The evidence add ups
An affix shift study of prefabs
In a usage-based model of the lexicon, linguistic elements that repeatedly
co-occur in sequence come to form a processing unit (Bybee 2002b). The present study supplements
previous psycholinguistic research finding that multiword sequences may form
a ‘prefabricated’ sequence. I describe a new experiment designed to elicit
‘affix shift’ speech errors (e.g., adds up →
add ups), which give evidence of holistic processing of
a multiword sequence. Analysis of the data indicates that affix positioning
errors are predicted by a confluence of factors – a high-frequency
stimulus sequence (settle down) in tandem with
low-frequency component words (settle,
down). These findings provide support for a usage-based
account, in which linguistic units are not fixed, but gradient and
changeable with experience.
Article outline
- 1.Introduction
- 1.1Errors of interest, and theoretical background
- 1.1.1Full affix shifts
- 1.1.2Double-marked affixes
- 1.1.3No-marking errors
- 1.2Predictions, and quantitative corpus metrics
- 2.Task design
- 2.1Materials and stimulus design
- 2.1.1Frequency x Mutual Dependency bins
-
2.1.2Bigram features matched across bins
- 2.1.3Bigram stimuli and distractors
-
2.2Participants and experiment setup
- 2.3Results and discussion: Affix shifts and other affixation errors on
bigram stimuli
- 2.3.1Participant accuracy
- 2.3.2Overview of errors: Affix shifts, double-marking errors,
no-marking errors
- 3.Data and analysis
-
3.1Post hoc analysis: Examining components of the MD metric
- 4.Discussion and conclusion
-
Notes
-
References