Of Minds and Language (44 page)

For the rest of this paper, I'll illustrate the explanatory power of this machinery for two kinds of case that pose principled problems for the idea that word-to-world pairing (though no doubt a necessary factor) is sufficient by itself as the information basis for vocabulary acquisition. The first case involves such perspective verb pairs as
give/get
,
chase/flee
,
buy/sell
, and the like, illustrated in
Fig. 16.1
. It depicts an action scene in which a dog is
chasing
a man. But literally by the same token it depicts a man who is
fleeing
(from) a dog. Suppose the adult utters a new verb – “Look! Pilking!” – in reference to such a scene. Is he or she speaking of chasing or of fleeing? Assuming that just these two interpretations come to mind, among the many that are really available and pertinent to the event, how is the listener to decide between them? At peril of belaboring the point, the next few hundreds of exposures to pursuit scenes are liable to embody the same ambiguity. Rarely do members of such pairs surface under real-world circumstances that differentiate between them. Which returns me to the problem that we are generally begging the questions at issue when we say that word-to-world pairing solves even the simplest cases of word learning –
that people acquire word meanings “from” observing the world. The difficulty from the outset is that, for word learning to occur, one has to conceive of the observed world
in the right way
, under the description that fits the word that is being used. But this requirement completes the circle.

Fig. 16.1. Dual conceptions: chasing and fleeing.

To escape from this circularity there has to be a way for the learner to focus (“zoom in” is our own favored technical term) on the right description (representation) of the scene without presupposing knowledge of the word whose acquisition is at issue. How could attention be focused on just one of these interpretations in the case of perspective verbs? For comparison, first consider another famous ambiguity, the duck-rabbit in
Fig. 16.2
. Perception psychologists

Fig. 16.2. Dual perceptions: duck and rabbit.

Georgiades and Harris (1997) showed that the chances of a naïve observer reporting seeing the duck versus the rabbit can be influenced by a subliminal visual attention-capture cue judiciously placed on such a figure. Perhaps more surprisingly, the same is true of chasing and fleeing depictions and other cases interpretable as one of two paired perspective verbs, including
give
vs.
get
,
win
vs.
lose
, and
buy
vs.
sell
(Gleitman, January, Nappa and Trueswell in press). Following Georgiades and Harris, we captured our subjects' visual attention on such pictures by briefly (60–80 msec) flashing a square on the computer screen just prior to the onset of the picture. This square was aligned with the upcoming position of one or the other character. Typically this caused eye movements to that character, even though the subjects were not able to report noticing the flashed square.
Fig. 16.3
exemplifies the procedure using the intended contrast
win/lose
. This manipulation reliably influenced the speaker's tendency in describing the scene. For the
chase/flee case
, the tendency to describe the scene as one of chasing was enhanced when the flash was where the dog subsequently appeared, and as one of fleeing if it was on the man. So how the speaker “attentionally approaches” an event like this does seem to affect its description and, consequently, verb choice.

It remains to ask how speaker choice might be related to the learning situation for such cases. We know from the work of Dare Baldwin (1991) that infants will attend to the direction of the speaker's gaze as a cue to the reference of a new noun. In preliminary studies we have shown adult subjects a version of these verbs in which a cartoon character (“John”) is looking down on the scene. “John's” eyes are directed either to the chaser or the fleer, as shown in
Fig. 16.4
; and again this influences the subject's report of what she thinks John would say, to describe the scene (Gleitman et al., in press). So here we have a hint that social-attentive cues from the speaker might direct the listener-learner toward a
particular choice of interpretations even in these cases where on the surface the scene itself seems to provide no basis for disambiguation.

Fig. 16.3. A subliminal attention manipulation: After visual fixation (panel 1), a block is briefly flashed, situated where (on different trials) the winner, the loser, or a place in between, will subsequently appear (panel 2). The picture then appears and the subject describes what is seen (panel 3).

Source
: Gleitman January, Nappa and Trueswell 2007

Fig. 16.4. Visual cueing of chase and flee: John's gaze direction influences subjects' utterance of chase (left image) or flee/run away (right image).

The effects of speaker-gaze direction on disambiguation of these pairs are by no means categorical even in this laboratory situation, and even with adult subjects. So I turn now to another attentional cue, evidently a more powerful one. In this experiment (Fisher, Hall, Rakowitz and Gleitman 1994) we showed 3- and 4-year-old children (and adult controls) videotaped puppet shows designed to exemplify perspective verbs, and we introduced an extra handheld puppet, telling the children that it was a Martian puppet that talks Martian talk. We asked them if they could help us figure out what Martian words this puppet was saying. One third of the children heard the Martian (who, in company with the child subjects, was viewing the puppet show) say
Look, gorping!
; the next third of the subjects heard
Look, the skunk is gorping the rabbit!
; and the final group heard
Look, the rabbit is gorping the skunk!
The results are rendered in
Fig. 16.5
, collapsing across several of the scenes that the children saw and responded to. Notice first that there is a cognitive bias in all these results toward source-to-goal interpretations. This shows up strongly for both the children and adults in the no-sentence (
Look! Gorping!
) condition which does not linguistically bias the subject. For instance,
give
is heavily preferred to
get
,
chase
is preferred to
flee
, and so forth. For the subjects who heard instead
The skunk is gorping the rabbit
, this effect is enhanced – it becomes essentially categorical because the form of the sentence supports the cognitive bias. But for those subjects who heard
The rabbit is gorping the skunk
, the results reverse. The adults shift completely to the goal-to-source verb (
flee
or
run away
) dispreferred by the prior subjects. You still see the residue of the cognitive bias with the children, but the modal response has for them too now
shifted to the goal-to-source interpretations.
¹
This pattern would be expected if the structural configuration chosen by a speaker is understood by the listener to reflect the speaker's attentional stance. Research on discourse coherence strongly suggests that subject position is often used to denote the current discourse center and to mark transitions from one center to another (e.g., Gordon, Grosz and Gillom 1993; Walker, Joshi and Prince 1998). This is why Fisher, Hall, Rakowitz and Gleitman (1994) described their effect as a “syntactic zoom lens” in which the structural configuration of the words in the utterance helps the child take the perspective necessary to achieve successful communication, and to infer the meaning of unknown elements in an utterance.

Fig. 16.5. Source versus goal by syntactic introducing context: The source is the preferred subject (e.g., the giver or chaser is preferred to the getter or evader) if the syntax is neutral. This tendency is enhanced or diminished in both adults and young children as a function of syntactic information.

Source
: (Fisher, Hall, Rakowitz and Gleitman 1994)

I want to emphasize a couple of points in wrapping up this part of the discussion. First was the idea that the word-to-world pairing procedure that is in place from earliest infancy is effective primarily for whole-object terms (Markman 1994; Gillette, Gleitman, Gleitman and Lederer 1999), accounting for the noundominated character of the novice vocabulary. My next ambition in this paper was to show how linguistic structure itself acts to redress these limitations once the novice (whether an infant or older language learner) has acquired its rudiments by considering the sequence of nouns against their contexts of use. To expose the problem and elements of the solution, I showed you how children and adults
discover the interpretation of novel terms – here, the perspective verbs – whose reference is just about always ambiguous and which therefore cannot be wholly explained as observation-based learning. Because the solution to this problem must be (somehow) to draw the observer's attention toward one of the two primary interpretations, it is reassuring that attentional cues of varying kinds, including subliminal flashes but also eye-gaze direction of a cartoon figure, materially influenced these interpretations in the laboratory. Perhaps more surprising, especially in its influence on young preschoolers, is that the strongest cue of all was implicit and linguistic. They interpreted the scene in accord with the semantic information latent in the structure of the introducing sentence, specifically, according to which character captured the subject position.