In: Proceedings of AMLaP 2004. Extending Wundt’s Principle for incremental processing

Language production is a complex task, and despite limited resources, humans can speak fluently at a constant rate. The means to achieve this performance is incremental processing; i.e., utterances and sequences of utterances are not planned completely in advance but are generated on the fly. Particularly suited for incremental processing are cascaded architectures. They consist of processes that are arranged in a fixed sequence, so that the output of one process is the input to its successor. This parallel processing of a sequential information stream is, together with the accompanying strict unidirectionality of the information flow, characteristic of incremental processing. As a principle for cognitively adequate incremental processing Levelt (1989: 26) proposes Wundt's Principle: Each processing component will be triggered into activity by a minimal amount of its characteristic input. (Input/output is characteristic if a process can operate on the information, e.g. a phonological process cannot use syntactic information.) I propose the following extension: Each processing component will be triggered into activity by a minimal amount of its characteristic input and produces characteristic output as soon as a minimal amount of output is available. (Guhe 2003: 82) This extension is in contrast to other principles put forward in the literature. For example, it contradicts principles that reason about the ‘when-to-say’. In the context of a language production system that generates self-repairs Kilger & Finkler (1995: 9), for example, state: ‘In order to both fulfill the time constraints and avoid overt repair incremental output production should be used to utter succeeding parts of the sentence [i.e. after the initial output increment] as soon as necessary […] rather than uttering them as soon as possible. ’ This means, for improving the quality of the output and for reducing the number of self-repairs th