{
  "id": 1759818,
  "title": "Why do you have reread some sentences and not others?",
  "url": "https://urgent.news/2026/08/18/why-do-you-have-reread-some-sentences-and-not-others",
  "topic": "science",
  "section": "Science",
  "published": "2026-08-18T17:29:05.000Z",
  "source": {
    "name": "Futurity",
    "slug": "futurity",
    "url": "https://www.futurity.org/reading-ai-language-processing-3342542/"
  },
  "original_language": "en",
  "account": "A recent study conducted by researchers from New York University and the University of Massachusetts Amherst has shed light on why some sentences in books or articles can be easily comprehended, while others require rereading. The researchers discovered that there is a partial explanation for this phenomenon in AI, specifically large language models (LLMs), but other aspects of the reading process cannot be explained by these technologies, highlighting the differences between human and AI language processing.\n\nThe study found that both humans and AI rely on next-word predictions during the initial moments of reading. However, as the complexity of the passages increases, humans' processing diverges from that of AI, which is solely built on next-word prediction. This divergence becomes particularly evident when dealing with garden-path sentences, which are grammatically correct but start in a way that leads to an incorrect initial interpretation, such as \"The old man the boat.\"\n\nThe researchers used eye-tracking technology to analyze 368 adult readers as they read and reread carefully designed sentences. While the AI models' next-word predictions could explain the first step of processing each word—identifying the word from a sequence of letters—it could not account for the next step of integrating that word into the larger context of the sentence. This process, which is particularly challenging in garden-path sentences, remains poorly understood.\n\nAccording to William Timkey, the lead author of the paper and a linguistics doctoral student at NYU, while LLMs can capture some aspects of language understanding, they fail to explain the challenges humans face when integrating words into the larger meaning of a sentence. This gap between AI predictions and human processing represents a crucial step in understanding how to improve language learning and address reading-related afflictions.\n\nBrian Dillon, a professor of linguistics at UMass Amherst and the paper's senior author, emphasizes that while AI can be valuable for cognitive science, it is not sufficient on its own. He notes that humans have unique cognitive processes, such as backward eye movements during reading, which AI models cannot explain. The researchers acknowledge that there is still much work to be done in creating computational models that more closely mimic the human mind, in order to better understand the complexities of reading.",
  "summary": "New research digs into why we breeze through some sentences in a book or article but have to reread others to comprehend their meaning.",
  "key_points": [
    "Humans and AI rely on next-word predictions during initial reading.",
    "Garden-path sentences challenge AI's ability to integrate words into sentence context.",
    "Humans use unique cognitive processes like backward eye movements not explained by AI."
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}