Rolf A Zwaan and Carol J Madden

There are two views of cognition in general and of language comprehension in particular. According to the traditional view (Chomsky, 1957; Fodor, 1983; Pylyshyn, 1986), the human mind is like a bricklayer, or maybe a contractor, who puts together bricks to build structures. The malleable clay of perception is converted to the neat mental bricks we call words and propositions, units of meaning, which can be used in a variety of structures. But whereas bricklayers and contractors presumably know how bricks are made, cognitive scientists and neuroscientists have no idea how the brain converts perceptual input to abstract lexical and propositional representations - it is simply taken as a given that this occurs (Barsalou, 1999).

According to an alternative and emerging view, there are no clear demarcations between perception, action, and cognition. Interactions with the world leave traces of experience in the brain. These traces are (partially) retrieved and used in the mental simulations that make up cognition. Crucially, these traces bear a resemblance to the perceptual/action processes that generated them (Barsalou, 1999) and are highly malleable. Words and grammar are viewed as a set of cues that activate and combine experiential traces in the mental simulation of the described events (Zwaan, 2004). The main purpose of this chapter is to provide a discussion of this view of language comprehension. To set the stage for this discussion we first analyze a series of linguistic examples that present increasingly larger problems for the traditional view. Consider the following sentences.

(1) The exterminator checked the room for bugs.

(2) The CIA agents checked the room for bugs.

It is clear that the bug in (1) is not the same as the bug in (2). In other words, "bug" is a homonym. The traditional view has no problem accounting for these two different interpretations of "bug" because it simply assumes that the sentential context disambiguates the homonym such that the correct meaning is selected (although the incorrect meaning may be briefly activated, e.g., Swinney, 1979). Thus, in these cases, one might assume a stronger lexical association between "exterminator" and the "insect" meaning of "bug" than between "CIA agent" and that meaning and vice versa for the "microphone" meaning of bug. The following two sentences already present more of a challenge to the traditional view.

(3) Fred stole all the books in the library.

(4) Fred read all the books in the library.

It is clear that "all the books" means two slightly different things in these two sentences. For example, (3) implies that Fred stole all 12 copies of War and Peace, whereas (4) in the most likely interpretation means that Fred read only one copy of War and Peace. But both words refer to the same thing: a physical object consisting of written pages, bound together and in a cover. This presents a problem for the traditional view of composi-tionality according to which concepts have atomic meanings that should remain unchangeable across contexts. However, the traditional view can be amended to account for interpretations such as these. Pustejovsky (1995) has proposed that words have different qualia, that is different interpretations, and that these interpretations are selected by other words in the sentence. For example, a book can be both a physical object and a source of information. Stealing typically involves physical objects (although one can steal glances, kisses, or ideas) and thus "steal" selects the physical-object quale of "book." Reading, on the other hand, involves information (even when reading facial expressions, tracks in the snow, or passes in a soccer game), and therefore "read" selects the information-source meaning of "book." In this sense, the bricklayer metaphor can be extended to that of a child playing with legos (a metaphor used in many linguistics courses). Some pieces, like wheels, have different shapes and different sites for attachment. For example, an axle fits into a flange on the inside of the wheel and a little square block fits on the hub. Similarly, some verbs select one quale of a noun, whereas other verbs will select another.

However, it is not clear whether this lego-extended view of comprehension can account for the following sentences.

(5) John pounded the nail into the wall.

(6) John pounded the nail into the floor.

Here, both sentences use the same verb, and in both sentences, this verb selects the same quale of the noun, "nail"; it is a slender usually pointed and headed fastener designed to be pounded in. What is different in the two sentences is the nail's orientation. There is nothing in the traditional view to suggest that the nail's orientation should be part of the comprehender's mental representation. For example, a common way to represent sentences (5) and (6) is (e.g., Kintsch & van Dijk, 1978):

Nothing in these propositional representations says anything about the nail's orientation, yet empirical evidence shows that comprehenders routinely represent this orientation (Stanfield & Zwaan, 2001). A similar point can be made about the shape of objects.

(10) He saw the eagle in the nest.

According to the traditional view the words in these sentences form the building blocks of meaning out of which the comprehender builds a structure that reflects his or her interpretation of the sentence. But how would this work for sentences (9) and (10)? Surely the eagle in (9) cannot be the same lego brick as the eagle in (10). In (9) the eagle has its wings stretched out, whereas the eagle in (10) most likely has its wings drawn in. Again, there is empirical evidence that comprehenders are sensitive to these differences (Zwaan, Stanfield, & Yaxley, 2002) and again, this is not predicted by the traditional view. The traditional view also does not seem to have a straightforward account for how the following sentences are interpreted.

(11) Jack looked across the room to see where the whisper/explosion came from.

(12) Jack looked across the valley to see where the whisper/explosion came from.

Clearly, whisper is a better fit for (11) and explosion for (12). But how does the traditional theory account for this? The Merriam-Webster dictionary defines "whisper" as follows: "to speak softly with little or no vibration of the vocal cords especially to avoid being overheard." Nothing in this definition points directly to the distance over which a whisper can be heard by the human ear. Similarly, nothing in the definition of valley - "an elongate depression of the earth's surface usually between ranges of hills or mountains" according to Merriam-Webster - provides explicit information about the typical width of a valley. It might be argued that both terms contain information that can be used to infer the respective distances (e.g., "softly," "overhear," and "mountain"). But if one looks up "mountain," for example, the most relevant meaning to be found is "a landmass that projects conspicuously above its surroundings and is higher than a hill." Of course, this does little to alleviate the interpretation problem. This is an example of what has become known as the Chinese Room Problem (Searle, 1980). Given that dictionary meanings are not grounded in perception and action, the comprehender who has to rely on a dictionary is given a perpetual runaround (see also Glenberg, 1997).

Finally, consider the following pair of sentences.

(13) The pitcher hurled the baseball to you.

(14) You hurled the baseball at the batter.

There is nothing different about the intrinsic properties of the baseball in (13) and (14), such as shape or orientation. The only difference between the baseball in (13) and that in (14) is the direction of motion. Although the traditional view would not predict any differences in the mental representations of baseballs formed during comprehension of these two sentences (and may even have trouble explaining it post-hoc in an elegant manner), there is evidence that the direction of motion is incorporated into the representations of the two baseballs, yielding distinct simulations (Zwaan, Madden, Yaxley, & Aveyard, 2004). We will discuss this evidence and other relevant evidence in more detail later.

In the rest of this chapter, we propose the beginnings of a theory of sentence comprehension that accounts for these findings in a straightforward way.

interconnected experiential traces

In our proposed theory, we assume that all mental representations are experiential. Within the category of experiential representations, we distinguish between referent representations and linguistic representations (see also Sadoski & Paivio, 2001). Referent representations are traces laid down in memory during perceptions of and interactions with the environment. These traces are multimodal (i.e., combining multiple senses).1 Because of attentional limitations, these traces are schematic (Barsalou, 1999). The second subcategory of experiential traces consists of linguistic traces. These traces are laid down as linguistic information is being received or produced. For example, there are perceptual traces of hearing, reading, seeing, and feeling (as in Braille) linguistic constructions. As well, there are motor representations of saying, signing, typing, and handwriting linguistic constructions. Not only are these constructions interconnected, they are also connected to referent representations, which are also interconnected (see Figure 10.1).

How are these interconnections established? The main mechanism is co-occurrence (e.g., Hebb, 1949). Certain entities in the environment tend

1 Damasio (1999, p. 160) describes object representations as stored in "dispositional form." Dispositions are records, which are dormant, rather than active and explicit, as images are. Dispositions include: records of sensory aspects, records of the motor adjustments necessary to gather sensory signals, obligate emotional reaction.

figure 10.1. Schematic depiction of experiential (visual) traces of leashes, dogs, and bones, and of the visual and sound patterns associated with them, as well as the links between these traces.

to co-occur. Ducks are generally found in or near ponds or lakes, monitors on desks, pacifiers in babies' mouths, clouds in the sky, and branches above roots. Events may co-occur or follow in close sequence. For example, a scratchy sound accompanies the action of striking of a match, and a flame typically follows it, along with a sulfuric smell, which we may find pleasurable. Because of these spatio-temporal co-occurrences, combinations of entities, actions, events, and bodily states become part of the same experiential trace.

Similarly, linguistic constructs co-occur and may therefore develop associations between themselves. First proposed by associationists such as Hume and Locke, but eschewed by transformational grammarians, the analysis of linguistic co-occurrences has made a recent comeback in the form of sophisticated computational linguistic analyses involving large corpora, such as latent semantic analysis (LSA, Landauer & Dumais, 1997).2 This idea that associations between representations are formed through co-occurrence of linguistic constructions is central to the current theory. For example, the words "nurse" and "doctor" often co-occur in language. As a result, the experiential traces for these linguistic units are associated, just as the words "peace" and "treaty" and the names Lennon and McCartney. Often sets of more than two words co-occur, as in "hail Mary pass," "internal revenue service," and "everything but the kitchen sink." As a

2 These analyses actually go significantly beyond co-occurrences by assessing the degree to which words occur in similar contexts, but this is not relevant to the current argument.

result, entire sequences of words can be treated as constructions (Goldberg, 1995, 2003).

The connections that are established between experiential traces for referents and experiential traces for linguistic constructions are of critical importance to the grounding of language in perception and action (see also Goldstone, Yeng, & Rogosky, Chapter 12, this volume). The initial mechanism by which these connections are forged is co-occurrence. When children learn to speak, parents and others point out objects to them in the environment. Moreover, even when children are not attending to the entity in question, they can use mentalistic cues such as the speaker's eye gazes and facial expressions to form associations between constructs and referents (Bloom, 2000). As a result, children learn to associate an experience of a referent with a particular sound pattern. In fact, children are surprisingly adept at learning word meaning this way, often needing only a few exposures (Carey & Bartlett, 1978).

Children do not only learn to associate constructs with objects, but also with actions and properties. For example when parents say, "give me the ball," the child will associate an action - grasping a ball, extending the arm, and then releasing the ball into the grasp of a parent - with a linguistic construction (and with encouraging sounds and facial expressions on the part of the parent). In fact, the child learns something more fundamental, namely that this syntactic construction can be applied in many other contexts - for instance, "throw me the ball," and even "tell me a story." As such, the syntactic structure can be thought of as a linguistic construction that conveys meaning (Goldberg, 1995,2003). The meaning of this construction, the double-object construction, is that an object or something more abstract moves from the agent to a recipient. This is what the different contexts in which the expression is used have in common. Importantly, however, this is only part of the meaning of an expression. For example, "throw me the ball" is associated with a different motor program than "give me the ball" and it is also associated with the salient pattern of an object getting smaller in one's visual field as it moves away from the thrower. On the other hand "tell me a story" is associated with cognitive effort and speech motor programs, as well as with certain encouraging or puzzled facial expressions on the part of the listener. As Hockett (1959) noted, one important feature of language - called displacement - is that it allows us to convey situations that are not part of our immediate environment. The connections between linguistic and referent traces enable this feature. For example, if we have never seen a zebra before and it is described to us as a "horse with black-and-white stripes," then we can form a new referent representation by combining the perceptual traces for horses, for stripes, and for black-and-white, based on their associations with the corresponding words (Harnad, 1990). This virtual experiential trace, constructed from a combination of other visual traces can now be stored in long-term memory. Along with it, an association is formed between the sound pattern of "zebra" and the new visual trace. This uniquely human way of learning about the environment through linguistic scaffolding significantly augments what we can learn by interacting directly with the environment.

One consequence of the development of connections between the two classes of experiential symbols is that co-occurrences in one domain will produce co-occurrences in the other. These effects are bidirectional. Just as the spatio-temporal proximity of certain objects or events enhances the likelihood that the linguistic constructions denoting them will co-occur, the co-occurrence of linguistic constructions will strengthen the connections between their referents. As a result, although there generally is not an analog first-order mapping between linguistic constructions and their referents, there is a strong second-order mapping. If the link between two constructs is strong, then it is very likely that the link between the corresponding referents is also strong (see Figure 10.1). This is one reason why techniques such as LSA are often surprisingly successful in capturing meaning. However, as we will show later, many experiential factors are not captured by linguistic co-occurrences.3

construal

Along with several other researchers, we conceive of language as a set of cues by which the speaker or writer manipulates the listener's or reader's attention on an actual or fictional situation (e.g., Langacker, 1987; Tomasello, 2003). The units in which this process takes place are atten-tional frames (Langacker, 2001). Attentional frames map onto intonation units, which are speech segments bounded by pauses or intonation shifts (Chafe, 1994). Because written language follows spoken language phyloge-netically as well as ontogenetically, the segmentation of spoken language provides the grounding for the segmentation of written language. We define construal as the mental simulation of an experience conveyed by an attentional frame. This mental simulation uses the experiential traces that are activated by the linguistic constructions in the intonation unit.

Figure 10.2 (from Zwaan, 2004) shows the components of construal. Each construal necessarily includes:

* a time at which the simulated situation occurs (as related to the moment of utterance, to the previously simulated event, and sometimes to some reference point);

* a spatial region in which the described event takes place;

* a perspective (spatial and psychological) from which the situation is experienced;

* a focal entity;

* a background entity.

3 There is much more to be said about this issue, but this is beyond the scope of this chapter.

C = construal T = time

S = spatial region (personal, action, vista) P = perspective F = focal entity R = relation B = background entity f = feature figure 10.2. The components of construal (from Zwaan, 2004).

In addition, the focal entity, relation, and background entity can have features (e.g., size, color, intensity, speed).

Here, our focus is on the process of construal. An important component of construal is the establishment of a focal entity and background entity. Language provides many cues for this, including syntactic, morphemic, and paralinguistic information. In English, for example, the focal entity is usually the first entity described (Langacker, 1987). This means that in active sentences the entity performing an action is the focal entity, whereas in passive constructions the entity undergoing the action is the focal entity. This means that passivization is not without semantic consequences, so that a passive sentence is not really a paraphrase of its active counterpart.

Intonation units typically describe what linguists have traditionally analyzed as events, processes, and states (e.g., Vendler, 1967; Ter Meulen, 1995). Viewing language comprehension as the modulation of attention compels us to treat each as a type of event. Consider the following sentences.

(15) The car pulled out of the driveway.

(16) The car was zooming along on the interstate.

Punctate events such as (15) are perhaps the most easily conceptualized as mental simulations. The focal entity (the car) initiates an event (pulling out of) that changes its location relative to the background entity (the driveway). But how about (16)? For all we know, the car may keep zooming along for many hours to come. It would be preposterous to claim that we keep our mental simulations zooming along with it for a similar duration. The notion of attentional frame provides an elegant solution to this problem. The situation conveyed by (16) can be conceptualized as the event of perceiving a car zooming along (either from inside or outside of the car). A perceptual event such as this would only take a short amount of time. Along similar lines, a static description like (17) can simply be understood as the event of seeing a blue car rather than as a mental tour de force in which the color blue is continuously being simulated.4

In an experiential representation, the perspective of the observer or agent vis a vis the described situation needs to be represented. This is illustrated in (18).

(18) The red squirrel jumped into the tree.

It is easy to see that the linguistic constructs in (18) by themselves do not provide sufficient perspectival constraints. For instance, is the squirrel jumping from left to right or from right to left? Our working assumption is that in case of underspecification, people will use default expectations, which may depend on environmental constraints, handedness, the direction of reading, or other cultural conventions.5

4 Often, a speaker or writer will insert a sentence later in the narrative to remind the listener/reader that the car is still in motion. When the reader's attention is directed back to the car, it should still be represented as zooming along.

5 For example, right-handers recognize objects more quickly when lit from top-left than when lit from other angles. Presumably this has to do with the fact that under this angle, their right

In many cases, however, perspective is automatically produced as part of the mental simulation. Activating relevant perceptual memory traces of experience with words and referents creates a simulation. As a result, the simulation will automatically adopt the perspective of the most frequent (or perhaps most recent) relevant memory traces. For example, clouds are most often seen from below. Therefore, most traces of seeing clouds will be from below. As a result, comprehenders of the sentence: "The farmer looked at the clouds" are likely to simulate clouds from below. In many cases, entity features and the range of the human sensory apparatus jointly place constraints upon the observer's distance from it. For example, human hearing is such that a whisper can only be heard from a relatively short distance. That is why a whisper in (10) sounds odd. Another illustration is (19), a sentence used by Morrow and Clark (1988).

(19) A mouse/tractor approached the fence.

This example not only shows that the interpretation of "approach" depends on the size of the focal entity - i.e., people place the mouse closer to the fence than they do the tractor (Morrow & Clark, 1988) - it also suggests that the observer is closer to the scene in the case of the mouse than in the case of the tractor. The main constraint here is the mouse's size, which makes that we can see mice only from relatively short distances. Importantly, this constraint is not just imposed by the mouse's size, but also by the limits of human vision. Hawks, for instance, are able to see mice from much longer distances than humans. In other words, our auditory traces of whispers and our visual traces of mice and tractors already contain relevant perspectival information grounded in human sensation and perception. It may therefore not be farfetched to say that perspective is part of the meaning of some words (Talmy, 2000a, 2000b; Zwaan, 2004). There already is some evidence that comprehenders interpret perspective in verbs such as "come" and "go" and "bring" and "take" (Black, Turner, & Bower, 1979). For example, "come" implies movement toward the observer while "go" implies movement away from the observer. In some cases, perspective is explicitly stated. This is the case in a sentence like (20), where the putative default perspective of "cloud" must be overridden.

(20) From the mountaintop, the clouds below looked like big balls of cotton.

hand does not cast a shadow over manipulated objects. Left-handers also have a left bias, though less strong than right-handers, presumably because they live in a world dominated by right-handers (Sun & Perona, 1998, see also Mamassian & Goutcher, 2001). Consistent with this left bias, pictures with a left-to-right directionality are judged as more aesthetically pleasing than pictures with a right-to-left directionality (Christman & Pinger, 1997). Recent evidence suggests that bias may be culturally determined (Maass & Russo, 2003).

To summarize, rather than simply constructing mental representations of who-did-what-to-whom out of mental Lego blocks, we perform experiential simulations that necessarily imply a spatio-temporal perspective on the described situation. The typical experiential perspective on the entity denoted by a linguistic construction plays a key role in establishing a perspective.

As noted earlier, experiential representations consist of multitudes of traces and are stored by the brain in what Damasio (1999) calls "dispositional form." Usually, they will only be partly relevant to the current context. Of particular importance therefore is the idea that traces activated by different words constrain each other during construal. For example, the feature "red" of the focal entity "squirrel" in (18, reprinted below) initially activates a range of traces of visual experiences of the color red.

(21) The red squirrel jumped into the tree.

Most of these traces turn out not to be relevant in the context of the sentence, but that does not become clear until the first noun is processed.6 Red squirrels are a particular kind of red (brownish red rather than fire truck red). So the noun will constrain what traces are relevant in the current context. But this is only the beginning. Just as red squirrels are a particular kind of red, they are a particular kind of squirrel. Unlike the gray squirrels typically found in North America, they have ear tufts (instead of mouse-like ears) and are smaller. In other words, the two concepts, "red" and "squirrel," constrain each other's representation. One way of conceptualizing this is that all the traces that make up a dispositional representation receive some degree of activation from the associated word, but that only one or a few of them will be activated above threshold and thus become incorporated in the mental simulation. In the next cycle, "jumped" provides further constraints on the ongoing simulation. For one, it provides some articulation of the shape of the squirrel; it is stretched out, rather than sitting on its hind legs, for example.7

During construal, the information is being integrated with previous con-struals, which form part of the context for the current construal in the comprehension of connected discourse. This is where the remaining two components of construal come into play: time frame and spatial region. When two construals pertain to the same time interval, they can be integrated more easily than when they pertain to two different intervals (Zwaan, 1996). To a certain extent, this is also true for spatial regions (see

6 As evidence shows, comprehension is incremental, rather than postponed until certain linguistic boundaries are reached (e.g., Chambers et al., 2001).

7 Also note that "squirrel" constrains the interpretation of "jumped." After all, the way a squirrel jumps is different from the way a human or an antelope jumps.

Zwaan & Radvansky, 1998, for a discussion). The topic of integration is beyond the scope of the present chapter, but is discussed in Zwaan (2004).

empirical evidence

What empirical evidence do we have for our claims? In this section we review the research from our lab; further relevant evidence is reviewed in other chapters in this volume. In most experiments, we used the same methodology. Subjects are exposed to linguistic materials and are then presented with one or more pictures. Their task consists in comprehending the sentences and performing speeded judgments on the pictures. The rationale is that sentence comprehension will involve a construal in which visual traces are activated and used that either match or mismatch the visual traces created by the pictures. 'Match' and 'mismatch' should be thought of in relative terms only. In our interactions with the environment there will probably never be a perfect match between a new visual trace and one already in memory. For example, we rarely if ever see objects under identical angles and lighting conditions and against identical backgrounds on different occasions. As a result, all visual traces are slightly different from each other. Some theories of object recognition account for this by assuming that some amount of interpolation between visual traces in memory occurs in order to obtain a match with the visual input (e.g., Bulthoff & Edelman, 1992; Tarr, 1995). In the context of our experiments, the claim is simply that the match condition provides a stronger match between construal and picture traces than the mismatch condition. We should also note that in our experiments, subjects are often not directly relating the picture to the sentence. In that sense, it cannot be argued that the subjects' responses are due to special imagery strategies.

Here is a list of claims we have made about construal:

1. Comprehenders represent perceptual aspects of referents or situations;

2. Comprehenders represent spatial relations between object parts;

3. Comprehenders represent dynamic aspects of events;

4. Comprehenders represent perspective.

How do these claims hold up against empirical scrutiny? Let's revisit sentences (3) and (4), which are included below as (22) and (23).

(22) John pounded the nail into the wall.

(23) John pounded the nail into the floor.

Our contention is that these sentences lead to different mental representations. Specifically, we predict that comprehenders will represent the orientation of the nail. In order to test this prediction, Stanfield and Zwaan (2001) presented subjects with sentences such as (22) and (23), followed by a line drawing of an object. The subjects simply decided whether the picture depicted a word mentioned in the sentence. Two things are important to note. First, a picture of a nail, irrespective of its orientation, should yield a "yes" response. Second, a picture of a horizontal nail would match a construal of (22) and one of a vertical nail would match a construal of (23). Stanfield and Zwaan found that responses were significantly faster in the match than in the mismatch condition. One suboptimal feature of these experiments was that the direction in which the nail points is indeterminate in case of the wall - it could point to the left or to the right (for obvious reasons, we don't have this problem with the floor). As mentioned earlier, it may be that people use a default assumption in cases such as these. But this means that the direction of the nail would in some cases mismatch that of the nail in the construal. Our counterargument is that the match condition still provides a better match than the mismatch condition because the visual trace of a horizontal nail, regardless of its orientation, provides a better match than a picture of a vertical nail.8

In a later series of experiments, we tested the claim that construal necessarily includes a representation of the shape of a focal entity (Zwaan, Stanfield, & Yaxley, 2002). Using the same logic as in our earlier experiments, we had subjects read sentences and then presented them with pictures. The sentences were of the type of (9) and (10) or as (24) and (25).

(24) He saw the lemon in the bowl.

(25) He saw the lemon in the glass.

In addition to a recognition task, we also employed a naming task. In a naming task, the subject sees a picture and simply names it. We used a naming task because it provides a more implicit measure than a recognition task. The recognition task calls for the subject to compare the picture with the sentence. Not so in the naming task-naming the picture does not directly involve reference to the sentence. Nonetheless, in both experiments we found that the match condition yielded faster responses than the mismatch condition. These findings suggest that comprehenders routinely represent the shape of the focal entity mentioned in a sentence.

In a more recent experiment, we investigated whether comprehenders form perceptual representations of described motion (Zwaan, Madden, Yaxley, & Aveyard, 2004). Subjects listened to sentences such as (26) or (27) over headphones.

(26) The shortstop hurled the softball at you.

(27) You hurled the softball at the shortstop.

8 MacWhinney (Chapter 9, this volume) points out that the orientation of the nail is only part of the mental simulation, not the whole simulation. We concur. It is, however, a diagnostic part of the simulation and as such is amenable to an empirical test.

Baseball For Boys

Baseball For Boys

Since World War II, there has been a tremendous change in the makeup and direction of kid baseball, as it is called. Adults, showing an unprecedented interest in the activity, have initiated and developed programs in thousands of towns across the United States programs that providebr wholesome recreation for millions of youngsters and are often a source of pride and joy to the community in which they exist.

Get My Free Ebook


Post a comment