CrossTowns: Automatically Generated Phonetic Lexicons of Cross-Lingual Pronunciation Variants of European City Names

Institut für Kommunikationsakustik (IKA) , Ruhr-Universität Bochum , D-44780 Bochum, Germany , stefan.schaden@rub.de

The CrossTowns lexicons are part of a study that focuses on the phonetic variants that occur when speakers of different native languages (L1) with varying degrees of target language (L2) proficiency pronounce foreign city names. Based on a collection of speech data from this domain, it is one of the aims to identify the most common pronunciation errors in a particular L1/L2 pair (language direction) and to model them by phonological rewrite rules. Although derived from only a small corpus of names, the rule sets already generate plausible variants when applied to unseen material. Yet there is a need for improvement. To demonstrate the current state of affairs, sample lexicons of 1.000 place names for English, French, and German were compiled and converted into various interlanguage pronunciation lexicons using the accent rule sets. In the paper, the procedures involved in the data collection, an outline of the rule-based accent generation technique, and a discussion of the problems involved in modelling non-native pronunciations on the lexicon level will be presented.