Running head: ASSOCIATIVE ASYMMETRY OF COMPOUND WORDS 1 Associative asymmetry of compound words

Early verbal-memory researchers assumed participants represent memory of a pairs of unrelated items with two independent, separately modifiable, directional associations. However, memory for pairs of unrelated words (A–B) exhibits associative symmetry: a near-perfect correlation between accuracy on forward (A→?) and backward (?←B) cued recall. This was viewed arguing against the independent-associations hypothesis, and in favor of the hypothesis that associations are remembered as holistic units. Here we test the Holistic Representation hypothesis further by examining cued recall of compound words. If we suppose pre-existing words are more unitized than novel associations, the Holistic Representation hypothesis predicts compound words (e.g., ROSE BUD) will have a higher forward–backward correlation than novel compounds (e.g., BRIEF TAX). We report the opposite finding: compound words, as well as non-compound words, exhibited less associative symmetry than novel compounds. This challenges the Holistic Representation account of associative symmetry. Moreover, pre-experimental associates (positional family size) influenced associative symmetry— but asymmetrically: increasing family size of the last constituent increasing decoupled forward and backward recall, but family size of the first constituent had no such effect. In short, highly practised, meaningful associations exhibit associative asymmetry, suggesting associative symmetry is not diagnostic of holistic representations, but rather, is a characteristic of ad-hoc associations. With additional learning, symmetric associations may be replaced by directional, independently modifiable associations as verbal associations become embedded within a rich knowledge structure.