A corpus-based study on foreign-accented Italian

This study proposed an empirical approach for the analysis of foreign-accented Italian. We focused on six foreign accents – as compared to native varieties of Italian – the perceptual dimension of which was explored by means of corpus-based approaches and large-scale experiments. The results were interpreted qualitatively and quantitatively with descriptive and inferential statistics. Furthermore, various computational techniques were used to better explore non-native speech. The findings of this work display though-provoking results: some accents are categorized as more prestigious than others; in most films, and especially in the Disney cartoons, standard or regional varieties of Italian are used to voice the villains or other originally foreign-accented characters; speaking with a foreign accent may have implications at personal, social, communicative, and professional levels; Italian listeners understand when a speech sample is uttered by a non-native speaker, but they are not always able to identify the speaker’s mother-tongue; all other listener and speaker variables being comparable, the accentedness rating revealed that some accents were perceived as being more marked than others; additionally, spontaneous speech was rated as being less accented than read speech; native speakers of Italian firstly identify segmental features and although this does not necessarily mean that these cues contribute more to the perception of accentedness, we believe that some of them are accent-dependent while others are common to all or some of the six accents; the performance of the automatic accent identification systems used in this study is comparable with the performance achieved by native speakers of Italian and it reflected similar (mis)-classification trends; however, the unsupervised clustering methods revealed that, based on their prosodic characteristics, speakers are generally placed into 2-4 large clusters that do not necessarily reflect their L1s; finally, according to the chi-square and ReliefF scoring methods, the best ranked cues of foreign accent were energy, f0, word duration, and the duration of filled pauses.