There is a difference between what linguists do and what speakers do. The goal of the linguist is to describe a language and to come up with some sort of system (be it syntactic, morphological, phonological, etc.) that accounts for the phenomena observed in the language. On the other hand, speakers are not concerned with any of this and their goal is to understand and be understood
The generalizations that linguists make about language are often times too broad and the desire to reach an “elegant” solution sometimes comes at the cost of the ability to account for all the data (as well as the psychological reality of their solution). Often problematic aspects of the language are left aside and deemed “irregular”, somehow being “outside” of the system being accounted for. Moreover, the data most linguists use describing linguistic phenomena (especially with well-studied languages) come from a standardized form of the language, which can be at times artificial and divorced from the way that people actually speak. A close, detailed examination of data shows that the generalizations that speakers make are not necessarily broad, and are usually local in nature (Joseph & Janda 1986, Joseph 1997).
I argue that the local nature of linguistic generalizations can be seen in the Latin and Romance conjugational class systems, the traditional views of which often try to boil down the differences to somewhere between three and five large conjugational classes. Given that there are many more than between three and five surface patterns of verb conjugation for Latin or any Romance language, it is necessary that some of these classes need to be broken down into several subclasses in order to account for the fact that different verbs normally considered to be of the same conjugational class often show inflectional patterns that differ in matters of detail. While the establishment of “subclasses” implies that there is a smaller generalization at play (verbs of subclass A are different from verbs of subclass B, even if they belong to the same superordinate class), I argue that the whole notion of “subclass” is unnecessary, and that it is just a artifact of attempts to make an elegant solution.
The data presented in this dissertation are from Romance varieties often seen as “peripheral” and come from dialect atlases, the internet, traditional grammars, and fieldwork undertaken in Sicily and Macedonia. These data show that speakers of Romance languages (as well as speakers of Latin) conjugate verbs that follow a large number of patterns. While it may not clear how these groups are organized mentally, a situation where different patterns that are clustered around features they share provides a more realistic situation than one of a small number of classes that have extensive subbranching.