Poetry Generation with PoeTryMe - Computational Creativity Research: Towards Creative Machines - page 243

12.4 An Incremental Approach to Poetry Generation
in Portuguese

Using PoeTryMe's platform, we devised a plan for the automatic generation of poetry,
in Portuguese. This section starts by presenting the lexical resources we used together
with PoeTryMe for this task. Moreover, given all the features that add value to this
kind of creative text, we present our approach to poetry generation step-by-step.
Instead of tackling all features at once, we describe how we coped with each feature,
towards our main goal. Different approaches, targeting different features, originated
different generation strategies, also described in this section.

12.4.1 Lexical Resources

CARTÃO [ 14 ] is a public lexical knowledge base for Portuguese, extracted auto-
matically from three Portuguese dictionaries. 2 It contains about 330,000 semantic
relation instances, held between words, which establish a semantic graph. A rela-
tion instance, represented as , indicates that one sense

of the word in the first argument ( arg1 ) is related to one sense of the word in the
second ( arg2 ), by means of a relation identified by PREDICATE. For example:

< arg1 PREDICATE arg2 >

animal HIPERONIMO_DE c ˜ a o

( animal HYPERNYM_OF dog )

CARTÃO includes relations as synonymy, hypernymy, part-of, causation, purpose
and property-of, amongst others. The name of the semantic relation also defines the
part-of-speech (POS) of its arguments.
SilabasPT 3 is a tool for syllabic division and stress identification in Portuguese.
It was developed for generating text based on rhythm, in the scope of Tra-la-Lyrics
[ 16 , 17 ], but it is independent, and can be integrated with other applications. Our
implementation of the Syllables Util module, used for evaluating poem lines accord-
ing to their size, relies on SilabasPT.
LABEL-LEX 4 is a lexicon of Portuguese, with 1,5 million inflected word forms,
automatically generated from about 120,000 lemmas. For each form, it provides
information such as the lemma, the POS and other morphological information. It
was used for handling the inflection of nouns and adjectives (number and gender).

2 CARTÃO is available from http://ontopt.dei.uc.pt/index.php?sec=downloads .
3 SilabasPT is available from http://code.google.com/p/silabaspt/ .
4 LABEL-LEX is available from http://label.ist.utl.pt/pt/labellex_pt.php .