Lydskrivning over landegrænser

En studie i interskandinavisk genbrug af sprogteknologi

Resumé

Danish, Norwegian, and Swedish phonetics are in fact so similar that trilateral transfer of lexical and phonetic resources seems a linguistically well-motivated enterprise. Not only is the overlap of the three phone inventories substantial, the toneme systems of Swedish and Norwegian correspond closely to the Danish stød (an instance of creaky voice), and the principles of stress assignment are almost identical. These conditions suggest that a single Scandinavian phonetics accompanied by a small number of language-specific transformations could be effective for several purposes, linguistic as well as technological. In this paper we present a pilot experiment in transferability. We produced a phonetic transcription of the 900k word Norwegian speech corpus NoTa using an extensive phonetic database for Danish as a catalyst. The core components of the transfer system were the two algorithms NO2DO (Norwegian Orthography to Danish Orthography) and DP2NP (Danish Phonetics to Norwegian Phonetics) developed with standard machine learning techniques.