The merge model: Speech perception is bottom-up
Dennis Graham Norris · The Journal of the Acoustical Society of America · 1999
Does feedback from the lexicon influence the processing of phonemic or phonetic information in normal speech perception? Arguments are presented that top-down feedback from the lexicon is no value in spoken word recognition. In fact, top-down feedback is as likely to be harmful as helpful. Consistent with this analysis, recent studies of phoneme identification and phonetic categorisation have generated data that are incompatible with interactive models like trace [McClelland and Elman (1986)] which employ top-down feedback. However, other data rule out bottom-up models like the Race model of Cutler and Norris (1979). A new, bottom-up model is presented, the Merge model. Merge can simulate these problematic results by merging information from both lexical and sublexical analyses while maintaining a strictly feed-forward data path for spoken word recognition. Merge builds on the principles of lexical competition and bottom-up processing incorporated in the Shortlist model [Norris (1994)].