Garnsey's Research Provides Insights into Language Comprehension

Earliest Detectable Effects of Linguistic Anomaly Using EROS

Beckman Institute researcher Susan Garnsey uses cutting-edge technology to study how people understand, and sometimes misunderstand, the true meaning of the words they are hearing or reading.

Imagine taking offense at someone's words and then being offered this apology: "I regret my remarks offended you." Put a period after "remarks" and you have a valid apology but as it stands, such an apology would, in the terminology of Beckman Institute researcher Susan Garnsey, be leading a listener down the garden path.

In her research into language comprehension Garnsey uses the phrase "garden pathing" to describe how grammatical elements such as word order, or the addition or subtraction of a single word, or the use of an unexpected word, can reroute the meaning of a sentence in an entirely different direction from what a listener or reader was expecting.

Garnsey says that people start predicting the meaning of the rest of a sentence in mid-sentence while they are still listening or reading, and that an unexpected turn can force a major revision of their understanding so far. She uses the example of an owner of a sports team from the 1990s who was taken to task for making offensive racial comments and responded with a pseudo-apology similar to the one above.

"Being here at Beckman has been very important for me because I work at that intersection between fields, between linguistics and psychology." - Susan Garnsey

"She didn't regret her remarks, she regretted how people reacted to them," Garnsey said. "English gives her the opportunity to suggest something that's ultimately not what she said. If she said she regretted her remarks and ended it right there, then her remarks would be what she regretted. But because it went on in a different way it has this twist.

"The main thing that I study is what people's very first thought is when there are temporary ambiguities like these about what people mean. What do you think it means at the first instant, and then what do you do when you find out 'oh that is not what it meant.'"

An Associate Professor of Psychology and member of the Cognitive Neuroscience group at Beckman, Garnsey has been studying language comprehension since joining the Institute in its first full year of operation in 1989. During that time she has shown two qualities essential for any long-term research success: one is an ability to incorporate and even pioneer the use of innovative techniques for measuring language comprehension, and the other is to go wherever the research leads, even if that's halfway around the world.

Garnsey was lead author on a paper, The Contributions of Verb Bias and Plausibility to the Comprehension of Temporarily Ambiguous Sentences, which went against conventional wisdom when it first appeared in the Journal of Memory and Language in 1997. Her research demonstrated, according to her Web site, that "people's cumulative prior experience with the particular words they are reading has a strong and immediate impact on their choices about how to interpret word combinations, while factors such as the relative plausibility of the combinations take a bit longer to come into play." It was a finding that didn't fit with the literature at the time her paper was published, or with what most people would predict.

"Now it's quite consistent with the literature," Garnsey said. "At the time there were claims out there that that cumulative prior experience with the words would also be slow to have an impact. So one of the things my research demonstrated is that information about how verbs in particular are likely to be used is the kind of information that people have access to very, very quickly. Now it's pretty well accepted that that's the case."

That paper helped Garnsey gain notoriety in the field and more chances to share her findings at conferences, which in turn led to a new research path. In English, the word order of a sentence is subject-verb-object (SVO), with listeners or readers gaining information on predicting overall sentence meaning from the verb. But in Japanese the order is SOV, so verbs can't be used in the same predictive way.

"I was tired of people, whenever I gave a talk about how English speakers use verbs to make predictions, asking 'well, what do Japanese speakers do?'" Garnsey said. "They would say 'well this works in English but you can't do it in Japanese.' So I said 'OK I've got to find out what Japanese speakers do.'"

For Garnsey, that meant going to the source. During a sabbatical in 1998 she took an intensive summer course in Japanese that incorporated two semesters' worth of study into one summer session, requiring meeting four hours a day for five days over eight weeks. She continued with the third semester of Japanese during the following fall semester, and also began doing research testing native speakers of the language. During a more recent sabbatical in 2007, she traveled to both Japan and Taiwan, speaking at seminars and visiting with researchers at places like the well-known Riken Brain Science Institute in Tokyo and the Academia Sinica in Taipei.

"It gave me ideas about what other kinds of sentences I wanted to look at in both Japanese and Mandarin Chinese in the future," she said. "I also met a lot of people and heard about not-yet published research projects that are related to projects some of my students are currently working on. One graduate student I met on my travels decided to apply for a Beckman Fellowship to work with me, and I also had lots of good comments on my talks as well. I even got to practice my Japanese at times.

"Mostly I discovered," she added with a laugh, "how bad it was."

Garnsey returned from her trip with a research line that is still flourishing. She has expanded the research to include Mandarin Chinese and Korean, but is still focusing heavily on Japanese (and is actively looking for both undergraduate and graduate students who are native Japanese speakers to work in her lab). By comparing across these different languages, Garnsey has been able to gain new insight into how we process language.

"English speakers have learned, 1), that verbs are very informative, and 2), that they come early enough in the sentence that that information is useful for making predictions," Garnsey said. "Japanese and Korean speakers have learned that because the verb comes at the very end that they can't rely on that, so they rely on other kinds of information instead. One of the things that they seem to rely on is how well the meanings of the nouns fit together and constrain what the upcoming verb might be."

And that reliance on using nouns or verbs for predictive purposes is related to the speed at which our brains process language.

"Because it comes in so fast, you have to learn what you can rely on in the particular language you happen to learn," Garnsey said. "People are really, really good at figuring out what is reliable in their language and then being able to use that information fast and automatically to make predictions about what's going to come next."

That speed of processing makes measuring language comprehension difficult. Garnsey uses electrical recordings of event-related brain potentials (ERPs), a technique which only a handful of researchers were doing at the time when she began using them upon her arrival at Beckman; later she employed eye-tracking and functional Magnetic Resonance Imaging (fMRI) methods. Most recently, Garnsey took part in a collaboration that has added a powerful new measuring tool to the field of language comprehension.

The successful application of a fast optical imaging method called EROS (Event-Related Optical Signal) to measure language processing was detailed in a 2007 paper by Garnsey and Beckman collaborators Monica Fabiani, Gabriele Gratton, and Gary Dell, and their graduate students. The paper and a discussion of the method by Garnsey at a conference in Finland caused a stir in the field of language comprehension because of what the new technique offers to researchers.

"The really crucial thing this technique adds is the ability to look at the relative timing of different aspects of language comprehension," Garnsey said. "Words typically come in at a rate of about four to five per second; that's true for both reading and listening, so things are changing really fast. Any other kind of auditory input that came in as fast as the sounds of language do would just sound like a blur to you.

"Optical imaging has the kind of temporal resolution that's going to let us look at comprehension millisecond by millisecond."

Garnsey said the capability of EROS to measure language comprehension in a dynamic way will allow researchers to gain information about how and when different parts of the brain are responsible for processing information.
"Are there timing differences in terms of how quickly different kinds of information come into play, are there strength differences in different kinds of information?" she said. "I'm very excited about the optical imaging tool being able to look at those questions."

While she is excited about using EROS in future projects, the use of ERPs is still an integral part of Garnsey's research, as it is for several Beckman researchers.

"This is an incredible place for that," she said. "When I first came here in 1989, this was one of the two or three places in the world to do that sort of thing. Now it's one of the two or three places to do optical imaging as well."

Advances in technology are just one of many changes Garnsey has seen since first coming to the Beckman Institute in 1989.

"I was very excited to have space here, not just in my home department, especially once I saw the building," she said with a laugh. "It was empty back then but there is a lot more competition for the space now.

"Being here at Beckman has been very important for me because I work at that intersection between fields, between linguistics and psychology. We have our Language Processing Brownbag meeting every other Thursday here and today the speaker was a computer scientist. We have people from computer science, from linguistics, from psychology, from educational psychology, and from particular language departments. Just being here where all those people have some presence has made things so much easier and given me more ideas."