CNL Lunch

Henk J. Haarmann & Orsi Kraut

Verbal Short-term Memory
for Noun-Verb Role bindings

Thursday April 25th, 12:30pm, 3416 Marie Mount Hall

     It is often assumed that our verbal short-term memory (STM) for words improves dramatically when words are presented in sentences instead of in lists of unconnected items, due to the chunking of words into higher-level phrases and propositions in sentences. Since by definition the internal structure of a chunk does not contribute to its memory load, this view predicts that a noun and verb bound by a thematic role (e.g., girl is-agent-of read) impose a memory load of only one item instead of two items and will thus lead to a two-fold increase in the storage capacity of verbal STM. We tested this prediction in two experiments with college-age adults, presenting a sequence of noun-verb pairs and lexically matched noun-noun pairs for immediate semantic- (Experiment 1) and order-cued recall (Experiment 2).

     The results from both experiments were virtually identical and did not support an extreme version of chunking. Instead, noun-verb role bindings increased the storage capacity only slightly compared to pairs of lexically matched unintegrated nouns and showed an almost identical forgetting function, displaying a recency effect. Moreover, we found an interaction between memory load (number of items) and serial position (of the probed item) for noun-verb and noun-noun pairs, suggesting displacement -based (and not decay-based) forgetting, which we had previously demonstrated in semantic STM for unintegrated word meanings (Haarmann & Usher, 2001).

     We will discuss boundary conditions for the chunking view and implications for current neurocomputational models of verbal STM for noun-verb role bindings proposed for sentence comprehension (Haarmann, Just, & Carpenter, 1997) and analogical reasoning (Hummel & Holyoak, 1997).