Initial Experiments with Multilingual Extraction of Rhetoric Figures by means of PERL-compatible Regular Expressions

A language-independent method of figure-ofspeech extraction is proposed in order to reinforce rhetoric-oriented considerations in natural language processing studies. The method is based upon a translation of a canonical form of repetition-based figures of speech into the language of PERL-compatible regular expressions. Anadiplosis, anaphora, antimetabole figures were translated into the form exploiting the backreference properties of PERL-compatible regular expression while epiphora was translated into a formula exploiting recursive properties of this very concise artificial language. These four figures alone matched more than 7000 strings when applied on dramatic and poetic corpora written in English, French, German and Latin. Possible usages varying from stylometric evaluation of translation quality of poetic works to more complex problem of semi-supervised figure of speech induction are briefly discussed.