Previous studies demonstrated that statistical properties of adult generated free associates predict the order of early noun learning. We investigate an explanation for this phenomenon that we call the associative structure of language: early word learning may be driven in part by contextual diversity in the learning environment, with contextual diversity in caregiver speech correlating with the cue-target structure in adult free association norms. To test this, we examined the co-occurrence of words in caregiver speech from the CHILDES database and found that a word’s contextual diversity—the number of unique word types a word co-occurs with in caregiver speech—predicted the order of early word learning and was highly correlated with the number of unique associative cues for a given target word in adult free association norms. The associative structure of language was further supported by an analysis of the longitudinal development of early semantic networks (from 16 to 30 months) using contextual co-occurrence. This analysis supported two growth processes: The lure of the associates, in which the earliest learned words have more connections with known words, and preferential acquisition, in which the earliest learned words are the most contextually diverse in the learning environment. We further discuss the impact of word class (nouns, verbs, etc.) on these results.

Since their inception, distributional models of semantics have been criticized as inadequate cognitive theories of human semantic learning and representation. A principal challenge is that the representations derived by distributional models are purely symbolic and are not grounded in perception and action; this challenge has led many to favor feature-based models of semantic representation. We argue that the amount of perceptual and other semantic information that can be learned from purely distributional statistics has been underappreciated. We compare the representations of three feature-based and nine distributional models using a semantic clustering task. Several distributional models demonstrated semantic clustering comparable with clustering-based on feature-based representations. Furthermore, when trained on child-directed speech, the same distributional models perform as well as sensorimotor-based feature representations of children's lexical semantic knowledge. These results suggest that, to a large extent, information relevant for extracting semantic categories is redundantly coded in perceptual and linguistic experience. Detailed analyses of the semantic clusters of the feature-based and distributional models also reveal that the models make use of complementary cues to semantic organization from the two data streams. Rather than conceptualizing feature-based and distributional models as competing theories, we argue that future focus should be on understanding the cognitive mechanisms humans use to integrate the two sources.

Neural approaches to automated essay scoring have recently shown state-of-theart performance. The automated essay scoring task typically involves a broad notion of writing quality that encompasses content, grammar, organization, and conventions. This differs from the short answer content scoring task, which focuses on content accuracy. The inputs to neural essay scoring models -ngrams and embeddings -are arguably well-suited to evaluate content in short answer scoring tasks. We investigate how several basic neural approaches similar to those used for automated essay scoring perform on short answer scoring. We show that neural architectures can outperform a strong nonneural baseline, but performance and optimal parameter settings vary across the more diverse types of prompts typical of short answer scoring.

Both increased and decreased temperature differences can be used to indicate reactive hyperemia or a Stage I pressure ulcer, but a tissue integrity problem may still exist despite the absence of a temperature difference.