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

Proceedings of the National Academy of Sciences, Journal Year: 2024, Volume and Issue: 121(10)

Published: Feb. 29, 2024

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

Language: Английский

---

Alignment of brain embeddings and artificial contextual embeddings in natural language points to common geometric patterns

Nature Communications, Journal Year: 2024, Volume and Issue: 15(1)

Published: March 30, 2024

Contextual embeddings, derived from deep language models (DLMs), provide a continuous vectorial representation of language. This embedding space differs fundamentally the symbolic representations posited by traditional psycholinguistics. We hypothesize that areas in human brain, similar to DLMs, rely on represent To test this hypothesis, we densely record neural activity patterns inferior frontal gyrus (IFG) three participants using dense intracranial arrays while they listened 30-minute podcast. From these fine-grained spatiotemporal recordings, derive for each word (i.e., brain embedding) patient. Using stringent zero-shot mapping demonstrate embeddings IFG and DLM contextual have common geometric patterns. The allow us predict given left-out based solely its geometrical relationship other non-overlapping words Furthermore, show capture geometry better than static embeddings. exposes vector-based code natural processing brain.

Language: Английский

---

Diverging neural dynamics for syntactic structure building in naturalistic speaking and listening

Proceedings of the National Academy of Sciences, Journal Year: 2024, Volume and Issue: 121(11)

Published: March 5, 2024

The neural correlates of sentence production are typically studied using task paradigms that differ considerably from the experience speaking outside an experimental setting. In this fMRI study, we aimed to gain a better understanding syntactic processing in spontaneous versus naturalistic comprehension three regions interest (BA44, BA45, and left posterior middle temporal gyrus). A group participants (n = 16) was asked speak about events episode TV series scanner. Another 36) listened spoken recall participant first group. To model processing, extracted word-by-word metrics phrase-structure building with top–down bottom–up parser make different hypotheses timing structure building. While anticipates structure, sometimes before it is obvious listener, builds integratory way after all evidence has been presented. comprehension, activity found be modeled by parser, while production, parser. We additionally two strategies were developed here predictions incrementality during speaking. for highly incremental anticipatory which confirmed converging analysis pausing patterns speech. Overall, study shows feasibility studying dynamics language production.

Language: Английский

---

Artificial Neural Network Language Models Predict Human Brain Responses to Language Even After a Developmentally Realistic Amount of Training

Neurobiology of Language, Journal Year: 2024, Volume and Issue: 5(1), P. 43 - 63

Published: Jan. 1, 2024

Abstract Artificial neural networks have emerged as computationally plausible models of human language processing. A major criticism these is that the amount training data they receive far exceeds humans during learning. Here, we use two complementary approaches to ask how models’ ability capture fMRI responses sentences affected by data. First, evaluate GPT-2 trained on 1 million, 10 100 or billion words against an benchmark. We consider 100-million-word model be developmentally in terms given this similar what children are estimated exposed first years life. Second, test performance a 9-billion-token dataset reach state-of-the-art next-word prediction benchmark at different stages training. Across both approaches, find (i) already achieve near-maximal capturing sentences. Further, (ii) lower perplexity—a measure performance—is associated with stronger alignment data, suggesting received enough sufficiently high also acquire representations predictive responses. In tandem, findings establish although some necessary for ability, realistic (∼100 million words) may suffice.

Language: Английский

---

A shared model-based linguistic space for transmitting our thoughts from brain to brain in natural conversations

Neuron, Journal Year: 2024, Volume and Issue: 112(18), P. 3211 - 3222.e5

Published: Aug. 2, 2024

Language: Английский

---

Robust Effects of Working Memory Demand during Naturalistic Language Comprehension in Language-Selective Cortex

Journal of Neuroscience, Journal Year: 2022, Volume and Issue: 42(39), P. 7412 - 7430

Published: Aug. 24, 2022

To understand language, we must infer structured meanings from real-time auditory or visual signals. Researchers have long focused on word-by-word structure building in working memory as a mechanism that might enable this feat. However, some argued language processing does not typically involve rich building, and/or apparent effects are underlyingly driven by surprisal (how predictable word is context). Consistent with alternative, recent behavioral studies of naturalistic control for surprisal shown clear effects. In fMRI study, investigate range theory-driven predictors demand during comprehension humans both sexes under rigorous controls. addition, address related debate about whether the mechanisms involved specialized domain general. do so, each participant, functionally localize (1) language-selective network and (2) “multiple-demand” network, which supports across domains. Results show robust surprisal-independent no effect multiple-demand network. Our findings thus support view involves computationally demanding operations memory, addition to any prediction-related mechanisms. Further, these appear be primarily conducted same neural resources store linguistic knowledge, evidence involvement brain regions known SIGNIFICANCE STATEMENT This study uses signatures (WM) story listening, using broad theoretically motivated estimates WM demand. strong distinct predictability. demands register regions, rather than previously been associated nonlinguistic core role incremental processing, language.

Language: Английский

---

Using Computational Models to Test Syntactic Learnability

Linguistic Inquiry, Journal Year: 2022, Volume and Issue: 55(4), P. 805 - 848

Published: Oct. 7, 2022

We studied the learnability of English filler-gap dependencies and “island” constraints on them by assessing generalizations made autoregressive (incremental) language models that use deep learning to predict next word given preceding context. Using factorial tests inspired experimental psycholinguistics, we found acquire not only basic contingency between fillers gaps, but also unboundedness hierarchical implicated in dependency. evaluated a model’s acquisition island demonstrating its expectation for is attenuated within an environment. Our results provide empirical evidence against argument from poverty stimulus this particular structure.

Language: Английский

---

Neural tracking as a diagnostic tool to assess the auditory pathway

Hearing Research, Journal Year: 2022, Volume and Issue: 426, P. 108607 - 108607

Published: Sept. 14, 2022

Language: Английский

---

Prediction during language comprehension: what is next?

Trends in Cognitive Sciences, Journal Year: 2023, Volume and Issue: 27(11), P. 1032 - 1052

Published: Sept. 11, 2023

Prediction is often regarded as an integral aspect of incremental language comprehension, but little known about the cognitive architectures and mechanisms that support it. We review studies showing listeners readers use all manner contextual information to generate multifaceted predictions upcoming input. The nature these may vary between individuals owing differences in experience, among other factors. then turn unresolved questions which guide search for underlying mechanisms. (i) Is prediction essential processing or optional strategy? (ii) Are generated from within system by domain-general processes? (iii) What relationship memory? (iv) Does comprehension require simulation via production system? discuss promising directions making progress answering developing a mechanistic understanding language.

Language: Английский

---

The Efficacy of Virtual Reality in Climate Change Education Increases with Amount of Body Movement and Message Specificity

Sustainability, Journal Year: 2023, Volume and Issue: 15(7), P. 5814 - 5814

Published: March 27, 2023

Climate change impacts are felt globally, and the increasing in severity intensity. Developing new interventions to encourage behaviors that address climate is crucial. This pre-registered field study investigated how design of a virtual reality (VR) experience about ocean acidification could impact participants’ learning, behavior, perceptions through manipulation message framing, sex voice-over pace experience, amount body movement. The was run 17 locations such as museums, aquariums, arcades U.S., Canada, U.K., Denmark. movement causal mechanism, eliciting higher feelings self-efficacy while hindering learning. Moreover, linking VR narrative linguistically impaired learning compared framing did not make connection. As participants learned more they perceived risks associated with higher, were likely engage pro-climate behavior. results shed light on mechanisms behind can teach influence

Language: Английский

---