Grounding Abstract Word Definitions In Prior Concrete Experience

Description/Abstract

Longman’s Dictionary of Contemporary English (LDOCE) and Cambridge International Dictionary of English (CIDE) have a defining vocabulary of 2000 words from which most but not all of the words are defined. We developed an algorithm, applicable to any dictionary, for recursively deleting words not used in any definition until the corpus is reduced to a subset – a grounding kernel – from which all the words of the complete dictionary are reachable by definition alone. We compared LDOCE’s and CIDE’s defining vocabulary (DV) and grounding kernel (GK) against the rest of its words, on (1) concreteness, (2) imagery and (3) age of acquisition based on the MRC psycholinguistic database [Wilson 1988]. Both GK and DV proved significantly higher on all three scales (p<0.001) for all but 2 of the 12 comparisons (no difference in concreteness in LDOCE for either GK or DV). The difference was also consistently greater using our automatically computed GK than using the DV defined by the compilers of LDOCE and CIDE.