Keystrokes: A practical exploration of semantic drift in timed word association tasks.

This study investigates the phenomena of semantic drift through the lenses of language and situated simulation (LASS) and the word frequency effect (WFE) within a timed word association task. Our primary objectives were to determine whether semantic drift can be identified over the short time (25 seconds) of a free word association task (a predicted corollary of LASS), and whether more frequent terms are generated earlier in the process (as expected due to the WFE). Respondents were provided with five cue words (tree, dog, quality, plastic and love), and asked to write as many associations as they could. We hypothesized that terms generated later in the task (fourth time quartile, the last 19-25 seconds) would be semantically more distant (cosine similarity) from the cue word than those generated earlier (first quartile, the first 1-7 seconds), indicating semantic drift. Additionally, we explored the WFE by hypothesizing that earlier generated words would be more frequent and less diverse. Utilizing a dataset matched with GloVe 300B word embeddings, BERT and WordNet synsets, we analysed semantic distances among 1569 unique term pairs for all cue words across time. Our results supported the presence of semantic drift, with significant evidence of within-participant, semantic drift from the first to fourth time (LASS) and frequency (WFE) quartiles. In terms of the WFE, we observed a notable decrease in the diversity of terms generated earlier in the task, while more unique terms (greater diversity and relative uniqueness) were generated in the 4th time quartile, aligning with our hypothesis that more frequently used words dominate early stages of a word association task. We also found that the size of effects varied substantially across cues, suggesting that some cues might invoke stronger and more idiosyncratic situated simulations. Theoretically, our study contributes to the understanding of LASS and the WFE. It suggests that semantic drift might serve as a scalable indicator of the invocation of language versus simulation systems in LASS and might also be used to explore cognition within word association tasks more generally. The findings also add a temporal and relational dimension to the WFE. Practically, our research highlights the utility of word association tasks in understanding semantic drift and the diffusion of word usage over a sub-minute task, arguably the shortest practically feasible timeframe, offering a scalable method to explore group and individual changes in semantic relationships, whether via the targeted diffusion of influence in a marketing campaign, or seeking to understand differences in cognition more generally. Possible practical uses and opportunities for future research are discussed.

Shared functional specialization in transformer-based language models and the human brain.

When processing language, the brain is thought to deploy specialized computations to construct meaning from complex linguistic structures. Recently, artificial neural networks based on the Transformer architecture have revolutionized the field of natural language processing. Transformers integrate contextual information across words via structured circuit computations. Prior work has focused on the internal representations ("embeddings") generated by these circuits. In this paper, we instead analyze the circuit computations directly: we deconstruct these computations into the functionally-specialized "transformations" that integrate contextual information across words. Using functional MRI data acquired while participants listened to naturalistic stories, we first verify that the transformations account for considerable variance in brain activity across the cortical language network. We then demonstrate that the emergent computations performed by individual, functionally-specialized "attention heads" differentially predict brain activity in specific cortical regions. These heads fall along gradients corresponding to different layers and context lengths in a low-dimensional cortical space.

Urdu translation and cross-cultural validation of the stroke self-efficacy questionnaire.

BACKGROUND: The Stroke Self-Efficacy Questionnaire (SSEQ) measures the self-confidence of the individual in functional activities after a stroke. The SSEQ is a self-report scale with 13 items that assess self-efficacy after a stroke in several functional domains. OBJECTIVE: The purpose was to translate the Stroke Self-Efficacy Questionnaire into Urdu Language and to find out the validity and reliability of Urdu SSEQ among stroke patients. METHODS: The cross-cultural validation study design was used. Following COSMIN guidelines, forward and backward translation protocols were adopted. After pilot testing on 10 stroke patients, the final Urdu version was drafted. A sample of 110 stroke patients was used to evaluate the validity and reliability of the SSEQ-U. Content and Concurrent validity were determined. The intraclass correlation coefficient and Cronbach's alpha were used to measure internal consistency and test-retest reliability. Data analysis was performed using SPSS 25. RESULTS: The final version was drafted after application on 10 stroke patients. Content validity was analyzed by a content validity index ranging from 0.87 to 1. The internal consistency was calculated by Cronbach's alpha (α > 0.80). Test-retest reliability was determined by the Intra-class correlation coefficient (ICC2,1=0.956). Concurrent validity was determined by correlations with other scales by using the Spearman correlation coefficient; moderate to strong correlations (positive and negative) were found with the Functional Independence Measure (r = 0.76), Beck Depression Inventory (r=-0.54), Short Form of 12-item Scale (r = 0.68) and Fall Efficacy Scale (r = 0.82) with p < 0.05. CONCLUSION: The Urdu version was linguistically acceptable and accurate for stroke survivors for determining self-efficacy. It showed good content and concurrent validity, internal consistency and test-retest reliability.

How funny is ChatGPT? A comparison of human- and A.I.-produced jokes.

Can a large language model produce humor? Past research has focused on anecdotal examples of large language models succeeding or failing at producing humor. These examples, while interesting, do not examine ChatGPT's humor production abilities in ways comparable to humans' abilities, nor do they shed light on how funny ChatGPT is to the general public. To provide a systematic test, we asked ChatGPT 3.5 and laypeople to respond to the same humor prompts (Study 1). We also asked ChatGPT 3.5 to generate humorous satirical headlines in the style of The Onion and compared them to published headlines of the satirical magazine, written by professional comedy writers (Study 2). In both studies, human participants rated the funniness of the human and A.I.-produced responses without being aware of their source. ChatGPT 3.5-produced jokes were rated as equally funny or funnier than human-produced jokes regardless of the comedic task and the expertise of the human comedy writer.

Development of a culturally enhanced caregiver-facilitated language nutrition intervention "+Language is Medicine" to address developmental delay in Diné (Navajo) toddlers.

Introduction: Developmental Delay (DD) is highly common in American Indian and Alaska Native (AI/AN; Indigenous) toddlers and leads to high numbers of AI/AN children who eventually need special education services. AI/AN children are 2.89 times more likely to receive special education compared to other children in the U.S., yet developmental disorders are more frequently under diagnosed and untreated in AI/AN infants and toddlers. DD, which can be identified as early as toddlerhood, can lead to negative impacts on developmental trajectories, school readiness, and long-term health. Signs of DD can be identified early with proper developmental screening and remediated with high quality early intervention that includes effective parent training. There are many evidence-based language facilitation interventions often used in Early Intervention programs. However, in communities in rural parts of the Navajo Nation where there are limited services and resources, infants and toddlers with early signs of DD are often missed and do not get the culturally responsive support and evidence-based intervention they deserve. Methods: The community-based +Language is Medicine (+LiM) study team partnered with tribal home visitors, community members, and a Diné linguist/elder using a collaborative virtual workgroup approach in 2021 and 2022 to present the +LiM pilot study aims and to discuss strategies for enhancing a language intervention for toddlers experiencing DD in their tribal community. This paper will detail the stages of community engagement, intervention enhancement and preparation for field testing of the +LiM intervention to address elevated rates of DD in toddlers in the Northern Agency of the Navajo Nation. Results: Two major outcomes from this collaborative workgroup included: (1) a team-initiated redefining of language nutrition to align with Indigenous values that center cultural connectedness and native language use and (2) a five-lesson caregiver-facilitated curriculum titled +Language is Medicine which includes caregiver lessons on language nutrition, language facilitation, shared book reading, pretend play, and incorporation of native language into home routines. These two workgroup outcomes were leveraged to develop a pilot pre-/post-intervention study to test the effectiveness of the +LiM intervention with caregiver-toddler dyads living on the Navajo Nation. Discussion: Delivering tailored child interventions through tribal home visiting are cost-effective and innovative methods for reaching reservation-based families who benefit from culturally responsive parent coaching and instruction. The +LiM team has applied a precision tribal home visiting approach to enhance methods of early intervention for children with DD. Our enhancement process was grounded in Indigenous community-based participatory research that centered culture and language.

Recruiting refugees to reduce labour shortages in health care professions: experimental evidence on the potential of foreign-language outreach on social media.

BACKGROUND: Many high-income countries are grappling with severe labour shortages in the healthcare sector. Refugees and recent migrants present a potential pool for staff recruitment due to their higher unemployment rates, younger age, and lower average educational attainment compared to the host society's labour force. Despite this, refugees and recent migrants, often possessing limited language skills in the destination country, are frequently excluded from traditional recruitment campaigns conducted solely in the host country's language. Even those with intermediate language skills may feel excluded, as destination-country language advertisements are perceived as targeting only native speakers. This study experimentally assesses the effectiveness of a recruitment campaign for nursing positions in a German care facility, specifically targeting Arabic and Ukrainian speakers through Facebook advertisements. METHODS: We employ an experimental design (AB test) approximating a randomized controlled trial, utilizing Facebook as the delivery platform. We compare job advertisements for nursing positions in the native languages of Arabic and Ukrainian speakers (treatment) with the same advertisements displayed in German (control) for the same target group in the context of a real recruitment campaign for nursing jobs in Berlin, Germany. Our evaluation includes comparing link click rates, visits to the recruitment website, initiated applications, and completed applications, along with the unit cost of these indicators. We assess statistical significance in group differences using the Chi-squared test. RESULTS: We find that recruitment efforts in the origin language were 5.6 times (Arabic speakers) and 1.9 times (Ukrainian speakers) more effective in initiating nursing job applications compared to the standard model of German-only advertisements among recent migrants and refugees. Overall, targeting refugees and recent migrants was 2.4 (Ukrainians) and 10.8 (Arabic) times cheaper than targeting the reference group of German speakers indicating higher interest among these groups. CONCLUSIONS: The results underscore the substantial benefits for employers in utilizing targeted recruitment via social media aimed at foreign-language communities within the country. This strategy, which is low-cost and low effort compared to recruiting abroad or investing in digitalization, has the potential for broad applicability in numerous high-income countries with sizable migrant communities. Increased employment rates among underemployed refugee and migrant communities, in turn, contribute to reducing poverty, social exclusion, public expenditure, and foster greater acceptance of newcomers within the receiving society.

An ERP Study on the Processing of Subject-Verb and Object-Verb Gender Agreement in Punjabi.

This study was conducted with the aim of exploring the general parsing mechanisms involved in processing different kinds of dependency relations, namely verb agreement with subjects versus objects in Punjabi, an SOV Indo-Aryan language. Event related brain potentials (ERPs) were recorded as twenty-five native Punjabi speakers read transitive sentences. Critical stimuli were either fully acceptable as regards verb agreement, or alternatively violated gender agreement with the subject or object. A linear mixed-models analysis confirmed a P600 effect at the position of the verb for all violations, regardless of whether subject or object agreement was violated. These results thus suggest that an identical mechanism is involved in gender agreement computation in Punjabi regardless of whether the agreement is with the subject or the object argument.

Word Forms Reflect Trade-Offs Between Speaker Effort and Robust Listener Recognition.

How do cognitive pressures shape the lexicons of natural languages? Here, we reframe George Kingsley Zipf's proposed "law of abbreviation" within a more general framework that relates it to cognitive pressures that affect speakers and listeners. In this new framework, speakers' drive to reduce effort (Zipf's proposal) is counteracted by the need for low-frequency words to have word forms that are sufficiently distinctive to allow for accurate recognition by listeners. To support this framework, we replicate and extend recent work using the prevalence of subword phonemic sequences (phonotactic probability) to measure speakers' production effort in place of Zipf's measure of length. Across languages and corpora, phonotactic probability is more strongly correlated with word frequency than word length. We also show this measure of ease of speech production (phonotactic probability) is strongly correlated with a measure of perceptual difficulty that indexes the degree of competition from alternative interpretations in word recognition. This is consistent with the claim that there must be trade-offs between these two factors, and is inconsistent with a recent proposal that phonotactic probability facilitates both perception and production. To our knowledge, this is the first work to offer an explanation why long, phonotactically improbable word forms remain in the lexicons of natural languages.

Learning English vowels: The effects of different phonetic training modes on Arabic learners' production and perceptiona).

This study investigated the effect of different types of phonetic training on potential changes in the production and perception of English vowels by Arabic learners of English. Forty-six Arabic learners of English were randomly assigned to one of three high variability vowel training programs: Perception training (High Variability Phonetic Training), Production training, and a Hybrid Training program (production and perception training). Pre- and post-tests (vowel identification, category discrimination, speech recognition in noise, and vowel production) showed that all training types led to improvements in perception and production. There was some evidence that improvements were linked to training type: learners in the Perception Training condition improved in vowel identification but not vowel production, while those in the Production Training condition showed only small improvements in performance on perceptual tasks, but greater improvement in production. However, the effects of training modality were complicated by proficiency, with high proficiency learners benefitting more from different types of training regardless of training mode than lower proficiency learners.

Clinical efficacy of pre-trained large language models through the lens of aphasia.

The rapid development of large language models (LLMs) motivates us to explore how such state-of-the-art natural language processing systems can inform aphasia research. What kind of language indices can we derive from a pre-trained LLM? How do they differ from or relate to the existing language features in aphasia? To what extent can LLMs serve as an interpretable and effective diagnostic and measurement tool in a clinical context? To investigate these questions, we constructed predictive and correlational models, which utilize mean surprisals from LLMs as predictor variables. Using AphasiaBank archived data, we validated our models' efficacy in aphasia diagnosis, measurement, and prediction. Our finding is that LLMs-surprisals can effectively detect the presence of aphasia and different natures of the disorder, LLMs in conjunction with the existing language indices improve models' efficacy in subtyping aphasia, and LLMs-surprisals can capture common agrammatic deficits at both word and sentence level. Overall, LLMs have potential to advance automatic and precise aphasia prediction. A natural language processing pipeline can be greatly benefitted from integrating LLMs, enabling us to refine models of existing language disorders, such as aphasia.

Generative AI enhances individual creativity but reduces the collective diversity of novel content.

Creativity is core to being human. Generative artificial intelligence (AI)-including powerful large language models (LLMs)-holds promise for humans to be more creative by offering new ideas, or less creative by anchoring on generative AI ideas. We study the causal impact of generative AI ideas on the production of short stories in an online experiment where some writers obtained story ideas from an LLM. We find that access to generative AI ideas causes stories to be evaluated as more creative, better written, and more enjoyable, especially among less creative writers. However, generative AI-enabled stories are more similar to each other than stories by humans alone. These results point to an increase in individual creativity at the risk of losing collective novelty. This dynamic resembles a social dilemma: With generative AI, writers are individually better off, but collectively a narrower scope of novel content is produced. Our results have implications for researchers, policy-makers, and practitioners interested in bolstering creativity.

Large language models are changing landscape of academic publications. A positive transformation?

The advent of large language models (LLMs) based on neural networks marks a significant shift in academic writing, particularly in medical sciences. These models, including OpenAI's GPT-4, Google's Bard, and Anthropic's Claude, enable more efficient text processing through transformer architecture and attention mechanisms. LLMs can generate coherent texts that are indistinguishable from human-written content. In medicine, they can contribute to the automation of literature reviews, data extraction, and hypothesis formulation. However, ethical concerns arise regarding the quality and integrity of scientific publications and the risk of generating misleading content. This article provides an overview of how LLMs are changing medical writing, the ethical dilemmas they bring, and the possibilities for detecting AI-generated text. It concludes with a focus on the potential future of LLMs in academic publishing and their impact on the medical community.

DriveLLaVA: Human-Level Behavior Decisions via Vision Language Model.

Human-level driving is the ultimate goal of autonomous driving. As the top-level decision-making aspect of autonomous driving, behavior decision establishes short-term driving behavior strategies by evaluating road structures, adhering to traffic rules, and analyzing the intentions of other traffic participants. Existing behavior decisions are primarily implemented based on rule-based methods, exhibiting insufficient generalization capabilities when faced with new and unseen driving scenarios. In this paper, we propose a novel behavior decision method that leverages the inherent generalization and commonsense reasoning abilities of visual language models (VLMs) to learn and simulate the behavior decision process in human driving. We constructed a novel instruction-following dataset containing a large number of image-text instructions paired with corresponding driving behavior labels, to support the learning of the Drive Large Language and Vision Assistant (DriveLLaVA) and enhance the transparency and interpretability of the entire decision process. DriveLLaVA is fine-tuned on this dataset using the Low-Rank Adaptation (LoRA) approach, which efficiently optimizes the model parameter count and significantly reduces training costs. We conducted extensive experiments on a large-scale instruction-following dataset, and compared with state-of-the-art methods, DriveLLaVA demonstrated excellent behavior decision performance. DriveLLaVA is capable of handling various complex driving scenarios, showing strong robustness and generalization abilities.

Paradoxical cognitive and language function recovery by zolpidem in a patient with traumatic brain injury: A case report.

BACKGROUND: Traumatic brain injury (TBI) is a significant public health issue, often resulting from traffic accidents and falls, leading to a wide spectrum of outcomes from mild concussions to severe brain damage. The neurorehabilitation of TBI focuses on enhancing recovery and improving quality of life. Zolpidem, traditionally used for short-term management of insomnia, has shown potential in improving cognitive functions and language in TBI patients. Advances in neuroimaging techniques, such as functional near-infrared spectroscopy (fNIRS), have facilitated the exploration of the effects of therapeutic interventions on brain activity and functional connectivity in TBI patients. CASE SUMMARY: We present the case of a 34-year-old male who sustained a TBI from a traffic collision. Despite severe impairments in cognitive and language functions, administration of 10 mg of zolpidem resulted in temporary but significant improvements in these areas, as evidenced by increased Mini-Mental State Examination scores and observed behavioral changes. fNIRS assessments before and after zolpidem administration revealed notable changes in cerebral cortex activity, including increased left hemisphere activation and a shift in functional connectivity to the bilateral frontal lobes, corresponding with the patient's improvement. CONCLUSION: This case study highlights the potential of zolpidem, a medication traditionally used for insomnia, in enhancing cognitive and verbal functions in a patient with TBI, suggesting a potential therapeutic role for zolpidem in neurorehabilitation, supported by changes in brain activity and connectivity observed through fNIRS. However, further investigation is warranted to validate these findings and elucidate zolpidem's long-term effects on cognitive and functional outcomes in TBI patients.

Large Language Model-Based Responses to Patients' In-Basket Messages.

Importance: Virtual patient-physician communications have increased since 2020 and negatively impacted primary care physician (PCP) well-being. Generative artificial intelligence (GenAI) drafts of patient messages could potentially reduce health care professional (HCP) workload and improve communication quality, but only if the drafts are considered useful. Objectives: To assess PCPs' perceptions of GenAI drafts and to examine linguistic characteristics associated with equity and perceived empathy. Design, Setting, and Participants: This cross-sectional quality improvement study tested the hypothesis that PCPs' ratings of GenAI drafts (created using the electronic health record [EHR] standard prompts) would be equivalent to HCP-generated responses on 3 dimensions. The study was conducted at NYU Langone Health using private patient-HCP communications at 3 internal medicine practices piloting GenAI. Exposures: Randomly assigned patient messages coupled with either an HCP message or the draft GenAI response. Main Outcomes and Measures: PCPs rated responses' information content quality (eg, relevance), using a Likert scale, communication quality (eg, verbosity), using a Likert scale, and whether they would use the draft or start anew (usable vs unusable). Branching logic further probed for empathy, personalization, and professionalism of responses. Computational linguistics methods assessed content differences in HCP vs GenAI responses, focusing on equity and empathy. Results: A total of 16 PCPs (8 [50.0%] female) reviewed 344 messages (175 GenAI drafted; 169 HCP drafted). Both GenAI and HCP responses were rated favorably. GenAI responses were rated higher for communication style than HCP responses (mean [SD], 3.70 [1.15] vs 3.38 [1.20]; P = .01, U = 12 568.5) but were similar to HCPs on information content (mean [SD], 3.53 [1.26] vs 3.41 [1.27]; P = .37; U = 13 981.0) and usable draft proportion (mean [SD], 0.69 [0.48] vs 0.65 [0.47], P = .49, t = -0.6842). Usable GenAI responses were considered more empathetic than usable HCP responses (32 of 86 [37.2%] vs 13 of 79 [16.5%]; difference, 125.5%), possibly attributable to more subjective (mean [SD], 0.54 [0.16] vs 0.31 [0.23]; P < .001; difference, 74.2%) and positive (mean [SD] polarity, 0.21 [0.14] vs 0.13 [0.25]; P = .02; difference, 61.5%) language; they were also numerically longer (mean [SD] word count, 90.5 [32.0] vs 65.4 [62.6]; difference, 38.4%), but the difference was not statistically significant (P = .07) and more linguistically complex (mean [SD] score, 125.2 [47.8] vs 95.4 [58.8]; P = .002; difference, 31.2%). Conclusions: In this cross-sectional study of PCP perceptions of an EHR-integrated GenAI chatbot, GenAI was found to communicate information better and with more empathy than HCPs, highlighting its potential to enhance patient-HCP communication. However, GenAI drafts were less readable than HCPs', a significant concern for patients with low health or English literacy.