MIT researchers develop model to explain difficulty in language comprehension

Researchers from MIT's Department of Brain and Cognitive Sciences (BCS) have developed a model to better predict the ease or difficulty at which individuals can comprehend sentences, building on the necessity for a unified account of difficulties in language comprehension, a recent report by MIT News said.

The study, built off a long desire by researchers to understand what makes sentences more or less difficult to understand, was published in the Proceedings of the National Academy of Sciences. Senior authors of the paper include BCS professors Roger Levy and Edward (Ted) Gibson, alongside lead author Michael Hahn—a professor at Saarland University and former visiting student to Levy and Gibson—and Richard Futrell, a professor at the University of California at Irvine and another former student of the two BCS professors. 

Researchers measure comprehension difficulty by the time it takes readers to respond to different comprehension tasks, with longer response times correlating to more difficulty understanding the sentence, the MIT news story said. 

The recent study builds upon two existing, competing models. "While these models successfully predict specific patterns of comprehension difficulties, their predictions are limited and don't fully match results from behavioral experiments. Moreover, until recently researchers couldn't integrate these two models into a coherent account," the MIT release said. 

The researchers built "a unified theoretical account of comprehension difficulty" using the two prior models. One of the older models identified difficulty in expectation, and the other difficulty in memory retrieval as a cause for frustrated comprehension. 

Difficulty in expectation means that a sentence does not give readers the opportunity to anticipate the words to come. Difficulty in memory retrieval comes when a sentence features a "complex structure of embedded clauses," making it hard to follow, the MIT news story said. 

Futrell had theorized in 2020 that "limits in memory don't affect only retrieval in sentences with embedded clauses but plague all language comprehension" and "our memory limitations don’t allow us to perfectly represent sentence contexts during language comprehension more generally," MIT News reports. With this, he said, memory constraints can present a difficulty in the ability to anticipate upcoming words in a sentence even if it should be easily predictable from the context alone. 

While this prior study was said to predict comprehension better than other models, it did not easily identify what parts of the sentence tend to be forgotten and how this obscures comprehension, MIT News reports. The recent study, led by Hahn, is able to fill in these gaps, according to the report, stating "the mind tends to deploy its limited memory resources in a way that optimizes its ability to accurately predict new word inputs in sentences."

Participants were asked to respond to AI-generated text with complex embedded clauses and asked to predict the outcome of the sentence when nouns in the sentence were replaced. The researchers found that less common, and thus, more confusing, wording made it more difficult to retain memory and comprehension, the report said.