Over a long-term time frame in a language with several discrete dialects, how far does grammaticalization theory elucidate the history of individual morphemes? This issue is addressed using the tense/mode prefix b-, found in Gulf/Najdi, Yemeni, Uzbekistan, Nigerian, and Egyptian/Levantine Arabic. It is argued that while standard grammaticalization theory correctly predicts its assumed origin, from a variant of the verb ‘want’ (yibġa, yiba, yibbi > *b-), it does little to predict its further development. This paper first examines the functions of the prefix *b-. Once integrated as a prefix, *b- takes odd twists and turns, sometimes a tense marker, sometimes a marker of deontic modality, sometimes a generalized modal/indicative marker. Grammaticalization theory says nothing about why *b- should have developed in one way in one dialect and in another way in another. As a step towards answering these questions, the idea of dialects as speech communities is introduced.