I think the text that Judy is referring to already exists.
>
> I know that the Dougals Biber group in England is accumulating a huge
> database of texts of all kinds which is making it possible to show
> grammar/text correlations with probabilities calculated. There is certainly
> no 1:1 absolute relation, but there are just as certainly quite high
> probabilities that certain linguistic features will show up in certain kinds
> of text in a certain likely order -- which would indicate that, for writing
> a certain kind of text, a student would benefit from knowing how to produce
> the correlated features in the expected order.

It is the Longman Grammar of Spoken and Written English.  Biber is the
first author.

I have a copy. The results have some interest, but I am not quite sure
that it would be helpful to a student, let alone a teacher.  Consider
what it says about Sentence/turn-initial coordination [defined in this
section as and, but, or, nor] (p 83-84).  Here are the two major
conclusions:

"Proportionally, turn-initial coordinators are considerably more common
in conversation than sentence-initial coordination than in the written
registers [defined as fiction, news, and academic prose]."

"Sentence-initial coordination are least common in academic prose."

In the figure next to these statements, it appears only 1% of the AND's
in academic prose is as a sentence-initial coordinator.  For BUT this
rate is 10%.  OR is like AND. For NOR, 30% of its occurrences is as a
sentence-initial coordinator. (These percentages are my interpretation
of the figure.  These numbers are not in the text.)

In the discussion, the authors note: "The prescription against initial
coordination seems most influential in academic prose.  The higher
frequencies in fiction and news reportage probably reflect the fact that
these registers often include more spontaneous discourse, including
fiction dialogue and quoted speech."  What they are referring to is that
in conversation about 20% of the AND's are sentence initial
coordinators, 50% of the BUT's, 8% of the OR's, and about 90% of the
NOR's.

Notice all of these numbers refer to percentages.  Several pages early,
there is a figure showing the distribution of coordinators across
registers.  In academic prose AND occurs 25,000 times per one million
words and BUT occurs about 6,000 per one million words.  So, out of
everyone one million words in academic prose 250 AND's will be as a
sentence-initial coordinator and about 600 BUT's will be.

What does a teacher of grammar do with such distributions, either
percentages or words per million words?  How do these data benefit a
student?  I modestly suggest some. In academic prose, AND is so rare as
a sentence-initial coordinator that a student should never use it.  Or,
AND can only be used as a sentence-initial coordinator once every 100
times AND is used.  Or, it is more acceptable to begin sentences with
BUT than with AND.

But, it might be the case that none of these suggestions are useful
because it would appear that the distributions are the result of editors
who have a commitment to prescriptivism.  It might be the case that it
does not matter anymore because this is strictly a prescriptivist
distribution and this rule is changing.

Bob Yates, Central Missouri State University

A final aspect of this issue.  FURTHERMORE occurs more than
100/1,000,000 words of academic prose and less than 50/1,000,000 words
in British English conversation. (p. 887)