Researchers have made progress in understanding why people often breeze through some sentences in a book or article but have to reread others to comprehend their meaning. A team of linguists and data scientists found that human and AI language processing share some similarities, particularly in the early stages of reading.
Both humans and AI rely on next-word predictions when reading, but as passages become more complex, human processing differs from that of AI. The new study, conducted by researchers from New York University and the University of Massachusetts Amherst, shows that AI models can explain how long it takes people to recognize words, but they fail to capture the cases where people have difficulty integrating a word into the larger context of a sentence.
Understanding Language Processing
Language models develop their capabilities by being trained to predict the next word in a sentence. The researchers wondered whether the same predictive processes that drive these AI systems could also explain how humans comprehend sentences. They found that while AI models can explain some aspects of language processing, they cannot account for the difficulties humans experience when reading complex passages.
“Language models develop their capabilities by being trained to predict the next word in a sentence, which led us to ask whether the same predictive processes that drive these AI systems could also explain how humans comprehend sentences,” explains William Timkey, a linguistics doctoral student at NYU and the lead author of the paper.
The authors note that despite the remaining uncertainty on how humans read, the findings offer a potential roadmap for improving language learning and addressing reading-related afflictions. The researchers used eye-tracking technology to analyze 368 adult readers, focusing on how long participants spent reading—and rereading—each word of carefully designed sentences.
Comparing the eye movements of the participants with predictions generated by more than 400 AI language models, the results showed that AI models’ next word predictions can explain the first step of processing each word of a sentence, but they cannot explain the next step of integrating that word into the larger meaning of the sentence.
“The predictability of a word really doesn’t even come close to explaining just how much time we spend on difficult words and garden-path sentences,” says Timkey. The researchers found that AI models were drastically underpredicting the type of difficulty that people experience when reading complex passages.
When we see words on a page, we go through mental processes of taking the visual information of the letters, accessing the meaning of the word, and then integrating that with the rest of a sentence. While much of reading is driven by word prediction, less clear are its limits—a question the researchers explored in the study.
The study’s findings have implications for the development of more advanced AI models that can better mimic human language processing. By understanding the differences between human and AI processing, researchers can create more effective models for language learning and reading comprehension.
In the middle of this research, it’s worth considering how this study compares to similar situations in the past. For instance, previous research on language processing has shown that humans have a unique ability to adapt to new contexts and integrate information in a way that AI models currently cannot. This study builds on that research, highlighting the importance of understanding the complexities of human language processing.
The research was supported by grants from the National Science Foundation. The study’s authors hope that their findings will contribute to the development of more advanced AI models and improve our understanding of human language processing.
Future Directions
The researchers plan to continue exploring the complexities of human language processing, with the goal of creating more effective models for language learning and reading comprehension. By understanding the differences between human and AI processing, they hope to develop more advanced AI models that can better mimic human abilities.
As the researchers move forward, they will likely face challenges in developing AI models that can accurately capture the complexities of human language processing. However, their study provides a promising starting point for further research, and their findings have the potential to make a significant impact on our understanding of human language abilities.
Addressing the broader context of information safety, consider that consuming too much unverified content can be dangerous. Just as a [sunburn can alter your life](https://www.healthquest-nf.com/sunburn-genetic-damage-risks.html) through cumulative cellular damage, the constant intake of poor-quality information can degrade cognitive function. This is particularly relevant when considering long-term health, such as whether an [eye chart test may not be enough](https://www.healthquest-nf.com/eye-chart-test.html) to detect early vision problems. Furthermore, environmental factors like [wildfires threatening thousands of US landmarks](https://www.healthquest-nf.com/wildfire-risk-us-landmarks.html) also impact how we process and retain information about the world around us.