There are many situations where automated alignment procedures can produce biologically incorrect aligments. An obvious challenge are distantly related input sequences where homologies at the primary sequence level may be obscured by spurious random similarities. Another notorious challenge for alignment programs are duplications within the input sequences. Here, tandem duplications are particularly hard to align, see e.g. [22]. Specialised software tools have been developed to cope with the problems caused by sequence duplications [23]. For the segment-based alignment program DIALIGN, the situation is as follows. As described in previous publications, the program constructs pairwise and multiple alignments from pairwise local sequence similarities, so-called fragment alignments or fragments [17,16]. A fragment is defined as an un-gapped pair of equal-length segments from two of the input sequences. Based on statistical considerations, the program assigns a weight score to each possible fragment and tries to find a consistent collection of fragments with maximum total score. For pairwise alignment, a chain of fragments with maximum score can be identified [24]. For multiple sequence sets, all possible pairwise alignments are performed and fragments contained in these pairwise alignments are integrated greedily into a resulting multiple alignment.