All those questions, and the status of "language" or "dialect" depend on two things: mutual intelligibility and politics. Why not just compute mutual intelligibility and leave the politics to the politicians?
Related to the issue of how to measure mutual intelligibility: at the 2008 LSA, I attended a fascinating paper by a linguist who was studying languages in Cameroon. He was revising all the mutual intelligibility findings based on new data; apparently the researchers who had been there before had asked villagers questions like, "Can you understand the people in the next village," and they had gotten answers like, "Who can understand those people? They do things so different over there." His work had implications for Greenberg's classification.
Similarly, there's the study by Rubin (1992) that found that comprehension is reduced when the speaker is perceived as "other." In that study, American students listened to a recording of a lecture read by "a doctoral student in speech communication, a native speaker of English raised in central Ohio, who was well regarded by her own undergraduate students for especially effective and clear classroom delivery." While listening, one group of students was shown a slide photograph of a "Caucasian" (White) woman, and the other group a photo of an "Asian (Chinese)" woman. The group shown the picture of the Asian woman not only claimed that it was more difficult to understand the lecture, but they scored lower on comprehension tests compared with the group shown the picture of the White woman (7.31 vs. 12.5).
In other words, mutual intelligibility is often a political matter.
-Angus B. Grieve-Smith
grvsmth at panix.com