Journal of memory and language最新文献

Why do children learn some words before others? A large body of behavioral research has identified properties of the language environment that facilitate word learning, emphasizing the importance of particularly informative language contexts that build on children’s prior knowledge. However, these findings have not informed research that uses distributional properties of words to predict vocabulary composition. In the current work, we introduce a predictor of word learning that emphasizes the role of prior knowledge. We investigate item-based variability in vocabulary development using lexical properties of distributional statistics derived from a large corpus of child-directed speech. Unlike previous analyses, we predicted word trajectories cross-sectionally across child age, shedding light on trends in vocabulary development that may not have been evident at a single time point. We also show that regardless of a word’s grammatical class, the best distributional predictor of whether a child knows a word is the number of other known words with which that word tends to co-occur.

The age at which a person acquires knowledge of an item is a strong predictor of item retrieval, hereon defined as the Age of Acquisition (AoA) effect. This effect is such that early-acquired words are processed more quickly and accurately than late-acquired items. One account to explain this effect is the integrated account, where the AoA effect occurs in the early processes of lexical retrieval and hence should increase in tasks necessitating greater semantic processing. Importantly, this account has been applied to lexical processing, but not, to date, memory tasks. The current study aimed to assess whether the integrated account could explain memory tasks, using compound words, which differ from monomorphemic words regarding ease of mapping and semantic processes. Four-hundred-and-eighty participants were split into four groups of 120 participants for each of four experiments. Participants were required to recall unspaced and spaced compound words (Experiments 1 and 2, respectively) or make a recognition decision for unspaced and spaced compound words (Experiments 3 and 4, respectively). This approach allowed us to establish how semantic processing was involved in recalling and recognising the items. We found that (AoA) was related to all tasks such that irrespective of space, early-acquired compound words were recalled more accurately than late-acquired compound words in free recall. In recognition memory, late-acquired compound words were recognised more accurately than early-acquired compound words. However, the slope for the AoA was semantic processing influenced free recall to a greater extent than the recognition memory, with the AoA effect being larger in free recall than recognition memory. In addition, the AoA effect for the compound word was larger in spaced compound words than unspaced compound words. This demonstrates that the AoA effect in memory has multiple sources.

Previous work has demonstrated that words are hyperarticulated on dimensions of speech that differentiate them from a minimal pair competitor. This phenomenon has been termed contrastive hyperarticulation (CH). We present a dynamic neural field (DNF) model of voice onset time (VOT) planning that derives CH from an inhibitory influence of the minimal pair competitor during planning. We test some predictions of the model with a novel experiment investigating CH of voiceless stop consonant VOT in pseudowords. The results demonstrate a CH effect in pseudowords, consistent with a basis for the effect in the real-time planning and production of speech. The scope and magnitude of CH in pseudowords was reduced compared to CH in real words, consistent with a role for interactive activation between lexical and phonological levels of planning. We discuss the potential of our model to unify an apparently disparate set of phenomena, from CH to phonological neighborhood effects to phonetic trace effects in speech errors.

The orthographic depth theory assumes that reading “deep” orthographies relies on lexical semantics more than “shallow” orthographies. Although Japanese kanji is a representative “deep” case, some scholars argue that kanji reading does not particularly recruit more lexical semantics than kana (the system of syllabic writing used for Japanese consisting of two forms). To reconcile this inconsistency, we ran a Monte Carlo simulation and found that orthographic neighbors in kanji had higher semantic similarities than those in kana. We further conducted a semantic space analysis (‘Word2Vec’) and showed that there was significant radical-level orthographic-semantic consistency in kanji characters. Furthermore, we demonstrated that this consistency had a positive effect on language performance in models (in terms of next-character prediction) and humans (in terms of semantic plausibility judgment). These findings suggest that radicals in kanji may help children to efficiently learn to use the vast number of characters present in Japanese.

Word learning is characterized by a bias for mapping meanings at the “basic” level (‘dog’), as opposed to a subordinate level (‘poodle’; Markman, 1986, 1990; Clark, 1987; Waxman et al., 1991, 1997). The fact that learners nevertheless acquire subordinate nouns has been attributed to properties of the referential world across multiple labelling events (e.g., Xu & Tanenbaum, 2007b; Spencer et al., 2011). Here we propose that the acquisition of subordinate-level meanings requires pragmatic reasoning that allows learners to take informative relevant alternatives into consideration. In support of this hypothesis, in a series of experiments we find that adult learners exploit information about semantic alternatives to generalize word meanings beyond the basic level. In Experiment 1, the introduction of a labelled alternative at the subordinate level eliminated the basic-level bias. In Experiment 2, this effect was found to be specific to labelled but not unlabeled alternatives. In Experiment 3, the availability of alternatives affected conjectures about subordinate-level word meanings even when these alternatives were presented well after the initial moment of ostensive labeling. Lastly, Experiment 4 replicated the semantic contrast effect using exclusively novel language input, highlighting the general communicative nature of these inferences. We conclude that the acquisition of subordinate nouns relies on pragmatic inferences about the informativity of labels as intentional linguistic-pragmatic acts, as opposed to simple word-to-world co-occurrences.

People sometimes predict upcoming words during language comprehension, but debate remains on when and to what extent such predictions indeed occur. The rational adaptation hypothesis holds that predictions develop with expected utility: people predict more strongly when predictions are frequently confirmed (low prediction error) rather than disconfirmed. However, supporting evidence is mixed thus far and has only involved measuring responses to supposedly predicted nouns, not to preceding articles that may also be predicted. The current, large-sample (N = 200) ERP study on written discourse comprehension in Dutch therefore employs the well-known ‘pre-nominal prediction effect’: enhanced N400-like ERPs for articles that are unexpected given a likely upcoming noun’s gender (i.e., the neuter gender article ‘het’ when people expect the common gender noun phrase ‘de krant’, the newspaper) compared to expected articles. We investigated whether the pre-nominal prediction effect is larger when most of the presented stories contain predictable article-noun combinations (75% predictable, 25% unpredictable) compared to when most stories contain unpredictable combinations (25% predictable, 75% unpredictable). Our results show the pre-nominal prediction effect in both contexts, with little evidence to suggest that this effect depended on the percentage of predictable combinations. Moreover, the little evidence suggesting such a dependence was primarily observed for unexpected, neuter-gender articles (‘het’), which is inconsistent with the rational adaptation hypothesis. In line with recent demonstrations (Nieuwland, 2021a,b), our results suggest that linguistic prediction is less ‘rational’ or Bayes optimal than is often suggested.

In studies of children’s categorization, researchers have typically studied how encoding characteristics of exemplars contribute to children’s generalization. However, it is unclear whether children’s internal cognitive processes alone, independent of new information, may also influence their generalization. Thus, we examined the role that one cognitive process, forgetting, plays in shaping children’s category representations by conducting three experiments. In the first two experiments, participants (N_Exp1 = 37, M_age = 4.02 years; N_Exp2 = 32, M_age = 4.48 years) saw a novel object labeled by the experimenter and then saw five new objects with between one and five features changed from the learned exemplar. The experimenter asked whether each object was a member of the same category as the exemplar; children saw the five new objects either immediately or after a 5-minute delay. Children endorsed category membership at higher rates at immediate test than at delayed test, suggesting that children’s category representations became narrower over time. In Experiment 3, we investigated forgetting as a key mechanism underlying the narrowing found in Experiments 1 and 2. We showed participants (N_Exp3 = 34, M_age = 4.20 years) the same exemplars used in Experiments 1 and 2; then, either immediately or after a 5-minute delay, we showed children seven individual object features and asked if each one had been part of the exemplar. Children’s accuracy was lower after the delay, showing that they did indeed forget individual features. Taken together, these results show that forgetting plays an important role in changing children’s newly-learned categories over time.

Readers progressed through a sentence in the Maze task (Forster et al., 2009), deciding at each word between a sensical and a non-sensical continuation. Contexts presented before these sentences manipulated whether words were linguistically focused and whether they were given or new (Experiment 1); focused targets were read more slowly even when they were given, and new targets were read slowly in general. This both replicated earlier results in which slowdowns were found in the reading of focus (Benatar and Clifton, 2014; Birch and Rayner, 1997; Lowder and Gordon, 2015), and demonstrated that focus slowdowns are not reducible to newness. To clarify earlier results in which speed-ups were found on focused words (Birch and Rayner, 2010; Morris and Folk, 1998), contexts manipulated whether contrastive alternatives to focused words were presented with a focus particle (Experiment 2) or in a cleft construction (Experiment 3). Focused targets were read less slowly when a contrastive alternative was present in the context. This effect of contrastive alternatives cannot be reduced to simple semantic associate priming: Contexts also manipulated whether a semantically associated expression was present independently of the presence of a contrastive alternative (Experiment 4). Readers slowed down less when an alternative was present in the context, even when this alternative was not semantically associated to the target. These results indicate that the processing of focus depends on contrastive alternatives, in their interaction with newness, semantic association, and focus construction.

A major strength of computational cognitive models is their capacity to accurately predict empirical data. However, challenges in understanding how complex models work and the risk of overfitting have often been addressed by trading off predictive accuracy with model simplification. Here, we introduce state-of-the-art model analysis techniques to show how a large number of parameters in a cognitive model can be reduced into a smaller set that is simpler to understand and can be used to make more constrained predictions with. As a test case, we created different versions of the Connectionist Dual-Process model (CDP) of reading aloud whose parameters were optimized on seven different databases. The results showed that CDP was not overfit and could predict a large amount of variance across those databases. Indeed, the quantitative performance of CDP was higher than that of previous models in this area. Moreover, sloppy parameter analysis, a mathematical technique used to quantify the effects of different parameters on model performance, revealed that many of the parameters in CDP have very little effect on its performance. This shows that the dynamics of CDP are much simpler than its relatively large number of parameters might suggest. Overall, our study shows that cognitive models with large numbers of parameters do not necessarily overfit the empirical data and that understanding the behavior of complex models is more tractable using appropriate mathematical tools. The same techniques could be applied to many different complex cognitive models whenever appropriate datasets for model optimization exist.