We study the impossible.

---

Introduction | 2-D PLP | 3-D PLP | Current Research

---

 

 

 

 

 

 

 

 

---

Introduction | 2-D PLP | 3-D PLP | Current Research

---

Introduction to Word Learning
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Picture this. The sunlight filters through the windows of a child's playroom, casting a warm contented glow on all the objects and creatures within. "Billy, look at the ball. Do you see the ball?" A mother smiles as she sees Billy looking with great purpose and intent at the ball, practically buried amongst all the other toys. "Great job! I knew you could do it." She hugs him to her, secure in the knowledge that even though Billy is not yet old enough to speak, he knows far more than he usually lets on.
     This little drama, playing itself out in countless houses across the world, illustrates a paradox of developmental psychology. While mothers are virtually certain that their children know and understand more than those children say, methods to test these intuitions have been few and far between until relatively recently within the history of developmental psychology.
     Indeed, these are exciting times for developmental psychologists. This excitement comes because of the development of new methods which make possible the study of human infants at earlier and earlier ages. Questions of emergent competencies and skills which could not have been answered even as recently as 20 years ago are reluctantly yielding up their secrets. Developmental psychologists now have at their disposal a powerful arsenal of tools and methods to examine early infant cognition and behavior in ways pioneers in the field could not have even dreamed.
     At the Temple University Infant Labs we make use of two such "new methods" to help us study children's emerging language competancies: the 2-D Preferential Looking Paradigm and the 3-D Preferential Looking Paradigm.

 

 

 

 

 

---

Introduction | 2-D PLP | 3-D PLP | Current Research

---

2-D Intermodal Preferential Looking Paradigm
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

The 2-D Intermodal Preferential Looking Paradigm was adapted from the work of Spelke (1979), who developed it to study intermodal perception. In it, the infant is seated on a blindfolded parent's lap in front of two television monitors. A concealed audio speaker plays a linguistic stimulus that is consonant with or "matches" only one of the displays shown on the screens. Mounted atop the speaker is a light that comes on when ever the screens are both black. The function of this light is both to hold the child's attention while the screens are black and to insure that the infant is looking to the center at the beginning of each trial. The infant's task is to look at one of the two video screens. An observer (either on-line or off line, watching the videotape) records all looking responses. The stimuli on both screens are matched for perceptual cues like, color, brightness, movement, etc.
     To see how this works let us consider a case of noun comprehension: a case very similar to that of the story above (Golinkoff, et.al. 1987). A center light gets the child's attention, reorients them towards the center, and a linguistic message (produced in child directed speech) is heard: "where's the apple?" Then on one screen, the child sees a bell and on the other an apple, and the linguistic message is: "Do you see the apple? Show me the apple!" The hypothesis, which was confirmed, was that children would look more quickly and longer at the screen displaying the apple than at the screen displaying the bell.
     Both because such a minimal response is required and because the TV's allow for presentation and child selection between dynamic stimuli, the 2-D Intermodal Preferential looking Paradigm has proved very useful in the study of early language. A sampling of some of the results thus far, we know that: 1) infants are sensitive to cues for constituent structure by 14 months of age, 2) to word order by 17 months of age; 3) to morphological word endings by about 20 months of age; 4) and to information about sentence frames by 24 months of age.

 

 

 

 

 

---

Introduction | 2-D PLP | 3-D PLP | Current Research

---

3-D Preferential Looking Paradigm
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Unlike the 2-D version of this paradigm, which uses television monitors, VCR's, an audio speaker, and a light, the 3-D Intermodal Preferential Looking Paradigm accomplishes a functionally similar objective by means of an experimenter visible to the baby, a video camera, and a special display board, the design of which was taken from a beautiful apparatus Fagan constructed for his infant test of attention. In the 3-D paradigm, infants are presented with actual, real "3-D" objects that are labeled in a play environment. These same objects are then presented side by side on the Fagan board during test trials. As in all such preferential looking paradigms, the infant's task is simply watch the displays. Any preference for one object over another will be detected by a video camera which records all looking responses. As in the 2-D Paradigm, it is expected that the children will look at that stimulus that is consonant with, or matches, the linguistic stimuli they are hearing. In this manner, it is possible to create very realistic (yet highly controled) labeling situations, and then study the child's abilities to map the labels onto various objects.

 

 

 

 

 

---

Introduction | 2-D PLP | 3-D PLP | Current Research

---

The Origins of Word Learning
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

As with many word learning stories, our tale starts with the now classic conundra from Quine. In this 1960 parable, Quine tells of a linguist who is trying to learn a foreign language by observing the speech of native speakers. "Gavagai," exclaims the native speaker as a rabbit scurries by. The linguist notes that "gavagai" must mean RABBIT. The linguist, however, acted in haste. There are, in fact, an indeterminate number of word-to-world mappings. "Gavagai" could just as easily referred to rabbit parts or rabbit fur, or for that matter, even the way in which the rabbit moved.
     This Quinean induction problem is the starting point for many who study the problem of word learning. How do infants sift through the myriad of potential meanings to learn words in their language so quickly and apparently, so effortlessly? By just 18 months of age, infants are virtual masters at the word learning game, acquiring an average of 9 new words per day (Carey, 1982) Moreover, there is some evidence that 11 month old children can acquire a new label for a new object with just one repetition.
     To address the question of how children learn words so rapidly, psychologists have proposed two apparently diametrically opposed classes of theories. One class of theories, "constraints" or "principles" theories, solve the induction problem by positing that children entertain only certain hypotheses for a word meaning. For example, children assume that words (though not music) refer to an object, action or event (Macnamara, 1982) and that words map onto whole objects rather than onto object parts (Markman, 1989). The second class of theories, social pragmatic theories, place the solution to Quine's problem in the social interaction between the apprentice (e.g. the child) and the sophisticated conversational partner (e.g.the mother).
     In general these two theories have had a polarizing effect on the word learning literature (Tucker, Hirsh-Pasek, & Hollich, in press). Indeed, in their 1994 paper, Golinkoff, Mervis, and Hirsh-Pasek endorsed the principles account and proposed 6 developmental lexical principles that children use to acquire a vocabulary. These principles were arranged on two developmental tiers. Three principles designed to "get word learning off the ground" occupied the first tier and were thought to be in place during the first half of the child's second year of life: the principles of reference (that a word refers), object scope (Markman, 1989, that a word refers to a whole object and not initially to object parts or attributes) and extendibility (that a word does not refer uniquely to any specific object).
     The three other principles were thought to emerge during the second half of the first year of life around the time of the"vocabulary spurt,"and they are: novel name nameless category (attach a novel name to the object or action for which you do not have a name), categorical scope (that words refer to related classes of objects), and conventionality (Clark, 1987, that words have conventionally agreed-upon meanings). The first tier principles were proposed as the foundation for the second tier, more sophisticated principles.
     In subsequent work, however, Golinkoff and Hirsh-Pasek developed a coalition view of this word learning principle perspective. Perhaps, the principles themselves are the products of development, not the engines of development. For example, children may not begin with a mature principle of reference. Rather, this principle might emerge on a continuum as the children acquire their very first words. Indeed, reference may begin as simple associative learning (a "goes with" kind of representation). Only at 17 to 19 months of age (shortly after the vocabulary spurt), would children abandon this associative principle in favor of one that is more sophisticated -- a 'stands for' relationship between word and referent.
     Thus this newer version of the model takes a midline position between nativism and constructivism: asserting that development is neither completely predescribed within the genes or purely the result of unbiased learning. Instead, development is the product of complex epigenetic interactions between weighted constraints and the external environment. Although present from the beginning of development, these constraints, or principles, may themselves mature over time. In 1991, Rochele Gelman called this approach a "rationalist constructivist" view. In 1994, Karmiloff-Smith talked about a similiar perspective as "representational redescription." Finally, in 1996, the "hot off the press" endorsement of this view can be found as Elman and his colleagues encourage readers to "Rethink Innateness."
     In the revised Hirsh-Pasek and Golinkoff story then, children are believed to be surrounded by a coalition of multiple input sources. As development progresses, constraints mature which predispose children to be sensitive to certain aspects of that input and not to others. For example, children might learn to pay greater attention to the shape of objects rather than their texture, in a labeling situation (Smith, Landau and Jones, 1994). Conversely, with time, function of the objects may come to play a larger role in categorization and labeling.
     These constraints or sensitivities to certain pieces of the input provide the groundwork that allows children to engage in a type of guided distributional learning in which they attend to certain cues in the input and then construct mature word learning principles. Thus, mature principles are an emergent phenomenon. This idea that the principles might be evolving heuristics is consistent with theorizing by Louis Bloom (1994) and Jones and Smith (1993). Importantly, under this cursory view of the theory, multiple sources are always available at the outset of language learning, and the differences that we see throughout development result because children are differentially sensitive to some of these inputs over others at different points in development.
     Among the coalition of cues, then, that children use to form these heuristics or principles for word learning are perceptual cues (movement, shape, texture), functional cues (what the object does), temporal cues (which objects they see first, or contigous with labeling) and last but not least, social cues. Thus, the social pragmatic view is no longer inherently inconsistent with the principles account of word learning. Rather, it is an intergral part of the foundation for the construction of those principles. In this coalition model of word learning, children differentially weight various perceptual, functional, temporal and social cues over time as they develop mature principles. While still endorsing a "principles-type" account, this newer examination of the constraints allows for an integrative approach to the theories in the field incorporating both the data from the social pragmatist and the lexical principle's camp.
     Thus one aspect of our current research within the Temple Infant Labs is focused on establishing this coalition of cues: discovering which cues can be used at various points in development, and discovering what effects the cues have as they are put in conflict with one another.

 

Word Reference Study 1:

Question: Is social eye gaze a cue to early word learning? That is, if presented with two toys, both of which are of equal saliency, do children "know" to label the toy to which the adult is paying attention? At what age does this ability first appear?

 

Word Reference Study 2:

How much do children "weight" social eye gaze in comparison to perceptual saliency? That is, if presented with two toys, one of which is perceptually very interesting, and one of which is reasonably non-descript, will children still "know" to label the toy to which the adult is paying attention: even if that toy is the

 

Word Reference Study 3:

Question: How much do children "weight" social cues like pointing an handling in comparison to perceptual saliency? That is, if presented with two toys, one of which is perceptually very interesting, and one of which is reasonably non-descript, will children still "know" to label the toy to which the adult is paying attention: even if that toy is the nondescript (boring) one, particularly if the adult is both pointing and handling the boring toy?

 

PostScript:

These are just three of the word learning studies, more are outlined within our NICHD grant and will be run sometime within the near future These include tests of the whole object assumption, novel name nameless category, and extendability questions. Recent results of these studies can be found in our 1998 ICIS Poster Presentation, and in our upcoming SRCD monograph.