Efficient learning and maximal generalization: A new view of the early development of basic reading skills

Efficient learning and maximal generalization: A new view of the early development of basic reading skills

First Author: Matthew Cooper Borkenhagen -- University of Wisconsin - Madison
Additional authors/chairs: 
Christopher R. Cox; Mark Seidenberg
Keywords: Computational modelling, Transfer, Curriculum Development, Instruction, Word reading skills
Abstract / Summary: 

Beginning readers face the task of acquiring massive amounts of information about orthography, phonology, and their relationship early in reading development. Given limited opportunities to learn these mappings, development relies heavily on generalization: applying knowledge to previously unseen cases. We used computational models to examine how to optimize learning and generalization. Given a large set of words to learn, we examined which subsets of words yield maximal generalization to the remaining untrained words.

Using an orthography-to-phonology reading model, we selected word sets from a corpus of 3000 monosyllabic English words, manipulating the size of each curriculum of words in increments of 100 (n; min = 100, max = 1000), and randomly selecting words from training (100,000 randomly drawn sets of size n).

Capacity to generalize to unlearned words increases dramatically as a function of an increase in the size of the set (overall average increase of 21% per 100 additional words trained). These gains begin decreasing around sets of size 600, with an optimal training set size of between 400 and 600 words (14% and 21% of total corpus, respectively) given trade-offs between adding more words and the capacity to read words not yet taught. Qualities of sets that enhance generalization and how sets can be learned efficiently will be discussed.

Early instruction may benefit from evidence about how many, which, and in what order words can be taught to optimize learning and generalization. Our approach may also provide a framework for allocating limited instruction time in educational contexts.