Better explanations of lexical and semantic cognition using networks derived from continued rather than single-word associations

Abstract

In this article, we describe the most extensive set of word associations collected to date. The database contains over 12,000 cue words for which more than 70,000 participants generated three responses in a multipleresponse free association task. The goal of this study was (1) to create a semantic network that covers a large part of the human lexicon, (2) to investigate the implications of a multiple-response procedure by deriving a weighted directed network, and (3) to show how measures of centrality and relatedness derived from this network predict both lexical access in a lexical decision task and semantic relatedness in similarity judgment tasks. First, our results show that the multiple-response procedure results in a more heterogeneous set of responses, which lead to better predictions of lexical access and semantic relatedness than do singleresponse procedures. Second, the directed nature of the network leads to a decomposition of centrality that primarily depends on the number of incoming links or in-degree of each node, rather than its set size or number of outgoing links. Both studies indicate that adequate representation formats and sufficiently rich data derived from word associations represent a valuable type of information in both lexical and semantic processing.Keywords Word associations . Semantic network . Lexical decision . Semantic relatedness . Lexical centrality Associative knowledge is a central component in many accounts of recall, recognition, and semantic representations in word processing. There are multiple ways to tap into this knowledge, but word associations are considered to be the most direct route for gaining insight into our semantic knowledge (Nelson, McEvoy, & Schreiber, 2004;Mollin, 2009) and human thought in general (Deese, 1965). The type of information produced by word associations is capable of expressing any kind of semantic relationship between words. Because of this flexibility, networks are considered the natural representation of word associations, where nodes correspond to lexicalized concepts and links indicate semantic or lexical relationships between two nodes. These networks correspond to an idealized localist representation of our mental lexical network. The properties derived from such a network have been instrumental in three different research traditions, which will be described below. These traditions have focused on (1) direct association strength, (2) second-order strength and distributional similarity, and (3) network topology and centrality measures.The first tradition has used word associations to calculate a measure of associative strength and was inspired by a behaviorist view of language in terms of stimulus-response patterns. This notion of associative strength plays an important role in studies that have focused on inhibition and facilitation in list learning (e.g., Roediger & Neely, 1982), studies on episodic memory (e.g., Nelson et al., 2004), and studies that have tried to distinguish semantic and associative priming (for a recent over...

From words-as-mappings to words-as-cues: the role of language in semantic knowledge

“…Consider the words "snow", "idea", "cheese", and "jealousy"-words with straightforward translations into Dutch. Figure 3 shows patterns of word associations for four words in English and their closest Dutch translations (De Deyne, Navarro, & Storms, 2013). Association patterns for "snow" and "idea," have very similar associates.…”

Quantifying the structure of free association networks across the life span.

“…One of the advantages of network analysis is that it allows researchers to study phenomena at scales that range from individual words (i.e., nodes) to the entire lexicon (i.e., the network). Network analyses have been widely used to study the large scale structures of associative networks (De Deyne, Navarro, & Storms, 2013;De Deyne & Storms, 2008a;Steyvers & Tenenbaum, 2005), the formation of categories in toddlers (Hills, Maouene, Maouene, Sheya, & Smith, 2009a), and the trajectories of network development during early life (Bilson, Yoshida, Tran, Woods, & Hills, 2015;Hills et al, 2009b).…”

Verb similarity: Comparing corpus and psycholinguistic data

“…Besides, Nelson et al (2000) and De Deyne et al (2013) suggest that these nonprimary responses contain useful information. In addition to these advantages, from the point of view of similarity calculation between verb senses, having more than one response per stimulus and participant helps in reducing data sparsity (De Deyne et al 2013). The total amount of responses collected was 11,617.…”

“…We chose to collect several responses to characterize each stimulus, as not limiting the responses to the first that came to mind has shown higher correlation with human similarity ratings (Steyvers et al 2004) and it has proven to perform better in lexical access tasks (Chumbley and Balota 1984). Besides, Nelson et al (2000) and De Deyne et al (2013) suggest that these nonprimary responses contain useful information. In addition to these advantages, from the point of view of similarity calculation between verb senses, having more than one response per stimulus and participant helps in reducing data sparsity (De Deyne et al 2013).…”

“…As for psycholinguistic data, we had five different formalizations for similarity: lemmas, hypernyms, SUMO, TCO and supersenses. Moreover, we experimented with three versions of word associations: 1) similarities calculated using only the first response of each participant; 2) similarities obtained using the first three responses at most, as in De Deyne et al (2013); and 3) all the responses obtained. We observed that taking into account the first three and all responses yielded higher correlations with corpus data than taking only the first response, in line with the findings of Nelson et al (2000) and De Deyne et al (2013).…”