Poster B32, Thursday, August 16, 3:05 – 4:50 pm, Room 2000AB
Do experiential semantic features predict automatic semantic priming in verbs?
Taylor Kalmus1, Lisa Conant2, Colin Humphries2, Jeffrey Binder2, Leonardo Fernandino2;1Carroll University, 2Medical College of Wisconsin
Prior research established that a model of concept representation based on a set of experiential attributes (CREA) predicts automatic semantic priming effects for nouns (Fernandino et al., 2016). Here, we assessed whether this effect is also present for verbs and whether the magnitude of the effect is similar for the two-word classes. We hypothesized that the magnitude of the semantic priming effect in a lexical decision task would correlate with a measure of semantic similarity between the prime and the target word derived from the CREA model. As a base for comparison, we also evaluated an alternative model of concept representation, Latent Semantic Analysis (LSA, Landauer & Dumais, 1997). In the noun condition, targets consisted of English nouns or matched pseudowords, with nouns used as primes. In the verb condition, targets consisted of English verbs or matched pseudowords, with verbs as primes. Nouns and verbs were presented in separate blocks. All words were previously rated on a set of 65 semantic features through Amazon Mechanical Turk (Binder et al., 2016). Words and pseudowords were matched for letter length, bigram and trigram frequency, and orthographic neighborhood size. Primes and targets were matched for word length. The possibility for forward association priming effects was eliminated by excluding any pairs with nonzero values in the USF Free Association Norms. For each model (CREA and LSA), semantic similarity between prime and target was estimated as the cosine between the vector representations of the two words. Prime-target pairs represented a continuous distribution of word similarity values ranging from 0 to 1. Each target word was presented twice (on separate days), each time preceded by a different prime. The priming effect for a given target word was calculated as the difference in response times between the two presentations. Each trial started with a central fixation cross (duration jittered 1-2 sec), followed by the prime (150 ms), a mask (hash marks, 50 ms), and the target (2 sec). The prime was presented in lowercase and the target in uppercase letters. Participants were instructed to ignore the prime and make a speeded lexical decision on the target. The task was performed in two sessions approximately one week apart. Twenty-six subjects completed the study. Results replicated the findings of Fernandino et al. (2016), showing a highly significant correlation between CREA cosine differences and priming for nouns (r=.45, p<.001). However, the correlation for verbs was much smaller (r=.18), not reaching significance across trials (p>.1), although it was marginally significant across participants (p=.06). Furthermore, there was a significant Model x Word Category interaction, in which priming for nouns was better predicted by CREA than by LSA while priming for verbs was better predicted by LSA than by CREA (p=.023). Correlation between CREA cosine differences and priming was significantly lower for verbs than for nouns (p= .05). These results show that, while ratings of experiential features strongly predict automatic semantic priming for nouns, their predictive power is considerably smaller for verbs, possibly due to the higher context-dependency of verb meaning.
Topic Area: Meaning: Lexical Semantics