In general, the main idea of EBMT is simple: from all that was translated in the past, use whatever is available to facilitate the translation of the next utterance. How translation data (examples) are stored and applied (in matching and recombination) can vary, as long as the examples are used in run-time. Being empirical in nature, the example base represents real language use, covers the constructions which really occur, and is relatively easy to extend – simply adding more examples. This is particularly true for examples stored as bilingual string pairs, which have benefited, and will continue to benefit from the massively growing number of documents on the web. The problem of developing MT systems for language pairs suffering from a scarcity of resources has been to some extent resolved.