Christianson Studies How We Interpret, and Misinterpret, Language

While Anna dressed the baby spit up on the bed.

The mind befuddles at the above sentence, used frequently by Beckman Institute faculty member Kiel Christianson in linguistics experiments. Read it again and try to figure out what happened. Did Anna dress the baby? Did the baby spit up on the bed?

Usually, more than half of test subjects in Christianson’s experiment say yes to both questions. However, by inserting a comma between “dressed” and “the” in that sentence to make it easier to parse, then the number of responders who say Anna dressed the baby are almost nil.

Those responses get to the heart of Christianson’s research, which seeks to understand how people integrate various information sources to interpret, and sometimes misinterpret, written and spoken language. A member of Beckman’s Cognitive Science group, Christianson’s perspective on language processing deviates from traditional thinking in the field in that he believes the brain sometimes takes a “good enough” approach to language comprehension.

Christianson said his use of the term “good enough” doesn’t mean sloppy or lazy.

“It means that once a certain level of interpretation that seems plausible is reached, there may be a disconnect between the interpretive processes and the more linguistic processes of building a grammatical structure,” he said. “All that might still go on, but by that time the interpretation has moved on because you have gotten to a level where everything seemed to make sense and any revision would be too costly for what it is you want to be doing with language.”

Christianson, who is director of the Educational Psychology Psycholinguistics Laboratory at Beckman and an Associate Professor of Educational Psychology at Illinois, says his research deals with the way that we interpret and comprehend language, either spoken or written, in real time.

“The cornerstone of what I am interested in is how we go from the input that decades of psycholinguistic work has suggested is processed pretty quickly and very well, how we go from a mental representation of that linguistic input, to not only comprehending language but how we interpret it within the context of everything we know about the world,” he said.

Christianson said that previous work has shown that when people are reading they usually take about a fifth of a second to extract enough information to recognize a word.

“One of my questions is related to this good enough idea: how much information is enough information, what is the critical information we need to recognize a word,” he said. “Traditionally speaking it has been considered that the true linguistic processing of the sound, the syntax, the morphology, is automatic and cost free in terms of processing; it just happens.

“If in fact this good enough processing is a real phenomenon it would suggest that there must be some costs in doing the linguistic work we do because sometimes we decide not to do it all. We are already at some threshold of understanding and full processing isn’t required to arrive at a plausible interpretation. Processing gets turned off or ignored because we’ve already moved on, so it couldn’t be quite as automatic or as cost free as we might imagine.”

Christianson’s focus as a psycholinguistics researcher is on misinterpretations in language processing that come from factors such as memory and attention allocation, syntactic constraints, and our knowledge of how and in what context words are usually used by people. Communicating language depends on mental representations that Christianson said may be underspecified due to these factors.

Christianson received his Ph.D. from Michigan State, where collaborative work with his advisor, Fernanda Ferreira, shaped the “good enough” concept. Christianson says his biggest contribution to the field is that “the mental representations of language that we hear or read – the interpretations that we take away from any linguistic input – are even more fragile than we had thought.

“They are far more prone to intrusions or to misinterpretation, in a systematic way, than previously thought. The sentence we have used a lot in our work over the years has been ‘While Anna dressed the baby spit up on the bed’ without the comma after dressed.”

Christianson said that when test subjects are asked ‘Did Anna dress the baby?’ – and a comma isn’t included in the sentence they were asked to read – that as many as 70 percent of them say ‘yes, she dressed the baby’.

“When there is a comma there, it goes down to close to zero,” he said. “This is called a garden path sentence because you are led down the garden path. By answering yes to the question did Anna dress the baby, it suggests to us that they didn’t go all the way back and clean everything up. So they did some work to find a subject for the second clause but they didn’t go back and figure out who Anna was really dressing.

“This is where the good enough idea comes from,” Christianson added. “This idea that getting a syntactic structure where the second clause has a subject, the baby, is pretty much good enough for the system at that point. They had an initial interpretation that is plausible where Anna was dressing the baby and everything is good. Oh, and the baby spit up on the bed, that makes sense too.”

But in order for the brain to put those two clauses together under the rules of English, Christianson said, it is going to have to do more work.

“As it turns out, both halves of those interpretations are good enough and they both kind of make sense. In some sense you’ve got a semantic or meaning-based interpretation that is perfectly fine. But the syntactic structure isn’t legal because if you want to make a legal sentence you have to go back and understand that Anna is not dressing the baby, she is dressing herself because she has to with that kind of a verb in English.

“So we suggest that the syntactic re-analysis that everyone has for decades assumed gets done, isn’t done or if it is done, it gets ignored. Either way, people are going with the good enough bit, they are not going back and cleaning everything up.”

Christianson’s interest in languages began at an early age in his native Minnesota. He has studied or learned to speak around a half-dozen languages over the years, including German, Yiddish, Spanish, Japanese, some Chinese, and an aboriginal North American language called Odawa. He studied that language while working on his dissertation, spending time at the tribe’s reserve on Manitoulin Island in the northern part of Lake Huron. Christianson said Odawa, unlike English, doesn’t depend on word order to communicate meaning in a sentence such as ‘The dog chased the cow.’

“Subject, object, and verb can be ordered in any six of those combinations in Odawa and not change the meaning at all,” Christianson said. “It offered a window into some of the information sources that we don’t have in English, a unique window into this information processing in understanding language.”

Information sources such as word order, whether a verb is passive or active, the context of how the information is being communicated, and visual cues all play a role in language processing and comprehension, Christianson said. They also can play a role when language is misinterpreted.

“What’s interesting to me is how we go from what is considered to be a relatively automatically computed syntactic and linguistic structure and how that maps onto an interpretation that some significant percent of the time is different from this linguistic structure that we probably should have built,” he said. “And what it is about the way that that gets done that sometimes allows for misinterpretations and how often those misinterpretations are actually created, but most of the time we don’t even notice them anyway.”

Christianson said there aren’t many theories about how we eventually arrive at an interpretation during communication.

“We’ve got a number of extremely good theories about syntactic parsing, including ones that overlap in certain ways with this good enough idea,” he said. “It is building on previous work by others in some ways but there are almost no theories about interpretation. They assume that a syntactic structure is perfectly interpreted in the way that it was intended and that doesn’t seem to be the case.”

In 2009 Christianson received a CAREER Award from the National Science Foundation to explore his theories about language processing, with a specific focus on misinterpretations. As part of that grant, he added an eye-tracker to his lab. His lab has shown though data gained from eye movement patterns and length of reading times that the processes that traditional research has shown people use to interpret and comprehend language still result in misinterpretations.

“Maybe they are doing all that work but the misinterpretation still persists, or maybe they are not doing all that work; they are only doing some of it, as I suggested with the Anna and the baby example,” Christianson said. “They are just doing enough and moving along.

“That’s the novel aspect of my research, this jump from just building a structure that is linguistically valid to what you actually take away from that and what you remember from it. There really aren’t a lot of theories of interpretation and I don’t have one yet, but that is the ultimate goal. We’re working on it.”  

Christianson said his research has translational aspects when it comes to education, second language learning, and in a collaboration with Patrick Vargas from the Department of Advertising, studying advertisement design and packaging of informational materials toward improving the interpretations people take away from those materials.

Christianson said misinterpretations have import for education and for parenting.

“The allocation of attentional resources to multiple types of information is also interesting especially in educational settings,” Christianson said. “How do you get kids to pay attention to the right stuff at the right time?

“I wish I knew,” he added with a laugh. “It would be a lot easier to get my kids through life.”