Even young children know what typical dogs and fish look like – and they apply that knowledge when they hear new words, reports a team from the Princeton Baby Lab, where researchers study how babies learn to see, talk and understand the world.
In a series of experiments with children 3 to 5 years old, the researchers found that when children are learning new nouns, they use what they know about these objects – how typical or unusual they are for their categories (such as fish, dog, bird or flower) – to help them figure out what these words mean.
This type of sophisticated reasoning was thought to only develop later.
The researchers’ work appears in the current issue of the Journal of Child Language.
“What we’re showing is that meaning matters!” said Adele Goldberg, professor of psychology at Princeton University and the senior author on the paper. “Children take the meaning of the objects that they are seeing into account when they learn new words.”
The researchers coined this tactic the “blowfish effect.”
If children see a blowfish (or a greyhound or an unusual tropical flower) and learn a new word to go with it, they will assume it refers to that specific type of object and not the broader category of fish (or dogs or flowers).
“This study helps to solve one of the big puzzles in language development,” said Lauren Emberson, assistant professor of psychology and first author on the paper.
Many years of studies have shown that when children learn new words, they assume that word means something fairly general: If taught a new word for a goldfish, children assume that it means “fish.”
“But children can learn these more specific terms,” like blowfish and greyhound, said Emberson, who is also one of the directors of the Princeton Baby Lab.
“How do they start to do that? We are showing that they use the objects themselves to do this. If they see an unusual fish and their parent calls it something, they will learn that it refers to that specific fish.”
Using a custom designed iPad program, the researchers taught children four new words: fep, zak, lat and galt.
Two of these terms were used for typical objects and two for unusual objects. The objects came from four categories that children are familiar with: fish, birds, dogs and flowers.
In each trial, a child saw either one or three examples at the top of the screen, identified by a new word: “This is a fep,” or “These are three feps.” With the press of an arrow, the child got 12 more images below: two that matched the examples, two that shared the category, and eight unrelated creatures. The experimenter then asked, “Can you find the feps?”
The researchers were curious whether children would decide a “fep” only meant the specific creature in the examples—a robin, for example, or a Dalmatian – or if the term was applied more generally to all birds or dogs.
Each child could choose as many images as she wanted to, at her own pace, before proceeding to the next trial by pressing the arrow again.
The order of the four categories – fish, birds, dogs and flowers – was randomized across participants.
The researchers ran the same experiment with college students; the only differences were that the undergraduates were told that this was an experiment intended for young children, and they were allowed to hold the iPads themselves.
The team found that both children and adults processed the new words in the same way. When any of them saw an unusual dog labeled a “fep,” they were more likely to interpret it narrowly—as meaning that type of dog, not “dogs” more generally.
These findings run counter to the idea that children will always assume that new words should be interpreted as general terms.
In addition, the researchers found that the more “typical” an example looks, the more likely children are to assume it’s a general term, unless it is repeated: A “zak” was likely to be interpreted as “fish” if it labeled a single salmon – a fairly typical-looking fish – but it was interpreted as “salmon” if illustrated by three salmon.
But if “zak” labeled even a single odd-looking fish—like a blowfish—the children were more likely to decide that the word meant “blowfish” than “fish.”
“The finding helps shed light on the mysteries and intricacies of language development,” Emberson said.
Many psychological experiments involve word learning tasks, in which participants—both adults and children – are taught names for objects either explicitly—for instance, through the use of social cues and ostensive feedback (Gauthier & Tarr, 1997; Horst & Samuelson, 2008) – or implicitly across several encounters (e.g., Axelsson & Horst, 2014; Yu & Smith, 2007).
In studies such as these, novelty is critical for ensuring that researchers are testing learning that has occurred as a function of the experimental manipulation and not merely tapping into knowledge acquired prior to the experiment (Ard & Beverly, 2004; Bornstein & Mash, 2010).
In addition, novelty is often critical for categorization studies in which participants must learn to extrapolate information from one category exemplar and generalize or apply that information to new exemplars (e.g., Homa, Hout, Milliken, & Milliken, 2011; J. D. Smith & Minda, 2002).
Previous research demonstrates that novelty (or the lack thereof) can have a profound effect on subsequent learning. For example, after toddlers explore novel objects for 1–2 min, they are significantly less likely to associate novel names with those objects than with still-novel objects (Horst, Samuelson, Kucker, & McMurray, 2011; see also Kucker & Samuelson, 2012).
Thus, even brief prior experience with stimuli can change subsequent behavior on critical test trials. Likewise, gradual, prior experience with stimuli can also influence subsequent behavior, such as looking times during the learning phase of an object-examination categorization task (Bornstein & Mash, 2010).
It is therefore ideal that visual stimuli not have been seen before, in order to ensure that any inferences made regarding learning were not actually due to participants’ exposure to the items prior to the experiment.
For many experimental designs it is important that objects also be easy to distinguish from each other (e.g., Twomey, Ranson, & Horst, 2014; Yu & Smith, 2007); however, for other designs it can be useful to have objects that are somewhat similar (e.g., Homa et al., 2011; Hout & Goldinger, 2015).
There are existing databases of known, familiar, real-world objects (e.g., Brady, Konkle, Alvarez, & Oliva, 2008; Dan-Glauser & Scherer, 2011; Hout, Goldinger, & Brady, 2014; Konkle, Brady, Alvarez, & Oliva, 2010; Migo, Montaldi, & Mayes, 2013) and human faces (e.g., Ebner, Riediger, & Lindenberger, 2010; Matheson & McMullen, 2011), as well as databases of related items uniquely optimal for use in categorization studies (e.g., Gauthier & Tarr, 1997; Marchewka, Żurawski, Jednoróg, & Grabowska, 2014). However, researchers investigating memory for objects and object names critically need a database of novel objects for use in such experiments, where it is critical that participants have no a priori knowledge of the stimuli and that the objects not already be associated with specific names. The NOUN database is such a collection of novel object images.
Why use the NOUN Database?
The NOUN Database offers several advantages for researchers requiring images of unusual objects. First, the images in the NOUN Database depict multipart, multicolored, real three-dimensional (3-D) objects, as opposed to simple geometric shape configurations (e.g., L. B. Smith & Yu, 2008; Wu, Gopnik, Richardson, & Kirkham, 2011) or seemingly animate objects (e.g., Gauthier & Tarr, 1997; Mather, Schafer, & Houston-Price, 2011; Rakison & Poulin-Dubois, 2002).
As such, these stimuli are ideal for researchers who need images of naturalistic, complex novel objects to present against images of real 3-D objects that are already familiar to participants (e.g., familiar distractors or known competitors). Indeed, complex novel objects are often presented against known objects—for example, in language research (e.g., Axelsson & Horst, 2014; Giezen, Escudero, & Baker, in press; Mather & Plunkett, 2009; Warren & Duff, 2014; Zosh, Brinster, & Halberda, 2013). In such cases, it is vital that the novel stimuli be just as “credible” as the familiar, known objects, which requires similar shading, colors, textures, and complexity. The stimuli in the NOUN Database have such properties because they are images of real objects (e.g., they are not “impossible” objects that might be created from a software package).
Second, researchers frequently choose their stimuli on the basis of their own intuitive judgments of novelty and similarity (Migo et al., 2013). This practice is especially prevalent in developmental psychology, where researchers make assumptions about objects that are unlikely to be familiar to young children without prior confirmation (but see Horst & Samuelson, 2008, for a quick confirmation method).
This can be problematic, because children may be implicitly learning about the object categories although they have not yet heard the category names. In experiments requiring images of novel objects, this problem can be avoided by using the preexisting NOUN Database; the novelty and similarity ratings we present can inform researchers’ decisions on which stimuli to use, depending on their research questions. Specifically, these ratings can be used to ensure that a subset of stimuli are equally novel and not already associated with a specific name, as well as to vary the novelty or similarity across items.
Relatedly, using an existing database facilitates comparison across experiments, which can be especially helpful when different experiments address unique but related research questions, or when one wants to compare related effects. For example, young children generalize names for nonsolid substances to other substances made of the same material (Imai & Gentner, 1997; Soja, Carey, & Spelke, 1992), but only when the substances share both the same material and color—if the stimuli only share the same material, this effect disappears (Samuelson & Horst, 2007).
Similarly, adults are faster to repeat nonwords from high-density lexical neighborhoods than those from low-density neighborhoods (Vitevitch & Luce, 1998), but this effect disappears with different stimuli—for example, when the leading and trailing silences in the audio files are removed to equate stimulus duration (Lipinski & Gupta, 2005). The use of existing stimuli is also consistent with the recent push in the psychology research community to share resources and to facilitate replicability (for a lengthier discussion, see Asendorpf et al., 2013).
Finally, using an existing set of stimuli saves time and reduces research expenses because data collection on the substantive experiment can begin quickly, without the need for additional preliminary experiments that ensure that the stimuli are in fact novel and unlikely to already be associated with a particular name.
The present Experiment 1 is effectively a preliminary experiment conducted on behalf of researchers who wish to use the NOUN Database.
This is valuable, because obtaining and selecting experimental stimuli is often a highly time-consuming phase of the research process (Dan-Glauser & Scherer, 2011; see also Umla-Runge, Zimmer, Fu, & Wang, 2012). Even using 3-D rendering software can take time to learn and can be expensive.
Moreover, using an existing database utilized by multiple researchers may expedite the ethical approval process for new studies.
Taken together, these time-saving aspects make the NOUN Database particularly useful to students who must conduct research projects quickly with a strict deadline, as well as early-career researchers who may especially benefit from a reduced time to publication (see A. K. Smith et al., 2011, for a related argument).
Although one advantage of the NOUN Database is its ability to save valuable time and money, researchers using the database may choose to conduct their own preliminary experiments to ensure that the stimuli that will be used in their main experiment are in fact novel to their participant pool. Researchers can also use the images in the NOUN Database as a supplement to their own stimuli, which offers even greater experimental flexibility. Note that a second advantage of the NOUN Database is its size:
It includes 64 items, which is many times more than is often required for most studies using images of novel objects (e.g., two novel objects, Bion, Borovsky, & Fernald, 2013; three novel objects, Rost & McMurray, 2009; one novel object, Werker, Cohen, Lloyd, Casasola, & Stager, 1998).
More information: Lauren L. EMBERSON et al. The blowfish effect: children and adults use atypical exemplars to infer more narrow categories during word learning, Journal of Child Language (2019). DOI: 10.1017/S0305000919000266
Provided by Princeton University