have to admit that many Tagmemic grammars are frustrating and difficult to read. One problem is that concepts like “contrast” and “minimal pair” do not apply as neatly to phrase or sentence patterns as they do to phonemes. The more fundamental issue is that the goal of a Tagmemic grammar was to list the inventory of contrastive units of various types, just as a major goal of phonemic analysis was to state the inventory of consonants and vowels. So Tagmemic grammars tend to be essentially lists of clause patterns, sentence patterns, word patterns, etc. What is often lacking is any statement about the generalizations, i.e. the rules that constitute the grammar of the language.
Even when our goal is to write a purely descriptive grammar, I feel it is important to adopt a rule-based perspective. We may never express these rules in formal syntactic notation, but the rules of the grammar are an important part of what we are trying to describe.
Of course, the focus on grammar as a system of rules is the defining characteristic of generative linguistics. But my impression is that a lot of people who do field linguistics and descriptive grammar have given up on generative syntax. In fact, a fair number seem to have given up on syntax in general, either ignoring it or adopting the view that most apparent syntactic regularities can be reduced to semantic and/or pragmatic generalizations. I suspect that a major reason for this is that the models of formal syntax that they have been exposed to seem so far removed from observable reality and impractical for descriptive purposes.
I believe that LFG offers a much better framework for descriptive grammar than recent transformational models. The modular design of the system means that c-structure representations are a direct and literal representation of the word order of the sentence as it is actually pronounced (WYSIWYG), and of constituency in the classical sense. These are basic facts that any descriptive grammar needs to spell out. (My impression is that a lot of current work in formal syntax, and even some more functional approaches, largely ignore these issues.)
Moreover, the modular design of the system means that problems or novel solutions in one area of grammar do not necessarily lead to complications in other parts of the analysis. For example, I have been interested for some time in the problem of “symmetric(al) voice”, as it has become known in Austronesian syntax. (The term was first used, I believe, by Bill Foley at LFG98 in Brisbane.) Essentially, this means a voice alternation (change in the assignment of the SUBJ function) without demotion. In a language like Balinese, there are two different transitive clause types: one in which the agent is SUBJ, and another in which the patient is SUBJ. Both of these are fully transitive, meaning that agent and patient are both terms (direct core arguments); neither gets demoted to oblique or