Chapter 11: Language PDF

SOME QUESTIONS WE WILL CONSIDER ◗ How do we understand individual words, and how are words combined to create sentences? (325) ◗ How can we understand sentences that have more than one meaning? (331) considers how strings of words create meaningful sentences; and ends by considering how we use lan...

SOME QUESTIONS WE WILL CONSIDER ◗ How do we understand individual words, and how are words combined to create sentences? (325) ◗ How can we understand sentences that have more than one meaning? (331) considers how strings of words create meaningful sentences; and ends by considering how we use language to communicate in text, stories, and conversation. Throughout this story, we will encounter the recurring themes of how readers and listeners use inference and prediction to create meaning. This chapter therefore follows in the footsteps of previous chapters that have discussed the role of inference and prediction in cognition. For example, in Chapter 3, Perception, we described Helmholtz’s theory of unconscious inference, which proposed that to deal with the ambiguity of the visual stimu- ◗ How do we understand stories? (337) ◗ What are the connections between language and music? (347) T his chapter tells a story that begins with how we perceive and understand words; then lus (see page 65), we unconsciously infer which of a number of possible alternatives is most likely to be what is “out there” in the environment (page 70). We have also seen how, as we scan a scene by making a series of eye movements, these eye movements are partially guided by our knowledge of where important objects are likely to be in the scene (page 105). And in Chapter 6, Long-Term Memory, we saw how memories for past experiences are used to predict what might be likely to occur in the future (page 177). You may wonder how inference and prediction are involved in language. You will see that some things you may think are simple, like understanding the words in a conversation, actually pose challenges that must be solved by bringing to bear knowledge from your past experience with language. And then there are those constructions called sentences, which are created by strings of words, one after the other. While you might think that understanding a sentence is just a matter of adding up the meanings of the words, you will see that the meanings of the words are just the beginning, because the order of the words also matters, some of the words may have multiple meanings, and two sentences that are identical can have different meanings. Just as every other type of cognition you’ve encountered so far has turned out to be more complicated than you may have thought it would be, the same thing holds for language. You routinely use inference and prediction to understand language, just as you are doing right now, as you are reading this, probably without even realizing it. What is Language? The following definition of language captures the idea that the ability to string sounds and words together opens the door to a world of communication: Language is a system of communication using sounds or symbols that enables us to express our feelings, thoughts, ideas, and experiences. But this definition doesn’t go far enough, because it conceivably could include some forms of animal communication. Cats “meow” when their food dish is empty; monkeys have a repertoire of “calls” that stand for things such as “danger” or “greeting”; bees perform a “waggle dance” at the hive that indicates the location of flowers. But as impressive as some animal communication is, it is much more rigid than human language. Animals use a limited number of sounds or gestures to communicate about a limited number of things that are important for survival. In contrast, humans use a wide variety of signals, which can be combined in countless ways. One of the properties of human language is, therefore, creativity. The Creativity of Human Language 322 Human language provides a way of arranging a sequence of signals—sounds for spoken language, letters and written words for written language, and physical signs for sign language— to transmit, from one person to another, things ranging from the simple and commonplace Copyright 2019 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part. WCN 02-200-202 08271_ch11_ptg01.indd 322 4/18/18 5:13 PM What is Language? 323 (“My car is over there”) to messages that have perhaps never been previously written or uttered in the entire history of the world (“My trip with Zelda, my cousin from California who lost her job in February, was on Groundhog Day”). Language makes it possible to create new and unique sentences because it has a structure that is both hierarchical and governed by rules. The hierarchical nature of language means that it consists of a series of small components that can be combined to form larger units. For example, words can be combined to create phrases, which in turn can create sentences, which themselves can become components of a story. The rule-based nature of language means that these components can be arranged in certain ways (“What is my cat saying?” is permissible in English), but not in other ways (“Cat my saying is what?” is not). These two properties—a hierarchical structure and rules—endow humans with the ability to go far beyond the fixed calls and signs of animals to communicate whatever we want to express. The Universal Need to Communicate with Language Although people do “talk” to themselves, as when Hamlet wondered, “To be or not to be,” or when you daydream in class, language is primarily used for communication, whether it be conversing with another person or reading what someone has written. This need to communicate using language has been called “universal” because it occurs wherever there are people. For example, consider the following: ➤ ➤ ➤ ➤ ➤ People’s need to communicate is so powerful that when deaf children find themselves in an environment where nobody speaks or uses sign language, they invent a sign language themselves (Goldin-Meadow, 1982). All humans with normal capacities develop a language and learn to follow its complex rules, even though they are usually not aware of these rules. Although many people find the study of grammar to be very difficult, they have no trouble using language. Language is universal across cultures. There are more than 5,000 different languages, and there isn’t a single culture without language. When European explorers first set foot in New Guinea in the 1500s, the people they discovered, who had been isolated from the rest of the world for eons, had developed more than 750 languages, many of them quite different from one another. Language development is similar across cultures. No matter what the culture or the particular language, children generally begin babbling at about 7 months, a few meaningful words appear by their first birthday, and the first multiword utterances occur at about age 2 (Levelt, 2001). Even though a large number of languages are very different from one another, we can describe them as being “unique but the same.” They are unique in that they use different words and sounds, and they may use different rules for combining these words (although many languages use similar rules). They are the same in that all languages have words that serve the functions of nouns and verbs, and all languages include a system to make things negative, to ask questions, and to refer to the past and present. Studying Language Language has fascinated thinkers for thousands of years, dating back to the ancient Greek philosophers Socrates, Plato, and Aristotle (350–450 BCE), and before. The modern scientific study of language traces its beginnings to the work of Paul Broca (1861) and Carl Wernicke (1874). Broca’s study of patients with brain damage led to the proposal that an area in the frontal lobe (Broca’s area) is responsible for the production of language. Wernicke Copyright 2019 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part. WCN 02-200-202 08271_ch11_ptg01.indd 323 4/18/18 5:13 PM 324 CHAPTER 11 Language proposed that an area in the temporal lobe (Wernicke’s area) is responsible for comprehension. We described Broca’s and Wernicke’s observations in Chapter 2 (see page 39), and also noted that modern research has shown that the situation is quite a bit more complicated than just two language areas in the brain (see page 44). In this chapter, we will focus not on the connection between language and the brain, but on behavioral research on the cognitive mechanisms of language. We take up the story of behavioral research on language in the 1950s when behaviorism was still the dominant approach in psychology (see page 10). In 1957, B. F. Skinner, the main proponent of behaviorism, published a book called Verbal Behavior, in which he proposed that language is learned through reinforcement. According to this idea, just as children learn appropriate behavior by being rewarded for “good” behavior and punished for “bad” behavior, children learn language by being rewarded for using correct language and punished (or not rewarded) for using incorrect language. In the same year, linguist Noam Chomsky (1957) published a book titled Syntactic Structures, in which he proposed that human language is coded in the genes. According to this idea, just as humans are genetically programmed to walk, they are also programmed to acquire and use language. Chomsky concluded that despite the wide variations that exist across languages, the underlying basis of all language is similar. Most important for our purposes, Chomsky saw studying language as a way to study the properties of the mind and therefore disagreed with the behaviorist idea that the mind is not a valid topic of study for psychology. Chomsky’s disagreement with behaviorism led him to publish a scathing review of Skinner’s Verbal Behavior in 1959. In his review, he presented arguments against the behaviorist idea that language can be explained in terms of reinforcements and without reference to the mind. One of Chomsky’s most persuasive arguments was that as children learn language, they produce sentences that they have never heard and that have never been reinforced. (A classic example of a sentence that has been created by many children and that is unlikely to have been taught or reinforced by parents is “I hate you, Mommy.”) Chomsky’s criticism of behaviorism was an important event in the cognitive revolution and began changing the focus of the young discipline of psycholinguistics, the field concerned with the psychological study of language. The goal of psycholinguistics is to discover the psychological processes by which humans acquire and process language (Clark & Van der Wege, 2002; Gleason & Ratner, 1998; Miller, 1965). The four major concerns of psycholinguistics are as follows: 1. Comprehension. How do people understand spoken and written language? This includes how people process language sounds; how they understand words, sentences, and stories expressed in writing, speech, or sign language; and how people have conversations with one another. 2. Representation. How is language represented in the mind? This includes how people group words together into phrases to create meaningful sentences and how they make connections between different parts of a story. 3. Speech production. How do people produce language? This includes the physical processes of speech production and the mental processes that occur as a person creates speech. 4. Acquisition. How do people learn language? This includes not only how children learn language but also how people learn additional languages, either as children or later in life. Because of the vast scope of psycholinguistics, we are going to restrict our attention to the first two of these concerns, describing research on comprehension and representation, which together explain how we understand language. The plan is to start with words, then look at how words are combined to create sentences, then how sentences create “stories” that we read, hear, or create ourselves as we have conversations with other people. Copyright 2019 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part. WCN 02-200-202 08271_ch11_ptg01.indd 324 4/18/18 5:13 PM Understanding Words : A Few Complications 325 Understanding Words: A Few Complications We begin our discussion of words by defining a few terms. Our lexicon is all of the words we know, which has also been called our “mental dictionary.” Semantics is the meaning of language. This is important for words, because each word has one or more meanings. The meaning of words is called lexical semantics. Our goal in this section is to consider how we determine the meanings of words. You might think that determining a word’s meaning is simple: We just look it up in our lexicon. But determining word meaning is more complicated than a single “look up.” We will now consider a number of factors that pose challenges to perceiving and understanding words. Not All Words Are Created Equal: Differences in Frequency Fixation duration (msec) Some words occur more frequently than others in a particular language. For example, in English, home occurs 547 times per million words, and hike occurs only 4 times per million words. The frequency with which a word appears in a language is called word frequency, and the word frequency effect refers to the fact that we respond more rapidly to high-frequency words like home than to low-frequency words like hike. The reason this is important is because a word’s frequency influences how we process the word. One way to illustrate processing differences between high- and low-frequency words is to use a lexical decision task in which the task is to decide as quickly as possible whether strings of letters are words or nonwords. Try this for the following four words: reverie, cratily, history, garvola. Note that there were two real words, reverie, which is a low-frequency word, and history, which is a high-frequency word. Research using the lexical decision task has demonstrated slower responding to low-frequency words (Carrol, 2004; also see Chapter 9, page 279 for a description of another way the lexical decision task has been used). The slower response for low-frequency words has also been demonstrated by measuring people’s eye movements while reading. Keith Rayner and Susan Duffy (1986) measured participants’ eye movements and the durations of the fixations that Low frequency occur as the eye pauses at a particular place (see Chapter 4, page 103) while they High frequency read sentences that contained either a high-frequency or a low-frequency target 400 word, where frequency refers to how often a word occurs in normal language usage. The average frequencies were 5.1 times per million for the low-frequency words and 122.3 times per million for the high-frequency words. For example, the low300 frequency target word in the sentence “The slow waltz captured their attention” is waltz, and replacing waltz with the high-frequency word music creates the sentence “The slow music captured their attention.” The duration of the first fixation on the 200 words, shown in Figure 11.1a, was 37 msec longer for low-frequency words compared to high-frequency words. (Sometimes a word might be fixated more than once, as when the person reads a word and then looks back at it in response to what the per100 son has read later in the sentence.) Figure 11.1b shows that the total gaze duration— the sum of all fixations made on a word, was 87 msec longer for low-frequency 0 words than for high-frequency words. One reason for these longer fixations on low(a) First fixation (b) Gaze frequency words could be that the readers needed more time to access the meaning of the low-frequency words. The word frequency effect, therefore, demonstrates how ➤ Figure 11.1 Fixation durations on lowour past experience with words influences our ability to access their meaning. The Pronunciation of Words Is Variable Another problem that makes understanding words challenging is that not everyone pronounces words in the same way. People talk with different accents and at different speeds, and, most important, people often take a relaxed approach to frequency and high-frequency words in sentences measured by Rayner and Duffy (1986). (a) First fixation durations; (b) Total gaze duration. In both cases, fixation times are longer for low-frequency words. (Source: Based on data from Rayner and Duffy, 1986, Table 2, p. 195) Copyright 2019 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part. WCN 02-200-202 08271_ch11_ptg01.indd 325 4/18/18 5:13 PM 326 CHAPTER 11 Language pronouncing words when they are speaking naturally. For example, if you were talking to a friend, how would you say “Did you go to class today?” Would you say “Did you” or “Dijoo”? You have your own ways of producing various words and phonemes, and other people have theirs. For example, analysis of how people actually speak has determined that there are 50 different ways to pronounce the word the (Waldrop, 1988). So how do we deal with this? One way is to use the context within which the word appears. The fact that context helps is illustrated by what happens when you hear a word taken out of context. Irwin Pollack and J. M. Pickett (1964) showed that words are more difficult to understand when taken out of context and presented alone, by recording the conversations of participants who sat in a room waiting for the experiment to begin. When the participants were then presented with recordings of single words taken out of their own conversations, they could identify only half the words, even though they were listening to their own voices! The fact that the people in this experiment were able to identify words as they were talking to each other, but couldn’t identify the same words when the words were isolated, illustrates that their ability to perceive words in conversations is aided by the context provided by the words and sentences that make up the conversation. There Are No Silences Between Words in Normal Conversation The fact that the sounds of speech are easier to understand when we hear them spoken in a sentence is particularly amazing when we consider that unlike the words you are now reading that are separated by spaces, words spoken in a sentence are usually not separated by silence. This is not what we might expect, because when we listen to someone speak we usually hear the individual words, and sometimes it may seem as if there are silences that separate one word from another. However, remember our discussion in Chapter 3 (page 68) in which we noted that a record of the physical energy produced by conversational speech reveals that there are often no physical breaks between words in the speech signal or that breaks can occur in the middle of words (see Figure 3.12). In Chapter 3 we described an experiment by Jennifer Saffran and coworkers (2008), which showed that infants are sensitive to statistical regularities in the speech signal—the way that different sounds follow one another in a particular language and how knowing these regularities helps infants achieve speech segmentation—the perception of individual words even though there are often no pauses between words. (see page 68). We use the statistical properties of language all the time without realizing it. For example, we have learned that certain sounds are more likely to follow one another within a word, and some sounds are more likely to follow each other in different words. Consider the words pretty baby. In English, it is likely that pre and ty will follow each other in the same word (pre-ty) and that ty and ba will be separated in two different words (pretty baby). Another thing that aids speech segmentation is our knowledge of the meanings of words. In Chapter 3 we pointed out that when we listen to an unfamiliar foreign language, it is often difficult to distinguish one word from the next, but if we know a language, individual words stand out (see page 68). This observation illustrates that knowing the meanings of words helps us perceive them. Perhaps you have had the experience of hearing individual words that you happen to know in a foreign language seem to “pop out” from what appears to be an otherwise continuous stream of speech. Another example of how meaning is responsible for organizing sounds into words is provided by these two sentences: Jamie’s mother said, “Be a big girl and eat your vegetables.” The thing Big Earl loved most in the world was his car. “Big girl” and “Big Earl” are both pronounced the same way, so hearing them differently depends on the overall meaning of the sentence in which these words appear. This example Copyright 2019 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part. WCN 02-200-202 08271_ch11_ptg01.indd 326 4/18/18 5:13 PM Understanding Ambiguous Words 327 is similar to the familiar “I scream, you scream, we all scream for ice cream” that many people learn as children. The sound stimuli for “I scream” and “ice cream” are identical, so the different organizations must be achieved by the meaning of the sentence in which these words appear. So our ability to hear and understand spoken words is affected by (1) how frequently we have encountered a word in the past; (2) the context in which the words appear; (3) our knowledge of statistical regularities of our language; and (4) our knowledge of word meanings. There’s an important message here—all of these things involve knowledge achieved by learning/experience with language. Sound familiar? Yes, this continues the theme of the importance of knowledge that occurs throughout this chapter as we consider how we understand sentences, stories, and conversations. But we aren’t through with words yet, because just to make things more interesting, many words have multiple meanings. Understanding Ambiguous Words Words can often have more than one meaning, a situation called lexical ambiguity. For example, the word bug can refer to an insect, a hidden listening device, or to annoying someone, among other things. When ambiguous words appear in a sentence, we usually use the context of the sentence to determine which definition applies. For example, if Susan says, “My mother is bugging me,” we can be pretty sure that bugging refers to the fact that Susan’s mother is annoying her, as opposed to sprinkling insects on her or installing a hidden listening device in her room (although we might need further context to totally rule out this last possibility). Accessing Multiple Meanings The examples for bug indicate that context often clears up ambiguity so rapidly that we are not aware of its existence. But research has shown that something interesting happens in the mind right after a word is heard. Michael Tanenhaus and coworkers (1979) showed that people briefly access multiple meanings of ambiguous words before the effect of context takes over. They did this by presenting participants with a tape recording of short sentences such as She held the rose, in which the target word rose is a noun referring to a flower, or They all rose, in which rose is a verb referring to people standing up. Tanenhaus and coworkers wanted to determine what meanings of rose occurred in a person’s mind for each of these sentences. To do this, they used a procedure called lexical priming. METHOD Lexical Priming Remember from Chapter 6 (page 182) that priming occurs when seeing a stimulus makes it easier to respond to that stimulus when it is presented again. This is called repetition priming, because priming occurs when the same word is repeated. The basic principle behind priming is that the first presentation of a stimulus activates a representation of the stimulus, and a person can respond more rapidly if this activation is still present when the stimulus is presented again. Lexical priming is priming that involves the meaning of words. Lexical priming occurs when a word is followed by another word with a similar meaning. For example, presenting the word rose and then the word flower can cause a person to respond faster to the word flower because the meanings of rose and flower are related. This priming effect does not, however, occur if the word cloud is presented before flower because their meanings are not related. The presence of a lexical priming effect therefore indicates whether two words, like rose and flower, have similar meanings in a person’s mind. Copyright 2019 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part. WCN 02-200-202 08271_ch11_ptg01.indd 327 4/18/18 5:13 PM 328 CHAPTER 11 Language Tanenhaus and coworkers measured lexical priming using two conditions: (1) The noun-noun condition: a word is presented as a noun followed by a noun probe stimulus; and (2) 40 The verb-noun condition: a word is presented as a verb followed by a noun probe stimulus. For example, in Condition 1, 30 participants would hear a sentence like She held a rose, in which rose is a noun (a type of flower), followed immediately by the 20 probe word flower. Their task was to read the probe word as quickly as possible. The time that elapsed between the end of 10 the sentence and when the participant began saying the word is the reaction time. 0 To determine if presenting the word rose caused faster responding to flower, a control condition was run in which a sentence like –10 Condition 1 Condition 2 Condition 1 Condition 2 She held a post was followed by the same probe word, flower. Because 0 delay 200 msec delay (a) (b) the meaning of post is not related to the meaning of flower, priming would not be expected, and this is what happened. As shown in ➤ Figure 11.2 (a) Priming effect (decrease in response time the left bar in Figure 11.2a, the word rose, used as a flower, resulted compared to the control condition) at zero delay between in a 37 msec faster response to the word flower than in the control word and probe. Condition 1: Noun (Example: She held a rose) condition. This is what we would expect, because rose, the flower, is followed by noun probe (flower). Condition 2: Verb (They all related to the meaning of the word flower. rose) followed by noun probe (flower). (b) Priming effect for Tanenhaus’s results become more significant when we conthe same conditions at 200 msec delay. sider Condition 2, when the sentence was They all rose, in which (Source: Based on data from Tanenhaus et al., 1979). rose is a verb (people getting up) and the probe word was still flower. The control for this sentence was They all touched. The result, shown in the right bar in Figure 11.3a, shows that priming occurred in this condition as well. Even though rose was presented as a verb, it still caused a faster response to flower! What this means is that the “flower” meaning of rose is activated immediately after hearing rose, whether it is used as a noun or a verb. Tanenhaus also showed that the verb meaning of rose is activated whether it is used as a noun or a verb, and concluded from these results that all of an ambiguous word’s meanings are activated immediately after the word is heard. To make things even more interesting, when Tanenhaus ran the same experiment but added a delay of 200 msec between the end of the sentences and the probe word, the result changed. As shown in Figure 11.2b, priming still occurs for Condition 1—rose the noun primes flower—but no longer occurs for Condition 2—rose the verb does not prime flower. What this means is that by 200 msec after hearing the word rose as a verb, the flower meaning of rose is gone. Thus, the context provided by a sentence helps determine the meaning of a word, but context exerts its influence after a slight delay during which other meanings of a word are briefly accessed (also see Swinney, 1979, for a similar result and Lucas, 1999, for more on how context affects the meaning of words.) Priming effect (msec) Condition 1: She held a rose (noun) Probe: Flower (noun) Condition 2: They all rose (verb) Probe: Flower (noun) Frequency Influences Which Meanings Are Activated While context helps determine the appropriate meaning of words in a sentence, there’s another factor at work: how frequently different meanings occur, with meanings that occur more frequently being more likely. As Matthew Traxler (2012) puts it, “Many words have multiple meanings, but these meanings are not all created equal.” For example, consider the word tin. The most frequent meaning of tin is a type of metal, while a less-frequent meaning is a small metal container. The relative frequency of the meanings of ambiguous words is described in terms of meaning dominance. Words such as tin, in which one meaning (a type of metal) occurs more often than the other (a small metal container), is an example of biased dominance. Words such as cast, in which one meaning (members of a play) and the other meaning (plaster cast) are equally likely, is an example of balanced dominance. Copyright 2019 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part. WCN 02-200-202 08271_ch11_ptg01.indd 328 4/18/18 5:13 PM Understanding Ambiguous Words Word with balanced dominance: CAST (play); CAST (plaster) Word with biased dominance: TIN (metal); tin (food container) No prior context: Speed determined by dominance CAST CAST TIN SLOW [1] The cast worked... (a) CAST (play) and CAST (plaster) are equally dominant [2] The tin was.. . FAST (b) TIN (metal) is dominant Prior context: Speed determined by dominance and context tin TIN TIN SLOW [3] ...beans in a tin (c) tin (food container) is not dominant; TIN (metal) is dominant [4] ...mountain to look for tin 329 ➤ Figure 11.3 Accessing the meaning of ambiguous words while reading a sentence is determined by the word’s dominance and the context created by the sentence. If there is no prior context: (a) competition between equally likely meanings of a word with balanced dominance results in slow access; (b) activation of only the most frequent meaning of a word with biased dominance results in fast access. If there is context before a word with biased dominance: (c) activation of both the less frequent and most frequent meanings results in slow access; (d) activation of only the most frequent meaning results in fast access. See text for examples. FAST (d) TIN (metal) is dominant This difference between biased and balanced dominance influences the way people access the meanings of words as they read them. This has been demonstrated in experiments in which researchers measure eye movements as participants read sentences and note the fixation time for an ambiguous word and also for a control word with just one meaning that replaces the ambiguous word in the sentence. Consider the following sentence, in which the ambiguous word cast has balanced dominance. The cast worked into the night. (control word: cook) As a person reads the word cast, both meanings of cast are activated, because cast (member of a play) and cast (plaster cast) are equally likely. Because the two meanings compete for activation, the person looks longer at cast than at the control word cook, which has only one meaning as a noun. Eventually, when the reader reaches the end of the sentence, the meaning becomes clear (Duffy et al., 1988; Rayner & Frazier, 1989; Traxler, 2012) (Figure 11.3a). But consider the following, with the ambiguous word tin: The tin was bright and shiny. (control word: gold) In this case, people read the biased ambiguous word tin just as quickly as the control word, because only the dominant meaning of tin is activated, and the meaning of tin as a metal is accessed quickly (Figure 11.3b). Copyright 2019 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part. WCN 02-200-202 08271_ch11_ptg01.indd 329 4/18/18 5:13 PM 330 CHAPTER 11 Language But meaning frequency isn’t the only factor that determines the accessibility of the meaning of a word. Context can play a role as well. Consider, for example, the following sentence, in which the context added before the ambiguous word tin indicates the less-frequent meaning of tin: The miners went to the store and saw that they had beans in a tin. (control word: cup) In this case, when the person reaches the word tin, the less frequent meaning is activated at increased strength because of the prior context, and the more frequent meaning of tin is activated as well. Thus, in this example, as with the first sentence that contained the word cast, two meanings are activated, so the person looks longer at tin (Figure 11.3c). Finally, consider the sentence below, in which the context indicates the more frequent meaning of tin: The miners went under the mountain to look for tin. (control word: gold) In this example, only the dominant meaning of tin is activated, so tin is read rapidly (Figure 11.3d). We’ve seen in this chapter that the process of accessing the meaning of a word is complicated and is influenced by multiple factors. First, the frequency of a word determines how long it takes to process its meaning. Second, if a word has more than one meaning, the context of the sentence influences which meaning we access. Finally, our ability to access the correct meaning of a word depends on both the word’s frequency and, for words with more than one meaning, a combination of meaning dominance and context. So simply identifying, recognizing, and knowing the meaning of individual words is a complex and impressive feat. However, except in rare situations in which words operate alone—as in exclamations such as Stop! or Wait!—words are used with other words to form sentences, and, as we will see next, sentences add another level of complexity to understanding language. T E ST YOUR SELF 11.1 1. What is the hierarchical nature of language? The rule-based nature of language? 2. Why has the need to communicate been called universal? 3. What events are associated with the beginning of the modern study of language in the 1950s? 4. What is psycholinguistics? What are its concerns, and what part of psycholinguistics does this chapter focus on? 5. What is semantics? The lexicon? 6. How does word frequency affect our processing of words? Describe the eye movement experiment that illustrates an effect of word frequency. 7. What is the evidence that context helps people deal with the variability of word pronunciation? 8. What is speech segmentation and why is it a problem? What are some of the factors that help us achieve speech segmentation? 9. What is lexical ambiguity? Describe the experiment that used lexical priming to show that (a) all of the multiple meanings of a word are accessed immediately after the word is heard; and (b) context determines the appropriate meaning of an ambiguous word within about 200 msec. 10. What is meaning dominance? Biased dominance? Balanced dominance? 11. How do frequency and context combine to determine the correct meaning of ambiguous words? Copyright 2019 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part. WCN 02-200-202 08271_ch11_ptg01.indd 330 4/18/18 5:13 PM Understanding Sentences 331 Understanding Sentences When we considered words, we saw how sentences create context, which makes it possible to (1) deal with the variability of word pronunciations, (2) perceive individual words in a continuous stream of speech, and (3) determine the meanings of ambiguous words.But now we are going to go beyond just considering how sentences help us understand words, by asking how combining words into sentences creates meaning. To understand how we determine the meaning of a sentence, we need to consider syntax—the structure of a sentence—and the study of syntax involves discovering cues that languages provide that show how words in a sentence relate to one another (Traxler, 2012). To start, let’s think about what happens as we hear a sentence. Speech unfolds over time, with one word following another. This sequential process is central to understanding sentences, because one way to think about sentences is meaning unfolding over time. What mental processes are occurring as a person hears a sentence? A simple way to answer this question would be to picture the meaning as being created by adding up the meanings of each word as they occur. But this idea runs into trouble right away when we consider that some words have more than one meaning and also that a sequence of words can have more than one meaning. The key to determining how strings of words create meaning is to consider how meaning is created by the grouping of words into phrases—a process called parsing. Parsing: Making Sense of Sentences Understanding the meaning of a sentence is a feat of mental pyrotechnics that involves understanding each word as it occurs (some of which may be ambiguous) and parsing words into phrases (Figure 11.4). To introduce parsing, let’s look at some sentences. Consider, for example, a sentence that begins: After the musician played the piano … [After the musician played the piano] [she left the stage] After the musician played the piano she left the stage Parsed sentence in mind Words in What do you think comes next? Some possibilities are: a. . . . she left the stage. b. . . . she bowed to the audience. c. . . . the crowd cheered wildly. All of these possibilities, which create sentences that are easy to understand and that make sense, involving grouping the words as follows: [After the musician played the piano] [the crowd cheered wildly]. But what if the sentence continued by stating ➤ Figure 11.4 Parsing is the process that occurs when a person hears or reads a string of words (Words in) and groups these words into phrases in their mind (Parsed sentence in mind). The way the words are grouped in this example indicates that the person has interpreted the sentence to mean that the musician played the piano and then left the stage. d. . . . was wheeled off of the stage. Reading the sentence ending in (d) as a whole, After the musician played the piano was wheeled off of the stage, might take you by surprise because the grouping of [After the musician played the piano] isn’t correct. The correct grouping is [After the musician played] [the piano was wheeled off of the stage.]. When written, adding a comma makes the correct parsing of this sentence clear: After the musician played, the piano was wheeled off the stage. Sentences like this one, which begin appearing to mean one thing but then end up meaning something else, are called garden path sentences (from the phrase “leading a person down the garden path,” which means misleading the person). Garden path sentences illustrate temporary ambiguity, because first one organization is adopted and then—when the error is realized—the person shifts to the correct organization. Copyright 2019 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part. WCN 02-200-202 08271_ch11_ptg01.indd 331 4/18/18 5:13 PM 332 CHAPTER 11 Language The Garden Path Model of Parsing Language researchers have used sentences with temporary ambiguity to help understand the mechanisms that operate during parsing. One of the early proposals to explain parsing, and garden path sentences in particular, is called the garden path model of parsing. This approach, proposed by Lynn Frazier (1979, 1987), states that as people read a sentence, their grouping of words into phrases is governed by a number of processing mechanisms called heuristics. As we will see when we discuss reasoning and decision making, a heuristic is a rule that can be applied rapidly to make a decision. The decisions involved in parsing are decisions about the structure of a sentence as it unfolds in time. Heuristics have two properties: On the positive side, they are fast, which is important for language, which occurs at about 200 words per minute (Traxler, 2012). On the negative side, they sometimes result in the wrong decision. These properties become apparent in a sentence like After the musician played the piano was wheeled off the stage, in which the initial parse of the sentence turns out to be incorrect. The garden path model proposes that when this happens, we reconsider the initial parse and make appropriate corrections. The garden path model specifies not only that rules are involved in parsing, but that these rules are based on syntax—the structural characteristic of language. We will focus on one of these syntax-based principles, which is called late closure. The principle of late closure states that when a person encounters a new word, the person’s parsing mechanism assumes that this word is part of the current phrase, so each new word is added to the current phrase for as long as possible (Frazier, 1987). Let’s return to sentence about the musician to see how this works. The person begins reading the sentence: After the musician played … So far, all the words are in the same phrase. But what happens when we reach the words the piano? According to late closure, the parsing mechanism assumes that the piano is part of the current phrase, so the phrase now becomes After the musician played the piano … So far, so good. But when we reach was, late closure adds this to the phrase to create After the musician played the piano was … And then, when wheeled is added to create an even longer phrase, it becomes obvious that something is wrong. Late closure has led us astray (down the garden path!) by adding too many words to the first phrase. We need to reconsider, taking the meaning of the sentence into account, and reparse the sentence so “the piano” is not added to the first phrase. Instead, it becomes part of the second phrase to create the grouping [After the musician played] [the piano was wheeled off the stage]. The garden path model generated a great deal of research, which resulted in support for the model (Frazier, 1987). However, some researchers questioned the proposal that syntactic rules like late closure operate alone to determine parsing until it becomes obvious that a correction is needed (Altmann et al., 1992; Tanenhaus & Trueswell, 1995). These researchers have provided evidence to show that factors in addition to syntax can influence parsing right from the beginning. The Constraint-Based Approach to Parsing The idea that information in addition to syntax participates in processing as a person reads or hears a sentence is called the constraint-based approach to parsing. As we consider some examples that show how parsing can be influenced by factors in addition to syntax, Copyright 2019 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part. WCN 02-200-202 08271_ch11_ptg01.indd 332 4/18/18 5:13 PM Understanding Sentences we will encounter a theme we introduced at the beginning of the chapter: Information contained in the words of a sentence, and in the context within which a sentence occurs, is used to make predictions about how the sentence should be parsed (Kuperberg & Jaeger, 2015). Defendent examining something Influence of Word Meaning Here are two sentences that illustrate how the meaning of words in a sentence can influence parsing right from the beginning. They differ in how hard they are to figure out because of the meanings of the second words in each sentence. 1. The defendant examined by the lawyer was unclear. 2. The evidence examined by the lawyer was unclear. Which one was easier to figure out as you were reading along? The process that occurs as sentence (1) is unfolding is illustrated in Figure 11.5a. After reading The defendant examined, two possibilities present themselves: (1) the defendant could be examining something or (2) the defendant could be being examined by someone else. It’s only after reading the rest of the sentence by the attorney that it is possible to definitely determine that the defendant is being examined. In contrast, only one possibility presents itself after reading The evidence examined in sentence (2) because it is unlikely that the evidence will be doing any examining. (Figure 11.5b). Here are two more examples: 1. The dog buried in the sand was hidden. 2. The treasure buried in the sand was hidden. Which one of these is more likely to initially lead to the wrong conclusion, and why? Influence of Story Context Consider the following sentence proposed by Thomas Bever (1970), which has been called the most famous garden path sentence because of the confusion it causes: 333 Defendent being examined Read: “The defendent examined” (a) Evidence being examined Read: “The evidence examined” (b) ➤ Figure 11.5 (a) Two possible predictions that could be made after reading or hearing The defendant examined in sentence (1) The defendant is going to either examine something (top) or be examined by someone else (bottom). (b) The only possible reading of The evidence examined in sentence (2) is that the evidence is examined by someone. The possibility that the evidence was going to examine something is highly unlikely. The horse raced past the barn fell Whoa! What’s going on here? For many people, everything is fine until they hit fell. Readers are often confused, and may even accuse the sentence of being ungrammatical. But let’s look at the sentence in the context of the following story: There were two jockeys who decided to race their horses. One raced his horse along the path that went past the garden. The other raced his horse along the path that went past the barn. The horse raced past the barn fell. Of course, the confusion could have been avoided by simply stating that the horse that was raced past the barn fell, but even without these helpful words, context wins the day and we parse the sentence correctly! Influence of Scene Context Parsing of a sentence is influenced not only by the context provided by stories but also by context provided by scenes. To investigate how observing objects in a scene can influence how we interpret a sentence, Michael Tanenhaus and coworkers (1995) developed a technique called the visual world paradigm, which involves determining how information in a scene can influence how a sentence is processed. Participants’ eye movements were measured as they saw objects on a table, as in Copyright 2019 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part. WCN 02-200-202 08271_ch11_ptg01.indd 333 4/18/18 5:14 PM 334 CHAPTER 11 Language Ambiguous 3 4 1 Put the apple (1) in the box (3, 4) Unambiguous 0.8 0.6 0.4 0.2 0 (a) One-apple condition (b) Eye movements (c) Bruce Goldstein 2 1.0 Proportion of trials with eye movements to incorrect destination on the towel (2) ➤ Figure 11.6 (a) One-apple scene similar to the one viewed by Tanenhaus et al.’s (1995) participants. (b) Eye movements made while comprehending the task. (c) Proportion of trials in which eye movements were made to the towel on the right for the ambiguous sentence. (Place the apple on the towel in the box) and the unambiguous sentence (Place the apple that’s on the towel in the box). Figure 11.6a. As participants looked at this display, they were told to carry out the following instructions: Place the apple on the towel in the box. When participants heard the phrase Place the apple, they moved their eyes to the apple, then hearing on the towel, they looked at the other towel (Figure 11.6b). They did this because at this point in the sentence they were assuming that they were being told to put the apple on the other towel. Then, when they heard in the box, they realized that they were looking at the wrong place and quickly shifted their eyes to the box. The reason participants looked first at the wrong place was that the sentence is ambiguous. First it seems like on the towel means where the apple should be placed, but then it becomes clear that on the towel is referring to where the apple is located. When the ambiguity was removed by changing the sentence to Move the apple that’s on the towel to the box, participants immediately focused their attention on the box. Figure 11.6c shows this result. When the sentence was ambiguous, participants looked at the other towel on 55 percent of the trials; when it wasn’t ambiguous, participants didn’t look at the other towel. Tanenhaus also ran another condition in which he presented the two-apple display like the one in Figure 11.7a. Because there are two apples, participants interpreted on the towel to be indicating which apple they should move, and so looked at the apple and then at the box (Figure 11.7b). Figure 11.7c shows that participants looked at the other towel on only about 10 percent of the trials for both place the apple on the towel (the ambiguous sentence) and place the apple that’s on the towel (the non-ambiguous sentence) when looking at this display. The fact that the eye movement patterns were the same for the ambiguous and non-ambiguous sentences means that in this context the participants were not led down the garden path. The important result of this study is that the participants’ eye movements occur as they are reading the sentence and are influenced by the contents of the scene. Tanenhaus therefore showed that participants take into account not only information provided by the syntactic structure of the sentence, but also by what Tanenhaus calls non-linguistic information— in this case, information provided by the scene. This result argues against the idea proposed Copyright 2019 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part. WCN 02-200-202 08271_ch11_ptg01.indd 334 4/18/18 5:14 PM Understanding Sentences 335 Ambiguous on the towel (2) 1 in the box (3) 2 Put the apple (1) 0.8 0.6 0.4 0.2 0 (a) Two-apple condition (b) Eye movements Unambiguous (c) Bruce Goldstein 3 Proportion of trials with eye movements to incorrect destination 1.0 ➤ Figure 11.7 (a) Two-apple scene similar to the one viewed by Tanenhaus et al.’s (1995) subjects. (b) Eye movements while comprehending the task. (c) Proportion of trials in which eye movements were made to the towel on the right for the ambiguous sentence (Place the apple on the towel in the box) and the unambiguous sentence (Place the apple that’s on the towel in the box). by the garden path model that syntactic rules are the only thing taken into account as a sentence is initially unfolding. Influence of Memory Load and Prior Experience with Language two sentences: Consider these 1. The senator who spotted the reporter shouted 2. The senator who the reporter spotted shouted These sentences have the same words, but they are arranged differently to create different constructions. Sentence (2) is more difficult to understand, as indicated by research that shows that readers spend longer looking at the part of the sentence following who in sentences with structures like sentence (2) (Traxler et al., 2002). To understand why sentence (2) is more difficult to understand, we need to break these sentences down into clauses. Sentence (1) has two clauses: Main clause: The senator shouted. Embedded clause: The senator spotted the reporter. The embedded clause is called embedded, because who spotted the reporter is inside the main clause. The senator is the subject of both the main clause and the embedded clause. This construction is called a subject-relative construction. Sentence [2] also contains two clauses: Main clause: The senator shouted. Embedded clause: The reporter spotted the senator. In this case, the senator is the subject of the main clause, as before, and is also replaced by who in the embedded clause, but is the object in this clause. The senator is the object because he is the target who was spotted. (The reporter is the subject of this clause, because he did the spotting.) This construction is called an object-relative construction. Copyright 2019 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part. WCN 02-200-202 08271_ch11_ptg01.indd 335 4/18/18 5:14 PM 336 CHAPTER 11 Language One reason the object-relative construction is more difficult to understand is because it demands more of the reader’s memory. In sentence [1] we find out who did the “spotting” right away. It was the senator. But in sentence [2] “spotted” is near the end of the sentence so we need to hold the early part of the sentence in memory until we find out that the reporter did the “spotting.” This higher memory load slows down processing. The second reason object-relative construction is more difficult to understand is that it is more complicated, because while the senator is the subject in both the main and embedded clauses in sentence [1], it is the subject of the main clause and the object of the embedded clause in sentence [2]. This more complex construction not only makes the object-relative construction more difficult to process, it may be a reason that it is less prevalent in English. Subjectrelative constructions account for 65 percent of relative clause constructions (Reali & Christiansen, 2007), and being more prevalent has an important effect—we have more exposure to subject-relative constructions, so we have more practice understanding these constructions. In fact, we have learned to expect that in sentences of this type, pronouns like who, which, or that are usually followed by a verb (spotted in sentence 1). So when the pronoun isn’t followed by a verb, as in sentence 2, we have to reconsider and adapt to the different construction. Does this sound familiar? As we have seen from examples like the defendant examined and the horse raced, making predictions during a sentence that turn out to be wrong slows down sentence processing. Prediction, Prediction, Prediction… The examples we’ve considered so far—the defendant being examined, the falling horse, the apple on the towel, and the shouting senator—all have something in common. They illustrate how people make predictions about what is likely to happen next in a sentence. We predict that The defendant examined means that the defendant is going to examine something but instead it turns out that the defendant is being examined! Oops! Our incorrect prediction has led us down the garden path. Similarly, we predict that The horse raced is going to say something about how the horse raced (The horse raced faster than it ever had before), but instead we find that raced refers to which horse was racing. We predict that we are being asked to place the apple on the other towel, but it turns out to be otherwise. But even though incorrect predications can temporarily throw us off track, most of the time prediction is our friend. We are constantly making predictions about what is likely to happen next in a sentence, and most of the time these predictions are correct. These correct predictions help us deal with the rapid pace of language. And prediction becomes even more important when language is degraded, as in a poor phone connection, or is heard in a noisy environment, or when you are trying to understand someone with a foreign accent. Gerry Altmann and Yuki Kamide (1999) did an experiment that showed that the participants were making predictions as they were reading a sentence by measuring their eye movements. Figure 11.8 shows a picture similar to one from the experiment. Participants heard either The boy will move the cake or The boy will eat the cake while viewing this scene. For both of these sentences, cake is the target object. Participants were told to indicate whether the sentence they read could be applied to the pictures. Altmann and Kamide didn’t care how they responded in this task. What they did care about was how they were processing the information as they were ➤ Figure 11.8 A picture similar to one used in Altman and hearing the sentences. Kamide’s (1999) experiment in which they measured eye Let’s consider what might be happening as the sentences unmovements that occurred as participants heard a sentence fold: First, The boy will move . . . What do you think the boy is while looking at the picture. Copyright 2019 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part. WCN 02-200-202 08271_ch11_ptg01.indd 336 4/18/18 5:14 PM Understanding Text and Stories 337 going to move? The answer isn’t really clear, because the boy could move the car, the train, the ball, or even the cake. Now consider The boy will eat . . . This one is easy. The boy will eat the cake. Measurement of participants’ eye movements as they were hearing these sentences indicated that eye movements toward the target object (cake in this example) occurred 127 msec after hearing the word cake for the move sentences and 87 msec before hearing the word cake for the eat sentences. Thus, hearing the word eat causes the participant to begin looking toward the cake before he or she even hears the word. Eat leads to the prediction that cake will be the next word. This kind of prediction is likely occurring constantly as we hear or read sentences. As we will see in the next sections, predictions also play an important role in understanding stories and having conversations. T E ST YOUR SELF 11.2 1. What is syntax? 2. What is parsing? What are garden path sentences? 3. Describe the garden path model of parsing. Be sure you understand what a heuristic is and the principle of late closure. 4. Describe the constraint-based approach to parsing. How does it differ from the garden path approach? 5. Describe the following lines of evidence that support the constraint-based approach to parsing: ➤ How meanings of words in a sentence affect parsing. ➤ How story context affects parsing. ➤ How scene context affects processing. Be sure you understand the visual world paradigm. ➤ How memory load and predictions based on knowledge of language structure affect parsing. Be sure you understand the difference between subject-relative and object-relative constructions and why object-relative constructions are harder to understand. 6. How can garden path sentences be related to prediction? 7. How is prediction important for understanding sentences? Understanding Text and Stories Just as sentences are more than the sum of the meanings of individual words, stories are more than the sum of the meanings of individual sentences. In a well-written story, sentences in one part of the story are related to sentences in other parts of the story. The reader’s task is to use these relationships between sentences to create a coherent, understandable story. An important part of the process of creating a coherent story is making inferences— determining what the text means by using our knowledge to go beyond the information provided by the text. We have seen how unconscious inference is involved in perception (Chapter 3, page 70), and when we described the constructive nature of memory in Chapter 8, we saw that we often make inferences, often without realizing it, as we retrieve memories of what has happened in the past (page 240). Making Inferences An early demonstration of inference in language was an experiment by John Bransford and Marcia Johnson (1973), in which they had participants read passages and then tested them Copyright 2019 Cengage Le