Several experiments have been developed around a bidirectional island-driven chart parser. The system follows basically the approach of Stock, Satta and Corazza, and the experiments have been designed and performed with the purpose of examining several ways of improvement: basic strategy of the algorithm (pure island-driven versus mixed island-driven/bottom-up approaches), strategies for extension of the islands, strategies for selecting the initial islands, ways of scoring the possible extensions, etc. Both the system and the results obtained up to date are presented in this paper.