Since the generation component (EGEN) [Bontcheva 95] depends heavily on the underlying semantic representation, most applied algorithms are strongly influenced by some features and their operations. We have focused mainly on generating sentences from conceptual graphs but we also tackle some problems of organising these sentences into a coherent text.
In order to produce a coherent explanation, the generator orders the input CGs by applying text organisation schemas, similar to these in [McKeown 85]. EGEN supports three schemas, where the main one is for definitions. At the next stage, the graphs are checked for similarities and such graphs are joined to avoid repetitions. Finally, each graph is verbalised into a sentence by a modification of the utterance path approach proposed in [Sowa 84]. The algorithm searches for a cyclic path which visits each node and relation at least once. If a node is visited more than once, grammar rules decide when and how it is verbalised. All uttered concepts are kept in a stack which allows the generation of pronouns and definite noun phrases. The stack is usually cleared at the end of the explanation.
As we already mentioned, the original utterance path algorithm was extended to: