Poetry Generation with PoeTryMe - Computational Creativity Research: Towards Creative Machines - page 240

Fig. 12.1 PoeTryMe's architecture

#haiku
estrofe { verso(5);verso(7);verso(5) }

#sonnet
estrofe { verso(10:A);verso(10:B);verso(10:B);verso(10:A) }

estrofe { verso(10:A);verso(10:B);verso(10:B);verso(10:A) }

estrofe { verso(10:C);verso(10:D);verso(10:C) }

estrofe { verso(10:D);verso(10:C);verso(10:D) }

Fig. 12.2 Templates with the structure of a haiku and a sonnet with a rhyme pattern

12.3.1 Generation Strategies

A Generation Strategy organises sentences according to some heuristics, such that
they suit, as much as possible, a target template of a poetic form, and that they exhibit
certain features. A poem template contains the poem's structure, including the num-
ber of stanzas, the number of lines per stanza and the number of syllables of each
line. Some templates might as well use a symbol for denoting the target rhyme for
each line. Figure 12.2 shows the representation of poem structure templates, for gen-
erating a haiku (5-7-5) and a sonnet (14*10-syllable verses). There, the Portuguese
word estrofe indicates a stanza and verso indicates a line. There is no rhyme pattern
specified for the haiku, but each line of the sonnet has a symbol that results in the
following rhyme pattern: ABBA ABBA CDC DCD .
Each strategy uses the Sentence Generator module to retrieve natural language
sentences, which might be selected as poem lines. For each generation, a set of seed
words may be provided and used to narrow the set of possible generations, this way
defining the generation domain. To evaluate the suitability of a sentence towards a