The Sentences Computers Can't Understand, But Humans Can
It’s the final video in this round of Language Files, that I collaborated on with Tom Scott and Molly Ruhl, and it’s about a kind of sentence known as the Winograd Schema.
#phm#ryland grace#rocky the eridian#project hail mary spoilers





seen from United States
seen from Russia
seen from Maldives
seen from United States
seen from United States
seen from United States

seen from United States
seen from Canada
seen from China
seen from Indonesia
seen from China
seen from United States
seen from United States
seen from T1
seen from United States
seen from Netherlands
seen from Brazil
seen from United States
seen from China

seen from United States
The Sentences Computers Can't Understand, But Humans Can
It’s the final video in this round of Language Files, that I collaborated on with Tom Scott and Molly Ruhl, and it’s about a kind of sentence known as the Winograd Schema.
Madly Ambiguous is a fill-in-the-blank game that teaches you about ambiguity while you try to trick a computer.
If you click on “Find out how it’s done”, it’s also a nice intro to automatic sentence processing.
Researchers discover that the brain proactively builds sentence structures during speech using predictive processing, explaining why second-language listening is difficult.
An interesting report of a study by Clara Cohen on using eye tracking to test how people process plurals in different languages:
Cohen suggests that for English speakers, determining whether or not a noun is plural might actually occur before a person even hears the "s" at the end of the word. In fact, there are subtle cues in the duration of singular and plural words that could help a listener predict plurality.
"One pattern that people will have heard throughout their lives as they speak English is that before a plural suffix, the stem of a noun is a little bit shorter," Cohen said. "So 'cats' with a suffix sounds shorter than 'cat.'"
Although English speakers have developed this fine-tuned language processing, speakers of many other languages, like Spanish, might not need to pay attention to subtle changes in duration since the article before a noun indicates whether it's singular or plural (e.g., el for singular or los for plural).
To test how quickly English and non-English speakers process plurality, Cohen monitors study participants with an eye-tracking device. The device works by shining a harmless and invisible infrared light on the subject's eye. Based on the position of the light's reflection, the attached optical video sensor uses an algorithm to determine the direction and duration of a subject's gaze.
For the purposes of the study, Cohen is interested in where a subject's eye travels while listening to a speaker say a sentence with either singular or plural words.
Study participants sit in a soundproof room and watch a computer screen while wearing headphones and resting their heads on a chin rest (similar to one you'd find at the eye doctor). Four images appear on the screen—a seal, a bun, a bunny and a herd of seals. Cohen's voice comes on through the headphones: "The man looked at the seals."
As the participant's eyes look toward the image of the seals, the reflection of the infrared light gives the sensor real-time information on how fast the participant processed the sentence.
(Read the whole thing.)
A gif showing how we might process garden path sentences from this website on human sentence processing. From the description:
Though we easily understand most sentences in written texts, we also stumble upon constructions such as "The horse raced past the barn fell" or "While Anna dressed the baby spit up" that challenge us cognitively. When this is the case, a garden path effect occurs: our parsing or interpretation strategy fails, causing us to first misinterpret the sentence (it leads us on the garden path) and forcing us to reanalyse the temporarily ambiguous sentence so that it can make sense (Bever, 1970; Milne, 1982; Ferreira, Christianson & Hollingworth, 2003).
The site overall has quite an accessible summary of research and methods in sentence processing, including notably a list of departments and researchers working on the topic. A useful addition to the psycholinguistics portion of the protolinguist series.
Have any of you worked with eyetracking?
I'm interested in visual world paradigms using printed words.
What have your experiences been like?