Developmental Changes in Cross-Situational Word Learning: The Inverse Effect of Initial Accuracy

Non-linear development in statistical learning of visual orthographic regularities

Statistical learning is a core ability for individuals in extracting and integrating regularities and patterns from linguistic input. Yet, the developmental trajectory of visual statistical learning has not been fully examined in the orthographic learning domain. Employing an artificial orthographic learning task, we manipulated three levels of positional consistency of radicals, i.e., high (100%), moderate (80%), and low (60%), embedded in pseudocharacters to investigate visual statistical learning across a wide age range between 4-12-year-olds and adults. The non-linear power-function models indicated that the rates of improvement in acquiring varying positional consistencies increased with age, particularly for high and moderate levels. Specifically, we observed a significant enhancement in statistical learning abilities between the ages of 4-5 years and 5-6 years, followed by a stabilization of performance after 8-9 years. Our findings support the age-dependent perspective that individuals' visual statistical learning ability improves significantly in early childhood and then decelerates its improvement progressively until adulthood.

What have we learned from 15 years of research on cross-situational word learning? A focused review

In 2007 and 2008, Yu and Smith published their seminal studies on cross-situational word learning (CSWL) in adults and infants, showing that word-object-mappings can be acquired from distributed statistics despite in-the-moment uncertainty. Since then, the CSWL paradigm has been used extensively to better understand (statistical) word learning in different language learners and under different learning conditions. The goal of this review is to provide an entry-level overview of findings and themes that have emerged in 15 years of research on CSWL across three topic areas (mechanisms of CSWL, CSWL across different learner and task characteristics) and to highlight the questions that remain to be answered.

Improvements of Statistical Learning Skills Allow Older Children to Go Beyond Single-Hypothesis Testing When Learning Words

Children learn words in ambiguous situations, where multiple objects can potentially be referents for a new word. Yet, researchers debate whether children maintain a single word-object hypothesis - and revise it if falsified by later information - or whether children establish a network of word-object associations whose relative strengths are modulated with experience. To address this issue, we presented 4- to 12-year-old children with sets of mutual exclusivity (fast-mapping) trials: offering them with obvious initial hypotheses (that the novel object is the referent for the novel word). We observe that children aged six years and above, despite showing a novelty bias and retaining this novel word - novel object association, also formed an association between the novel word and the name-known object, thereby suggesting that older children attend to more than one word-object association, in a manner similar to associative learning. We discuss our findings in the context of competing theoretical accounts related to word learning.

Here come the nouns: Czech two-year-olds use verb number endings to predict sentence subjects

Two-year-old children can use gender or number agreement to predict upcoming words in phrases or sentences. However, most findings showed prediction from freestanding grammatical words, such as articles or copulas. While this shows knowledge of agreement relations, it might be limited to a narrow set of grammatical words. We examined the possibility that children at this age can use grammatical number agreement independently of specific closed-class words, testing whether they predict nouns from bound morphemes on lexical verbs. If this were the case, the emerging grammatical knowledge is unlikely to be lexically specific. Our first experiment replicated existing findings using number-marked copula, while the second experiment marked number on endings of four different verbs. Two-year-old children watched pairs of pictures showing single or multiple items while listening to sentences whose sentence-final subject referred to one of the two pictures. The grammatical Czech sentences contained a copula (Experiment 1: where is/are in the picture car/s?) or one of four number-marked lexical verbs (Experiment 2: Here jump/s the frog/s in the picture). Children in both experiments anticipated the subject from the verb or copula form. Children thus used number agreement predictively in the complex Czech copula system and lexical verbs marked by endings. This suggests that children understand grammatical number independently of specific grammatical words and supports the view that early knowledge of grammar is not lexically specific.

Multiple components of statistical word learning are resource dependent: Evidence from a dual-task learning paradigm

It is increasingly understood that people may learn new word/object mappings in part via a form of statistical learning in which they track co-occurrences between words and objects across situations (cross-situational learning). Multiple learning processes contribute to this, thought to reflect the simultaneous influence of real-time hypothesis testing and graduate learning. It is unclear how these processes interact, and if any require explicit cognitive resources. To manipulate the availability of working memory resources for explicit processing, participants completed a dual-task paradigm in which a cross-situational word-learning task was interleaved with a short-term memory task. We then used trial-by-trial analyses to estimate how different learning processes that play out simultaneously are impacted by resource availability. Critically, we found that the effect of hypothesis testing and gradual learning effects showed a small reduction under limited resources, and that the effect of memory load was not fully mediated by these processes. This suggests that neither is purely explicit, and there may be additional resource-dependent processes at play. Consistent with a hybrid account, these findings suggest that these two aspects of learning may reflect different aspects of a single system gated by attention, rather than competing learning systems.

Prediction error boosts retention of novel words in adults but not in children

How do we update our linguistic knowledge? In seven experiments, we asked whether error-driven learning can explain under what circumstances adults and children are more likely to store and retain a new word meaning. Participants were exposed to novel object labels in the context of more or less constraining sentences or visual contexts. Both two-to-four-year-olds (M_age = 38 months) and adults were strongly affected by expectations based on sentence constraint when choosing the referent of a new label. In addition, adults formed stronger memory traces for novel words that violated a stronger prior expectation. However, preschoolers' memory was unaffected by the strength of their prior expectations. We conclude that the encoding of new word-object associations in memory is affected by prediction error in adults, but not in preschoolers.

The temporal structure of naming events differentially affects children's and adults' cross-situational word learning

To acquire novel words, learners often need to integrate information about word meanings across ambiguous learning events distributed in time. How does the temporal structure of those word learning events affect what learners encode? How do the effects of temporal structure differ in children and adults? In the current experiments, we asked how 4- to 7-year-old children's (N = 110) and adults' (N = 90) performance on a cross-situational word learning task is influenced by the temporal distribution of learning events. We tested participants in three training conditions, manipulating the number of trials that separated naming events for specific objects. In the Unstructured condition, the temporal distribution was varied; in the Massed condition, naming events occurred with few interleaved trials; and in the Interleaved condition, naming events occurred with many interleaved trials. Adults showed substantially larger benefits from the Massed condition than children, whereas children were equally successful at learning in the Massed and Interleaved conditions. These results provide evidence that adults differ from children in how they exploit temporal structure during cross-situational word learning.

Syntactic Cues Help Disambiguate Objects Referred to With Count Nouns: Illustration With Malay Children

Children employ multiple cues to identify the referent of a novel word. Novel words are often embedded in sentences and children have been shown to use syntactic cues to differentiate between types of words (adjective vs. nouns) and between types of nouns (count vs. mass nouns). In this study, we show that children learning Malay (N = 67), a numeral classifier language, can use syntactic cues to perform even finer-grained disambiguation-between count nouns. The manipulation of congruence between lexical and syntactic cues reveals a clear developmental trajectory: while 5-year-olds use predominantly lexical cues, older children increasingly rely on syntactic cues, such that by 7 years of age, they disambiguate between objects referred to with count nouns using syntactic rather than lexical cues.

Cross-Situational Statistical Learning of New Words Despite Bilateral Hippocampal Damage and Severe Amnesia

Word learning requires learners to bind together arbitrarily-related phonological, visual, and conceptual information. Prior work suggests that this binding can be robustly achieved via incidental cross-situational statistical exposure to words and referents. When cross-situational statistical learning (CSSL) is tested in the laboratory, there is no information on any given trial to identify the referent of a novel word. However, by tracking which objects co-occur with each word across trials, learners may acquire mappings through statistical association. While CSSL behavior is well-characterized, its brain correlates are not. The arbitrary nature of CSSL mappings suggests hippocampal involvement, but the incremental, statistical nature of the learning raises the possibility of neocortical or procedural learning systems. Prior studies have shown that neurological patients with hippocampal pathology have word-learning impairments, but this has not been tested in a statistical learning paradigm. Here, we used a neuropsychological approach to test whether patients with bilateral hippocampal pathology (N = 3) could learn new words in a CSSL paradigm. In the task, patients and healthy comparison participants completed a CSSL word-learning task in which they acquired eight word/object mappings. During each trial of the CSSL task, participants saw two objects on a computer display, heard one novel word, and selected the most likely referent. Across trials, words were 100% likely to co-occur with their referent, but only 14.3% likely with non-referents. Two of three amnesic patients learned the associations between objects and word forms, although performance was impaired relative to healthy comparison participants. Our findings show that the hippocampus is not strictly necessary for CSSL for words, although it may facilitate such learning. This is consistent with a hybrid account of CSSL supported by implicit and explicit memory systems, and may have translational applications for remediation of (word-) learning deficits in neurological populations with hippocampal pathology.

Mark my words: High frequency marker words impact early stages of language learning

High frequency words have been suggested to benefit both speech segmentation and grammatical categorization of the words around them. Despite utilizing similar information, these tasks are usually investigated separately in studies examining learning. We determined whether including high frequency words in continuous speech could support categorization when words are being segmented for the first time. We familiarized learners with continuous artificial speech comprising repetitions of target words, which were preceded by high-frequency marker words. Crucially, marker words distinguished targets into 2 distributionally defined categories. We measured learning with segmentation and categorization tests and compared performance against a control group that heard the artificial speech without these marker words (i.e., just the targets, with no cues for categorization). Participants segmented the target words from speech in both conditions, but critically when the marker words were present, they influenced acquisition of word-referent mappings in a subsequent transfer task, with participants demonstrating better early learning for mappings that were consistent (rather than inconsistent) with the distributional categories. We propose that high-frequency words may assist early grammatical categorization, while speech segmentation is still being learned. (PsycINFO Database Record (c) 2019 APA, all rights reserved).

Symbolic flexibility during unsupervised word learning in children and adults

Considerable debate in language acquisition concerns whether word learning is driven by domain-general (symbolically flexible) or domain-specific learning mechanisms. Prior work has shown that very young children can map objects to either words or nonlinguistic sounds, but by 20 months of age this ability narrows to only words. This suggests that although symbolically flexible mechanisms are operative early, they become more specified over development. However, such research has been conducted only with young children in ostensive teaching contexts. Thus, we investigated symbolic flexibility at later ages in more referentially ambiguous learning situations. In Experiment 1, 47 6- to 8-year-olds acquired eight symbol-object mappings in a cross-situational word learning paradigm where multiple mappings are learned based only on co-occurrence. In the word condition participants learned with novel pseudowords, whereas in the sound condition participants learned with nonlinguistic sounds (e.g., beeps). Children acquired the mappings, but performance did not differ across conditions, suggesting broad symbolic flexibility. In Experiment 2, 41 adults learned 16 mappings in a comparable design. They learned with ease in both conditions but showed a significant advantage for words. Thus, symbolic flexibility decreases with age, potentially due to repeated experiences with linguistic materials. Moreover, trial-by-trial analyses of the microstructure of both children's and adults' performance did not reveal any substantial differences due to condition, consistent with the hypothesis that learning mechanisms are generally employed similarly with both words and nonlinguistic sounds.