You might be nonplussed to find out just how fragile the meaning of words are—or you might not be, or you might actually be nonplussed, but think that you aren't because "nonplussed" for some reason means both "surprised and confused" and also "not surprised or perturbed."With so much of our communication happening online, computer scientists now have more resources than ever for tracking when and how a word's meaning changes. While it's unlikely that one person can shift a meaning alone—like poor Gretchen, trying so badly to get "fetch" to happen—sometimes single events or new products can cause a word's meaning to shift.
Advertisement
For example, when Hurricane-turned-Super Storm Sandy hit, it moved more than just the beach at Fire Island. The meaning of "sandy" expanded, moving from an adjective to a proper noun in order to include a really expensive natural disaster.
Researchers at Stony Brook University have been working out how to best track when and how the meaning of words shifts, and are drawing from some familiar sources. They just published a paper that mapped the shift by measuring the frequency with which the word is used, the patterns in which it is used and the part-of-speech its being used as.To do any of these things, though, the research drew from decades of computer science."Part of speech detection is a classical problem in natural language processing (NLP)," said Steven Skiena, the Stony Brook computer science professor who worked on the project. "The best programs learn to detect part of speech as identified in example sentences tagged by human experts. Part of speech tagging is generally quite reliable, with accuracies over 95 percent."Tracing the context in order to track the meaning is slightly more complicated, but also is a stronger indicator of a shift in meaning."'Deep Learning' methods can be used to find a point in a, say, 200-dimensional space for every word such that nearby words have similar meanings or usage patterns over a particular set of training texts," Skiena said. "These high dimensional representations capture a basic picture of what each word 'means.' Training each word over texts written in different years constructs a set of points in 'meaning space' that the word traces over time. By watching the word/point move, we can see how its usage has changed over time."
Advertisement
For a computers to "learn" to recognize that "big" is used in similar contexts as "huge" required giving said computers a ton of contextual evidence to work with. So the researchers drew from our deep linguistic wells: The Google Books Ngram Corpus was their source for big, long-term shifts, like those that have happened to "gay" over the last half century or more, when the word shifted from meaning "dapper and happy" to being homosexual.
Image: Steven Skiena/Stony Brook
For more recent and faster moving media, they looked to Twitter, obvs, and Amazon Movie Reviews, where they saw how "twilight" evolved from a noun to a proper noun thanks to certain vampire-related books (presumably the research will also be able to track the books' decline in popularity, and the return of a beautiful word to its crepuscular origins), and the popularity of "Candy Crush" shifting the usage of "candy" in statistically significant ways. (This is notable and amusing considering the developers of Candy Crush tried to trademark "candy" earlier this year.)Lest you think that only in the internet age could we be so shallow as to let a new game or product change usage, note that the researchers found that Candy Crush and Twilight weren't unprecedented."The introduction of Polystyrene in 1950 popularized the term 'plastic' as a synthetic polymer, which was once used only denote the physical property of 'flexibility.'" the paper states. "The popularity of books on dieting started with the best selling book Dr. Atkins' Diet Revolution by Robert C. Atkins in 1972. This changed the use of the word diet to mean a lifestyle of food consumption behavior and not only the food consumed by an individual or group."These are the early fruits of the research, more proof of what it can do than anything, but the potential is pretty exciting. Skiena told me that "by looking at language usage in published books, we have over 200 years of history to study, and so [we] can make an accurate call as to whether things are changing faster now than they did in past times."
Mostly I just wanted to know if we could use their research to predict where teenage slang is going next, so I can perpetually stay hip, without ever having to talk to any teenagers. So I asked him what we all wondered: is "fetch" ever going to happen?"We can tell whether fetch is happening or has happened, but not whether it is going to happen," Skiena said, crushing me. "But our methods could detect rapidly changing words on social media, so we might be the first to know it is happening…"
ONE EMAIL. ONE STORY. EVERY WEEK. SIGN UP FOR THE VICE NEWSLETTER.
By signing up, you agree to the Terms of Use and Privacy Policy & to receive electronic communications from Vice Media Group, which may include marketing promotions, advertisements and sponsored content.
Researchers at Stony Brook University have been working out how to best track when and how the meaning of words shifts, and are drawing from some familiar sources. They just published a paper that mapped the shift by measuring the frequency with which the word is used, the patterns in which it is used and the part-of-speech its being used as.To do any of these things, though, the research drew from decades of computer science."Part of speech detection is a classical problem in natural language processing (NLP)," said Steven Skiena, the Stony Brook computer science professor who worked on the project. "The best programs learn to detect part of speech as identified in example sentences tagged by human experts. Part of speech tagging is generally quite reliable, with accuracies over 95 percent."Tracing the context in order to track the meaning is slightly more complicated, but also is a stronger indicator of a shift in meaning."'Deep Learning' methods can be used to find a point in a, say, 200-dimensional space for every word such that nearby words have similar meanings or usage patterns over a particular set of training texts," Skiena said. "These high dimensional representations capture a basic picture of what each word 'means.' Training each word over texts written in different years constructs a set of points in 'meaning space' that the word traces over time. By watching the word/point move, we can see how its usage has changed over time."For a computers to "learn" to recognize that "big" is used in similar contexts as "huge" required giving said computers a ton of contextual evidence to work with. So the researchers drew from our deep linguistic wells: The Google Books Ngram Corpus was their source for big, long-term shifts, like those that have happened to "gay" over the last half century or more, when the word shifted from meaning "dapper and happy" to being homosexual.
Mostly I just wanted to know if we could use their research to predict where teenage slang is going next, so I can perpetually stay hip, without ever having to talk to any teenagers. So I asked him what we all wondered: is "fetch" ever going to happen?"We can tell whether fetch is happening or has happened, but not whether it is going to happen," Skiena said, crushing me. "But our methods could detect rapidly changing words on social media, so we might be the first to know it is happening…"