On the Mathematical Relationship Between Contextual Probability and N400 Amplitude

Open Mind, Journal Year: 2024, Volume and Issue: 8, P. 859 - 897

Published: Jan. 1, 2024

Accounts of human language comprehension propose different mathematical relationships between the contextual probability a word and how difficult it is to process, including linear, logarithmic, super-logarithmic ones. However, empirical evidence favoring any these over others mixed, appearing vary depending on index processing difficulty used approach taken calculate probability. To help disentangle results, we focus relationship corpus-derived N400, neural difficulty. Specifically, use 37 contemporary transformer models stimuli from 6 experimental studies test whether N400 amplitude best predicted by super-logarithmic, or sub-logarithmic transformation probabilities calculated using models, as well combinations transformed metrics. We replicate finding that some datasets, combination linearly logarithmically-transformed can predict better than either metric alone. In addition, find overall, single predictor sub-logarithmically-transformed probability, which for almost all datasets explains variance in otherwise explained linear logarithmic transformations. This novel not current theoretical accounts, thus one argue likely play an important role increasing our understanding statistical regularities impact comprehension.

Language: Английский

---

Moving beyond word frequency based on tally counting: AI-generated familiarity estimates of words and phrases are a better index of language knowledge

Published: Aug. 28, 2024

This study investigates the potential of large language models (LLMs) to estimate familiarity words and multi-word expressions (MWEs). We validated LLM estimates for isolated using existing human ratings found strong correlations. were perform even better in predicting lexical decisions naming performance megastudies than best available word frequency measures. then applied MWEs, also finding their effectiveness measuring these expressions.We created a list over 400,000 English MWEs with LLM-generated estimates, valuable resource researchers. There is cleaned-up 150,000 words, excluding lesser-known stimuli, streamline research.Our findings highlight advantages LLM-based including traditional measures (particularly recognition accuracy), ability generalize availability lists ease obtaining new all types stimuli.

Language: Английский

---

Grammar and Expectation in Active Dependency Resolution: Experimental and Modeling Evidence From Norwegian

Cognitive Science, Journal Year: 2024, Volume and Issue: 48(10)

Published: Oct. 1, 2024

Abstract Filler‐gap dependency resolution is often characterized as an active process. We probed the mechanisms that determine where and why comprehenders posit gaps during incremental processing using Norwegian our test language. First, we investigated filler‐gap suspended inside island domains like embedded questions in some languages. Processing‐based accounts hold resource limitations prevent gap‐filling across languages, while grammar‐based predict only blocked languages are grammatical islands. In a self‐paced reading study, find participants exhibit filled‐gap effects questions, which not islands The findings consistent with grammar‐based, but processing, accounts. Second, asked if can be understood special case of probabilistic ambiguity within expectation‐based framework. To do so, tested whether word‐by‐word surprisal values from neural language model could location magnitude behavioral data. accurately tracks severely underestimates their magnitude. This suggests either above beyond required to fully explain behavior or derived long‐short term memory good proxies for humans' expectations resolution.

Language: Английский

---

Prediction in reading: A review of predictability effects, their theoretical implications, and beyond

Psychonomic Bulletin & Review, Journal Year: 2024, Volume and Issue: unknown

Published: Oct. 31, 2024

Abstract Historically, prediction during reading has been considered an inefficient and cognitively expensive processing mechanism given the inherently generative nature of language, which allows upcoming text to unfold in infinite number possible ways. This article provides accessible comprehensive review psycholinguistic research that, over past 40 or so years, investigated whether readers are capable generating predictions reading, typically via experiments on effects predictability (i.e., how well a word can be predicted from its prior context). Five theoretically important issues addressed: What is best measure predictability? functional relationship between difficulty? stage(s) does affect? Are ubiquitous? processes do actually reflect? Insights computational models about manifests itself facilitate also discussed. concludes by arguing that can, certain extent, taken as demonstrating evidence but flexible component real-time language comprehension, line with broader predictive accounts cognitive functioning. However, converging evidence, especially concurrent eye-tracking brain-imaging methods, necessary refine theories prediction.

Language: Английский

---

Readers have to work harder to understand a badly translated text: an eye-tracking study into the effects of translation errors

Perspectives, Journal Year: 2024, Volume and Issue: unknown, P. 1 - 21

Published: Nov. 13, 2024

Texts are translated to be read and provide access otherwise inaccessible information or experiences. Scant empirical interest in how translations received by readers is surprising the context of our knowledge about features translations, systematic ways which they differ from originally written texts. In this paper, we explore impact translation quality on reading experience analysing cognitive effort involved text comprehension. Two groups participants (n = 64) were eye-tracked as either a low-quality (with errors) high-quality (without same source text. Overall, errors contributed longer dwell time when entire but did not significantly affect participants' comprehension scores. A more in-depth analysis shows that it depends amount confusion cause reader building coherent model

Language: Английский

---