Ode to a Keatsian Turn: Creating Meaningful and Poetic Instances of Rhetorical Forms - Computational Creativity Research: Towards Creative Machines - page 205

Consider this poem, elaborated around the metaphor China is a rival :

No Rival Is More Bitterly Determined

Inspire me with your determined battle
The most dogged defender scarcely struggles so much
Stir me with your spirited challenge
Let your competitive threat reward me
Was ever a treaty negotiated by a more competitive rival?
You compete with me like a competitively determined athlete
Does any rival test more competitively than this China?
You oppose me with your bitter battle
Can a bitter rival suffer from such sweet jealousies?
O China, you oppress me with your hated fighting

Stereotypes are most eye-catching when subverted, as in the second-last line
above. The Google 2-gram “ sweet jealousies ” catches Stereotrope 's eye (and ours)
because it up-ends the belief that jealousy is a bitter emotion. This subversion comple-
ments the stereotype that rivals are bitter, allowing Stereotrope to impose a thought-
provoking opposition onto the banal norm rivals suffer from jealousy .
Stereotype emphasizes meaning and intent over sound and form, and does not
(yet) choose lines for their rhyme or metre. However, given a choice of renderings,
it does choose the form that makes best use of the Keats heuristic, by favoring lines
with alliteration and internal symmetry.

10.8 Quality Considerations

Stereotrope is a knowledge-based approach to poetry, one that crucially relies on
three sources of inspiration: a large roster of stereotypes, which maps a slew of
familiar ideas to their most salient properties; a large body of normative relation-
ships which relate these stereotypes to each other; and the Google n-grams, a vast
body of language snippets. The first two are derived from attested language use on the
Web, while the third is a reduced view of the linguistic Web itself. Stereotrope repre-
sents approx. 10,000 stereotypes in terms of approx. 75,000 stereotype-to-property
mappings, where each of these is supported by a real Web simile that attests to the
accepted salience of a given property. In addition, Stereotrope represents over 50,000
norms, each derived from a presupposition-laden question on the Web.
The reliability of Stereotrope 's knowledge has been demonstrated in recent stud-
ies. For instance, [ 23 ] shows that Stereotrope 's simile-derived representations are
balanced and unbiased, as the positive/negative affect of a stereotype T can be reli-
ably estimated as a function of the affect of the contents of typical ( T ) . In addition,

[ 24 ] further shows that typical can be reliably partitioned into sets of positive

or negative properties as needed, to reflect an affective “ spin ” imposed by any given

( T )