ABSTRACT
In this chapter, we have presented a number of simulation results that demonstrate how multiple-cue integration in a connectionist network, such as the SRN, can provide a solid basis for solving the speech segmentation problem. We have also discussed how the process of integrating multiple cues may facilitate learning, and have reviewed evidence for the existence of a plethora of probabilistic cues for the learning of word meanings, grammatical class, and syntactic structure. We conclude by drawing attention to the kind of learning mechanism needed for multiple-cue integration.
