Pub Date : 2025-06-16DOI: 10.1016/j.jml.2025.104665
Chong Zhao , Edward K. Vogel
Working memory predicts cognitive abilities like fluid intelligence (gF) and source memory. This suggests these abilities depend on working memory and attentional control. When attentional resources were occupied by a secondary task, previous research shows that source memory performance is more impaired than recognition memory, implying that working memory abilities exert less influence on recognition memory performance than source memory performance. Here, we directly tested if working memory and attentional control differences predict visual recognition memory performance across four experiments (n = 841 in total). Surprisingly, we found that working memory and attentional control nearly always predicted recognition memory performance as robustly as source memory (Studies 1, 3 and 4), with the exception of when rapid presentation rates exceeded the temporal limits of attention during encoding (Study 2). Additionally, source memory and recognition memory, regardless of encoding presentation rates across experiments, remained highly correlated across individuals. Together, our findings suggest that working memory and attention control resources play a role in performance of both recognition and source memory tests of visual long-term memory.
{"title":"Working memory and attentional control abilities predict individual differences in visual long-term memory tasks","authors":"Chong Zhao , Edward K. Vogel","doi":"10.1016/j.jml.2025.104665","DOIUrl":"10.1016/j.jml.2025.104665","url":null,"abstract":"<div><div>Working memory predicts cognitive abilities like fluid intelligence (gF) and source memory. This suggests these abilities depend on working memory and attentional control. When attentional resources were occupied by a secondary task, previous research shows that source memory performance is more impaired than recognition memory, implying that working memory abilities exert less influence on recognition memory performance than source memory performance. Here, we directly tested if working memory and attentional control differences predict visual recognition memory performance across four experiments (n = 841 in total). Surprisingly, we found that working memory and attentional control nearly always predicted recognition memory performance as robustly as source memory (Studies 1, 3 and 4), with the exception of when rapid presentation rates exceeded the temporal limits of attention during encoding (Study 2). Additionally, source memory and recognition memory, regardless of encoding presentation rates across experiments, remained highly correlated across individuals. Together, our findings suggest that working memory and attention control resources play a role in performance of both recognition and source memory tests of visual long-term memory.</div></div>","PeriodicalId":16493,"journal":{"name":"Journal of memory and language","volume":"144 ","pages":""},"PeriodicalIF":2.9,"publicationDate":"2025-06-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144291428","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"心理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-06-13DOI: 10.1016/j.jml.2025.104660
Simón Ruiz , Padraic Monaghan , Wensi Zhang , Jiayi Li , Chaofan Jiang , Siqi Yang , Patrick Rebuschat
Individual differences in cognitive abilities and explicit instruction can affect language learning. Understanding how individual differences and instruction interact, however, requires us to determine the points in the language learning process that are open to influence. One hundred and eleven adults were exposed to an artificial language comprising transitive sentences occurring with action scenes and were either instructed or not in the language structure. Learning proceeded by determining the cross-situational correspondences between words and scene features. We found that declarative memory ability related strongly and positively but procedural memory related weakly and negatively to overall immediate learning. Rule-search instruction also positively influenced short-term learning, but not of the structure that was explicitly highlighted, and this was most pronounced in those with high declarative memory. The results highlight which features of language learning are accessible to information about language structure, and how that is affected by the learners’ cognitive abilities, with practical implications for personalised design of language learning programmes.
{"title":"The role of memory and instruction in the acquisition of vocabulary and grammar: An aptitude–treatment interaction study","authors":"Simón Ruiz , Padraic Monaghan , Wensi Zhang , Jiayi Li , Chaofan Jiang , Siqi Yang , Patrick Rebuschat","doi":"10.1016/j.jml.2025.104660","DOIUrl":"10.1016/j.jml.2025.104660","url":null,"abstract":"<div><div>Individual differences in cognitive abilities and explicit instruction can affect language learning. Understanding how individual differences and instruction interact, however, requires us to determine the points in the language learning process that are open to influence. One hundred and eleven adults were exposed to an artificial language comprising transitive sentences occurring with action scenes and were either instructed or not in the language structure. Learning proceeded by determining the cross-situational correspondences between words and scene features. We found that declarative memory ability related strongly and positively but procedural memory related weakly and negatively to overall immediate learning. Rule-search instruction also positively influenced short-term learning, but not of the structure that was explicitly highlighted, and this was most pronounced in those with high declarative memory. The results highlight which features of language learning are accessible to information about language structure, and how that is affected by the learners’ cognitive abilities, with practical implications for personalised design of language learning programmes.</div></div>","PeriodicalId":16493,"journal":{"name":"Journal of memory and language","volume":"144 ","pages":"Article 104660"},"PeriodicalIF":2.9,"publicationDate":"2025-06-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144271167","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"心理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-06-12DOI: 10.1016/j.jml.2025.104664
Andy L. Fordyce, Thomas S. Redick, Joseph P. Bedwell, Jeffrey D. Karpicke
Previous research on the association between individual differences in working memory and the benefit of retrieval practice has yielded mixed results, with various studies showing no differential retrieval practice benefit as a function of working memory ability, and others finding either more or less retrieval practice benefit for individuals lower in working memory. The current studies addressed how (a) variations in the learning task procedure and (b) measurement of working memory might influence the presence and/or strength of the relationship between working memory and retrieval practice. To ensure high initial retrieval success in Experiments 1 and 2, we used a learning-to-criterion procedure which had not been used in previous retrieval practice studies that examined individual differences. Experiment 3 extended the results of Experiments 1 and 2 to different learning task materials, while attempting to replicate a previous study that had shown a specific retrieval practice benefit for individuals with lower working memory. Additionally, separate analyses were conducted using partial and absolute scoring methods for operation span to address variability in previous research. Across all three experiments, retrieval practice outperformed restudying, and this benefit held regardless of individual differences in working memory ability.
{"title":"Individual differences in working memory and the benefit of retrieval practice","authors":"Andy L. Fordyce, Thomas S. Redick, Joseph P. Bedwell, Jeffrey D. Karpicke","doi":"10.1016/j.jml.2025.104664","DOIUrl":"10.1016/j.jml.2025.104664","url":null,"abstract":"<div><div>Previous research on the association between individual differences in working memory and the benefit of retrieval practice has yielded mixed results, with various studies showing no differential retrieval practice benefit as a function of working memory ability, and others finding either more or less retrieval practice benefit for individuals lower in working memory. The current studies addressed how (a) variations in the learning task procedure and (b) measurement of working memory might influence the presence and/or strength of the relationship between working memory and retrieval practice. To ensure high initial retrieval success in Experiments 1 and 2, we used a learning-to-criterion procedure which had not been used in previous retrieval practice studies that examined individual differences. Experiment 3 extended the results of Experiments 1 and 2 to different learning task materials, while attempting to replicate a previous study that had shown a specific retrieval practice benefit for individuals with lower working memory. Additionally, separate analyses were conducted using partial and absolute scoring methods for operation span to address variability in previous research. Across all three experiments, retrieval practice outperformed restudying, and this benefit held regardless of individual differences in working memory ability.</div></div>","PeriodicalId":16493,"journal":{"name":"Journal of memory and language","volume":"144 ","pages":"Article 104664"},"PeriodicalIF":2.9,"publicationDate":"2025-06-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144271166","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"心理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-06-10DOI: 10.1016/j.jml.2025.104662
Juliana Gerard , Dana McDaniel
This study tests the predictions of domain-general and language-specific accounts for children’s interpretations of adjunct control, as in “John called Bill before running to the store.” While adults only allow a subject interpretation for these sentences – that John ran to the store – children have allowed non-subject interpretations at various rates across studies. In particular, we consider how these interpretations may arise due to incomplete working memory development. This contrasts with language-specific accounts (e.g. grammatical and pragmatic accounts), which predict that children’s adjunct control will resemble other structures which allow non-subject interpretations – for example, the referentially ambiguous subject pronoun in “John called Bill before he ran to the store.” Our results support a domain general account: adjunct control is predicted by working memory, and does not pattern with ambiguous pronouns. We consider the implications for a more fine-grained account of children’s interpretations, and for interactions between working memory development and language acquisition.
{"title":"Domain-specificity and the development of syntactic dependencies: The role of working memory in the acquisition of adjunct control","authors":"Juliana Gerard , Dana McDaniel","doi":"10.1016/j.jml.2025.104662","DOIUrl":"10.1016/j.jml.2025.104662","url":null,"abstract":"<div><div>This study tests the predictions of domain-general and language-specific accounts for children’s interpretations of adjunct control, as in “John called Bill before running to the store.” While adults only allow a subject interpretation for these sentences – that John ran to the store – children have allowed non-subject interpretations at various rates across studies. In particular, we consider how these interpretations may arise due to incomplete working memory development. This contrasts with language-specific accounts (e.g. grammatical and pragmatic accounts), which predict that children’s adjunct control will resemble other structures which allow non-subject interpretations – for example, the referentially ambiguous subject pronoun in “John called Bill before he ran to the store.” Our results support a domain general account: adjunct control is predicted by working memory, and does not pattern with ambiguous pronouns. We consider the implications for a more fine-grained account of children’s interpretations, and for interactions between working memory development and language acquisition.</div></div>","PeriodicalId":16493,"journal":{"name":"Journal of memory and language","volume":"144 ","pages":"Article 104662"},"PeriodicalIF":2.9,"publicationDate":"2025-06-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144254706","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"心理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-06-05DOI: 10.1016/j.jml.2025.104661
Mieke Sarah Slim , Peter Lauwers , Robert J. Hartsuiker
A doubly-quantified sentence like Every bear approached a tent is ambiguous: Did every bear approach a different tent, or did they approach the same tent? These two interpretations are assumed to be mentally represented as logical representations, which specify how the different quantifiers are assigned scope with respect to each other. Based on a structural priming study, Feiman and Snedeker (2016) argued that logical representations capture quantifier-specific combinatorial properties (e.g., the specification of every differs from the specification of each in logical representations). We re-examined this conclusion by testing logical representation priming in Dutch. Across four experiments, we observed that priming of logical representations emerged if the same quantifiers are repeated in prime and target, but also if the prime and target contained different quantifiers. However, logical representation priming between quantifiers emerged less consistently than priming within the same quantifier. More specifically, our results suggest that priming between quantifiers emerges more robustly if the participant is presented with quantifier variation in the prime trials. When priming between quantifiers emerged, however, its strength was comparable to priming within the same quantifier. Therefore, we conclude that logical representations do not specify quantifier-specific biases in the assignment of scope.
{"title":"Revisiting the logic in language: The scope of each and every universal quantifier is alike after all","authors":"Mieke Sarah Slim , Peter Lauwers , Robert J. Hartsuiker","doi":"10.1016/j.jml.2025.104661","DOIUrl":"10.1016/j.jml.2025.104661","url":null,"abstract":"<div><div>A doubly-quantified sentence like <em>Every bear approached a tent</em> is ambiguous: Did every bear approach a different tent, or did they approach the same tent? These two interpretations are assumed to be mentally represented as logical representations, which specify how the different quantifiers are assigned scope with respect to each other. Based on a structural priming study, <span><span>Feiman and Snedeker (2016)</span></span> argued that logical representations capture quantifier-specific combinatorial properties (e.g., the specification of <em>every</em> differs from the specification of <em>each</em> in logical representations). We re-examined this conclusion by testing logical representation priming in Dutch. Across four experiments, we observed that priming of logical representations emerged if the same quantifiers are repeated in prime and target, but also if the prime and target contained different quantifiers. However, logical representation priming between quantifiers emerged less consistently than priming within the same quantifier. More specifically, our results suggest that priming between quantifiers emerges more robustly if the participant is presented with quantifier variation in the prime trials. When priming between quantifiers emerged, however, its strength was comparable to priming within the same quantifier. Therefore, we conclude that logical representations do not specify quantifier-specific biases in the assignment of scope.</div></div>","PeriodicalId":16493,"journal":{"name":"Journal of memory and language","volume":"144 ","pages":"Article 104661"},"PeriodicalIF":2.9,"publicationDate":"2025-06-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144222439","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"心理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-05-29DOI: 10.1016/j.jml.2025.104651
Elise Oltrogge , João Veríssimo , Umesh Patil , Sol Lago
Memory retrieval and prediction are typically studied separately, so little is known about their interaction. To address this gap, we studied a construction known to simultaneously trigger antecedent retrieval and possessee prediction processes: German possessive pronouns. We examined the comprehension of possessive pronouns using eye-tracking and computational modeling. Specifically, we chose an existing cue-based retrieval model that formalized prediction as a memory retrieval process. We used the model to generate predicted fixation patterns for a novel linguistic configuration, which replaced the possessive pronoun with an indefinite determiner. This allowed maintaining the prediction process—as German determiners agree in gender with a following noun—while effectively removing the antecedent retrieval process—as indefinite determiners, unlike pronouns, do not presuppose but rather introduce a new discourse referent. The eye-tracking results showed that participants’ predictions were affected by similarity-based interference, a well-known marker of memory processes. However, the timecourse of the novel determiner condition was different than predicted by the computational model. To better capture the behavioral data, we extended the model by introducing a process motivated by the semantics of indefinite determiners. Our results support the claim that linguistic predictions can be formalized as feature-driven processes that operate on representations shared by predictive and retrieval mechanisms.
{"title":"Memory retrieval and prediction interact in sentence comprehension: An experimental evaluation of a cue-based retrieval model","authors":"Elise Oltrogge , João Veríssimo , Umesh Patil , Sol Lago","doi":"10.1016/j.jml.2025.104651","DOIUrl":"10.1016/j.jml.2025.104651","url":null,"abstract":"<div><div>Memory retrieval and prediction are typically studied separately, so little is known about their interaction. To address this gap, we studied a construction known to simultaneously trigger antecedent retrieval and possessee prediction processes: German possessive pronouns. We examined the comprehension of possessive pronouns using eye-tracking and computational modeling. Specifically, we chose an existing cue-based retrieval model that formalized prediction as a memory retrieval process. We used the model to generate predicted fixation patterns for a novel linguistic configuration, which replaced the possessive pronoun with an indefinite determiner. This allowed maintaining the prediction process—as German determiners agree in gender with a following noun—while effectively removing the antecedent retrieval process—as indefinite determiners, unlike pronouns, do not presuppose but rather introduce a new discourse referent. The eye-tracking results showed that participants’ predictions were affected by similarity-based interference, a well-known marker of memory processes. However, the timecourse of the novel determiner condition was different than predicted by the computational model. To better capture the behavioral data, we extended the model by introducing a process motivated by the semantics of indefinite determiners. Our results support the claim that linguistic predictions can be formalized as feature-driven processes that operate on representations shared by predictive and retrieval mechanisms.</div></div>","PeriodicalId":16493,"journal":{"name":"Journal of memory and language","volume":"144 ","pages":"Article 104651"},"PeriodicalIF":2.9,"publicationDate":"2025-05-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144166685","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"心理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-05-26DOI: 10.1016/j.jml.2025.104652
Andrew Olson , Claudia Galluzzi , Ivana Bureca , Cristina Romani
Organising and producing a sequence of events is a basic human cognitive capacity. It occurs across a wide variety of domains including speech, writing, memory, planning and almost every type of skilled action. Errors involving sequences have been widely studied and often present two kinds of profiles: performance either declines across positions or it declines and then improves in the final positions (a U-shaped pattern). Studies of errors in aphasia have also reported these patterns with letters (in spelling) or phonemes (in speech). Another pattern, with more difficulty initiating speech, has been reported in apraxia of speech. Contrasting declines and increases in performance, however, have not been described in studies using the same methodology and evidence of performance linearly improving is very limited. We document all three patterns using statistical models in a case series of 23 people with aphasia (PwA) who make speech errors when repeating single words. We found that the declining pattern and the U-shape patterns occurred across patients, independent of whether their main impairment was a phonological impairment or apraxia of speech. Only people with apraxia of speech, however, showed the inverse pattern of linearly improving performance. Upward and downward patterns were not the consequence of a general factor like severity. Importantly, further exploration with statistical models revealed that phoneme position in the word was not, in fact, the dominant factor determining the visual patterns. Instead, performance was determined by either the number of previous errors (for declining performance) or the number of previous phonemes correct (for improving performance). Errors were almost never governed by serial position or word length per se. Our results support an important role for evolving context in the serial production mechanisms supporting single word production and we discuss implications for current models of speech production and, more generally, for models of serial performance. We suggest that temporary retention of novel sequences may rely more on an explicit representation of position, while stored articulatory representations may benefit from a contextual format (of the chaining type) where activation of previous units helps to support retrieval of units further along in the sequence.
{"title":"Serial position effects in spoken word production are determined by previous context: Evidence from aphasia","authors":"Andrew Olson , Claudia Galluzzi , Ivana Bureca , Cristina Romani","doi":"10.1016/j.jml.2025.104652","DOIUrl":"10.1016/j.jml.2025.104652","url":null,"abstract":"<div><div>Organising and producing a <em>sequence</em> of events is a basic human cognitive capacity. It occurs across a wide variety of domains including speech, writing, memory, planning and almost every type of skilled action. Errors involving sequences have been widely studied and often present two kinds of profiles: performance either declines across positions or it declines and then improves in the final positions (a U-shaped pattern). Studies of errors in aphasia have also reported these patterns with letters (in spelling) or phonemes (in speech). Another pattern, with more difficulty initiating speech, has been reported in apraxia of speech. Contrasting declines and increases in performance, however, have not been described in studies using the same methodology and evidence of performance linearly improving is very limited. We document all three patterns using statistical models in a case series of 23 people with aphasia (PwA) who make speech errors when repeating single words. We found that the declining pattern and the U-shape patterns occurred across patients, independent of whether their main impairment was a phonological impairment or apraxia of speech. Only people with apraxia of speech, however, showed the inverse pattern of linearly improving performance. Upward and downward patterns were not the consequence of a general factor like severity. Importantly, further exploration with statistical models revealed that phoneme position in the word was not, in fact, the dominant factor determining the visual patterns. Instead, performance was determined by either the number of previous errors (for declining performance) or the number of previous phonemes correct (for improving performance). Errors were almost never governed by serial position or word length <em>per se</em>. Our results support an important role for evolving context in the serial production mechanisms supporting single word production and we discuss implications for current models of speech production and, more generally, for models of serial performance. We suggest that temporary retention of novel sequences may rely more on an explicit representation of position, while stored articulatory representations may benefit from a contextual format (of the chaining type) where activation of previous units helps to support retrieval of units further along in the sequence.</div></div>","PeriodicalId":16493,"journal":{"name":"Journal of memory and language","volume":"144 ","pages":"Article 104652"},"PeriodicalIF":2.9,"publicationDate":"2025-05-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144138452","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"心理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-05-23DOI: 10.1016/j.jml.2025.104650
Ethan Gotlieb Wilcox , Michael Y. Hu , Aaron Mueller , Alex Warstadt , Leshem Choshen , Chengxu Zhuang , Adina Williams , Ryan Cotterell , Tal Linzen
When trained to place high probability on a training corpus, neural network language models can learn a surprising amount about language. Recent work has demonstrated that large performance improvements can arise from simply increasing, i.e., scaling, the size of the corpora they are trained on and the number of parameters in those models. Accordingly, many contemporary systems are trained on trillions of words. While largely beneficial to performance on language applications, scaling has several downsides for both computational psycholinguistics and natural language processing research. We discuss the scientific challenges presented by the scaling paradigm, as well as the benefits that would result from language models that can learn from human-scale data. In the second half of this paper, we report on findings from a recent effort to bring about human-scale language model pretraining: the first iteration of the BabyLM Challenge, a shared task organized by the authors that invited participants to train a language model on 100 million words or less. The challenge produced several concrete best practices for practitioners interested in small-scale language modeling. For cognitive scientists, the challenge demonstrated that robust linguistic generalizations can be learned by models trained on a human-scale dataset, though this is not yet achieved through cognitively plausible mechanisms. Furthermore, it established a population of “BabyLMs” that are all effective at data-efficient language learning. Studying such models can help us identify hypotheses for the computational mechanisms that underlie human language acquisition.
{"title":"Bigger is not always better: The importance of human-scale language modeling for psycholinguistics","authors":"Ethan Gotlieb Wilcox , Michael Y. Hu , Aaron Mueller , Alex Warstadt , Leshem Choshen , Chengxu Zhuang , Adina Williams , Ryan Cotterell , Tal Linzen","doi":"10.1016/j.jml.2025.104650","DOIUrl":"10.1016/j.jml.2025.104650","url":null,"abstract":"<div><div>When trained to place high probability on a training corpus, neural network language models can learn a surprising amount about language. Recent work has demonstrated that large performance improvements can arise from simply increasing, i.e., scaling, the size of the corpora they are trained on and the number of parameters in those models. Accordingly, many contemporary systems are trained on trillions of words. While largely beneficial to performance on language applications, scaling has several downsides for both computational psycholinguistics and natural language processing research. We discuss the scientific challenges presented by the scaling paradigm, as well as the benefits that would result from language models that can learn from human-scale data. In the second half of this paper, we report on findings from a recent effort to bring about human-scale language model pretraining: the first iteration of the BabyLM Challenge, a shared task organized by the authors that invited participants to train a language model on 100 million words or less. The challenge produced several concrete best practices for practitioners interested in small-scale language modeling. For cognitive scientists, the challenge demonstrated that robust linguistic generalizations can be learned by models trained on a human-scale dataset, though this is not yet achieved through cognitively plausible mechanisms. Furthermore, it established a population of “BabyLMs” that are all effective at data-efficient language learning. Studying such models can help us identify hypotheses for the computational mechanisms that underlie human language acquisition.</div></div>","PeriodicalId":16493,"journal":{"name":"Journal of memory and language","volume":"144 ","pages":"Article 104650"},"PeriodicalIF":2.9,"publicationDate":"2025-05-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144124437","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"心理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-05-23DOI: 10.1016/j.jml.2025.104653
Cassandra L. Jacobs , Ryan J. Hubbard , Loïc Grobol , Kara D. Federmeier
Psycholinguistic researchers often collect cloze probabilities in order to measure the predictability of upcoming words but have largely discarded the variability in the structure of responses people provide. This variability in the semantic structure of responses may be important for understanding selection during language production; however, it has proven difficult to model the semantic variability of participants’ responses, and thus upcoming semantic uncertainty. Recent advances in large language models (LLMs) permit us to approximate the degree of semantic variability in cloze responses, but most methods are restricted to symbolic or hand-crafted meaning representations. We show in two studies that Bayesian Gaussian mixture models can cluster LLM representations of participants’ responses and produce coherent, taxonomically similar clusters. We apply these clustering algorithms to response time data in a serial cloze task and show that the semantic structure of cloze responses influences how quickly people are able to provide a response. We show clear effects of semantic competition on production speed. In addition to providing novel operationalizations of what semantic competition might look like in the cloze task, we explain how this clustering method is extensible to other datasets and applications of interest to researchers of semantic processing in psycholinguistics.
{"title":"Uncovering patterns of semantic predictability in sentence processing","authors":"Cassandra L. Jacobs , Ryan J. Hubbard , Loïc Grobol , Kara D. Federmeier","doi":"10.1016/j.jml.2025.104653","DOIUrl":"10.1016/j.jml.2025.104653","url":null,"abstract":"<div><div>Psycholinguistic researchers often collect cloze probabilities in order to measure the predictability of upcoming words but have largely discarded the variability in the structure of responses people provide. This variability in the semantic structure of responses may be important for understanding selection during language production; however, it has proven difficult to model the semantic variability of participants’ responses, and thus upcoming semantic uncertainty. Recent advances in large language models (LLMs) permit us to approximate the degree of semantic variability in cloze responses, but most methods are restricted to symbolic or hand-crafted meaning representations. We show in two studies that Bayesian Gaussian mixture models can cluster LLM representations of participants’ responses and produce coherent, taxonomically similar clusters. We apply these clustering algorithms to response time data in a serial cloze task and show that the semantic structure of cloze responses influences how quickly people are able to provide a response. We show clear effects of semantic competition on production speed. In addition to providing novel operationalizations of what semantic competition might look like in the cloze task, we explain how this clustering method is extensible to other datasets and applications of interest to researchers of semantic processing in psycholinguistics.</div></div>","PeriodicalId":16493,"journal":{"name":"Journal of memory and language","volume":"144 ","pages":"Article 104653"},"PeriodicalIF":2.9,"publicationDate":"2025-05-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144124438","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"心理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-05-21DOI: 10.1016/j.jml.2025.104643
Kumiko Fukumura
Models of human sentence production often propose a clear distinction between syntactic and semantic processes. We examined this assumption by investigating the interaction between animacy and thematic roles in active–passive structural priming. Study 1 found that the active or passive structure of a preceding sentence (prime) influenced structural choice in a subsequent sentence (target). This priming effect increased when the prime and target sentences shared the same animacy features in their thematic roles, which affected the persistence of the prime subject’s animacy. While verb repetition enhanced active–passive priming, the persistence of the prime subject’s animacy was not affected by lexical repetition. Studies 2 and 3 demonstrated that repeated animacy features in the thematic roles increase the likelihood of preserving both the thematic role order of the prime (e.g., maintaining the agent-first order in It was the thief that chased the lorry) and its argument structure (e.g., assigning the agent as the subject) in English cleft constructions. In Japanese declarative sentences, where particles indicate the sentential topic, the repeated animacy features strengthened argument structure persistence but not the persistence of thematic role order. These findings suggest that thematic role animacy repetition boosts structural priming by reinforcing thematic emphasis.
人类句子生成模型经常提出句法和语义过程之间的明确区分。我们通过调查主动-被动结构启动中动画和主题角色之间的相互作用来检验这一假设。研究1发现，前句（启动句）的主动或被动结构会影响后句（目标句）的结构选择。当启动句和目标句在主位角色中具有相同的活力特征时，这种启动效应会增强，从而影响启动主语活力的持久性。动词重复可增强主-被动启动效应，但词汇重复对启动主语活力的持久性没有影响。研究2和研究3表明，主位角色的重复活泼特征增加了英语裂隙结构中主位角色顺序（如在It was the thief that chase the lorry中保持主体优先顺序）和主位结构（如指定主体为主语）的可能性。在日语陈述句中，小品句表示句子的主题，重复的动画表现为论点结构的持久性增强，而不是主位角色顺序的持久性。这些发现表明，主题角色动画重复通过强化主题强调来促进结构启动。
{"title":"The interplay of animacy and thematic role in structural persistence","authors":"Kumiko Fukumura","doi":"10.1016/j.jml.2025.104643","DOIUrl":"10.1016/j.jml.2025.104643","url":null,"abstract":"<div><div>Models of human sentence production often propose a clear distinction between syntactic and semantic processes. We examined this assumption by investigating the interaction between animacy and thematic roles in active–passive structural priming. Study 1 found that the active or passive structure of a preceding sentence (<em>prime</em>) influenced structural choice in a subsequent sentence (<em>target</em>). This priming effect increased when the prime and target sentences shared the same animacy features in their thematic roles, which affected the persistence of the prime subject’s animacy. While verb repetition enhanced active–passive priming, the persistence of the prime subject’s animacy was not affected by lexical repetition. Studies 2 and 3 demonstrated that repeated animacy features in the thematic roles increase the likelihood of preserving both the thematic role order of the prime (e.g., maintaining the agent-first order in <em>It was the thief that chased the lorry</em>) and its argument structure (e.g., assigning the agent as the subject) in English cleft constructions. In Japanese declarative sentences, where particles indicate the sentential topic, the repeated animacy features strengthened argument structure persistence but not the persistence of thematic role order. These findings suggest that thematic role animacy repetition boosts structural priming by reinforcing thematic emphasis.</div></div>","PeriodicalId":16493,"journal":{"name":"Journal of memory and language","volume":"144 ","pages":"Article 104643"},"PeriodicalIF":2.9,"publicationDate":"2025-05-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144108075","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"心理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号