Garden Path Repair: Diagnosis and Triage

Abstract

Meng and Bader(2000b)have shown that for garden-path repair in German, case is a more effective cue than number. They argue that sensitivity to the nature of the cue supports a diagnosis model of garden path repair such as we have proposed. However, in making this argument Meng and Bader introduced a new notion of diagnosis. Retaining the original "basic" diagnosis system,they added a new function that we call triage. Triage determines the probable revisability of a structure, in order to decide whether to make the effort of trying to reanalyze it. To determine whether triage is a feature of human parsing, its scope must be established. We compare four hypotheses about how much work triage could do. A powerful triage component would succeed where the human parser fails. Limited $F_local triage does no work if symptom effectiveness varies within a $F_language. An intermediate hypothesis discriminates better between easy and difficult revisions. However, this variant of triage with the strongest claim to psychological reality is the most difficult to distinguish from no triage at all, since it shares important properties with basic diagnosis. We identify some empirical predictions that differentiate diagnosis with triage from simple basic diagnosis. What $F_little evidence there is at present suggests that the human sentence parser has no triage capability.