RKCP appears to be significantly more successful than other simpler poetry generation
programs. It often generates rich poetic language and some reasonable fraction of its
output appear to be at least coherent poems. Because of the language modeling and
recursive poetry generation techniques used by RKCP, there is significant
"linkage" between all the words in a given generated poem. The goals used by
RKCP are "stronger" in terms of local linkages, that is between words or lines
that are relatively near each other. Thus RKCP is particularly successful in writing
shorter poems such as Haiku although longer poems are often successful as well. Additional
modeling techniques have been added that facilitate writing longer poems with thematic
RKCP is most effective at producing interesting lines of poetry and turns of phrase,
and is less effective in maintaining coherency through an entire poem. This strength is
exploited in the "poets assistant" authoring tool. The purpose of this
tool is not primarily to write entire poems, but rather to assist and stimulate a (human)
poet in finding the right verbal images and phrases. These results are often intriguing
Most other attempts at computer written poetry have used relatively simple algorithms
involving straightforward combinations of randomly chosen words, or "templates"
in which a word is randomly chosen from various lists for each point in a structured poem.
RKCPs techniques are significantly more sophisticated and thus its results are more
varied and more interesting than those produced by these simpler approaches.