Semantic similarity computation for abstract and concrete nouns using network-based distributional semantic models

Motivated by cognitive lexical models, network-based distributional semantic models (DSMs) were proposed in [Iosif and Potamianos (2013)] and were shown to achieve state-of-the-art performance on semantic similarity tasks. Based on evidence for cognitive organization of concepts based on degree of concreteness, we investigate the performance and organization of network DSMs for abstract vs. concrete nouns. Results show a "concreteness effect" for semantic similarity estimation. Network DSMs that implement the maximum sense similarity assumption perform best for concrete nouns, while attributional network DSMs perform best for abstract nouns. The performance of metrics is evaluated against human similarity ratings on an English and a Greek corpus.