Do dominance effects for polysemous words depend on sense probabilities?

Abstract

Polysemous words have multiple related senses and represent one of the most widespread phenomena in language (Rodd et al., 2004). When presented in isolation in a word recognition task (e.g. visual lexical decision task), such words are typically processed faster than unambiguous words due to sense relatedness facilitating the activation of all senses. However, when presenting a word in a task that demands activation of one particular sense, or within a context that primes one of the senses, the processing becomes slower (Armstrong & Plaut, 2016), as a restriction to one sense occurs, leading to competition between senses. This increase in processing time depends on sense probability (dominance effect). Priming the most probable (dominant) sense leads to a smaller processing lag, compared to priming the less probable (subordinate) sense. We hypothesise that this effect will not depend solely on the dominance status of the primed sense but on the full sense probability distribution described by entropy (Filipović Đurđević & Kostić, 2021). If a polysemous word is presented within a neutral context, we expect a facilitatory entropy effect as if it were presented in isolation. Priming the dominant sense was expected to annul the entropy effect, as low competition occurs. Priming the subordinate sense was expected to lead to an inversion of the entropy effect. To test the cross between dominance and entropy effects we presented each of the 102 Serbian polysemous nouns in three sentences. One where the context was neutral regarding any of the senses, one sentence priming the dominant sense, and one sentence priming one of the subordinate senses. These 306 sentences were split into three lists using a Latin square design, such that each participant saw all 102 words, but the exact condition varied between participants. In addition, we included 60 filler sentences, with comprehension questions to keep participants’ attention on reading. Sentences were presented in the moving window self-paced reading task to 196 participants. Mixed-effect regression revealed that priming dominant senses led to slower processing compared to neutral (ꞵ = -.07, S.E. = .02, df = .03, t = -2.64, p = .008, baseline: dominant priming condition). No significant differences were found between the dominant and subordinate priming conditions (ꞵ = .04, S.E. = .02, df = .03, t = -1.67, p = .095), although the trend suggests our hypotheses regarding the dominance effect were correct. Neither the entropy effect nor the interaction between the dominance priming condition and entropy were significant. Our results partially support our hypotheses. Priming a particular sense did lead to slower processing of polysemous words, but differences between probabilities of the primed senses did not clearly point to a larger slowing down of processing for subordinate senses. No entropy effects were significant, opening further questions on in-context processing of polysemous words.