A CONNECTIONIST MODEL FOR BOOTSTRAP LEARNING OF SYLLABIC STRUCTURE

We report on a series of experiments with simple recurrent networks (S RNs) solving phoneme prediction in continuous phonemic data. The purpo se of the experiments is to investigate whether the network output cou ld function as a source for syllable boundary detection. We show that this is possible, using a generalisation of the network resembling the linguistic sonority principle. We argue that the primary generalisati on of the network, that is, the fact that sonority varies in a hat-sha ped way across phonemic strings, ending and starting at syllable bound aries, is an indication that sonority might be a major cue in discover ing the essential building bricks of language when confronted with uns egmented running speech. The segment which is most directly related to sonority patterns, the syllable, has received considerable attention in psycholinguistics as being an element of natural language that is e asily grasped by language learners. The phoneme prediction network pre sents a simulation of the necessary bootstrap to arrive at the discove ry of syllabic segmentation in unsegmented speech, which can be used a s a basis for the segmentation of larger structures like words.