Adult semantic memory has been traditionally conceptualized as a relatively static memory system that consists of knowledge about the world, concepts, and symbols. Considerable work in the past few decades has challenged this static view of semantic memory, and instead proposed a more fluid and flexible system that is sensitive to context, task demands, and perceptual and sensorimotor information from the environment. This paper (1) reviews traditional and modern computational models of semantic memory, within the umbrella of network (free association-based), feature (property generation norms-based), and distributional semantic (natural language corpora-based) models, (2) discusses the contribution of these models to important debates in the literature regarding knowledge representation (localist vs. distributed representations) and learning (error-free/Hebbian learning vs. error-driven/predictive learning), and (3) evaluates how modern computational models (neural network, retrievalbased, and topic models) are revisiting the traditional "static" conceptualization of semantic memory and tackling important challenges in semantic modeling such as addressing temporal, contextual, and attentional influences, as well as incorporating grounding and compositionality into semantic representations. The review also identifies new challenges regarding the abundance and availability of data, the generalization of semantic models to other languages, and the role of social interaction and collaboration in language learning and development. The concluding section advocates the need for integrating representational accounts of semantic memory with process-based accounts of cognitive behavior, as well as the need for explicit comparisons of computational models to human baselines in semantic tasks to adequately assess their psychological plausibility as models of human semantic memory.

We examined 3 different network models of representing semantic knowledge (5,018-word directed and undirected step distance networks, and an association-correlation network) to predict lexical priming effects. In Experiment 1, participants made semantic relatedness judgments for word pairs with varying path lengths. Response latencies for judgments followed a quadratic relationship with network path lengths, replicating and extending a recent pattern reported by Kenett, Levi, Anaki, and Faust (2017) for an 800-word association-correlation network in Hebrew. In Experiment 2, participants identified target words in a progressive demasking task, immediately following a briefly presented prime (120 ms). Response latencies to identify the target showed a linear trend for all network path lengths. Importantly, there were statistically significant differences between relatively distant words in the step distance networks, for example, path lengths 4 and beyond, suggesting that association networks can indeed capture distant functional semantic relationships. Additional comparisons with 2 distributional models (LSA and word2vec) suggested that distributional models also successfully predicted response latencies, although there appear to be fundamental differences in the types of semantic relationships captured by the different models.

Some of the earliest work on understanding how concepts are organized in memory used a network‐based approach, where words or concepts are represented as nodes, and relationships between words are represented by links between nodes. Over the past two decades, advances in network science and graph theoretical methods have led to the development of computational semantic networks. This review provides a modern perspective on how computational semantic networks have proven to be useful tools to investigate the structure of semantic memory as well as search and retrieval processes within semantic memory, to ultimately model performance in a wide variety of cognitive tasks. Regarding representation, the review focuses on the distinctions and similarities between network‐based (based on behavioral norms) approaches and more recent distributional (based on natural language corpora) semantic models, and the potential overlap between the two approaches. Capturing the type of relation between concepts appears to be particularly important in this modeling endeavor. Regarding processes, the review focuses on random walk models and the degree to which retrieval processes demand attention in pursuit of given task goals, which dovetails with the type of relation retrieved during tasks. Ultimately, this review provides a critical assessment of how the network perspective can be reconciled with distributional and machine‐learning‐based perspectives to meaning representation, and describes how cognitive network science provides a useful conceptual toolkit to probe both the structure and retrieval processes within semantic memory.

The present experiments investigated the influence of combined phonological and semantic information on lexical retrieval, metacognitive retrieval states, and selection in an immediate multiple-choice task. Younger and older adults attempted to retrieve words (e.g., abdicate) from low-frequency word definitions. Retrieval was preceded by primes that were “both” semantically and phonologically related (e.g., abandon), phonologically related (e.g., abdomen), semantically related (e.g., resign), or unrelated (e.g., pink). Younger and older adults benefited from phonological primes in retrieval, and also showed reduced, but reliable, facilitation from “both” primes. Younger and older adults also indicated that they were likely to “know” the answer more often after any related primes compared with unrelated primes. Because there was no facilitation in actual retrieval after semantic primes, this reflects a false “knowing” response. After each retrieval attempt, participants were given the correct answer along with the 4 primes in a multiple-choice test. Both younger and older adults were likely to false alarm to the “both” and semantic alternatives. When instructed that the prime was not the answer, younger adults decreased their false alarms, but not the older adults. With masked, briefly presented primes, younger adults mimicked the false alarms shown by older adults, suggesting that the high false alarm rates in older adults reflect an inability to discriminate the source of activation. The present experiments provide strong evidence for age-invariant phonological facilitation, and also suggest that overlapping semantic information moderates the facilitatory effect of phonological information on retrieval, and also produces age-related differences on an immediate multiple-choice task.

Considerable work during the past two decades has focused on modeling the structure of semantic memory, although the performance of these models in complex and unconstrained semantic tasks remains relatively understudied. We introduce a two-player cooperative word game, Connector (based on the boardgame Codenames), and investigate whether similarity metrics derived from two large databases of human free association norms, the University of South Florida norms and the Small World of Words norms, and two distributional semantic models based on large language corpora (word2vec and GloVe) predict performance in this game. Participant dyads were presented with 20-item word boards with word pairs of varying relatedness. The speaker received a word pair from the board (e.g., exam-algebra) and generated a one-word semantic clue (e.g., math), which was used by the guesser to identify the word pair on the board across three attempts. Response times to generate the clue, as well as accuracy and latencies for the guessed word pair, were strongly predicted by the cosine similarity between word pairs and clues in random walk-based associative models, and to a lesser degree by the distributional models, suggesting that conceptual representations activated during free association were better able to capture search and retrieval processes in the game. Further, the speaker adjusted subsequent clues based on the first attempt by the guesser, who in turn benefited from the adjustment in clues, suggesting a cooperative influence in the game that was effectively captured by both associative and distributional models. These results indicate that both associative and distributional models can capture relatively unconstrained search processes in a cooperative game setting, and Connector is particularly suited to examine communication and semantic search processes.