Phonotactic probability and neighborhood density in visual and

spoken word recognition: Psychophysical and MEG evidence.

 

Diogo A. de Azevedo e Almeida (1), Andrew Nevins (2) & David

Poeppel (1,3)

 

1 CNL Lab, Department of Linguistics, University of Maryland,

College Park

2 Department of Linguistics, Harvard University

3 Department of Biology, University of Maryland, College Park

 

Some recent models of word recognition distinguish between (at

least) two different processing levels, one involving lexical

and the other sublexical information. Behavioral data point to

a complex interaction between the putative levels that appears

to be modulated by task demands. For instance, processing at

these two levels can have opposite effects on behavioral

responses depending on the task (Vitevich & Luce 1999). The

present study seeks to fractionate lexical processing by using

magnetoencephalography (MEG) to investigate the time-course of

word recognition in both auditory and visual lexical decision

tasks. The goal of the study was twofold: i) to isolate evoked

MEG responses indexing distinct processing stages and ii) to

investigate how modality-dependent each processing level is.

We constructed a set of CV.CVC words and non-words that had

either high phonotactic probabilty and dense lexical

neighborhoods or low phonotactic probabilty and sparse lexical

neighborhoods. The materials were controlled across a large

number of possible metrics, in both their phonetic and

orthographic forms. We predicted that effects of sublexical

processing should be observable in early electrophysiological

responses (~150-350 ms), whereas effects of lexical

competition might be observable in later evoked responses

(~400 ms) as well as at the behavioral level. Preliminary MEG

data suggest facilitatory effects for

low-probability/sparse-neighborhood items, regardless of

lexicality and modality of presentation. Psychophysical data

suggest facilitatory effects for

low-probability/sparse-neighborhood non-words, but not for

words, regardless of modality of presentation. The data

implicate two distinct levels of analysis.

 

Supported by NIH-DC 05660 to DP.