Lexical-Semantic Content, Not Syntactic Structure, Is the Main Contributor to ANN-Brain Similarity of fMRI Responses in the Language Network.

Representations from artificial neural network (ANN) language models have been shown to predict human brain activity in the language network. To understand what aspects of linguistic stimuli contribute to ANN-to-brain similarity, we used an fMRI data set of responses to n = 627 naturalistic English sentences (Pereira et al., 2018) and systematically manipulated the stimuli for which ANN representations were extracted. In particular, we (i) perturbed sentences' word order, (ii) removed different subsets of words, or (iii) replaced sentences with other sentences of varying semantic similarity. We found that the lexical-semantic content of the sentence (largely carried by content words) rather than the sentence's syntactic form (conveyed via word order or function words) is primarily responsible for the ANN-to-brain similarity. In follow-up analyses, we found that perturbation manipulations that adversely affect brain predictivity also lead to more divergent representations in the ANN's embedding space and decrease the ANN's ability to predict upcoming tokens in those stimuli. Further, results are robust as to whether the mapping model is trained on intact or perturbed stimuli and whether the ANN sentence representations are conditioned on the same linguistic context that humans saw. The critical result-that lexical-semantic content is the main contributor to the similarity between ANN representations and neural ones-aligns with the idea that the goal of the human language system is to extract meaning from linguistic strings. Finally, this work highlights the strength of systematic experimental manipulations for evaluating how close we are to accurate and generalizable models of the human language network.

Large-scale evidence for logarithmic effects of word predictability on reading time.

During real-time language comprehension, our minds rapidly decode complex meanings from sequences of words. The difficulty of doing so is known to be related to words' contextual predictability, but what cognitive processes do these predictability effects reflect? In one view, predictability effects reflect facilitation due to anticipatory processing of words that are predictable from context. This view predicts a linear effect of predictability on processing demand. In another view, predictability effects reflect the costs of probabilistic inference over sentence interpretations. This view predicts either a logarithmic or a superlogarithmic effect of predictability on processing demand, depending on whether it assumes pressures toward a uniform distribution of information over time. The empirical record is currently mixed. Here, we revisit this question at scale: We analyze six reading datasets, estimate next-word probabilities with diverse statistical language models, and model reading times using recent advances in nonlinear regression. Results support a logarithmic effect of word predictability on processing difficulty, which favors probabilistic inference as a key component of human language processing.

Rational Sentence Interpretation in Mandarin Chinese.

Previous work has shown that English native speakers interpret sentences as predicted by a noisy-channel model: They integrate both the real-world plausibility of the meaning-the prior-and the likelihood that the intended sentence may be corrupted into the perceived sentence. In this study, we test the noisy-channel model in Mandarin Chinese, a language taxonomically different from English. We present native Mandarin speakers sentences in a written modality (Experiment 1) and an auditory modality (Experiment 2) in three pairs of syntactic alternations. The critical materials are literally implausible but require differing numbers and types of edits in order to form more plausible sentences. Each sentence is followed by a comprehension question that allows us to infer whether the speakers interpreted the item literally, or made an inference toward a more likely meaning. Similar to previous research on related English constructions, Mandarin participants made the most inferences for implausible materials that could be inferred as plausible by deleting a single morpheme or inserting a single morpheme. Participants were less likely to infer a plausible meaning for materials that could be inferred as plausible by making an exchange across a preposition. And participants were least likely to infer a plausible meaning for materials that could be inferred as plausible by making an exchange across a main verb. Moreover, we found more inferences in written materials than spoken materials, possibly a result of a lack of word boundaries in written Chinese. Overall, the fact that the results were so similar to those found in related constructions in English suggests that the noisy-channel proposal is robust.

How adults understand what young children say.

Children's early speech often bears little resemblance to that of adults, and yet parents and other caregivers are able to interpret that speech and react accordingly. Here we investigate how adult listeners' inferences reflect sophisticated beliefs about what children are trying to communicate, as well as how children are likely to pronounce words. Using a Bayesian framework for modelling spoken word recognition, we find that computational models can replicate adult interpretations of children's speech only when they include strong, context-specific prior expectations about the messages that children will want to communicate. This points to a critical role of adult cognitive processes in supporting early communication and reveals how children can actively prompt adults to take actions on their behalf even when they have only a nascent understanding of the adult language. We discuss the wide-ranging implications of the powerful listening capabilities of adults for theories of first language acquisition.

HLA Alleles in a Behçet Disease Multiethnic Population With and Without Ophthalmic Manifestations.

OBJECTIVE: The aim of this study was to analyze HLA alleles in patients with Behçet disease (BD) and their correlation with ophthalmic manifestations (OMs) in a multiethnic Brazilian population. METHODS: This case-control study compared 72 BD patients with or without OM who underwent a thorough ophthalmologic evaluation, including best-corrected visual acuity, bino-ophthalmoscopy, and HLA analysis, with 144 matched healthy controls. Fluorescein angiography was also performed in the patients with BD and OM. HLA class I (A, B, and C) and II (DRB1, DQB1, and DQA1) typing were performed using PCR-SSO. RESULTS: Of 72 patients with BD, 42 (58%) had OM. The HLA-B*51 and -A*26 alleles were more frequent in patients with BD than in controls (23.6% vs 14.6% and 12.5% vs 4.3%, respectively), but could not differentiate OM risk. The HLA alleles of BD patients that differentiated those with and without OM were HLA-B*15 (40.5% vs 20.7%; odds ratio [OR], 2.59; p = 0.0059), HLA-C*02 (33.3% vs 13.4%; OR, 3.20; p = 0.0024), and HLA-DQB1*03 (64.3% vs 45.7%, p = 0.017), whereas HLA-A*03 (0.0% vs 13.3%, p = 0.006) and HLA-DRB1*15 (4.8% vs 19.5%; OR, 0.21; p = 0.0121) were protective against OM. CONCLUSIONS: In this study of a Brazilian multiethnic BD population, alleles were similar between groups of BD patients with and without OM. We described HLA-B*15, -C*02, and -DQB1*03 as risk factors and -A*03 and -DRB1*15 as protective factors for OM in BD, which could function as biomarkers for predicting disease phenotypes.

Development of a Conceptual Model to Understand Disease Burden in Patients With Systemic Lupus Erythematosus and Organ Damage.

Background: Systemic lupus erythematosus (SLE) is a chronic autoimmune disease that can lead to irreversible organ damage (OD). Data describing the patient burden of OD, as compared with SLE without OD, are limited. Objective: To develop a comprehensive conceptual model describing the burden experienced by patients living with SLE-associated OD. Methods: There were three phases to this qualitative study. First, a targeted literature review was conducted to inform a draft conceptual model. Second, key opinion leaders (KOLs) were interviewed to assess the draft conceptual model and help shape patient interview materials. Third, patients of different demographic backgrounds from across the United States were interviewed individually to gather their perspectives on living with SLE-associated OD. Data from concept elicitation interviews with KOLs and patients were coded and analyzed using NVivo software to identify the key concepts of the overall patient burden of SLE-associated OD. Findings from the KOL and patient interviews were used to finalize the conceptual model. Results: KOLs highlighted that SLE-associated OD carried a higher rate of mortality than SLE alone. Participants with SLE-associated OD (n = 40) experienced detrimental impacts across 4 areas of their lives: physical, cognitive, psychosocial functioning, and economic and work-related well-being. Physical impacts were described by all participants, often affecting their ability to perform everyday tasks. Many also described deterioration of cognitive functioning. Almost all participants experienced emotional impacts and challenges to their relationships and social lives resulting from living with SLE-associated OD. Additionally, SLE-associated OD imposed an economic burden including increased healthcare costs. SLE-associated OD had a more severe and debilitating impact on all aspects of the patient's quality of life than SLE prior to OD development, including further limitations in activities of daily living after the development of OD. Discussion: Study findings guided the development of a comprehensive conceptual model that fully represents the patient experience of living with SLE-associated OD, highlighting the additional burden of OD when compared with SLE alone. Conclusions: The conceptual model will inform improvements in disease management, which may result in better patient outcomes and aid development of clinical outcome assessments of disease burden.

Lupus spectrum ambiguity has long-term negative implications for patients.

Lupus is a complex disease that is often difficult to diagnose. Risks of diagnostic delays include non-specific signs and symptoms that mimic other diseases and a lack of diagnostic criteria and referral pathways for non-specialists. To address these issues, we convened a series of virtual meetings with members of our Addressing Lupus Pillars for Health Advancement clinical care team. Meeting participants included lupus physicians, treatment developers from biotechnology, patient advocacy group representatives from the Lupus Foundation of America and advocacy/government consultants. Causes and consequences of ambiguity in diagnosis and diagnostic delays were evaluated through historical, experiential and evidence-based accounts (survey data, literature reviews and patient testimonials). Discussions highlighted the need for a clearer understanding of the definition of lupus, the natural history of the disease and the need for advancements in biotechnology to support an accurate and timely diagnosis with the potential development of a lupus spectrum.

The 2023 ACR/EULAR Antiphospholipid Syndrome Classification Criteria.

OBJECTIVE: To develop new antiphospholipid syndrome (APS) classification criteria with high specificity for use in observational studies and trials, jointly supported by the American College of Rheumatology (ACR) and EULAR. METHODS: This international multidisciplinary initiative included 4 phases: 1) Phase I, criteria generation by surveys and literature review; 2) Phase II, criteria reduction by modified Delphi and nominal group technique exercises; 3) Phase III, criteria definition, further reduction with the guidance of real-world patient scenarios, and weighting via consensus-based multicriteria decision analysis, and threshold identification; and 4) Phase IV, validation using independent adjudicators' consensus as the gold standard. RESULTS: The 2023 ACR/EULAR APS classification criteria include an entry criterion of at least one positive antiphospholipid antibody (aPL) test within 3 years of identification of an aPL-associated clinical criterion, followed by additive weighted criteria (score range 1-7 points each) clustered into 6 clinical domains (macrovascular venous thromboembolism, macrovascular arterial thrombosis, microvascular, obstetric, cardiac valve, and hematologic) and 2 laboratory domains (lupus anticoagulant functional coagulation assays, and solid-phase enzyme-linked immunosorbent assays for IgG/IgM anticardiolipin and/or IgG/IgM anti-ß2 -glycoprotein I antibodies). Patients accumulating at least 3 points each from the clinical and laboratory domains are classified as having APS. In the validation cohort, the new APS criteria versus the 2006 revised Sapporo classification criteria had a specificity of 99% versus 86%, and a sensitivity of 84% versus 99%. CONCLUSION: These new ACR/EULAR APS classification criteria were developed using rigorous methodology with multidisciplinary international input. Hierarchically clustered, weighted, and risk-stratified criteria reflect the current thinking about APS, providing high specificity and a strong foundation for future APS research.

2023 ACR/EULAR antiphospholipid syndrome classification criteria.

OBJECTIVE: To develop new antiphospholipid syndrome (APS) classification criteria with high specificity for use in observational studies and trials, jointly supported by the American College of Rheumatology (ACR) and EULAR. METHODS: This international multidisciplinary initiative included four phases: (1) Phase I, criteria generation by surveys and literature review; (2) Phase II, criteria reduction by modified Delphi and nominal group technique exercises; (3) Phase III, criteria definition, further reduction with the guidance of real-world patient scenarios, and weighting via consensus-based multicriteria decision analysis, and threshold identification; and (4) Phase IV, validation using independent adjudicators' consensus as the gold standard. RESULTS: The 2023 ACR/EULAR APS classification criteria include an entry criterion of at least one positive antiphospholipid antibody (aPL) test within 3 years of identification of an aPL-associated clinical criterion, followed by additive weighted criteria (score range 1-7 points each) clustered into six clinical domains (macrovascular venous thromboembolism, macrovascular arterial thrombosis, microvascular, obstetric, cardiac valve, and hematologic) and two laboratory domains (lupus anticoagulant functional coagulation assays, and solid-phase enzyme-linked immunosorbent assays for IgG/IgM anticardiolipin and/or IgG/IgM anti-ß2-glycoprotein I antibodies). Patients accumulating at least three points each from the clinical and laboratory domains are classified as having APS. In the validation cohort, the new APS criteria vs the 2006 revised Sapporo classification criteria had a specificity of 99% vs 86%, and a sensitivity of 84% vs 99%. CONCLUSION: These new ACR/EULAR APS classification criteria were developed using rigorous methodology with multidisciplinary international input. Hierarchically clustered, weighted, and risk-stratified criteria reflect the current thinking about APS, providing high specificity and a strong foundation for future APS research.

Eye Movement Traces of Linguistic Knowledge in Native and Non-Native Reading.

The detailed study of eye movements in reading has shed considerable light into how language processing unfolds in real time. Yet eye movements in reading remain inadequately studied in non-native (L2) readers, even though much of the world's population is multilingual. Here we present a detailed analysis of the quantitative functional influences of word length, frequency, and predictability on eye movement measures in reading in a large, linguistically diverse sample of non-native English readers. We find many similar qualitative effects as in L1 readers, but crucially also a proficiency-sensitive "lexicon-context tradeoff". The most proficient L2 readers' eye movements approach an L1 pattern, but as L2 proficiency diminishes, readers' eye movements become less sensitive to a word's predictability in context and more sensitive to word frequency, which is context-invariant. This tradeoff supports a rational, experience-dependent account of how context-driven expectations are deployed in L2 language processing.

Real-World Immunosuppressant Treatment Patterns for Patients with Lupus Nephritis in the United States.

INTRODUCTION: Lupus nephritis (LN) treatment aims to control and prevent flares and irreversible kidney damage. Around 30% of patients are unresponsive to treatment; however, real-world LN treatment patterns have not been reported. Objectives of this retrospective cohort study (GSK 209758) were to quantify the time to switching/re-initiating induction therapy in patients with LN initiating immunosuppressant therapy and conversion from induction to maintenance immunosuppressant therapy, and to assess corticosteroid use. METHODS: Patients with LN initiating induction or maintenance immunosuppressant therapy were identified using claims data. Patients were followed up from the index date (immunosuppressant initiation date) until treatment discontinuation, death, disenrollment, administrative censoring, or the end of follow-up period. The cumulative incidence of switching/re-initiating induction therapy and conversion to maintenance therapy was estimated using outpatient pharmacy claims and procedure codes. Corticosteroid use was estimated using pharmacy claims; a mean daily dose of ≥ 7.5 mg/day was considered high. RESULTS: In total, 5000 patients with LN contributed 5516 treatment episodes (induction cohort, N = 372; maintenance cohort, N = 5144). In the induction cohort, the cumulative incidence (95% confidence interval) of switching between induction therapies was 24.6% (20.1-30.0) at 12 months, while 59.6% (52.4-66.1) of patients converted to maintenance therapy at 12 months. In the maintenance cohort, 21.2% (19.9-22.5) re-initiated induction therapy at 12 months. Oral corticosteroid use decreased during the follow-up in both cohorts, but 21.5% of patients remained on a high dose at 12 months in the induction cohort, while 15.8% in the maintenance cohort were taking a high dose at 24 months. CONCLUSIONS: Around a quarter of patients with LN initiating immunosuppressant therapy switched within 12 months, while a fifth re-initiated induction therapy within 12 months. Use of high corticosteroid doses were observed during 24 months of follow-up. These data suggest that many patients do not respond to existing standard LN therapies.

Effect of belimumab on kidney-related outcomes in patients with lupus nephritis: post hoc subgroup analyses of the phase 3 BLISS-LN trial.

BACKGROUND: Data on belimumab efficacy in patients with lupus nephritis (LN) according to diagnosis duration or induction therapy are limited. Post hoc analyses of the phase 3, randomized, double-blind BLISS-LN study (GSK BEL114054; NCT01639339) were performed to assess belimumab efficacy on kidney-related outcomes in newly diagnosed and relapsed LN subgroups and according to the use of glucocorticoid (GC) pulses at induction. METHODS: BLISS-LN randomized 448 patients with active LN to monthly intravenous belimumab 10 mg/kg or placebo plus standard therapy. Post hoc analyses assessed primary efficacy renal response (PERR) and complete renal response (CRR) at week 104, time to kidney-related event or death and time to first LN flare from week 24 in newly diagnosed and relapsed patients and patients with/without GC pulses at induction. RESULTS: A greater proportion of patients achieved a PERR with belimumab versus placebo in the newly diagnosed {69/148 [46.6%] versus 55/148 [37.2%]; odds ratio [OR] 1.36 [95% confidence interval (CI) 0.85-2.20]} and relapsed [27/75 (36.0%) versus 17/75 (22.7%); OR 2.31 (95% CI 1.07-5.01)] subgroups. Similarly for CRR [newly diagnosed: 50/148 (33.8%) versus 36/148 (24.3%); OR 1.49 (95% CI 0.88-2.51) and relapsed: 17/75 (22.7%) versus 8/75 (10.7%); OR 3.11 (95% CI 1.16-8.31)]. The probability of kidney-related event or death, or LN flare was lower with belimumab versus placebo in both subgroups. Belimumab was associated with improved kidney outcomes versus placebo with or without GC pulses at induction. CONCLUSION: Data suggest consistent benefits of belimumab on kidney outcomes for newly diagnosed and relapsed patients, and irrespective of GC pulses at induction.

Belimumab versus anifrolumab in adults with systemic lupus erythematosus: an indirect comparison of clinical response at 52 weeks.

OBJECTIVE: To generate comparative efficacy evidence of belimumab versus anifrolumab in SLE that can inform treatment practices. METHODS: The SLE Responder Index (SRI)-4 response at 52 weeks of belimumab versus anifrolumab was evaluated with an indirect treatment comparison. The evidence base consisted of randomised trials that were compiled through a systemic literature review.A feasibility assessment was performed to comprehensively compare the eligible trials and to determine the most appropriate indirect treatment comparison analysis method. A multilevel network meta-regression (ML-NMR) was implemented that adjusted for differences across trials in four baseline characteristics: SLE Disease Activity Index-2K, anti-double-stranded DNA antibody positive, low complement (C)3 and low C4. Additional analyses were conducted to explore if the results were robust to different sets of baseline characteristics included for adjustment, alternative adjustment methods and changes to the trials included in the evidence base. RESULTS: The ML-NMR included eight trials: five belimumab trials (BLISS-52, BLISS-76, NEA, BLISS-SC, EMBRACE) and three anifrolumab trials (MUSE, TULIP-1, TULIP-2). Belimumab and anifrolumab were comparable in terms of SRI-4 response (OR (95% credible interval), 1.04 (0.74-1.45)), with the direction of the point estimate slightly favouring belimumab. Belimumab had a 0.58 probability of being the more effective treatment. The results were highly consistent across all analysis scenarios. CONCLUSIONS: Our results suggest that the SRI-4 response of belimumab and anifrolumab are similar at 52 weeks in the general SLE population, but the level of uncertainty around the point estimate means we cannot rule out the possibility of a clinically meaningful benefit for either treatment. It remains to be seen if specific groups of patients could derive a greater benefit from anifrolumab or from belimumab, and there is certainly an unmet need to identify robust predictors towards more personalised selection of available biological agents in SLE.

Lexical semantic content, not syntactic structure, is the main contributor to ANN-brain similarity of fMRI responses in the language network.

Representations from artificial neural network (ANN) language models have been shown to predict human brain activity in the language network. To understand what aspects of linguistic stimuli contribute to ANN-to-brain similarity, we used an fMRI dataset of responses to n=627 naturalistic English sentences (Pereira et al., 2018) and systematically manipulated the stimuli for which ANN representations were extracted. In particular, we i) perturbed sentences' word order, ii) removed different subsets of words, or iii) replaced sentences with other sentences of varying semantic similarity. We found that the lexical semantic content of the sentence (largely carried by content words) rather than the sentence's syntactic form (conveyed via word order or function words) is primarily responsible for the ANN-to-brain similarity. In follow-up analyses, we found that perturbation manipulations that adversely affect brain predictivity also lead to more divergent representations in the ANN's embedding space and decrease the ANN's ability to predict upcoming tokens in those stimuli. Further, results are robust to whether the mapping model is trained on intact or perturbed stimuli, and whether the ANN sentence representations are conditioned on the same linguistic context that humans saw. The critical result-that lexical-semantic content is the main contributor to the similarity between ANN representations and neural ones-aligns with the idea that the goal of the human language system is to extract meaning from linguistic strings. Finally, this work highlights the strength of systematic experimental manipulations for evaluating how close we are to accurate and generalizable models of the human language network.

Immunology of pregnancy and reproductive health in autoimmune rheumatic diseases. Update from the 11th International Conference on Reproduction, Pregnancy and Rheumatic Diseases.

Autoimmune rheumatic diseases (ARD) can affect women and men during fertile age, therefore reproductive health is a priority issue in rheumatology. Many topics need to be considered during preconception counselling: fertility, the impact of disease-related factors on pregnancy outcomes, the influence of pregnancy on disease activity, the compatibility of medications with pregnancy and breastfeeding. Risk stratification and individualized treatment approach elaborated by a multidisciplinary team minimize the risk of adverse pregnancy outcomes (APO). Research has been focused on identifying biomarkers that can be predictive of APO. Specifically, preeclampsia and hypertensive disorders of pregnancy tend to develop more frequently in women with ARD. Placental insufficiency can lead to intrauterine growth restriction and small-for-gestational age newborns. Such APO have been shown to be associated with maternal disease activity in different ARD. Therefore, a key message to be addressed to the woman wishing for a pregnancy and to her family is that treatment with compatible drugs is the best way to ensure maternal and fetal wellbeing. An increasing number of medications have entered the management of ARD, but data about their use in pregnancy and lactation are scarce. More information is needed for most biologic drugs and their biosimilars, and for the so-called small molecules, while there is sufficient evidence to recommend the use of TNF inhibitors if needed for keeping maternal disease under control. Other issues related to the reproductive journey have emerged as "unmet needs", such as sexual dysfunction, contraception, medically assisted reproduction techniques, long-term outcome of children, and they will be addressed in this review paper. Collaborative research has been instrumental to reach current knowledge and the future will bring novel insights thanks to pregnancy registries and prospective studies that have been established in several Countries and to their joint efforts in merging data.

Belimumab use during pregnancy: a summary of birth defects and pregnancy loss from belimumab clinical trials, a pregnancy registry and postmarketing reports.

OBJECTIVE: Describe available data on birth defects and pregnancy loss in women with systemic lupus erythematosus (SLE) exposed to belimumab. METHODS: Data collected from belimumab clinical trials, the Belimumab Pregnancy Registry (BPR), and postmarketing/spontaneous reports up to 8 March 2020 were described. Belimumab exposure timing, concomitant medications and potential confounding factors were summarised descriptively. RESULTS: Among 319 pregnancies with known outcomes (excluding elective terminations), 223 ended in live births from which birth defects were identified in 4/72 (5.6%) in belimumab-exposed pregnancies and 0/9 placebo-exposed pregnancies across 18 clinical trials, 10/46 (21.7%) belimumab-exposed pregnancies in the BPR prospective cohort (enrolled prior to pregnancy outcome) and 0/4 belimumab-exposed pregnancies in the BPR retrospective cohort (enrolled after pregnancy outcome), and 1/92 (1.1%) in belimumab-exposed pregnancies from postmarketing/spontaneous reports. There was no consistent pattern of birth defects across datasets. Out of pregnancies with known outcomes (excluding elective terminations), pregnancy loss occurred in 31.8% (35/110) of belimumab-exposed women and 43.8% (7/16) of placebo-exposed women in clinical trials; 4.2% (2/48) of women in the BPR prospective cohort and 50% (4/8) in the BPR retrospective cohort; and 31.4% (43/137) of belimumab-exposed women from postmarketing/spontaneous reports. All belimumab-exposed women in clinical trials and the BPR received concomitant medications and had confounding factors and/or missing data. CONCLUSIONS: Observations reported here add to limited data published on pregnancy outcomes following belimumab exposure. Low numbers of exposed pregnancies, presence of confounding factors/other biases, and incomplete information preclude informed recommendations regarding risk of birth defects and pregnancy loss with belimumab use.

CELER: A 365-Participant Corpus of Eye Movements in L1 and L2 English Reading.

We present CELER (Corpus of Eye Movements in L1 and L2 English Reading), a broad coverage eye-tracking corpus for English. CELER comprises over 320,000 words, and eye-tracking data from 365 participants. Sixty-nine participants are L1 (first language) speakers, and 296 are L2 (second language) speakers from a wide range of English proficiency levels and five different native language backgrounds. As such, CELER has an order of magnitude more L2 participants than any currently available eye movements dataset with L2 readers. Each participant in CELER reads 156 newswire sentences from the Wall Street Journal (WSJ), in a new experimental design where half of the sentences are shared across participants and half are unique to each participant. We provide analyses that compare L1 and L2 participants with respect to standard reading time measures, as well as the effects of frequency, surprisal, and word length on reading times. These analyses validate the corpus and demonstrate some of its strengths. We envision CELER to enable new types of research on language processing and acquisition, and to facilitate interactions between psycholinguistics and natural language processing (NLP).

A resource-rational model of human processing of recursive linguistic structure.

A major goal of psycholinguistic theory is to account for the cognitive constraints limiting the speed and ease of language comprehension and production. Wide-ranging evidence demonstrates a key role for linguistic expectations: A word's predictability, as measured by the information-theoretic quantity of surprisal, is a major determinant of processing difficulty. But surprisal, under standard theories, fails to predict the difficulty profile of an important class of linguistic patterns: the nested hierarchical structures made possible by recursion in human language. These nested structures are better accounted for by psycholinguistic theories of constrained working memory capacity. However, progress on theory unifying expectation-based and memory-based accounts has been limited. Here we present a unified theory of a rational trade-off between precision of memory representations with ease of prediction, a scaled-up computational implementation using contemporary machine learning methods, and experimental evidence in support of the theory's distinctive predictions. We show that the theory makes nuanced and distinctive predictions for difficulty patterns in nested recursive structures predicted by neither expectation-based nor memory-based theories alone. These predictions are confirmed 1) in two language comprehension experiments in English, and 2) in sentence completions in English, Spanish, and German. More generally, our framework offers computationally explicit theory and methods for understanding how memory constraints and prediction interact in human language comprehension and production.