This paper is a continuation of earlier work on the controversial genetic classification of Songhay as Nilo-Saharan (Nicolaï 2003), where I show that the results thus far (cf. Bender 1995, Ehret 2001) are unsatisfactory. This is attributable to the authors' models and methods, the assumptions these rest upon, the nature of the data, and a priori factors in research. I discuss theoretical hypotheses regarding the ways in which languages change, the methods and procedures used by scholars, and the ways in which empirical data are used. This leads me to ask the following questions at the outset: 1) Are we entitled to make explicit use of virtual or real multilingual/multidialectal factors as normal theoretical parameters in building models of language change and proposing long-distance relationships? (This would imply changes to the classical tree representations and facilitate the incorporation of areal and contact phenomena.) 2) Would it be helpful to postulate an anthropologically defined setting for the interaction of structural linguistic forms, basic c ognitive processes, and punctual input from historical contingencies, resulting in rearrangements of norms of use and formal structures? (This question is vital to the study of areal phenomena and situations lying beyond the theoretical constraints of the standard genetic model.) These questions are in line with others raised during the Symposium with regard to the relations between theories of language structure and theories of cognition, the degree of conscious motivation for processes of language change and grammaticalization, and adabtability in language change and functionalization. In answering them, I make a number of suggestions based on my overall views on the practice of comparative linguistics, the construction of theoretical frameworks, and my recent empirical results in Nilo-Saharan and Afroasiatic.