Computational cognitive modeling of predictive sentence processing in a second language

Proceedings of the 26th Conference on Computational Natural Language Learning (CoNLL), pages 339 - 349 December 7-8, 2022 ©2022 Association for Computational Linguistics Computational cognitive modeling of predictive sentence processing in a second language Umesh Patil (umesh.patil@gmail.com) University of Cologne 50923 Cologne, Germany Sol Lago (sollago@em.uni-frankfurt.de) Goethe University Frankfurt 60629 Frankfurt, Germany Abstract We propose an ACT-R cue-based retrieval model of the real-time gender predictions dis- played by second language (L2) learners. The model extends a previous model of native (L1) speakers according to two central accounts in L2 sentence processing: (i) the Interference Hypothesis, which proposes that retrieval inter- ference is higher in L2 than L1 speakers; (ii) the Lexical Bottleneck Hypothesis, which proposes that problems with gender agreement are due to weak gender representations. We tested the predictions of these accounts using data from two visual world experiments, which found that the gender predictions elicited by German pos- sessive pronouns were delayed and smaller in size in L2 than L1 speakers. The experiments also found a “match effect”, such that when the antecedent and possessee of the pronoun had the same gender, predictions were earlier than when the two genders differed. This match ef- fect was smaller in L2 than L1 speakers. The model implementing the Lexical Bottleneck Hypothesis captured the effects of smaller pre- dictions, smaller match effect and delayed pre- dictions in one of the two conditions. By con- trast, the model implementing the Interference Hypothesis captured the smaller prediction ef- fect but it showed an earlier prediction effect and an increased match effect in L2 than L1 speakers. These results provide evidence for the Lexical Bottleneck Hypothesis, and they demonstrate a method for extending computa- tional models of L1 to L2 processing. 1 Introduction Although the world population is quickly becoming bilingual, there are very few computational models of bilingual sentence processing. Because most of these models were developed for technological applications—e.g., automatic translation—, this re- sults in a scarcity of models that are cognitively realistic or even evaluable with human data (Frank, 2021; Frank et al., 2016; Hinaut et al., 2015; Hen- driks and Vogelzang, 2020). However, such mod- els are crucial to develop computational research that is informed by state-of-the-art psycholinguistic work. With this goal, we propose a computational cognitive model of bilingual processing built in an architecture, ACT-R, which is designed to model human cognition and can be evaluated with human data (Anderson, 2007; Ritter et al., 2019). The ACT-R architecture has also been used to model a number of linguistic phenomena, such as retrieval interference effects in linguistic dependency resolu- tion (Vasishth et al., 2008), the influence of promi- nence on pronoun resolution (Patil et al., 2016b; Patil and Schumacher, 2022), the effect of memory load on sentence processing (van Rij et al., 2013), sentence processing in patients with aphasia (Cres- centini and Stocco, 2005; Patil et al., 2016a), the interaction of sentence processing and eye move- ments (Engelmann et al., 2013), and incremental formal semantic processing (Brasoveanu and Dot- laˇcil, …

Content truncated to 500 words