Using statistical smoothing to estimate the psycholinguistic acceptability of novel phrases

The statistical technique of smoothing, which interpolates unknown values in a distribution, at once addresses twin problems faced by natural language processing (NLP) engineers and psycholinguists as a result of sparse data. The primary purpose of these data is to facilitate the exploration of smoothing as a psycholinguistic process responsible for the differential acceptability of nonoccurring phrases and to facilitate the comparison of smoothing algorithms in ways that will shed light on the aspects of their operation that are germane to the explanation of psycholinguistic acceptability. second, therefore, to illustrate the potential of the data set hi these respects, we present a short series of experiments that utilize a novel methodology for comparing several smoothing algorithms' capacity to predict acceptability judgments.
Matthew A J Roberts, Nick Chater. Behavior Research Methods. Austin: Feb 2008. Vol. 40, Iss. 1; pg. 84, 10 pgs
DOWNLOAD HERE