Mutual Exclusivity Bias Research Articles

Mutual Exclusivity in Pragmatic Agents

One of the great challenges in word learning is that words are typically uttered in a context with many potential referents. Children's tendency to associate novel words with novel referents, which is taken to reflect a mutual exclusivity (ME) bias, forms a useful disambiguation mechanism. We study semantic learning in pragmatic agents-combining the Rational Speech Act model with gradient-based learning-and explore the conditions under which such agents show an ME bias. This approach provides a framework for investigating a pragmatic account of the ME bias in humans but also for building artificial agents that display an ME bias. A series of analyses demonstrates striking parallels between our model and human word learning regarding several aspects relevant to the ME bias phenomenon: online inference, long-term learning, and developmental effects. By testing different implementations, we find that two components, pragmatic online inference and incremental collection of evidence for one-to-one correspondences between words and referents, play an important role in modeling the developmental trajectory of the ME bias. Finally, we outline an extension of our model to a deep neural network architecture that can process more naturalistic visual and linguistic input. Until now, in contrast to children, deep neural networks have needed indirect access to (supposed to be novel) test inputs during training to display an ME bias. Our model is the first one to do so without using thismanipulation.

Where's the Advantage? Mutual Exclusivity Promotes Children's Initial Mapping, but Not Long-Term Memory, for Words Compared to Other Strategies.

Children frequently apply a novel label to a novel object, a behavior known as the mutual exclusivity bias (MEB). This study examined how MEB affects children’s retention for word mappings. In Experiment 1, preschoolers (N = 39; Mage = 46.62 months) and adults (N = 24; Mage = 21.63 years) completed an immediate word mapping task and a delayed retention test. Both samples used MEB during referent selection, but neither group displayed higher retention for words mapped via MEB than words mapped via other referent selection strategies at test. Experiment 2 replicated Experiment 1 with preschoolers (N = 85; Mage = 47.78 months) and provided evidence against the possibility that interference from multiple words contributed to children’s faster forgetting of word mappings when using MEB. Experiment 3 presented children (N = 30; Mage = 51.13 months) with an abbreviated version of the task, providing evidence against the alternative hypothesis that cognitive load during learning caused the forgetting observed in Experiments 1 and 2. Taken together, these experiments suggest that MEB supports initial word mapping but may not provide an advantage for long-term retention.

Bayesian Word Learning in Multiple Language Environments.

Infant language learners are faced with the difficult inductive problem of determining how new words map to novel or known objects in their environment. Bayesian inference models have been successful at using the sparse information available in natural child-directed speech to build candidate lexicons and infer speakers' referential intentions. We begin by asking how a Bayesian model optimized for monolingual input (the Intentional Model; Frank etal., 2009) generalizes to new monolingual or bilingual corpora and find that, especially in the case of the bilingual input, the model shows a significant decrease in performance. In the next experiment, we propose the ME Model, a modified Bayesian model, which approximates infants' mutual exclusivity bias to support the differential demands of monolingual and bilingual learning situations. The extended model is assessed using the same corpora of real child-directed speech, showing that its performance is more robust against varying input and less dependent than the Intentional Model on optimization of its parsimony parameter. We argue that both monolingual and bilingual demands on word learning are important considerations for a computational model, as they can yield significantly different results than when only one such context is considered.

The Role of Novelty in Early Word Learning

What mechanism implements the mutual exclusivity bias to map novel labels to objects without names? Prominent theoretical accounts of mutual exclusivity (e.g., Markman, 1989, 1990) propose that infants are guided by their knowledge of object names. However, the mutual exclusivity constraint could be implemented via monitoring of object novelty (see Merriman, Marazita, & Jarvis, 1995). We sought to discriminate between these contrasting explanations across two preferential looking experiments with 22-month-olds. In Experiment 1, infants viewed three objects: one name-known, two name-unknown. Of the two name-unknown objects, one was novel, and the other had been previously familiarized. The infants responded to hearing a novel label by increasing attention only to the novel, name-unknown object. In a second experiment in which the name-known object was absent, a novel label increased infants' attention to a novel object beyond baseline preference for novelty. The experiments provide clear evidence for a novelty-based mechanism. However, differences in the time course of disambiguation across experiments suggest that novelty processing may be influenced by contextual factors.

Language Experience Shapes the Development of the Mutual Exclusivity Bias.

Halberda (2003) demonstrated that 17-month-old infants, but not 14- or 16-month-olds, use a strategy known as mutual exclusivity (ME) to identify the meanings of new words. When 17-month-olds were presented with a novel word in an intermodal preferential looking task, they preferentially fixated a novel object over an object for which they already had a name. We explored whether the development of this word-learning strategy is driven by children's experience of hearing only one name for each referent in their environment by comparing the behavior of infants from monolingual and bilingual homes. Monolingual infants aged 17-22 months showed clear evidence of using an ME strategy, in that they preferentially fixated the novel object when they were asked to "look at the dax." Bilingual infants of the same age and vocabulary size failed to show a similar pattern of behavior. We suggest that children who are raised with more than one language fail to develop an ME strategy in parallel with monolingual infants because development of the bias is a consequence of the monolingual child's everyday experiences with words.

Novel labels support 10-month-olds’ attention to novel objects

What is the source of the mutual exclusivity bias whereby infants map novel labels onto novel objects? In an intermodal preferential looking task, we found that novel labels support 10-month-olds’ attention to a novel object over a familiar object. In contrast, familiar labels and a neutral phrase gradually reduced attention to a novel object. Markman (1989, 1990) argued that infants must recall the name of a familiar object to exclude it as the referent of a novel label. We argue that 10-month-olds’ attention is guided by the novelty of objects and labels rather than knowledge of the names for familiar objects. Mutual exclusivity, as a language-specific bias, might emerge from a more general constraint on attention and learning.

On having complex representations of things: preschoolers use multiple words for objects and people.

Applying several names to an entity (polynomy) reflects the ability to categorize entities in different ways. Two experiments demonstrate preschoolers' abilities to apply multiple labels to representational objects and to people. In Experiment 1, 3- and 4-year-olds labeled representational objects and verified labels for story characters. In both tasks children reliably produced or accepted several words per entity and accepted a high percentage of both class-inclusive and overlapping word pairs. These results were replicated in Experiment 2; 3- to 5-year-olds also completed appearance-reality and receptive vocabulary tests. The mean number of words produced in the labeling task was significantly related to receptive vocabulary, but not to appearance-reality performance. The results indicate that preschoolers represent an entity as belonging to multiple categories (e.g., dinosaur and crayon). Implications for cognitive and language development, particularly the appearance-reality distinction and the mutual exclusivity bias, are discussed.

On having complex representations of things: Preschoolers use multiple words for objects and people.

Applying several names to an entity (polynomy) reflects the ability to categorize entities in different ways. Two experiments demonstrate preschoolers' abilities to apply multiple labels to representational objects and to people. In Experiment 1, 3- and 4-year-olds labeled representational objects and verified labels for story characters. In both tasks children reliably produced or accepted several words per entity and accepted a high percentage of both class-inclusive and overlapping word pairs. These results were replicated in Experiment 2; 3- to 5-year-olds also completed appearance-reality and receptive vocabulary tests. The mean number of words produced in the labeling task was significantly related to receptive vocabulary, but not to appearance-reality performance. The results indicate that preschoolers represent an entity as belonging to multiple categories (e.g., dinosaur and crayon). Implications for cognitive and language development, particularly the appearance-reality distinction and the mutual exclusivity bias, are discussed.

Restricting a Familiar Name in Response to Learning a New One: Evidence for the Mutual Exclusivity Bias in Young Two-Year-Olds

Children under 2 1/2 years old tend to interpret novel words in accordance with the Mutual Exclusivity Principle, but tend not to reinterpret familiar words this way. Because alternative principles have been proposed that only predict the novel word effects, and because tests of the familiar word effects may have been flawed, a new test was administered. In Experiment 1 (N = 32), 24- to 25-month-olds heard stories in which a novel noun was used for an atypical exemplar of a familiar noun. When asked to select exemplars of the familiar noun, they showed a small but reliable tendency to avoid the object from the story. In Experiment 2 (N = 16), the novel nouns in the stories were replaced by pronouns and proper names, and the children did not avoid the story object in the test of the familiar noun. Thus, the aversion to this object that was observed in Experiment 1 was not due to its greater exposure or its being referenced immediately before testing, but to toddlers' Mutual Exclusivity bias. Their bias is hypothesized to be a form of implicit probabilistic knowledge that derives from the competitive nature of category retrieval.

Restricting a Familiar Name in Response to Learning a New One: Evidence for the Mutual Exclusivity Bias in Young Two-Year-Olds

Restricting a Familiar Name in Response to Learning a New One: Evidence for the Mutual Exclusivity Bias in Young Two-Year-Olds

How shall a deceptive thing be called?

Thirty-six three-, four- and five-year-olds were asked to select labels for deceptive stimuli (e.g. for an eraser that looked like a pencil). Three types of labelling were investigated--simple (e.g. 'is an eraser'); appearance-predicated (e.g. 'looks like an eraser'); and reality-predicated (e.g. 'is really and truly an eraser'). An age-related appearance--reality shift was observed in simple labelling (e.g. older children were more likely than younger ones to accept eraser and reject pencil as simple names for the pencil-eraser). This trend was robust over method and semantic domain, though weaker with object than with colour labels. As in previous research, older children were more likely than younger ones to map different appearance- than reality-predicated labels onto an item (e.g. to accept that the pencil-eraser looks like a pencil, but is really and truly an eraser); however, all age groups were reluctant to extend more than one name to a stimulus via a common predicate (e.g. to accept two reality-predicated labels for the same object). This one-label-perpredicate pattern was observed more frequently within reality than within appearance predicates; more frequently with colour than with object names, and with questions blocked by predicate than by name. It is argued that younger children maintained this pattern because of inflexible encoding, but that older ones did so because of better understanding of the appearance-reality distinction, greater reality dominance, and a Mutual Exclusivity bias.

The mutual exclusivity bias in children's word learning: By W. E. Merriman and L. L. Bowman. In Monographs of the Society for Research in Child Development, 54, Serial No. 220, 1989

Merriman and Bowman (1989) explore one proposed constraint on word learning, mutual exclusivity, the assumption that each object has only one label. Chief among their contributions is treatment of mutual exclusivity as a default assumption—a probabilistic bias which can be overridden. We elaborate this view to address misconceptions of the notions of biological constraints that pervade recent discussions of constraints on word learning. Misconstruing constraints as rigid, absolute responses instead of probabilistic biases has led researchers to interpret any violation as invalidating a given constraint. More confusion surrounds questions about the origins of the constraints. We dispute the idea that the age of appearance of lexical constraints reveals whether the constraints are innate, and argue that current discussions of the innateness of constraints are over-simplified. In this case, we also question the appropriateness of Merriman and Bowman's methodology for use with 2-year-olds and challenge their conclusion that mutual exclusivity is absent in children under 2 1 2 . Merriman and Bowman's thoughtful conceptual analysis establishes several distinct ways in which mutual exclusivity can be manifested. Thus, putative counterexamples occur when an investigator tests for only one consequence of mutual exclusivity and ignores its other possible implications. Merriman and Bowman's studies with children from 2 1 2 on document that each of these alternative ways of preserving mutual exclusivity guides word learning.

The mutual exclusivity bias in children's word learning: A reply to Woodward and Markman

My reply to Woodward and Markman's (1990) commentary on the monograph that I coauthored with Bowman ( Merriman & Bowman, 1989) has three parts. First, I elaborate on the aspect of our position that they liked the most—namely, our conceptualization of the Mutual Exclusivity bias in children's word learning as a default assumption that can be maintained by four strategies. Second, I present an updated review of the factors that affect whether a particular strategy will be used by children who are 2 1 2 years or older. Woodward and Markman accepted some of our claims about the way the bias is manifest after toddlerhood, but made a few mistakes in summarizing evidence and ignored some of our proposals. Finally, I consider what was Woodward and Markman's least favorite part of the monograph—namely, our conclusion that children who are less than 2 1 2 years old lack the bias. I retrace the steps that Bowman and I took in reaching this conclusion, respond to specific criticisms, and acknowledge that the case for the bias in toddlers is stronger than we thought, though still not completely convincing. I emphasize that, at most, toddlers have a weak bias that has little impact on their solution to the problem of overextension.

The Mutual Exclusivity Bias in Children's Word Learning

Nearly every recent account of children's word learning has addressed the claim that children are biased to construct mutually exclusive extensions, that is, that they are disposed to keep the set of referents of one word from overlapping with those of others. Three basic positions have been taken--that children have the bias when they first start to learn words, that they never have it, and that they acquire it during early childhood. A review of diary and test evidence as well as the results of four experiments provide strong support for this last view and indicate that the bias develops in the months following the second birthday but does not gain full strength or become accessible to consciousness until sometime after the third birthday. Several studies also show that, after this point, it can still be counteracted by information in input or by a strong belief that something belongs to the extension of a particular word. The full body of evidence is compatible with the view that mutual exclusivity is the default option in children's and adults' procedures for integrating the extensions of new and old words. We present several arguments for the adaptive value of this kind of bias.