Elsevier

Cognition

Volume 87, Issue 1, February 2003, Pages B23-B34
Cognition

Brief article
The development of a word-learning strategy

https://doi.org/10.1016/S0010-0277(02)00186-5Get rights and content

Abstract

Two studies investigated young infants' use of the word-learning principle Mutual Exclusivity. In Experiment 1, a linear relationship between age and performance was discovered. Seventeen-month-old infants successfully used Mutual Exclusivity to map novel labels to novel objects in a preferential looking paradigm. That is, when presented a familiar and a novel object (e.g. car and phototube) and asked to “look at the dax”, 17-month-olds increased looking to the novel object (i.e. phototube) above baseline preference. On these trials, 16-month-olds were at chance. And, 14-month-olds systematically increased looking to the familiar object (i.e. car) in response to hearing the novel label “dax”. Experiment 2 established that this increase in looking to the car was due solely to hearing the novel label “dax”. Several possible interpretations of the surprising form of failure at 14 months are discussed.

Introduction

Word-learning is a problem of induction. For any new word, there are a myriad of possible meanings and possible referents. To help pare down this large hypothesis-space, children may use principles which constrain what a new word can refer to (Carey, 1978, Golinkoff et al., 1994, Markman and Hutchinson, 1984, Markman, 1992, Mervis, 1987; but see Gathercole, 1989, Nelson, 1988). By looking at ever younger children, researchers hope to discover both the structure and the scope of these constraints. This article investigates the status of one such constraint, children's tendency to map novel labels to novel objects, in infants who are just beginning to learn words.

There have been multiple proposals as to the particular principle that guides children to map novel labels (i.e. labels with no known referent) to novel objects (i.e. objects for which the child does not know a name). To see how these principles may help constrain the possible referents of a new word consider one such principle, Mutual Exclusivity. Mutual Exclusivity (ME), simply stated, is the assumption that every object has just one name (Markman and Wachtel, 1988, Merriman and Bowman, 1989). Imagine that a child is presented with two toys. One is a toy car, and the child already knows the name “car”. The other has never been seen before. If the experimenter asks the child to “Hand me the car”, children as young as 16 months comply (Mervis & Bertrand, 1994). But, if the experimenter instead asks, “Hand me the dax”, how is the child to decide on the correct referent? Utilizing ME, the child may be driven to preferably map the novel label “dax” to the novel toy because the car already has a known label (i.e. “car”) and, according to ME, it cannot receive another.

Each of the different principles proposed to motivate the mapping of novel labels to novel objects (of which ME is one example) makes some commitment to an underlying computational structure. ME motivates the mapping of novel labels to novel objects via the rejection of objects that already have a known label (e.g. “car”) from consideration (Markman, 1990). The principles of Contrast (Clark, 1990, Clark, 1993) and a Pragmatic Account (Diesendruck and Markson, 2001, Tomasello and Barton, 1994) share this computational structure but offer a different motivation for the rejection of objects with known labels. Contrast proposes that children wish to avoid synonyms. Because known objects already have known labels (e.g. “car”), children will reject them as possible referents of a new word.1 A Pragmatic Account offers that children reject known objects because of the social pragmatics against such a construal. That is, children might reason as follows: “If you had meant for me to give you the car you would have said ‘car’. But you said ‘dax’, so you must mean the novel toy.” (Diesendruck & Markson, 2001).

A fourth proposal is that children might simply “map novelty to novelty”. Proponents of the Novel-Name Nameless-Category principle (N3C) contend that children do not reject known objects from consideration (Mervis, Golinkoff, & Bertrand, 1994). Instead, children are simply positively motivated to map novel labels to novel objects (Golinkoff et al., 1992, Mervis and Bertrand, 1994).

Whichever particular principle guides children's behavior, two of the important open questions concerning all such word-learning constraints are: (1) do they help get word-learning off the ground by constraining children's earliest word meanings? and (2) what are the computations that subserve constraints throughout word-learning, from early through mature use? Data from young infants who are just beginning to utilize a constraint may contribute to answering both of these questions.

I present a variation on the preferential looking paradigm (Allopenna et al., 1998, Fernald et al., 1998, Hirsh-Pasek and Golinkoff, 1996) which will allow the status of the constraint on novel labels to be assessed at younger ages than previously reported. At approximately 17.5 months, children are able to successfully map novel labels to novel objects in a pointing game (Mervis & Bertrand, 1994). But developmental research has found that looking-time is often a more sensitive measure than reaching or pointing. Therefore, instead of asking young babies to “hand me the dax”, infants' change in looking to a novel object after hearing a novel label (e.g. “dax”) is taken as a measure of success at mapping the novel label to the novel object.

Section snippets

Participants

The participants were 38 full-term infants (16 male) from predominantly English-speaking families (mean age 16 months, 0 days; range 14-7 to 17-25). Fifteen additional infants were tested but not included in the final sample due to: fussiness (2), side-bias (6), parental interference (4), and equipment failure (3). One subject was removed from the final analyses for performing more than two standard deviations below the mean, leaving 37 infants in the final sample.

Stimuli

Visual stimuli consisted of

Experiment 2

Beyond the above internal controls, there are several possible explanations for the 14-month-olds' negative difference scores on Novel Label trials (“dax”). Experiment 2 sought to control for all incidental aspects of the experiment which might have caused the 14-month-olds' surprising performance.

Although an overall preference for the known object (car) cannot explain why 14-month-olds significantly increased looking to the car above baseline preference in Experiment 1, perhaps these infants

Discussion

The present study demonstrates two theoretically important results. First, Experiment 1 shows the development of a word-learning strategy within a single paradigm. This work contributes to the debate on whether constraints in general help get word-learning off the ground. Experiment 1 presents evidence from a looking-time procedure that 17 months is the age of onset of the strategy studied here. This converges with findings from a pointing procedure (Mervis & Bertrand, 1994). These results

Acknowledgements

This work was supported by a National Institutes of Health (NIH) grant HD-38338-01 to Susan Carey, and by a National Science Foundation (NSF) predoctoral fellowship awarded to the author. The author wishes to thank Susan Carey without whom this work would not have been possible. The author also thanks four anonymous reviewers and Lisa Feigenson, Gary Marcus, and Brian McElree for their helpful comments, Alex Todorov for statistical advice, and Matthias Boltz, Bill Carson, Erik Cheries, Delia

References (31)

  • A Fernald et al.

    Rapid gains in speed of verbal processing by infants in the 2nd year

    Psychological Science

    (1998)
  • V.C Gathercole

    Contrast: a semantic constraint?

    Journal of Child Language

    (1989)
  • R.M Golinkoff et al.

    Young children and adults use lexical principles to learn new nouns

    Journal of Developmental Psychology

    (1992)
  • R.M Golinkoff et al.

    Early object labels: the case for a developmental lexical principles framework

    Journal of Child Language

    (1994)
  • Halberda, J. (2002). Is this a dax I see before me? A computational analysis of mutual exclusivity, in...
  • Cited by (237)

    • Interaction of phonological biases and frequency in learning a probabilistic language pattern

      2022, Cognition
      Citation Excerpt :

      We are unaware of any experimental studies specifically comparing the learnability of homophony-creating and non-homophony-creating phonological rules in young children. Given that the well-known mutual exclusivity bias (e.g., Merriman, Bowman, & MacWhinney, 1989), which could be related to homophony avoidance (Yin & White, 2018), has been displayed by children as young as 17 months old (Halberda, 2003), it is an interesting question for future work to confirm whether this homophony avoidance effect also occurs when children learn phonological rules. Our second conclusion is that adult learners have a preference to avoid phonological alternations, and this preference could influence which morphological constructions are used when multiple constructions are possible.

    View all citing articles on Scopus
    View full text