Psychology Language: How We Understand

How We Understand

How do listeners decipher the drama of who-did-what-to-whom from the myriad com-plex and ambiguous sentence forms in which this drama can be expressed? A major por-tion of the answer lies in a complex, rapid, nonconscious processing system churning away to recover the structure—and hence the semantic roles—as the speaker’s utterance arrives word by word at the listener’s ear (Marcus, 2001; Savova et al., 2007; Tanenhaus Trueswell, 2006). To get a hint of how this system operates, notice first that the vari-ous function morphemes in speech help mark the boundaries between phrases and propositions (for instance, but or and) and reveal the roles of various content words even when their order changes. For instance, when we reorder the phrases in a so-called

passive-voice sentence, we say, The cheese was eaten by the mouse instead of The mouse atethe cheese. The telltale morphemes -en and by cue the fact that the done-to rather than thedo-er has become the subject (Bever, 1970). Sometimes the rhythmic structure of speech (or the hyphen punctuation in written English) helps to disambiguate the utterance, as in the distinction between a black bird-house and a black-bird house (L. R. Gleitman & H. Gleitman, 1970; Snedeker & Trueswell, 2003; see Figure 10.16). Listeners also are sensi-tive to many clues from background knowledge and plausibility that go beyond syntax to discern the real communicative intents of speakers. We next discuss some examples of how these kinds of clues work together to account for the remarkable speed and accuracy of human language comprehension despite the apparent complexity and ambiguity of the task that the listener faces.

THE FREQUENCY WITH  WHICH THINGS HAPPEN

Often a listener’s interpretation of a sentence is guided by background knowledge, knowledge that indicates the wild implausibility of one interpretation of an otherwise ambiguous sentence (G. Altmann & Steedman, 1988; Sedivy, Tanenhaus, Chambers, & Carlson, 1999). For example, no sane reader is in doubt over the punishment meted out to the perpetrator after seeing the headline Drunk Gets Six Months in Violin Case (Pinker, 1994). But in less extreme cases, the correct interpretation is not immediately obvious. Most of us have had the experience of being partway through hearing or reading a sen-tence and realizing that somewhere we went wrong. For example, we may make a word-grouping error, as in reading a sentence that begins The fat people eat . . . The natural inclination is to take the fat people as the subject noun phrase and eat as the beginning of the verb phrase (Bever, 1970). But suppose the sentence continues:

The fat people eat accumulates on their hips and thighs.

Now one must go back and reread. (Notice that this sentence would have been much easier if, as is certainly allowed in English, the author had placed the function word that before the word people: The fat that people eat accumulates on their hips and thighs.) The partial misreading (or “mishearing” in the case of spoken language) is termed a gardenpath (in honor of the cliché phrase “led down the garden path,” in which someone canbe deceived without noticing it). Because of the misleading content or structure at the beginning of the sentence, the reader is enticed toward one interpretation, but he must then retrace his mental footsteps to find a grammatical and understandable alternative.

Psycholinguists have various ways of detecting when people are experiencing a garden path during reading. One is to use a device that records the motion of the reader’s eyes as they move across a page of print. Slowdowns and visible regressions of these eye move-ments tell us where and when the reader has gone wrong and is rereading the passage (MacDonald, Pearlmutter, & Seidenberg, 1994; Rayner, Carlson, & Frazier, 1983; Trueswell, Tanenhaus, & Garnsey, 1994). Using this technique, one group of investigators looked at the effects of plausibility on readers’ expectations of the structure they were encountering (see Figure 10.17). Suppose that the first three words of a test sentence are

The detectives examined . . .

Participants who read these words typically assume that The detectives is the subject of the sentence and that examined is the main verb. They therefore expect the sentence to end with some noun phrase—for example: the evidence. As a result, they are thrown off track when they dart their eyes forward and instead read that the sentence continues

. . . by the reporte.

The readers’ puzzlement is evident in their eye movements: They pause and look back at the previous words, obviously realizing that they need to revise their notion that examined was the main verb. After this recalculation, they continue on, putting all of thepieces together in a new way, and so grasp the entire sentence:

The detectives examined by the reporter revealed the truth about the robbery.

The initial pause at by the reporter showed that readers had been led down the gar-den path and now had to rethink what they were reading. But what was it exactly that led the participants off course with this sentence? Was the difficulty just that passive-voice sentences are less frequent than active-voice sentences? To find out, the experi-menters also presented sentences that began

The evidence examined by the reporter . . .

Now the participants experienced little or no difficulty, and read blithely on as the sen-tence ended as it had before (. . . revealed the truth about the robbery). Why? After all, this sentence has exactly the same structure as the one starting The detectives . . . and so, apparently, the structure itself was not what caused the garden path in this case. Instead, the difficulty seems to depend on the plausible semantic relations among the words. The noun detectives is a “good subject” of verbs like examined because detectives often do examine things—such as footprints in the garden, spots of blood on the snow, and so on. Therefore, plausibility helps the reader to believe that the detectives in the test sentence did the examining—thus leading the reader to the wrong interpretation. Things go differently, though, when the sentence begins The evidence, because evidence, of course, is not capable of examining anything. Instead, evidence is a likely object of someone’s examination, and so a participant who has read The evidence examined . . . is not a bit surprised that the next word that comes up is by. This is the function mor-pheme that signals that a passive-voice verb form is on its way—just what the reader expected given the meanings of the first three words of the sentence (Trueswell & Tanenhaus, 1992).

It seems, then, that the process of understanding makes use of word meanings and sentence structuring as mutual guides. We use the meaning of each word (detectives ver-sus evidence) to guide us toward the intended structure, and we use the expected struc-ture (active versus passive) to guess at the intended meanings of the words.

WHAT IS HAPPENING RIGHT NOW ?

Humans often talk about the future, the past, and the altogether imaginary. We devour books on antebellum societies that are now gone with the wind, and tales that speak of . . . a Voldemort universe that we hope never to experience. But much of our conversation is focused on more immediate concerns, and in these cases the listener can often see what is being referred to and can witness the actions being described in words. This sets up a two-way influence—with the language we hear guiding how we perceive our surround-ings, and the surroundings in turn shaping how we interpret the heard speech.

For example, in one experiment, on viewing an array of four objects (a ball, a cake, a toy truck, and a toy train), participants listening to the sentence Now I want you to eatsome cake turned their eyes toward the cake as soon as they heard the verb eat (Altmann Kamide, 1999) and before hearing cake. After all, it was unlikely that the experi-menter would be requesting the participants to ingest the toy train. In this case, the meaning of the verb eat focused listeners’ attention on only certain aspects of the world in view—the edible aspects!

Just as powerful are the reverse phenomena: effects of the visually observed world on how we interpret a sentence (consider the array of toy objects in Figure 10.18A). These include a beanie-bag frog sitting on a napkin and another napkin that has no toy on it. When study participants look at such scenes and hear the instruction Put thefrog on the napkin into the box, most of them experience a garden path, thinking (whenthey hear the first six words) that on the napkin is the destination where a frog should next be placed. After all, the empty napkin seems a plausible destination for the frog. But three words later (upon hearing into the box) they are forced to realize that the intended destination is really the box and not the empty napkin after all. This double-take reaction is evident in the participants’ eye movements: On hearing napkin, they look first to the empty napkin, and then look around the scene in confusion when they hear into the box. Of course, adult participants rapidly recover from this momentary boggle, and go on to execute the instruction correctly, picking up the frog and putting it in the box. But the tell-tale movement of the eyes has identified the temporary mis-interpretation, a garden path.*

But now consider the array of objects in Figure 10.18B. It differs from the array in Figure 10.18A, for now there are two frogs, only one of which is on a napkin. This has a noticeable effect on participants’ eye movements. Now, on hearing the same instruc-tion, most of the participants immediately look to the frog that’s already on a napkin when they hear napkin, and they show no subsequent confusion on hearing into the box.

What caused the difference in reaction? In the array of Figure 10.18A, with only one frog, a listener does not expect the speaker to identify it further by saying the green frog or the frog to the left or the frog on the napkin, for there would be no point in doing so. Though such descriptions are true of that particular frog, there is no need to say so— it is obvious which frog is being discussed, because only one is in view. When the lis-tener hears on the napkin, therefore, she assumes (falsely) that this is a destination, not a further specification of the frog.

Things are different, though, with the array shown in Figure 10.18B. Now there is a risk of confusion about which frog to move, and so listeners expect more information. In this two-frog situation, therefore, the listener correctly assumes that on the napkin is the needed cue to the uniquely intended frog, and so he does not wander down the mental garden path (Crain & Steedman, 1985; Tanenhaus, Spivey-Knowlton, Eberhard, & Sedivy, 1995; Trueswell, Sekerina, Hill, & Logrip, 1999).

CONVERSATIONAL INFERENCE : FILLING IN THE BLANKS

The actual words that pass back and forth between people are merely hints about the thoughts that are being conveyed. In fact, talking would take just about forever if speak-ers literally had to say all, only, and exactly what they meant. It is crucial, therefore, that the communicating pair take the utterance and its context as the basis for making a series of complicated inferences about the meaning and intent of the conversation (P. Brown & Dell, 1987; Grice, 1975; Noveck & Sperber, 2005; Papafragou, Massey, & Gleitman, 2006). For example, consider this exchange:

·              Do you own a Cadillac?

·              I wouldn’t own any American car.

Interpreted literally, Speaker B is refusing to answer Speaker A’s yes/no question. But Speaker A will probably understand the response more naturally, supplying a series of plausible inferences that would explain how her query might have prompted B’s retort. Speaker A’s interpretation might go something like this: “Speaker B knows that I know that a Cadillac is an American car. He’s therefore telling me that he does not own a Cadillac in a way that both responds to my question with a no and also tells me some-thing else: that he dislikes all American cars.”

Such leaps from a speaker’s utterance to a listener’s interpretation are commonplace. Listeners do not usually wait for everything to be said explicitly. On the contrary, they often supply a chain of inferred causes and effects that were not actually contained in what the speaker said, but that nonetheless capture what was intended (H. H. Clark, 1992).

LANGUAGE COMPREHENSION

We have seen that the process of language comprehension is marvelously complex, influenced by syntax, semantics, the extralinguistic context, and inferential activity, all guided by a spirit of communicative cooperation. These many factors are uncon-sciously processed and integrated “on line” as the speaker fires 14 or so phonemes (about 3 words, on average) a second toward the listener’s ear. Indeed this immediate use of all possible cues is what sometimes sends us down the garden path with false and often hilarious temporary misunderstandings (Grodner & Gibson, 2005). But in the usual case, the process of understanding would be too slow and cumbersome to sustain conversation if the mind reacted to each sentence only at its very end, after all possible information had been delivered, word by word, to the ear. The mind thus makes a trade-off between rate and accuracy of comprehension—the small risk of error is compensated for by the great gain in speed of everyday understanding. When

we hear These missionaries are ready to eat or Will you join me in a bowl of soup, our common sense and language skill combine in most cases to save us from drowning in confusion (Gibson, 2006). Most of the time we don’t even consciously register the various zany interpretations that the language “theoretically” makes available for many sentences (Altmann &Steedman, 1988; Carpenter, Miyake, & Just, 1995; Dahan Tanenhaus, 2004; MacDonald, Pearlmutter, & Seidenberg, 1994; Marslen-Wilson, 1975; Tanenhaus & Trueswell, 2006).