260 likes | 404 Views
The Role of Phonological Distance and Relative Support on the Productivity of the Dutch Simple Past Tense. Bram Vandekerckhove, Emmanuel Keuleers, & Dominiek Sandra Center for Psycholinguistics University of Antwerp. Wilf ing on the web, the new British pastime Wilf t u wel eens?
E N D
The Role of Phonological Distance and Relative Support on the Productivity of the Dutch Simple Past Tense Bram Vandekerckhove, Emmanuel Keuleers, & Dominiek Sandra Center for Psycholinguistics University of Antwerp
Wilfing on the web, the new British pastime Wilft u wel eens? Maar liefst zeven op de tien mensen erkennen dat ze wel eens 'wilfen’ dus dan moet er gewilft worden
How do we inflect novel forms? • Dual-Mechanism Model • Regular inflection: • Governed by rules that operate over abstract symbols • Fully productive • Default • Irregular inflection: • Database of stored word forms • Limited productivity • Similarity-based analogies • Blocks the default V V suffix -edpast wilf
How do we inflect novel forms? • Similarity-based single-mechanism models • Regular and irregular inflection: • Analogy-based productivity kill => killed will => would wilf =>? delve => delved fill => filled
Prasada & Pinker 1993 • S. Prasada & S. Pinker (1993). Generalisation of Regular and Irregular Morphological Patterns. Language and Cognitive Processes 8, 1-56. • Past tense acceptability ratings • Rate on a scale from 1 to 7 • Today, I spling, yesterday I splinged • Today, I spling, yesterday I splung • Today, I plip, yesterday I plipped • Today, I plip, yesterday I plup
Prasada & Pinker 1993, stimuli • Type: distance from irregular/regular verbs • Distances: Prototypical/Intermediate/Distant • Pattern: Vowel change/suffix
Prasada & Pinker 1993: Predictions • Dual-Mechanism Model • Irregular inflection: • Negative effect of Phonological Distance from existing irregular verbs • Regular inflection: • No effect of Phonological Distance from existing regular verbs • Analogy-based models • Regular and irregular inflection: • According to Prasada & Pinker (1993) (!): Effect of Phonological Distance from existing regular and irregular verbs
However ... Is a similarity-based model really bound to fail in strong dissimilarity conditions?
Memory-Based Modelling spring cling spling swing sting
Memory-Based Modelling fill will wilf fill kill
Keuleers & Sandra (submitted). Similarity and Productivity in the English Past Tense.
Implications • Findings of Prasada & Pinker can be explained as the result of Relative Support effects in a k-nn model. • The question whether Phonological Distance from existing verbs in itself has a negative effect on regular/irregular productivity is still unresolved.
A new experiment • Investigation of the effects of Phonological Distance and Relative Support on the productivity of regular and irregular patterns of the Dutch simple past tense.
Method • Past tense acceptability rating task • Stimuli selected on the basis of the output of a memory-based simulation, with model trained on monomorphemic verbs (CELEX): • Phonological Distance of the nearest neighbor • Relative Support for one of 6 past tense classes (3 regular allomorphs, 3 largest vowel-change classes)
Predictions • Dual-Mechanism Model • Regular inflection • No effect of Phonological Distance • Possible postive effect of Relative Support (partial blocking) • Irregular inflection • Negative effect of Phonological Distance • Positive effect of Relative Support • K-nn model • Regular and irregular inflection • Positive effect of Relative Support • No effect of Phonological Distance
Results and Discussion • Positive effect of Relative Support on mean ratings for both regular and irregular forms • Negative effect of Phonological Distance on mean ratings for irregular forms • No significant effect of Phonological Distance on mean ratings for regular forms
Results and discussion • Regular inflection • Both models make the correct predictions. • Preference for k-nn model. • Irregular inflection • K-nn model makes the wrong prediction. • Preference for Dual-Mechanism Model?
An alternative explanation • K-nn model is a rather crude implementation of the memory-based approach to cognitive/linguistic modelling. • Generalized Context Model: (Nosofsky, R. M. (1988). Similarity, frequency and category representations. Journal of Experimental Psychology: Learning, Memory and Cognition, 14, 54-65.): • k nearest neighbors = the whole lexicon • Exemplars are weighted by an exponential decay function
An alternative explanation k1 k7 k1 k7 k1 k7
An alternative explanation • In the k-nn model, irregular nearest neighbors that had a relatively high Phonological Distance from the test item might have been too influential in comparison with items that were somewhat further away • We have not yet explored this possibility.
Conclusions • A k-nn model is very good model of regular productivity. • At the moment, a dual-mechanism model seems to account best for the productivity of the Dutch simple past tense. • However, a more sophisticated implementation of the memory-based approach might be a better model for this pattern of results.