive script by an
explicit modeling of the inter-character patterns called ligatures. The mod
el is given in terms of both the spatial relations among the ligature and t
he adjacent characters, and the ligature shape. The former allows for detec
ting plausible regions of connection between successive characters within t
he word, while the latter accounts for discriminating, among those regions,
the ones that actually correspond to ligatures.