Enhancing translation of science into non-English languages.

Translating scientific findings from English to other native languages is essential to make sure that they can be integrated into timely and informed dialogue with policymakers and a diverse range of audiences who are affected by the science. Here, we present innovative approaches how to enhance access to scientific knowledge in non-English languages.

Language in Brains, Minds, and Machines.

It has long been argued that only humans could produce and understand language. But now, for the first time, artificial language models (LMs) achieve this feat. Here we survey the new purchase LMs are providing on the question of how language is implemented in the brain. We discuss why, a priori, LMs might be expected to share similarities with the human language system. We then summarize evidence that LMs represent linguistic information similarly enough to humans to enable relatively accurate brain encoding and decoding during language processing. Finally, we examine which LM properties-their architecture, task performance, or training-are critical for capturing human neural responses to language and review studies using LMs as in silico model organisms for testing hypotheses about language. These ongoing investigations bring us closer to understanding the representations and processes that underlie our ability to comprehend sentences and express thoughts in language.

Evolution of the Human Nervous System Function, Structure, and Development.

The nervous system-in particular, the brain and its cognitive abilities-is among humans' most distinctive and impressive attributes. How the nervous system has changed in the human lineage and how it differs from that of closely related primates is not well understood. Here, we consider recent comparative analyses of extant species that are uncovering new evidence for evolutionary changes in the size and the number of neurons in the human nervous system, as well as the cellular and molecular reorganization of its neural circuits. We also discuss the developmental mechanisms and underlying genetic and molecular changes that generate these structural and functional differences. As relevant new information and tools materialize at an unprecedented pace, the field is now ripe for systematic and functionally relevant studies of the development and evolution of human nervous system specializations.

A Theoretical Framework for Human and Nonhuman Vocal Interaction.

Vocal communication is a critical feature of social interaction across species; however, the relation between such behavior in humans and nonhumans remains unclear. To enable comparative investigation of this topic, we review the literature pertinent to interactive language use and identify the superset of cognitive operations involved in generating communicative action. We posit these functions comprise three intersecting multistep pathways: (a) the Content Pathway, which selects the movements constituting a response; (b) the Timing Pathway, which temporally structures responses; and (c) the Affect Pathway, which modulates response parameters according to internal state. These processing streams form the basis of the Convergent Pathways for Interaction framework, which provides a conceptual model for investigating the cognitive and neural computations underlying vocal communication across species.

The genetic legacy of the expansion of Bantu-speaking peoples in Africa.

The expansion of people speaking Bantu languages is the most dramatic demographic event in Late Holocene Africa and fundamentally reshaped the linguistic, cultural and biological landscape of the continent1-7. With a comprehensive genomic dataset, including newly generated data of modern-day and ancient DNA from previously unsampled regions in Africa, we contribute insights into this expansion that started 6,000-4,000 years ago in western Africa. We genotyped 1,763 participants, including 1,526 Bantu speakers from 147 populations across 14 African countries, and generated whole-genome sequences from 12 Late Iron Age individuals8. We show that genetic diversity amongst Bantu-speaking populations declines with distance from western Africa, with current-day Zambia and the Democratic Republic of Congo as possible crossroads of interaction. Using spatially explicit methods9 and correlating genetic, linguistic and geographical data, we provide cross-disciplinary support for a serial-founder migration model. We further show that Bantu speakers received significant gene flow from local groups in regions they expanded into. Our genetic dataset provides an exhaustive modern-day African comparative dataset for ancient DNA studies10 and will be important to a wide range of disciplines from science and humanities, as well as to the medical sector studying human genetic variation and health in African and African-descendant populations.

Persistent interaction patterns across social media platforms and over time.

Growing concern surrounds the impact of social media platforms on public discourse1-4 and their influence on social dynamics5-9, especially in the context of toxicity10-12. Here, to better understand these phenomena, we use a comparative approach to isolate human behavioural patterns across multiple social media platforms. In particular, we analyse conversations in different online communities, focusing on identifying consistent patterns of toxic content. Drawing from an extensive dataset that spans eight platforms over 34 years-from Usenet to contemporary social media-our findings show consistent conversation patterns and user behaviour, irrespective of the platform, topic or time. Notably, although long conversations consistently exhibit higher toxicity, toxic language does not invariably discourage people from participating in a conversation, and toxicity does not necessarily escalate as discussions evolve. Our analysis suggests that debates and contrasting sentiments among users significantly contribute to more intense and hostile discussions. Moreover, the persistence of these patterns across three decades, despite changes in platforms and societal norms, underscores the pivotal role of human behaviour in shaping online discourse.

Language is primarily a tool for communication rather than thought.

Language is a defining characteristic of our species, but the function, or functions, that it serves has been debated for centuries. Here we bring recent evidence from neuroscience and allied disciplines to argue that in modern humans, language is a tool for communication, contrary to a prominent view that we use language for thinking. We begin by introducing the brain network that supports linguistic ability in humans. We then review evidence for a double dissociation between language and thought, and discuss several properties of language that suggest that it is optimized for communication. We conclude that although the emergence of language has unquestionably transformed human culture, language does not appear to be a prerequisite for complex thought, including symbolic thought. Instead, language is a powerful tool for the transmission of cultural knowledge; it plausibly co-evolved with our thinking and reasoning capacities, and only reflects, rather than gives rise to, the signature sophistication of human cognition.

AI generates covertly racist decisions about people based on their dialect.

Hundreds of millions of people now interact with language models, with uses ranging from help with writing1,2 to informing hiring decisions3. However, these language models are known to perpetuate systematic racial prejudices, making their judgements biased in problematic ways about groups such as African Americans4-7. Although previous research has focused on overt racism in language models, social scientists have argued that racism with a more subtle character has developed over time, particularly in the United States after the civil rights movement8,9. It is unknown whether this covert racism manifests in language models. Here, we demonstrate that language models embody covert racism in the form of dialect prejudice, exhibiting raciolinguistic stereotypes about speakers of African American English (AAE) that are more negative than any human stereotypes about African Americans ever experimentally recorded. By contrast, the language models' overt stereotypes about African Americans are more positive. Dialect prejudice has the potential for harmful consequences: language models are more likely to suggest that speakers of AAE be assigned less-prestigious jobs, be convicted of crimes and be sentenced to death. Finally, we show that current practices of alleviating racial bias in language models, such as human preference alignment, exacerbate the discrepancy between covert and overt stereotypes, by superficially obscuring the racism that language models maintain on a deeper level. Our findings have far-reaching implications for the fair and safe use of language technology.

The language network as a natural kind within the broader landscape of the human brain.

Language behaviour is complex, but neuroscientific evidence disentangles it into distinct components supported by dedicated brain areas or networks. In this Review, we describe the 'core' language network, which includes left-hemisphere frontal and temporal areas, and show that it is strongly interconnected, independent of input and output modalities, causally important for language and language-selective. We discuss evidence that this language network plausibly stores language knowledge and supports core linguistic computations related to accessing words and constructions from memory and combining them to interpret (decode) or generate (encode) linguistic messages. We emphasize that the language network works closely with, but is distinct from, both lower-level - perceptual and motor - mechanisms and higher-level systems of knowledge and reasoning. The perceptual and motor mechanisms process linguistic signals, but, in contrast to the language network, are sensitive only to these signals' surface properties, not their meanings; the systems of knowledge and reasoning (such as the system that supports social reasoning) are sometimes engaged during language use but are not language-selective. This Review lays a foundation both for in-depth investigations of these different components of the language processing pipeline and for probing inter-component interactions.

Human-like systematic generalization through a meta-learning neural network.

The power of human language and thought arises from systematic compositionality-the algebraic ability to understand and produce novel combinations from known components. Fodor and Pylyshyn1 famously argued that artificial neural networks lack this capacity and are therefore not viable models of the mind. Neural networks have advanced considerably in the years since, yet the systematicity challenge persists. Here we successfully address Fodor and Pylyshyn's challenge by providing evidence that neural networks can achieve human-like systematicity when optimized for their compositional skills. To do so, we introduce the meta-learning for compositionality (MLC) approach for guiding training through a dynamic stream of compositional tasks. To compare humans and machines, we conducted human behavioural experiments using an instruction learning paradigm. After considering seven different models, we found that, in contrast to perfectly systematic but rigid probabilistic symbolic models, and perfectly flexible but unsystematic neural networks, only MLC achieves both the systematicity and flexibility needed for human-like generalization. MLC also advances the compositional skills of machine learning systems in several systematic generalization benchmarks. Our results show how a standard neural network architecture, optimized for its compositional skills, can mimic human systematic generalization in a head-to-head comparison.

Like-minded sources on Facebook are prevalent but not polarizing.

Many critics raise concerns about the prevalence of 'echo chambers' on social media and their potential role in increasing political polarization. However, the lack of available data and the challenges of conducting large-scale field experiments have made it difficult to assess the scope of the problem1,2. Here we present data from 2020 for the entire population of active adult Facebook users in the USA showing that content from 'like-minded' sources constitutes the majority of what people see on the platform, although political information and news represent only a small fraction of these exposures. To evaluate a potential response to concerns about the effects of echo chambers, we conducted a multi-wave field experiment on Facebook among 23,377 users for whom we reduced exposure to content from like-minded sources during the 2020 US presidential election by about one-third. We found that the intervention increased their exposure to content from cross-cutting sources and decreased exposure to uncivil language, but had no measurable effects on eight preregistered attitudinal measures such as affective polarization, ideological extremity, candidate evaluations and belief in false claims. These precisely estimated results suggest that although exposure to content from like-minded sources on social media is common, reducing its prevalence during the 2020 US presidential election did not correspondingly reduce polarization in beliefs or attitudes.

Indigenous Australian genomes show deep structure and rich novel variation.

The Indigenous peoples of Australia have a rich linguistic and cultural history. How this relates to genetic diversity remains largely unknown because of their limited engagement with genomic studies. Here we analyse the genomes of 159 individuals from four remote Indigenous communities, including people who speak a language (Tiwi) not from the most widespread family (Pama-Nyungan). This large collection of Indigenous Australian genomes was made possible by careful community engagement and consultation. We observe exceptionally strong population structure across Australia, driven by divergence times between communities of 26,000-35,000 years ago and long-term low but stable effective population sizes. This demographic history, including early divergence from Papua New Guinean (47,000 years ago) and Eurasian groups1, has generated the highest proportion of previously undescribed genetic variation seen outside Africa and the most extended homozygosity compared with global samples. A substantial proportion of this variation is not observed in global reference panels or clinical datasets, and variation with predicted functional consequence is more likely to be homozygous than in other populations, with consequent implications for medical genomics2. Our results show that Indigenous Australians are not a single homogeneous genetic group and their genetic relationship with the peoples of New Guinea is not uniform. These patterns imply that the full breadth of Indigenous Australian genetic diversity remains uncharacterized, potentially limiting genomic medicine and equitable healthcare for Indigenous Australians.

Genetic continuity and change among the Indigenous peoples of California.

Before the colonial period, California harboured more language variation than all of Europe, and linguistic and archaeological analyses have led to many hypotheses to explain this diversity1. We report genome-wide data from 79 ancient individuals from California and 40 ancient individuals from Northern Mexico dating to 7,400-200 years before present (BP). Our analyses document long-term genetic continuity between people living on the Northern Channel Islands of California and the adjacent Santa Barbara mainland coast from 7,400 years BP to modern Chumash groups represented by individuals who lived around 200 years BP. The distinctive genetic lineages that characterize present-day and ancient people from Northwest Mexico increased in frequency in Southern and Central California by 5,200 years BP, providing evidence for northward migrations that are candidates for spreading Uto-Aztecan languages before the dispersal of maize agriculture from Mexico2-4. Individuals from Baja California share more alleles with the earliest individual from Central California in the dataset than with later individuals from Central California, potentially reflecting an earlier linguistic substrate, whose impact on local ancestry was diluted by later migrations from inland regions1,5. After 1,600 years BP, ancient individuals from the Channel Islands lived in communities with effective sizes similar to those in pre-agricultural Caribbean and Patagonia, and smaller than those on the California mainland and in sampled regions of Mexico.

What neural oscillations can and cannot do for syntactic structure building.

Understanding what someone says requires relating words in a sentence to one another as instructed by the grammatical rules of a language. In recent years, the neurophysiological basis for this process has become a prominent topic of discussion in cognitive neuroscience. Current proposals about the neural mechanisms of syntactic structure building converge on a key role for neural oscillations in this process, but they differ in terms of the exact function that is assigned to them. In this Perspective, we discuss two proposed functions for neural oscillations - chunking and multiscale information integration - and evaluate their merits and limitations taking into account a fundamentally hierarchical nature of syntactic representations in natural languages. We highlight insights that provide a tangible starting point for a neurocognitive model of syntactic structure building.

The Anglo-Saxon migration and the formation of the early English gene pool.

The history of the British Isles and Ireland is characterized by multiple periods of major cultural change, including the influential transformation after the end of Roman rule, which precipitated shifts in language, settlement patterns and material culture1. The extent to which migration from continental Europe mediated these transitions is a matter of long-standing debate2-4. Here we study genome-wide ancient DNA from 460 medieval northwestern Europeans-including 278 individuals from England-alongside archaeological data, to infer contemporary population dynamics. We identify a substantial increase of continental northern European ancestry in early medieval England, which is closely related to the early medieval and present-day inhabitants of Germany and Denmark, implying large-scale substantial migration across the North Sea into Britain during the Early Middle Ages. As a result, the individuals who we analysed from eastern England derived up to 76% of their ancestry from the continental North Sea zone, albeit with substantial regional variation and heterogeneity within sites. We show that women with immigrant ancestry were more often furnished with grave goods than women with local ancestry, whereas men with weapons were as likely not to be of immigrant ancestry. A comparison with present-day Britain indicates that subsequent demographic events reduced the fraction of continental northern European ancestry while introducing further ancestry components into the English gene pool, including substantial southwestern European ancestry most closely related to that seen in Iron Age France5,6.

Evaluating large language models on medical, lay-language, and self-reported descriptions of genetic conditions.

Large language models (LLMs) are generating interest in medical settings. For example, LLMs can respond coherently to medical queries by providing plausible differential diagnoses based on clinical notes. However, there are many questions to explore, such as evaluating differences between open- and closed-source LLMs as well as LLM performance on queries from both medical and non-medical users. In this study, we assessed multiple LLMs, including Llama-2-chat, Vicuna, Medllama2, Bard/Gemini, Claude, ChatGPT3.5, and ChatGPT-4, as well as non-LLM approaches (Google search and Phenomizer) regarding their ability to identify genetic conditions from textbook-like clinician questions and their corresponding layperson translations related to 63 genetic conditions. For open-source LLMs, larger models were more accurate than smaller LLMs: 7b, 13b, and larger than 33b parameter models obtained accuracy ranges from 21%-49%, 41%-51%, and 54%-68%, respectively. Closed-source LLMs outperformed open-source LLMs, with ChatGPT-4 performing best (89%-90%). Three of 11 LLMs and Google search had significant performance gaps between clinician and layperson prompts. We also evaluated how in-context prompting and keyword removal affected open-source LLM performance. Models were provided with 2 types of in-context prompts: list-type prompts, which improved LLM performance, and definition-type prompts, which did not. We further analyzed removal of rare terms from descriptions, which decreased accuracy for 5 of 7 evaluated LLMs. Finally, we observed much lower performance with real individuals' descriptions; LLMs answered these questions with a maximum 21% accuracy.

Negation mitigates rather than inverts the neural representations of adjectives.

Combinatoric linguistic operations underpin human language processes, but how meaning is composed and refined in the mind of the reader is not well understood. We address this puzzle by exploiting the ubiquitous function of negation. We track the online effects of negation ("not") and intensifiers ("really") on the representation of scalar adjectives (e.g., "good") in parametrically designed behavioral and neurophysiological (MEG) experiments. The behavioral data show that participants first interpret negated adjectives as affirmative and later modify their interpretation towards, but never exactly as, the opposite meaning. Decoding analyses of neural activity further reveal significant above chance decoding accuracy for negated adjectives within 600 ms from adjective onset, suggesting that negation does not invert the representation of adjectives (i.e., "not bad" represented as "good"); furthermore, decoding accuracy for negated adjectives is found to be significantly lower than that for affirmative adjectives. Overall, these results suggest that negation mitigates rather than inverts the neural representations of adjectives. This putative suppression mechanism of negation is supported by increased synchronization of beta-band neural activity in sensorimotor areas. The analysis of negation provides a steppingstone to understand how the human brain represents changes of meaning over time.

Triangulation supports agricultural spread of the Transeurasian languages.

The origin and early dispersal of speakers of Transeurasian languages-that is, Japanese, Korean, Tungusic, Mongolic and Turkic-is among the most disputed issues of Eurasian population history1-3. A key problem is the relationship between linguistic dispersals, agricultural expansions and population movements4,5. Here we address this question by 'triangulating' genetics, archaeology and linguistics in a unified perspective. We report wide-ranging datasets from these disciplines, including a comprehensive Transeurasian agropastoral and basic vocabulary; an archaeological database of 255 Neolithic-Bronze Age sites from Northeast Asia; and a collection of ancient genomes from Korea, the Ryukyu islands and early cereal farmers in Japan, complementing previously published genomes from East Asia. Challenging the traditional 'pastoralist hypothesis'6-8, we show that the common ancestry and primary dispersals of Transeurasian languages can be traced back to the first farmers moving across Northeast Asia from the Early Neolithic onwards, but that this shared heritage has been masked by extensive cultural interaction since the Bronze Age. As well as marking considerable progress in the three individual disciplines, by combining their converging evidence we show that the early spread of Transeurasian speakers was driven by agriculture.

Genomic insights into the formation of human populations in East Asia.

The deep population history of East Asia remains poorly understood owing to a lack of ancient DNA data and sparse sampling of present-day people1,2. Here we report genome-wide data from 166 East Asian individuals dating to between 6000 BC and AD 1000 and 46 present-day groups. Hunter-gatherers from Japan, the Amur River Basin, and people of Neolithic and Iron Age Taiwan and the Tibetan Plateau are linked by a deeply splitting lineage that probably reflects a coastal migration during the Late Pleistocene epoch. We also follow expansions during the subsequent Holocene epoch from four regions. First, hunter-gatherers from Mongolia and the Amur River Basin have ancestry shared by individuals who speak Mongolic and Tungusic languages, but do not carry ancestry characteristic of farmers from the West Liao River region (around 3000 BC), which contradicts theories that the expansion of these farmers spread the Mongolic and Tungusic proto-languages. Second, farmers from the Yellow River Basin (around 3000 BC) probably spread Sino-Tibetan languages, as their ancestry dispersed both to Tibet-where it forms approximately 84% of the gene pool in some groups-and to the Central Plain, where it has contributed around 59-84% to modern Han Chinese groups. Third, people from Taiwan from around 1300 BC to AD 800 derived approximately 75% of their ancestry from a lineage that is widespread in modern individuals who speak Austronesian, Tai-Kadai and Austroasiatic languages, and that we hypothesize derives from farmers of the Yangtze River Valley. Ancient people from Taiwan also derived about 25% of their ancestry from a northern lineage that is related to, but different from, farmers of the Yellow River Basin, which suggests an additional north-to-south expansion. Fourth, ancestry from Yamnaya Steppe pastoralists arrived in western Mongolia after around 3000 BC but was displaced by previously established lineages even while it persisted in western China, as would be expected if this ancestry was associated with the spread of proto-Tocharian Indo-European languages. Two later gene flows affected western Mongolia: migrants after around 2000 BC with Yamnaya and European farmer ancestry, and episodic influences of later groups with ancestry from Turan.