Informativity of phonemic tiers: evidence from Dutch and English corpus analyses

Abstract

Both adults and children show an impressive statistical learning ability when segmenting continuous speech streams, and the Martian language experiment has proven itself an invaluable tool in better understanding this ability. Such experiments have shown clear differences in how our statistical learning ability attends to consonants and vowels in segmentation tasks. The current thesis aims to better understand this difference through measurement of the informativity of phonemic tiers. To this end, computational corpus analysis is employed to evaluate an entropy based informativity measurement of these tiers in Dutch and English. Results show that English and Dutch measure similarly, and add to the debate on an innate consonantal preference.