Czech Verse Processing System KVĚTA – Phonetic and Metrical Components

The following paper describes the algorithms of phonetic and metrical components of the Czech verse processing system KVĚTA, updating information contained in previous reports (Ibrahim and Plecháč 2011; Plecháč et al. 2013a; Ibrahim and Plecháč 2014). The system is being used in the building of the Corpus of Czech Verse (hereinafter CCV), which at present contains 1 689 Czech books of poetry (over 2.5 million lines) from the nineteenth and early twentieth centuries. In contrast to standard language corpora, in each lexical unit are not only the lemma and morphological tag attributes assigned but they also contain a phonetic transcription; furthermore, the attributes metre (iamb, trochee…), length (number of feet), ending (feminine, masculine…) and metrical pattern are assigned to each verse line. At higher levels rhyme pairs (or n -some) and fixed forms (sonnet, rondel, etc.) are annotated. Here we will focus on components providing phonetic and metrical annotation: (1) the F-component, whose task is to derive the phonetic transcription from the input data, (2) the G-component, whose task is to generate a set of all possible metrical interpretations of these data, and (3) the M-component, whose task is to select from this set the final interpretation. Automatic analysis has so far been limited to accentual-syllabic (hereinafter AS) and monometric poems – i. e., poems consisting of repetitions and variations of a single metrical pattern (though AS imitations of some quantitative meters are recognized).