|
A Common Benchmark for Human Language Prediction Across 118 Language Models| title | A Common Benchmark for Human Language Prediction Across 118 Language Models |
|---|
| start_date | 2026/10/22 |
|---|
| schedule | 11h30-13h |
|---|
| online | yes |
|---|
| location_info | online |
|---|
| summary | We present HumanScore-Language, a benchmark for evaluating cognitively motivated language models against behavioral and neural human language data. The benchmark covers 118 models—spanning standard LLMs, memory-constrained models, developmentally constrained models, and their combination—evaluated across 18 behavioral and neural datasets with two predictor tracks (contextual embeddings and surprisal). We describe the design and we report group comparisons. We release the benchmark and the evaluation code to support further work on cognitively motivated language modeling. |
|---|
| oncancel | https://us06web.zoom.us/j/85182424734?pwd=gZWO7nSI5lS1TXvcTACM1pPP9FLEsN.1 Password: 485583 |
|---|
| responsibles | Chemla |
|---|
Workflow history| from state (1) | to state | comment | date |
|---|
| submitted | published | | 2026/10/19 13:16 UTC |
| |
|