Abstract
Covariation among words is certainly related to meaning, meaning similarity, and psychological processing. We argue, however, that the causal arrow is from meaning (and meaning similarity) to covariation, not vice versa. Consequently, covariation is not meaning, it is unlikely to provide an accurate metric for similarity of meanings, and embodied learning mechanisms, rather than computation of statistics, underlie effects of covariation on psychological processing. We report the results from two experiments that provide the first empirical test of the strong covariation claim that meaning can be derived from covariation structure. In the experiments, people studied the covariation among unnamed features taken from a familiar domain. In the first experiment, after learning the covariation structure of the features, participants were unable to choose the correct domain on a forced choice test, and they were unable to use the learned structure to grossly classify unnamed features even after the domain and majority of features were named. In the second experiment, the majority of the features was named during the study of the covariance structure. Nonetheless, participants were unable to use the learned structure to classify the few remaining unnamed features. Thus, contrary to the strong covariance claim, covariance structure alone is not particularly useful for deriving meaning.
Original language | English (US) |
---|---|
Pages (from-to) | 241-264 |
Number of pages | 24 |
Journal | Italian Journal of Linguistics |
Volume | 20 |
Issue number | 1 |
State | Published - 2008 |
Keywords
- Connectionism
- Covariation
- Distribution hypothesis
- Embodiment
- Meaning
ASJC Scopus subject areas
- Language and Linguistics
- Linguistics and Language