.png){kind=icon}
Closing the “thirty million word gap”: What should an intervention program look like?
Relates to Chapter 5: Learning Words
The following is a guest post by James Morgan, Professor of Cognitive, Linguistic& Psychological Sciences at Brown University. In this post, Dr. Morgan comments on Providence Talks, an intervention program aimed at reducing the language input gap for low-income children. The following PBS video describes the program and contains clips from an interview with Dr. Morgan:
http://www.pbs.org/newshour/bb/coaching-parents-toddler-talk-address-low-income-word-gap/
Providence Talks is an ambitious intervention program. Its goal is to narrow the considerable gap in linguistic input that is heard by children from lower-income families relative to their more affluent peers, and ultimately, to accelerate the language development of these disadvantaged children. The plan is to equip participating families with a LENA Foundation “word pedometer” device that provides estimates of the number of words that have been spoken by adults in the child’s presence (while filtering out sounds from television or radio). The child’s caregivers will receive feedback on the quantity of their speech, as well as coaching sessions to encourage them to be more talkative with their children.
My concern is that as originally planned, the program is missing an opportunity to collect data that will be informative enough to be truly helpful. Here, I discuss why rich data—richer than the superficial information provided by the LENA devices—are critical for the success of Providence Talks. In short, it is important for us to know not only how much language, but also what kind of language children are hearing.
Hart and Risley (1995) conducted a longitudinal study involving 42 families with varying income levels to explore why children from disadvantaged families tend to show early and persistent deficits in both language development and school achievement. They had been engaged in attempting interventions with preschool-age children, but noted, “We were also among the many who saw that our results, however promising at the start, washed out fairly early and fairly completely as children aged.” The purpose of their longitudinal study was to explore the roots of these individual differences. They observed that parents in more advantaged families in their sample directed larger volumes of speech to their young children and moreover observed a significant correlation between the quantity of speech heard in infancy and later levels of linguistic and academic achievement. Hart and Risley’s basic finding has entered the public consciousness under the rubric of “the thirty million word gap.”
Hart and Risley did not only observe differences in quantity of language across socioeconomic status (SES) strata, however; they also observed differences in quality of language and of parent-child interaction. Indeed, much of their book is devoted to quantifying “quality features of language and interaction” and analyzing their effects on developmental outcomes.
First, Hart and Risley observed that more privileged families used a much greater variety of words than did poorer families. The parents of “professional” families had an average of 2,176 different recorded words in the speech samples, whereas the parents of “welfare” families had an average of only 974 different words—less than 45% of the “professional” parents. Strikingly, by age 3, children of the “professional” families averaged 1,116 different words—more than the parents of the poorest families! Hart and Risley found that parental vocabulary range (“language diversity”) was very strongly related to language and IQ outcomes at ages 3 and 9–10.
Second, Hart and Risley observed syntactic differences in the language used by “welfare” and “professional” families—the former, for example, used larger proportions of imperative utterances, while the latter used more questions. Classic findings in studies of relations between input and language (e.g., Newport, Gleitman&Gleitman, 1977) have shown that larger proportions of imperatives are correlated with slower development of several aspects of syntax. Hart and Risley found that both syntactic diversity (“symbolic emphasis”) and reliance on questions (“guidance style”) were very strongly related to language and IQ outcomes at ages 3 and 9–10.
Third, Hart and Risley observed differences in discourse style across income groups. They noted that children of “professional” families heard about six times more parental affirmations than prohibitions, while children of “welfare” families heard twice as many prohibitions as affirmations. Hart and Risley found that parental use of affirmations (“feedback tone”) was also very strongly related to language and IQ outcomes at ages 3 and 9–10.
Finally, although Hart and Risley did not explicitly measure topical content, it is undoubtedly the case that “professional” families talked about wider ranges of topics than did “welfare” families. Thus, there are differences in the lexical, syntactic, and discourse properties of the language heard by children of families in varying SES strata, and these differences are related to developmental outcomes.
Hart and Risley’s observations were all correlational. In planning and conducting an intervention, however, it is critical to ask about causal linkages between aspects of early experience and later developmental outcomes: focusing efforts on aspects of input that turn out to have at best weak causal linkages to outcomes will be fruitless.
Quantity of language is the easiest property of input to measure, particularly with the advent of automated analysis such as that provided by the LENA devices. Encouraging parents to talk more to their young children increases parental engagement, which is an unqualified good. However, with respect to language development more specifically, while increased quantities of speech may provide infants and toddlers with expanded opportunities to practice language, in and of itself, increased quantity of language is not likely to have much effect on those abilities that lead to success in early literacy. No one really expects that inducing a parent who says “Doggie” to instead say “Doggie, doggie, doggie, doggie, doggie” will be of significant benefit to any child.
Increasing the range of words heard, on the other hand, is likely to have beneficial effects. Language development rests upon the early acquisition of certain processing skills, of which arguably the most important is the ability to recognize words in fluent speech. Fernald, Marchman&Weisleder (2013) have shown that differences in speed of word recognition across SES strata can be observed as early as 18 months; such differences have cascading effects on other aspects of language and cognitive development. Development of spoken word recognition skills involves building robust representations of possible words and making fine-grained distinctions among speech sounds; variation in experience—hearing a range of different words and listening to more than one speaker—is an important factor in acquiring these skills. Similarly, exposure to a range of syntactic patterns is important for mastering the grammatical patterns of the native language.
Will intervening to increase quantity of parental talk also serve to increase these quality features? Simply put: we do not know. Unfortunately, the current plan for gathering data in the Providence Talks program eliminates any possibility of finding out.
References
Fernald, A., Marchman, V. A., &Weisleder, A. (2013) SES differences in language processing skill and vocabulary are evident at 18 months. Developmental Science, 16, 234–248.
Hart, B., &Risley, T. R. (1995) Meaningful differences in the everyday experience of young American children. Baltimore: Paul H. Brookes.
Newport, E. L., Gleitman, H., &Gleitman, L. R. (1977) Mother, I’d rather do it myself: Some effects and noneffects of maternal speech style. In C. E. Snow & C. A. Ferguson (Eds.), Talking to children: Language input and acquisition. Cambridge, UK: Cambridge University Press, pp. 109–149.
Learning language from machines
Relates to Chapter 5: Learning Words
The evidence pointing to the important social aspects that guide word learning suggests that plunking children down in front of the TV is probably not the best way to help them to learn language—even though there’s lots of information in the signal, the cues about referential intent are sorely lacking. Interacting with other humans (or, at the very least, watching other humans interacting with each other) seems to be fairly important to the process of learning about language meaning.
But more and more, we live in a world where many of our interactions that used to involve humans are taking place with machines instead—often machines that are programmed to act like humans in many respects. We book airline tickets by talking to a computer-generated ticket agent, and we access our bank accounts though automated bank machines or Internet-based programs instead of human tellers.
As adults, we seem surprisingly willing to treat our machines as if they had human desires and goals—I once heard, for example, someone state that her word processor “had it in” for her. This eagerness to attribute human-like qualities to machines underlies the success of computer programs for simple interactions like checking out your groceries orof artificial conversational agents called “chat bots.” Chat bots have little in the way of humanoid intelligence—for example, the earliest one, ELIZA, was programmed to respond to any human-uttered sentence containing the word mother with a sympathetic Tell me about your family. But users of these programs can easily be fooled that they’re interacting with a human agent, presumably because they are so willing to project human-like goals and communicative intentions (see “Chat bot chatter” for a snippet of conversation with a chat bot). It turns out that even when we know we’re interacting with a machine, we treat them as we might a human—for instance, people routinely apologize to computer-generated voices, knowing full well they are not real people.
All of this raises an interesting question in light of our discussion of word learning: Would young children be willing to attribute referential intent to robots? And would doing so allow them to learn new words from the machine?
A clever study by Laura O’Connell and her colleagues(2009) set out to find out whether 18-month-olds would assume that, like humans, robots are likely to be referring to objects that they aim their (mechanical) eyeballs at. Using an experimental setup similar to the one used by Baldwin (1993), the researchers had a robot use a new label (for example,dax) for one of two novel objects while varying which of the two objects it was “looking” at when it uttered the word (using a recorded human voice). As when listening to human speakers, the babies checked out the eye gaze of the robot, and spent more time looking at the object that was seemingly holding the robot’s attention. But they failed to learn the word from the robot—when they were later asked by a human which object was a dax, they performed randomly. So, while they were clearly able to follow the robot’s eye gaze, it seems they didn’t take the extra step of attributing referential intent—at least not strongly enough to actually learn the word. Nor did it help matters any if the children first watched the experimenter and the robot chatting briefly. The babies still weren’t willing to see the robots as fully communicative beings (or possibly, they weren’t willing to buy into the idea that the robot belonged to the same linguistic community that they did). A later study by Yusuke Moriguchi and colleagues (2011) found similar results with older kids. Four-year-olds were completely unwilling to learn a new word from a robot, while some 5-year-olds did learn new words from a robot, but not as many as they did from a human speaker.
There’s still a lot that we don’t know about the nature of human interactions with intelligent machines, so it’s hard to know what it would take to get children to learn language from robots as if they were human conversational partners. Perhaps there are subtle clues other than eye gaze that we humans give to infants that increase their confidence in our purposeful use of language. Perhaps it’s about in-group status. Maybe babies treat all outsiders with some suspicion, not quite willing to treat them as reliable speakers of the same language, whether it’s because they dress differently, belong to a different ethnic group, or are made out of metal parts. Or maybe children need to develop a more nuanced theory of how robots work—they’re not alive, but they are intelligent and can implement programs that are purposeful and reliably consistent.
As we get a better handle on some of these questions, at some point it may become possible to create robots that serve as good language teachers for small children. But for the time being, parents and caregivers are not obsolete in this role.
Chat bot chatter
Here’s a conversation I had with Jabberwock, a chat bot. Can you tell which is the human and which is the bot?
How are you?
Great. How are you?
Coolio. How’s your boyfriend?
I don’t have a boyfriend.
I thought you said you did. Aren’t you lonely?
Not really.
Really?
I like spending time alone.
Why do you like spending time alone?
It’s just the way I’m built.
How are you drawn?
Actually, I’m a live person. Flesh and blood and all that.
I am a human and you are a bot.
I hesitate to correct you, but you are wrong.
No. You are wrong.
Never mind. How’s your girlfriend?
I don’t have one.
Straining the parser for literary effect
Relates to Chapter 9: Understanding Sentence Structure and Meaning
What makes for good writing? A sensible editorial response might be: writing that puts as little stress as possible on the language processing system. If that’s the case, no one seems to have told the novelist Henry James, famous for his convoluted syntax. Here are some typical Jamesian sentences from his 1903 novel The Ambassadors:
One of the features of the restless afternoon passed by him after Mrs. Peacock’s visit was an hour spent, shortly before dinner with Maria Gostrey, whom of late, in spite of so sustained a call on his attention from other quarters, he had by no means neglected.
Chad offered him, as always, a welcome in which the cordial and the formal—so far as the formal was the respectful—handsomely met; and after he had expressed a hope that he would let him put him up for the night, Strether was in full possession of the key, as it might have been called, to what had lately happened.
Why would a writer make a conscious choice to create sentences that wind their way through multiple embeddings, stretch syntactic dependencies over dizzying distances, and follow bizarre twists of structure, all of which make the prose harder to read? Author and critic Zadie Smith (2009) suggests that the technique was part of James’s attempt to cultivate a more acute consciousness in his reader, that his syntactic choices were “intended to make you aware, to break the rhythm that excludes thinking.”
Like most literary writers, James (and Smith) likely relied on intuitions about language. But from a scientific perspective, the idea’s not crazy. It seems strange and counterintuitive, but a number of studies suggest that when information is too easy to process fluently, people are somewhat prone to thinking less deeply, to falling back more readily on fast-but-dumb cognitive heuristics.
One intriguing example comes from a study led by Adam Alter (2007), in which subjects had to answer math problems similar to this: In a lake there is a patch of lily pads. Every day, the patch doubles in size. If it takes 48 days for the patch to cover the entire lake, how long would it take for the patch to cover half the lake?Many people mistakenly answered 24, but they were less likely to do so if they’d read the problem in a nearly illegible font. One interpretation of this intriguing finding is that the font made the problems feel harder, thereby kick-starting a more careful mode of thinking. (Or, if you will, the hard-to-read font had the effect of “breaking the rhythm that excludes thinking.”)
There are other fascinating cases that might be part of the same phenomenon. One of these is a study by Boaz Keysar and colleagues (2012) in which bilingual subjects were confronted with what’s known as “the Asian disease problem,” first studied by Amos Tversky and Daniel Kahneman (1981):
A new disease, recently emerged in Asia, has begun to spread. In the U.S., without medicine, 600,000 people will die from it. Two new medicines have been developed, but only one of them can be manufactured and distributed. You must choose which medicine to use.
If you choose medicine A, 200,000 people will be saved.
If you choose medicine B, there is a 33.33% chance that 600,000 people will be saved, and a 66.66% chance that no one will be saved.
Which medicine do you choose?
The interesting finding from this experiment is that people’s choices tend to depend heavily on the wording of the problem, a phenomenon known as the framing effect.When the problem is worded as it is above, people are more likely to choose medicine A. But they switch and are more likely to choose medicine B if they are informed that the outcome of medicine A is that 400,000 people will die, but that with medicine B there is a 33.33% chance that no one will die and a 66.66% chance that 600,000 people will die.
The careful reader will have noticed that this scenario is exactly the same as the earlier one, so it seems somewhat irrational that people would have such a different response to it. Keysar and colleagues found that their bilingual subjects showed the usual framing effects if they heard the problem in their native language.But when the problem was presented in their second language—which required more effort to process—the bias introduced by the wording vanished, and their subjects showed a more “rational” pattern of responding to both versions in the same way. Again, one possible interpretation is that the extra processing challenge of reading the problem in a foreign language triggered a more thoughtful processing mode.
Whether or not it leads to deeper thinking, there’s no doubt that reading Henry James feels more strenuous than reading the easy, breezy style of much contemporary popular writing. And this sense of effort can be leveraged for literary effect. One exceptional example can be found in “The Depressed Person,” a short story by David Foster Wallace (2007). I’ll leave it to you to identify the various structures and long dependencies that strain the parser. Add to that the actual content, and the effect is a passage that feels as exhausting and overwhelming to read as depression is itself:
The friends whom the depressed person reached out to for support and tried to open up to and share at least some contextual shape of her unceasing psychic agony and feelings of isolation with numbered around half a dozen and underwent a certain amount of rotation. The depressed person’s therapist—who had earned both a terminal graduate degree and a medical degree, and who was the self-professed exponent of a school of therapy which stressed the cultivation and regular use of a supportive peer community in any endogenously depressed adult’s journey toward self-healing—referred to these friends as the depressed person’s Support System. … The excruciating feelings of shame and inadequacy which the depressed person experienced about calling supportive members of her Support System long-distance late at night and burdening them with her clumsy attempts to articulate at least the overall context of her emotional agony were an issue on which the depressed person and her therapist were doing a great deal of work in their time together.
President George W. Bush: A modern-day Reverend Spooner?
Relates to Chapter 10: Speaking: From planning to articulation
Almost nothing is known as of yet about whether some people are simply more prone to making certain kinds of speech errors, or why that might be the case. But it’s unlikely that a tendency to make speech errors is a sign of an overall lack of intellectual endowment. It’s more likely that some disruption occurs in choosing among multiply activated representations. We know, for example, that the Reverend Spooner was a highly successful academic, but was exceedingly absentminded and susceptible to distraction.
Most of our own speech errors are evanescent. We misspeak, catch and correct ourselves, and everyone moves on. But today’s public figures, who live in a spotlight where their utterances are not only preserved but widely circulated over the Internet, often suffer unwarranted consequences of their minor lapses in language production. In many cases, speech errors are held up as examples of their specific ignorance or a general lack of intelligence.
Not many recent political figures have been as broadly ridiculed for their slips of the tongue as former U.S. president George W. Bush. It does seem as if Bush was more prone than the average politician to making speech errors, many of which probably reflect a language production glitch as opposed to ignorance, lack of education, or muddled thinking. As we’ve seen, a great deal happens in the space between thinking and speaking. To paraphrase an old proverb, “There’s many a slip twixt the thought and the lip.”
Comb through these “Bushisms” and see if you can make some guesses as to where the breakdowns occurred:
“I’m going to put people in my place, so when the history of this administration is written at least there’s an authoritarian voice saying exactly what happened.”—on what he hopes to accomplish with his memoir, as reported by the Associated Press, Calgary, Canada, March 17, 2009
“And they have no disregard for human life.”—on the brutality of Afghan fighters, Washington, DC, July 15, 2008
“Soldiers, sailors, Marines, airmen, and Coastmen—Coast Guardmen, thanks for coming, thanks for wearing the uniform.”—at the Pentagon, March 19, 2008
“And so, General, I want to thank you for your service. And I appreciate the fact that you really snatched defeat out of the jaws of those who are trying to defeat us in Iraq.”—to Army General Ray Odierno, Washington, DC, March 3, 2008
“There’s a lot of people in the Middle East who are desirous to get into the Mitchell process. And—but first things first. The—these terrorist acts and, you know, the responses have got to end in order for us to get the framework—the groundwork—not framework, the groundwork to discuss a framework for peace, to lay the—all right.”—referring to former senator George Mitchell’s report on Middle East peace, Crawford, Texas, August 13, 2001
“I know how hard it is for you to put food on your family.”—Greater Nashua, NH, January 27, 2000
“And so during these holiday seasons, we thank our blessings.”—at Fort Belvoir, VA, December 10, 2004
“It’s a time of sorrow and sadness when we lose a loss of life.”—Washington, DC, December 21, 2004
“The enemy understands a free Iraq will be a major defeat in their ideology of hatred. That’s why they’re fighting so vociferously.”—first presidential debate, Coral Gables, FL, September 30, 2004
“Let me put it to you bluntly. In a changing world, we want more people to have control over your
own life.”—Annandale, VA, August 9, 2004
“And free societies will be allies against these hateful few who have no conscience, who kill at the whim of a hat.”—Washington, DC, September 17, 2004
“The illiteracy level of our children are appalling.”—Washington, DC, January 23, 2004
“I am mindful not only of preserving executive powers for myself, but for predecessors as well.”—Washington, DC, January 29, 2001
“The law I sign today directs new funds and new focus to the task of collecting vital intelligence on terrorist threats and on weapons of mass production.”—Washington, DC, November 27, 2002
“We cannot let terrorists and rogue nations hold this nation hostile or hold our allies hostile.”—Des Moines, Iowa, August 21, 2000
“Rarely is the question asked: Is our children learning?”—Florence, SC, January 11, 2000
“And there is distrust in Washington. I am surprised, frankly, at the amount of distrust that exists in this town. And I’m sorry it’s the case, and I’ll work hard to try to elevate it.”—interview on National Public Radio, January 29, 2007
“Families is where our nation finds hope, where wings take dream.”—La Crosse, WI, October 18, 2000
“You cannot lead if you send mexed missages.”—September 30, 2004
Presuppositions and President Clinton’s re-election
Relates to Chapter 11: Discourse and Inference
Political strategist Dick Morris was in charge of Bill Clinton’s 1996 re-election campaign. In his 1997 book Behind the Oval Office, he related the following anecdote:
[Clinton’s] achievements were a problem. In strategy meetings, he often complained that he had created seven million jobs and cut the deficit but nobody seemed to notice. In speeches, he referred to the achievements awkwardly. Our polls showed audiences either already knew about them or didn’t believe they were true.
At one strategy session, Bob Squier suggested a better way to draw attention to what he had done. The key, Squier explained, was to cite the achievement while talking about something he was going to do. For example: “The hundred thousand extra police we put on the street can’t solve the crime problem by themselves; we need to keep anti-drug funding for schools in the budget and stop Republicans from cutting it.” Or: “The seven million jobs we’ve created won’t be much use if we can’t find educated people to fill them. That’s why I want a tax deduction for college tuition to help kids go on to college to take those jobs.”
Morris claims that this approach was much more effective with voters.
If you look closely at the reworked language in the new-and-improved statements, you’ll notice that they’re loaded with presuppositions. Here is some of the information that’s being presented as presupposed: Clinton’s administration put 100,000 extra police on the street; there’s a crime problem; anti-drug funding is set aside in the budget; Republicans are trying to cut this funding; Clinton’s administration created 7 million jobs; Clinton wants a tax deduction for college tuition.
Using presuppositional language provides an ideal solution to the problems that were highlighted in the campaign polls. Its function is to signal that the presupposed material is already presumed to be in the hearer’s mental model, or at the very least, to be highly consistent with it. This aligns neatly with those viewers who were already aware of Clinton’s achievements—this information is not being presented as new, but as simply referring back to something voters were already assumed to know.
But what about the other viewers, those whose mental models didn’t include these achievements, and who were in fact skeptical of Clinton’s boasts? By coding the accomplishments as presupposed, rather than directly asserting them as new information that should be added to the hearer’s mental model, the statements were subtly sending the signal that this information was generally accepted by other people, that there wasn’t anything especially controversial about it. And not surprisingly, information about what other people think can have a strong influence on opinion. For example, scores of social psychology studies have shown that people modulate their own opinions and responses quite strongly if they’re presented with evidence of a consensus of others’ opinions that collides directly with their own. Moreover, they may be less likely to critically evaluate opinions that are presented as the majority view than opinions that are claimed to be held only by a minority of people.
Presuppositions in persuasive language are discussed in detail in a book I coauthored with linguist Greg Carlson (Sedivy & Carlson, 2011). There, we use the following analogy in thinking about how Clinton’s campaign team solved the problem of getting past voters’ resistance to his claims: Think about how you might get into a room or building that you had no permission to enter. Your best strategy might be to saunter in casually and behave as if your presence there were the most unremarkable thing in the world, as if it had never occurred to you that anyone might question your right to be there—that’s what presuppositions accomplish linguistically.
Clinton is not the only politician whose boasts tend to be met with resistance. His own vice president, Al Gore, met with a great deal of derision over his assertion that his policy initiatives had helped to create the Internet—in fact, people typically misremember him as claiming that he “invented” the Internet. Who knows, perhaps the claim would have been more gently and unobtrusively entered into voters’ mental models if he’d slid it in as presupposed information, using it to preface some new idea, something like, “The policy initiatives I put in place to help start the Internet will provide a foundation for … .”
Read Language in Mind in one day? A closer look at a speed-reading app
Relates to Chapter 11: Discourse and Inference
So much to read, so little time. This wistful and widespread regret contains great commercial opportunity—since the 1960s, speed reading courses have been marketed as promises to double or quadruple the rates at which text can be absorbed and understood. In 2014, a company named Spritz offered a new twist on the same promise, by creating an app that streams text within a small viewing window one word at a time, at speeds ranging from 250 to 1000 words per minute. (On average, people normally read text at about 250–300 words per minute.) The app was presented in part as a solution for displaying text on very small digital screens. But the company also claimed that this method of presenting text allows readers to process written material at a much faster speed than is possible using traditional methods in which the eye roams over a page of text. According to the company, the problem with free-range reading (whether on paper or on screen) is that you waste a lot of time in physically moving your eyes around the text—if the text is automatically streamed within a small viewing window, this removes the need to make eye movements while reading, thereby eliminating all the time devoted to this “wasteful” activity. In fact, the company’s website asserts that “when reading, only about 20% of your time is spent processing content. The remaining 80% is spent physically moving your eyes from word to word.” Under this view, increasing the efficiency of reading essentially boils down to solving a transit problem—if only the visual information could be delivered to the brain more effectively, reading could zip along at an accelerated pace.
Spritz claims to have solved the transit problem by tinkering with a method known as Rapid Serial Visual Presentation (RSVP). RSVP is a technique that has been used by psycholinguists in the lab since the 1970s and involves presenting single words at a fixed location on a screen using a controlled rate of presentation. (For example, this method has been useful in ERP studies, because it allows for brain activity to be temporally linked to the presentation of specific words, and it eliminates any additional “noise” from brain activity that results from planning and executing eye movements during reading.) Spritz’s adaptations of RSVP for everyday reading include the use of specific algorithms to align words of varying lengths inside the viewing window rather than simply centering each word within the window. This was done because readers of languages like English, which are written from left to right, are able to process more letters to the right than to the left of where their eyes are fixating. As a result, when viewing longer words, the optimal location for eye fixation is toward the left edge of the word rather than at its center. The Spritz app takes this into consideration. The app also varies the amount of time that a word spends on the screen depending on its length, and it allowed for extra time at the ends of sentences, where readers exhibit well-known “wrap up” effects that slow down their processing, as readily seen in self-paced reading tasks.
These adaptations probably do make reading a bit smoother than the typical lab experiment that relies on RSVP. Still, few psycholinguists would recommend Spritz as an optimal reading experience. Before you read any further, see if you can anticipate some plausible objections based on what you already know about how people process language while reading.
One difficulty with the Spritz method is that the speed at which a word can be read depends on a great many factors other than the length of the word or whether it appears at the end of a sentence. The frequency of a word matters, as does its predictability in that particular context. The complexity of the syntactic structure in which it appears can also affect how quickly it can be processed. Even if the app could take these factors into consideration (a programming feat which, while sophisticated, wouldn’t be impossible), there are other sources of variability for the reading times of individual words that would be much harder to predict and build into the app. For example, the degree to which a reader is familiar with a certain topic can affect reading times, as can individual differences in memory span or cognitive control. This makes it unlikely that a one-size-fits-all algorithm could ever provide the best relative rate of presentation of all the words of a sentence for any given context or reader. (Sure, the reader can choose an overall faster or slower rate of presentation when using the app, but will the amount of time that each individual word spends on the screen really represent the amount of time needed to process that particular piece of information by that particular reader?)
Especially dubious is the claim made by the company that a mere 20% of reading time is typically spent “processing content” as opposed to moving the eyes around the page. The company’s website provides no references for this claim, nor does it precisely explain what it means by “processing content.” Under a very generous interpretation of the claim, perhaps what is meant is that individual words can be recognized in a fraction of the time that people normally spend gazing at them in a sentence. For example, people show behavioral evidence of having recognized a word even if it’s presented subliminally at a rate faster than 50 milliseconds (though they’re not usually consciously aware of having recognized the word at this speed). This might give the impression that it’s the sluggishness of eye movements that creates a bottleneck for reading speed. But reading involves a great deal more than recognizing the individual words that are strung together in a sentence: readers have to integrate each word into a meaningful structure; build a mental model of the events it describes, and generate inferences about meaning that may not even be linguistically encoded. As you’ve seen throughout Chapters 9 and 11, some of these processes can take considerable time, and there’s some evidence that when readers slow down in order to generate richer inferences and more detailed mental models, they may retain more information. It seems more than a stretch to assert that the time it takes to program eye movements is what limits the speed of reading.
In fact, far from representing a wasteful activity, eye movements may provide readers with a tremendously useful tool for language comprehension. Reading allows us to do something that is impossible to do when comprehending live spoken language: slow down or speed up the flow of information, and even backtrack to revisit earlier information in a sentence. About 10–15% of eye movements during reading go backwards, and these regressive eye movements, as they’re called, likely play an important role in reading comprehension. Elizabeth Schotter and her colleagues demonstrated this in a study in which the words in a sentence were transformed into strings of dummy characters once the eyes of the reader had alighted on the them and moved on. (For example, once read, the word sentence would become xxxxxxxx). Essentially, the act of reading through a sentence left behind a wake of meaningless symbols, rendering any regressive eye movements completely uninformative. The researchers found that the comprehension of complex sentences suffered when readers were prevented from getting meaningful information from regressive eye movements compared with conditions in which the content of the sentence was undisturbed after it had been viewed. Based on this study, they suggested that methods such as RSVP, which preclude backtracking to repair lapses in comprehension, suffer from a fatal design flaw that inevitably impairs comprehension.
Some form of RSVP may well turn out to be a viable way to read through certain kinds of texts—in particular, texts with simpler sentences, sentences in which the complexity of information is evenly distributed, sentences that avoid unexpected twists or turns, and sentences which explicitly lay out their intended meanings, without requiring readers to mentally fill in the blanks through inference. But a Spritz-style app is unlikely to provide a satisfying method for reading texts that involve complex or unfamiliar ideas, that require deep processing, or that use language in innovative ways. Some might argue that this app is especially unsuitable for many of the texts that are actually worth reading.
References
Schotter, E.R. , Tran, R., & Rayner, K. (2014) Don't believe what you read (only once): Comprehension is supported by regressions during reading. Psychological Science, 25.
How do actors remember all those words?
Relates to Chapter 12: The Social Side of Language
What’s the longest piece of text you can recite from memory? These days, few of us ever have to remember and reproduce lengthy stretches of language. Actors (especially stage actors), who do so on a regular basis, appear to be the elite athletes of verbal memory and are often asked how they manage to learn their lines. But surprisingly, they tend not to devote much time to the specific task of memorizing dialogue in the form of focused repetition and rehearsal of their lines. How is it that they can remember so much with so little apparent effort?
The key word here is “apparent.” Actors exert a great deal of effort. It’s just that very little of it is directed at memorization itself. Instead, remembering their lines is a byproduct of the deep analysis of the text they perform as they prepare to put on a play. An actor’s main task is to discern and project the internal mental states and intentions of the character she is portraying. In art, perhaps even more than in life, the connection between motive and utterance can be very indirect. When was the last time you saw a play in which one character declared to another, “I feel utter contempt for you!” It’s more likely that you, the audience, is meant to infer the character’s attitude from veiled comments and subtle body language. In order for you to be able to do that, the actor has to give a coherent presentation, one in which the character’s words, manner of speaking, and body language all align to reflect the underlying mental state. This means that the actor has to develop a deep sense of the character’s interior life by paying very close attention to the words that character utters throughout the play. The script itself doesn’t come annotated with overt explanations of the characters’ motivations—all of the hidden emotional agendas, relational tensions, and deep-seated drives that make the play interesting are indirectly communicated by the playwright to the actors in the form of carefully-crafted dialogue. The actor’s job is to decipher and amplify these inner states of mind.
Tony and Helga Noice (1997) studied what actors did when they were asked to learn a role in preparation for audition. They found that the actors focused their efforts on explaining their character’s underlying thoughts and goals. They segmented the script into “beats,” with each beat reflecting a distinct sub-goal for the character in that scene. In comparison with a group of non-actors who were given the same script to prepare, they generated much more detailed analyses, with a greater focus on their character’s perspective. Not surprisingly, they remembered more of the script than the non-actors. What’s especially interesting is that, left to their own devices to pick apart the script as they normally would, the actors remembered more of the script verbatim than when they were explicitly instructed to memorize it through rote repetition.
These results line up nicely with research on memory, which shows that people remember content better if they elaborate it in their minds—thinking about verbal material is usually a much better strategy for remembering it than specifically trying to memorize it. What’s striking is that in the study by the Noices, the detailed analysis performed by the actors helped them to remember not only the gist of their lines, but also the verbatim detail.
But maybe this shouldn’t be surprising, if you think carefully about the actor’s job of reconstructing the character’s mental state. As you read in the discussion on conversational implicature in Chapter 12, how something is expressed can be just as revealing of the speaker’s intent as the overall gist of what is said. Consider the different messages and attitudes that a character might be trying to convey in each of the following pairs of similar sentences:
This is my mother.
This is my biological mother.
I thought you were so wise.
I used to think that you were so wise.
You look nice today.
Today, you look nice.
In normal conversation, we might implicitly ask ourselves “Now, why did the speaker choose to use those particular words in this specific context?” and conversational inferences represent our answers to such questions. In analyzing a script, actors explicitly ask themselves these questions. It’s not hard to see how, after dissecting their lines in this way, with such attentive focus on their character’s specific choice of words, actors would be likely to remember their lines, word for word.
In fact, Tony and Helga Noice discovered that actors actively resisted learning their lines through rote repetition. A number of the actors reported that they hated being in the experimental condition focused on rote memorization. Some confessed that even when they tried to just repeat the words to themselves, they couldn’t help thinking of the character’s possible motives and intentions. It seems that actors generally don’t think of memorization as an important part of their job description. One actor described it this way:
“This has been an interesting process for me, this memorizing a part—because it isn’t words. The actor creates all that stuff that’s underneath the words, and the words are just the topping, the words are the froth on top of the beer. The actor has to create the mug, and the hops, and the beer, and the fermentation of all that; and when that happens, the foam just gets there—by some natural ‘blump.’ It’s just there. It just happens.” (Noice, 1992, p. 422)
References
Noice, T., & Noice, H. (1997) The Nature of Expertise in Professional Acting. Mahwah, NJ: Erlbaum.
Noice, H. (1992) Elaborative memory strategies of professional actors. Applied Cognitive Psychology, 6, 417–427.
Absurdity, intent, and meaning in art
Relates to Chapter 12: The Social Side of Language
Have you ever read a surreal story, or stood in front of a strange piece of art, and wondered, “But what is it supposed to mean?” You may even have felt irritated—I know someone who’s prone to feeling actual rage after reading an avant-garde poem.
What’s this unsettled (or even angry) feeling all about? At its heart, it reflects a failure to get at the intent of the writer or artist. You assume he must have had one. But the norms you’ve come to expect of the genre have been violated. You begin to suspect that the true intent of the artist was to confuse you, to alienate you, or make you feel dumb for not “getting” it. Dang it, the fellow just seems so deliberately … uncooperative.
When confronted with a clearly intentional product like a poem or a painting, we humans seem driven to pursue its intended meaning until we reach a satisfactory conclusion. When that doesn’t happen, we exhibit some interesting reactions, as discovered by several social scientists who have made a career out of studying failures to recover meaning. For example, Travis Proulx, Steven Heine, and their colleagues have published a series of studies looking at how people responded to strange and unconventional art forms, including an absurdist story by Franz Kafka, a strange Monty Python parody, and the surreal painting Son of Man by René Magritte, which depicts a man wearing a bowler hat with his face obscured by a hovering green apple).
The researchers found that exposing people to bewildering art has some weird lingering effects. For example, people do better at finding the statistical patterns in a set of stimuli generated by an artificial grammar; they’re more likely to feel aligned with their ethnic or national identity; they’re more likely to say that a prostitute should be punished in a mock courtroom scenario. What do all of these effects have in common? According to Proulx and Heine (2010), these reactions all reflect a drive to reaffirm meaning when a sense of meaning has been threatened—whether that drive is expressed by finding patterns in abstract stimuli, valuing traditional identities, or aligning with cultural norms of acceptable behavior. The researchers argue that when a sense of coherence or meaning is undermined, people reach for whatever meaning structures are readily on hand to compensate for the unsettling experience. (Notice the irony: the sociopolitical impact of avant-garde art—not normally seen as a funding priority for right-of-center political groups—seems to be to trigger responses that are normally associated with a conservative worldview.)
Now, Proulx and Heine construe the notion of “meaning” very broadly, far beyond our discussion of conventional linguistic meaning and speakers’ intended meanings. Their research program consists of showing that various “meaning threats,” whether achieved by a Kafka story or by getting people to think about their own mortality, are equivalent in terms of their effects on people’s need to reaffirm meaning.
But their studies on absurdist art are relevant to our discussion here about communicative purpose, because the strange responses that people have to the absurdist art seem to vanish if they’re given some clues about the intent of the creator. For instance, in one study, some readers who were given an absurd story were first informed that it was meant to be a joke, a Monty Python parody of adventure stories for young boys. But other readers were led to believe that they were about to read a classic adventure story for young boys—only the latter group showed any signs of “meaning threat” (Proulx et al., 2010).
This suggests that being able to connect an author’s or artist’s purpose to the way in which the work is implemented is essential, and that when the audience can’t easily do this, the disorienting effect can ripple out in unexpected ways.