Linguistic register is defined as a variety of language shaped by different situational settings. Adapting to register is crucial for successful communication and involves the processing of language features related to register variation. Few studies have focused on the impact of linguistic register on language processing. Our research investigates whether register variation affects the detection of linguistic errors. To determine if linguistic background further impacts the way we deal with register, our sample includes monolingual, bilingual, and bidialectal participants. All groups completed an acceptability judgement task in Italian that features Subject-Verb agreement mismatches presented in high and low register. The results reveal a significant impact of linguistic register on accuracy: morphosyntactic errors are better detected in low-register stimuli. Furthermore, different trends characterize the tested groups. While monolinguals show more similar accuracy rates for low- and high-register sentences, the bilingual groups tend to better spot errors in low-register stimuli. Our findings suggest that register plays an important role in the processing of morphosyntactic errors, highlighting the need to consider both its cognitive and social dimensions. Moreover, the variation observed among the tested groups underscores that language processing can be influenced by factors related to the sociolinguistic dimensions of each linguistic community.

Modern Artificial Intelligence applications show great potential for language- related tasks that rely on next-word prediction. The current generation of Large Language Models (LLMs) have been linked to claims about human-like linguistic performance and their applications are hailed both as a step towards artificial general intelligence and as a major advance in understanding the cognitive, and even neural basis of human language. To assess these claims, first we analyze the contribution of LLMs as theoretically informative representations of a target cognitive system vs. atheoretical mechanistic tools. Second, we evaluate the models’ ability to see the bigger picture, through top-down feedback from higher levels of processing, which requires grounding in previous expectations and past world experience. We hypothesize that since models lack grounded cognition, they cannot take advantage of these features and instead solely rely on fixed associations between represented words and word vectors. To assess this, we designed and ran a novel ‘leet task’ (l33t t4sk), which requires decoding sentences in which letters are systematically replaced by numbers. The results suggest that humans excel in this task whereas models struggle, confirming our hypothesis. We interpret the results by identifying the key abilities that are still missing from the current state of development of these models, which require solutions that go beyond increased system scaling.