Textual Genre Analysis and Identification - Ambient Intelligence for Scientific Discovery - page 138

In sum, the idealized competent writers must call upon implicit rules for
rhetorical scrabble-playing that minimizes ambiguity and maximizes variety and
reusability. Our heuristics for coding indicate our best efforts to incorporate these
trade-offs. In this section, we briefly describe these heuristics.

5.1 Heuristic 1: Don't Over-Commit

Surface English is ambiguous and indeterminate and our conception of the ide-
alized competent writer projects a writer who exercises careful control anticipat-
ing the state of affairs for a reader. This meant that many potential words and
strings would, in the eyes of this writer, communicate nothing definite. Trans-
lated into coding rules, this meant to us that many strings would need to go
either un-coded or assigned to a “low commitment” coding category that im-
pacts the reader experience in generic ways, ways that are typically overridden
with more information. We came to rely on one low-commitment coding category
we call specification. A word or string in this category adds an information unit
(in the sense of information theory, a possible reduction in uncertainty) to the
reader's experience. However, this string coding was used as a default category,
the assignment for stings for which no other more committed codings seemed to
apply.

5.2 Heuristic 2: Prefer the Conservative

This heuristic applies when the surface stream is ambiguous between two or
more definite codings and both can apply in the context of the coded text.
Conservative codings are defined as codings that make the fewest assumptions
about the surrounding text in which the string appears. For example, while
coding a mystery story, a coder came across the string “her hands trembled.”
The larger context was that they were trembling with fear. The coder could
have coded this string as a negative affect string, indicating fear. But that coding
would have missed the effect the writer was most after when describing symptoms
of Parkinson's disease in a medical journal. The idea of motion as a coding
category remains a relevant description of trembling both in the mystery and
medical context. It is the more conservative coding and so the preferred coding
according to heuristic 2.

5.3 Heuristic 3: Maximize Diversity Everywhere

Coders finding a conservative code for “her hands trembled” would not stop
there. Having established different contexts for this string, the coder, like the
curious writer, would rely on a vast productive knowledge to generate variations
that could distinguish unambiguously the mystery from the medical context.
“Her hands trembling in fear” would not likely survive the Parkinson's context.
Symptoms marked by her hands trembling departs from the mystery context.
The diversity principle enjoins that coding decisions that restrict coding lead to
the invention and classification of many new strings.