Predicting concreteness and perceivability

C.W.J. van Miltenburg

Research output: Contribution to ConferenceAbstractOther research output

Abstract

Concreteness ratings are often used as a measure of readability. There have been several attempts to build a model that accurately predicts human judgments of concreteness, but none of them incorporate a measure of sensory perceivability. This is striking because of two reasons: (1) it is a salient aspect of concrete terms that they tend to denote objects that are more directly experienced, (2) recent literature shows that sensory perceivability is a strong predictor of readability.

We created a model to predict concreteness as well as sensory perceivability ratings. We looked at factors common in the literature, and in addition our model is enriched with corpus data indicating:
* The relative frequency of perception related modifiers (e.g. blue, tasty, rough) occurring with each noun. Perception-relatedness was determined using WordNet mappings from the SUMO ontology.
* The relative frequency of kind-level modifiers (e.g. academic, vegetarian, military) occurring with each noun. These modifiers show us that the noun involved is a higher-level noun that can have subclasses. Whether a modifier is a kind-level modifier is determined on the basis of their morphology (ending in –ic, -ary, -ian).

We show the performance of our model on human ratings, and discuss some hidden assumptions behind recent studies. For example: Feng et al.'s (2011) model assumes that concreteness ratings are based on all senses of a word. Is this true, or do participants in concreteness rating studies base their judgments only on the predominant sense?
Original languageEnglish
Publication statusPublished - 2015
EventCLIN 25 -
Duration: 6 Feb 20156 Feb 2015

Conference

ConferenceCLIN 25
Period6/02/156/02/15

Fingerprint Dive into the research topics of 'Predicting concreteness and perceivability'. Together they form a unique fingerprint.

Cite this