ABSTRACT
Recent work has shown that parsing processes can be influenced by detailed information recovered from the lexicon (e.g. the subcategorisation class of verbs). However, it is not clear exactly how this effect is mediated. There has been an increasing tendency to distinguish between parsing submechanisms responsible for proposing new structures and those that are mainly concerned with filtering or checking the structures generated in this way. An experiment was therefore carried out to determine whether lexical effects exert their major influence during the first or second of these phases of parsing. The results indicated that the primary effect occurs at the second (checking) stage. Indeed, the data suggest that the first processor, using “rough and ready” procedures, may sometimes set about its task by proposing (and assembling words into) structures that are actually illegitimate, leaving the second processor to check the details, note the inconsistencies, and trigger the process of revising the structural interpretation of the sentence. The results of the experiment also throw light on the effects of text format or layout and suggest that parafoveal information may play a role in guiding the parsing process. The data are discussed in terms of a general model of parsing.
