While searching for fragments of the currently-translated segment in translation memories, CafeTran comes across such fragments within the full segments and tries to determine the translation of the fragments. Since it is not an exact fragment match but only a hit inside the full segment, the program looks for a pattern across other segments that contain the same fragment. Based on the number of the hits, CafeTran establishes the accuracy of the pattern which becomes a translation for the given fragment. The larger number of the hits, the higher accuracy of the translation.

Where there is only one hit, a fragment match found inside one segment, the program displays the full target segment that contain the fragment. Such a hit is colored blue. Starting from two hits, the same fragment found in two segments, CafeTran performs "guessing" of the translation for the fragment and displays its translation. This is a low accuracy translation of the fragment with the hit colored orange. Finally, after a certain number of hits has been found, the "guess" becomes more accurate and is displayed in the purple color. Furthermore, CafeTran also analyses the context of the fragment to enhance the accuracy of the translation guess even when the number of the hits is low. The most accurate hits are listed in the panel of the terms and fragments used in auto-assembling (press the F2 key) for their quick transfer to the target segment. They are also displayed in the list of automatic suggestions during typing of the target segment.

Together with the translation of the fragments, CafeTran can display up to ten full segments that contain the fragment. Just hover the mouse over a number from 1 to 10 to see the full segment. If you click on this number, you will be able to edit the full segment containing the fragment.