Quantitative determinants of prefabs: A corpus-based, experimental study of multiword units in the lexicon

In recent years many researchers have been rethinking the Words and Rules' model of syntax (Pinker 1999), instead arguing that language processing relies on a large number of preassembled multiword units, or 'prefabs' (Bolinger 1976). A usage-based perspective predicts that linguistic units, including prefabs, arise via repeated use, and prefabs should thus be associated with the frequency with which words co-occur (Langacker 1987). Indeed, in several recent experiments, corpus analysis is found to be associated with behavioral measures for multiword sequences (Kapatsinski and Radicke 2009, Ellis and Simpson-Vlach 2009). This dissertation supplements such findings with two new psycholinguistic investigations of prefabs. Study 1 revisits a dictation experiment by Schmitt et al. (2004), in which participants are asked to listen to stretches of speech and repeat the input verbatim, after performing a distractor task intended to encourage reliance on prefabs. I describe the results of an updated experiment which demonstrates that participants are less likely to interrupt or partially alter high-frequency multiword sequences. Although the original study by Schmitt et al. (2004) reported null findings, the revised methodology suggests that frequency indeed plays a role in the creation of prefabs. Study 2 investigates the distribution of affix positioning errors (he go aheads) which give evidence that some multiword sequences (e.g., go ahead) are retrieved from memory as a unit. As part of this study, I describe a novel methodology which elicits the errors of interest in an experimental setting. Errors evincing holistic retrieval are induced more often among multiword sequences that are high in Mutual Dependency, a corpus measure that weighs a sequence's frequency against the frequencies of its component words. Followup analyses indicate that sequence frequency is positively associated with affix errors, but only if component-word frequencies are included as variables in the model. In sum, the studies in this dissertation provide evidence that prefabricated, multiword units are associated with high frequency of a sequence, in addition to statistical measures that take component words' frequency into account. These findings provide further support for a usage-based model of the lexicon, in which linguistic units are both gradient and changeable with experience.

Name thesis:degree_name

Linguistics

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Department of Linguistics

Year

2013

Author dc:creator

• Beckner, Clayton

Contributors dc:contributor

• Morford, Jill
• Bybee, Joan
• Croft, William
• Wedel, Andrew

• psycholinguistics
• usage-based theory
• holistic processing
• multiword units
• mental lexicon
• speech errors

Language dc:language

English

Repository record dc:identifier

https://digitalrepository.unm.edu/ling_etds/3

OAI identifier oai:identifier

oai:digitalrepository.unm.edu:ling_etds-1002