Peer Review #1 of "Towards a distributed connectionist account of cognates and interlingual homographs: evidence from semantic relatedness tasks (v0.1)"

Abstract

Background. Current models of how bilinguals process cognates (e.g. "wolf", which has the same meaning in Dutch and English) and interlingual homographs (e.g. "angel", meaning "insect's sting" in Dutch) are based primarily on data from lexical decision tasks. A major drawback of such tasks is that it is difficult-if not impossible-to separate processes that occur during decision making (e.g. response competition) from processes that take place in the lexicon (e.g. lateral inhibition). Instead, we conducted two English semantic relatedness judgement experiments. Methods. In Experiment 1, highly proficient Dutch-English bilinguals (N = 29) and English monolinguals (N = 30) judged the semantic relatedness of word pairs that included a cognate (e.g. "wolf"-"howl"; n = 50), an interlingual homograph (e.g. "angel"-"heaven"; n = 50) or an English control word (e.g. "carrot"-"vegetable"; n = 50). In Experiment 2, another group of highly proficient Dutch-English bilinguals (N = 101) read sentences in Dutch that contained one of those cognates, interlingual homographs or the Dutch translation of one of the English control words (e.g. "wortel" for "carrot") approximately 15 minutes prior to completing the English semantic relatedness task. Results. In Experiment 1, there was an interlingual homograph inhibition effect of 39 ms only for the bilinguals, but no evidence for a cognate facilitation effect. Experiment 2 replicated these findings and also revealed that cross-lingual long-term priming had an opposite effect on the cognates and interlingual homographs: recent experience with a cognate in Dutch speeded processing of those items 15 minutes later in English but slowed processing of interlingual homographs. However, these priming effects were smaller than previously observed using a lexical decision task. Conclusion. After comparing our results to studies in both the bilingual and monolingual domain, we argue that bilinguals appear to process cognates and interlingual homographs as monolinguals process polysemes and homonyms, respectively. In the monolingual domain, processing of such words is best modelled using distributed connectionist frameworks. We conclude that it is necessary to explore the viability of such a model for the bilingual case. Data, scripts, materials and pre-registrations. Experiment 1: www.osf.io/ndb7p; Experiment 2: www.osf.io/2at49.