Papers with surprisal estimates
Surprisal from Larger Transformer-based Language Models Predicts fMRI Data More Poorly (2026.eacl-short)
Copied to clipboard
| Challenge: | Recent work has observed an inverse scaling relationship between Transformers’ per-word estimated probability and the predictive power of their surprisal estimates on reading times. |
| Approach: | They conducted a more comprehensive evaluation using surprisal estimates from 17 pre-trained LMs on two functional magnetic resonance imaging datasets. |
| Outcome: | Recent work shows that surprisal from larger Transformer-based models is less predictive of reading times, resolving the inconclusive results and indicating that this trend is not specific to latency-based measures. |
Why Does Surprisal From Larger Transformer-Based Language Models Provide a Poorer Fit to Human Reading Times? (2023.tacl-1)
Copied to clipboard
| Challenge: | Existing studies have shown that larger pre-trained language models with more parameters and lower perplexity are less predictive of human reading times. |
| Approach: | They propose to use a transformer-based model with more parameters and lower perplexity to investigate why these models are less predictive of human reading times. |
| Outcome: | The results show that the larger models with more parameters and lower perplexity are less predictive of human reading times and eye-gaze durations collected during naturalistic reading. |
Transformer-Based Language Model Surprisal Predicts Human Reading Times Best with About Two Billion Training Tokens (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Recent studies have drawn conflicting conclusions about the relationship between the quality of a language model and the ability of its surprisal estimates to predict human reading times. |
| Approach: | They propose to evaluate surprisal estimates from Transformer-based language model variants that vary systematically in the amount of training data and model capacity on their ability to predict human reading times. |
| Outcome: | The proposed model variants with contemporary model capacities provide the best fit after seeing about two billion training tokens, while smaller models show a ‘tipping point’ at convergence after the decrease in language model perplexity . |
Frequency Explains the Inverse Correlation of Large Language Models’ Size, Training Data Amount, and Surprisal’s Fit to Reading Times (2024.eacl-long)
Copied to clipboard
| Challenge: | Recent studies have shown that as Transformer-based language models become larger and are trained on very large amounts of data, the fit of their surprisal estimates to naturalistic human reading times degrades. |
| Approach: | They present a series of analyses showing that word frequency is a key explanatory factor underlying these two trends. |
| Outcome: | The results show that word frequency is a key explanatory factor underlying these two trends. |
Surprisal Estimators for Human Reading Times Need Character Models (2021.acl-long)
Copied to clipboard
| Challenge: | Experimental results show that character models can be applied to a structural parser-based processing model to calculate word generation probabilities. |
| Approach: | They propose to use a character model to calculate word generation probabilities from a structural parser-based processing model. |
| Outcome: | The proposed model performs better on self-paced reading, eye-tracking, and fMRI data than large-scale language models trained on much more data. |
Language Model Quality Correlates with Psychometric Predictive Power in Multiple Languages (2023.emnlp-main)
Copied to clipboard
| Challenge: | Existing studies have found that higher quality language models provide more powerful predictors of human reading behavior, but empirical support for the QP hypothesis is mixed. |
| Approach: | They propose to test the quality–power hypothesis by using surprisal language models to test their ability to predict eye tracking data. |
| Outcome: | The proposed model is based on a set of language models with a 'quality-power' hypothesis. |
Linear Recency Bias During Training Improves Transformers’ Fit to Reading Times (2025.coling-main)
Copied to clipboard
| Challenge: | Recent research has shown a strong fit between surprisal values from Transformers and reading times. |
| Approach: | They evaluate a Transformer model that uses a recency bias added to attention scores to improve the fit to human reading times. |
| Outcome: | The proposed model improves on a Transformer that includes a recency bias added to attention scores. |
Temperature-scaling surprisal estimates improve fit to human reading times – but does it do so for the “right reasons”? (2024.acl-long)
Copied to clipboard
| Challenge: | a wide body of evidence shows that human language processing difficulty is predicted by the information-theoretic measure surprisal, a word’s negative log probability in context. |
| Approach: | They propose to use large language models to predict the surprisal of a word's negative log probability in context to test their predictive power. |
| Outcome: | The proposed model can be significantly more accurate than humans because it has more data. |
Reverse-Engineering the Reader (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing studies have sought to determine to what extent language models can serve as useful models of human cognition by aligning them to human psychometric data. |
| Approach: | They propose a method to fine-tune a language model to implicitly optimize parameters of a linear regressor that directly predicts humans’ reading times of in-context linguistic units. |
| Outcome: | The proposed technique improves language models’ psychometric predictive power but also its perplexity on held-out test data. |
Words, Subwords, and Morphemes: What Really Matters in the Surprisal-Reading Time Relationship? (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing studies using LLMs on psycholinguistic data have gone unverified . a growing body of research is using word-level prediction as a computational proxy . |
| Approach: | They compare morphological, morphologic, and BPE tokenization estimates with reading time data. |
| Outcome: | The proposed method could be used to evaluate morphological prediction. |