Cumulative Frequency Can Explain Cognate Facilitation in Language Models

Abstract

Cognates – words which share form and meaning across two languages – have been extensively studied to understand the bilingual mental lexicon. One consistent finding is that bilingual speakers process cognates faster than non-cognates, an effect known as cognate facilitation. Yet, there is no agreement on the underlying factors driving this effect. In this paper, we use computational modeling to test whether the effect can be explained by the cumulative frequency hypothesis. We train a computational language model on two language pairs (Dutch–English, Norwegian–English) under different conditions of input presentation and test it on sentence stimuli from two existing studies with bilingual speakers of those languages. We find that our model can exhibit a cognate effect, lending support to the cumulative frequency hypothesis. Further analyses reveal that the size of the effect in the model depends on its linguistic accuracy. We interpret our results within the literature on cognate processing.

Cognate facilitation in bilingual reading: The influence of orthographic and phonological similarity on lexical decisions and eye-movements

“…Taken together, there is abundant experimental evidence across different tasks and languages that multilingual language-users do not "switch off" a non-target language during language comprehension (Vanlangendonck et al, 2020), and that crosslanguage similarity facilitates word recognition, consistent with the assumption of language non-selectivity in theoretical and computational models of bilingual language processing (Dijkstra & Van Heuven, 2002;Van Heuven & Dijkstra, 2010;Winther et al, 2021). The apparent absence of a discrete difference in the expression of the cognate facilitation effect between identical and close cognates in word recognition and sentence reading in this study, together with the clear evidence for continuous linear orthographic similarity effects, supports the gradual facilitation account of cross-language overlap in word recognition (Van Assche et al, 2011).…”

“…Importantly, the degree of crosslanguage activation, and thus cognate facilitation, should be a function of the similarity of the lexical representations (Van Assche et al, 2011). The gradual facilitation effect of cross-language similarity is also compatible with frequency-based accounts of cognate facilitation (Midgley et al, 2011;Peeters et al, 2013;Voga & Grainger, 2007;Winther et al, 2021), assuming that exposure to identical cognates in multiple languages elicits a greater cumulative frequency effect than non-identical, close cognates. This study adds to the evidence that the cognate facilitation effect can be viewed as a continuous effect of orthographic similarity, which increases in magnitude with cross-language overlap of word representations.…”

“…An alternative explanation for the faster processing of words that share word-form and meaning across languages is their greater cumulative frequency of occurrence (Midgley et al, 2011;Peeters et al, 2013;Voga & Grainger, 2007;Winther et al, 2021). According to the cumulative frequency hypothesis, encountering words that share form-meaning associations across languages known to a multilingual language-user results in an accumulated activation benefit (e.g., Midgley et al, 2011).…”

“…In unbalanced bilinguals, the accumulated exposure explanation predicts a stronger cognate facilitation effect in L2 than L1, as the accumulated exposure in L1 is a greater influence on L2 processing than the lesser exposure in L2 on L1 processing (Midgley et al, 2011). Computational language models are able to simulate cognate facilitation effects based on the cumulative frequency of exposure to words that share characteristics in L1 and L2, although only when simulated exposure to L1 is higher than to L2 (Winther et al, 2021). Unlike other established models such as the BIA+ (Dijkstra & Van Heuven, 2002) and the Multilink model (Dijkstra & Rekké, 2010;Dijkstra et al, 2019), frequency-based explanations of the cognate facilitation effect do not assume that identical cognates have a special status in the bilingual mind, beyond their greater cumulative cross-language frequency.…”

Bilingual parafoveal processing: Children and adults preprocess orthographic information of the upcoming word during sentence reading in their first and second language.

“…However, it should be noted that cognates are typically processed faster than noncognates by multilingual language users (Kroll et al, 2016;Lauro & Schwartz, 2017;Lijewska, 2020). This cognate facilitation effect is attributed to multilingual language-users having an integrated mental lexicon, in which cognates enjoy a special status, because of their overlapping features (e.g., Dijkstra & Rekké, 2010;Dijkstra & Van Heuven, 2002;Dijkstra et al, 2019), and increased joint frequency of occurrence across languages (Midgley, Holcomb, & Grainger, 2011;Peeters et al, 2013;Voga & Grainger, 2007;Winther et al, 2021). This had three relevant consequences for the present study.…”