Word associations: Norms for 1,424 Dutch words in a continuous task

SIMON DE DEYNE 0 GERr Simms 0 0 University of Leuven , Leuven, Belgium This study describes the collection of a large set of word association norms. In a continuous word association task, norms for 1,424 Dutch words were gathered. For each cue, three association responses were obtained per participant. In total, an average of 268 responses were collected for each cue. We investigated the relationship with similar procedures, such as discrete association tasks and exemplar generation tasks. The results show that the use of a continuous task allows the study of weaker associations in comparison with a discrete task. The effects of the continuous tasks were investigated for set size and the availability characteristics of the responses, measured through word frequency, age of acquisition, and imageability. Finally, we compared our findings to those of a semantically constrained version of the association task in which participants generated responses within the domain of a semantic category. Results of this comparison are discussed. The Appendix cited in this article is available at www.psychonomic.org/archive. - The study and use of word associations has been widespread in many fields of psychology since its conception as an academic discipline (Boring, 1950; Deese, 1965). In the late nineteenth century, pioneers such as Galton and Wundt started systematic investigations with psychometric, classificatory studies (Boring, 1950). Later, research on associations was used in clinical studies of human pathologies and intelligence, to be adopted by behaviorists and finally reinstated by modem cognitive psychologists who study language and memory (Cramer, 1968). Language and memory researchers are interested in word associations because, from a theoretical point of view, they are necessary at the center of understanding the organization of word knowledge. Although the exact interpretation of word associations is certainly not a closed chapter in cognitive psychology or psychology in general, it is generally agreed that word associations reflect our lexical knowledge acquired through world experience by means of words and the relationships between them (Nelson, McEvoy, & Schreiber, 2004). Furthermore, these structures capture important aspects of meaning or the semantic representation of words (de Groot, 1988; Nelson et al., 2004). Besides the representation of meaning, the generation of word associations results in probabilities of responses that tell something about the retrieval mechanisms that underlie these responses (de Groot, 1988; Nelson, McEvoy, & Dennis, 2000) and memory performance in other tasks, such as cued recall experiments (Nelson et al., 2000; Nelson, McKinney, Gee, & Janczura, 1998; Steyvers, Shifirin, & Nelson, 2005). Word associations can also be used as a gold standard, built upon a psychological foundation. This is the case when free association norms are compared with other norms in order to evaluate the validity of different types of procedures for indexing proximity. For instance, Steyvers et al. (2005) compared proximities based on a word association model with proximities derived from a model based on text collocations in the spirit of latent semantic analysis (LSA; Landauer & Dumais, 1997) in cued recall and semantic similarity judgments. Semantic representations based on word associations might be used to evaluate vector models relying on text collocation data such as LSA (Landauer & Dumais, 1997) and HAL (Burgess & Lund, 1997), where it is often unclear how semantic similarity between words comes about. Finally, in empirical studies of word processing, associative norms are often used to avoid confounding word variables such as printed word frequency (Nelson et al., 2004). Clearly, the widespread use of association data in all these settings creates a considerable attention to the norms used in many previously reported findings. However, the largest sets of association norms in English (Nelson et al., 2004) and Dutch (de Groot, 1988) both employ a discrete version of the free association task. For both association norms, participants provide only one association for each cue. Although there are good reasons to restrict the collection of word associations to one response per cue, there are cases in which the continuous version of the word association task provides a meaningful way to gather these responses. One of the advantages of the continuous approach is that it allows a larger variability in associationsfor instance, in the case of a very strong first association (e.g., blood and red). The collection of multiple responses leads to better estimates of the probabilities of weak associations. A continuous task, in which participants generate more than one associate, is of particular interest to researchers who try to disentangle priming studies with mediated, associative, or semantic priming. Moreover, the representation of words in vector models based on multiple associations is much richer than vectors based on discrete associations that are the stock data in most studies. Finally, differences in early and late responses in word association tasks might give additional insight in the conceptual structure and the processes that drive the generation of word associates (Barsalou, Santos, Simmons, & Wilson, in press). In previous large-scale association studies, such a procedure has been shunned since the generation of multiple responses, in comparison with a single, discrete response, evokes chaining and retrieval inhibition (McEvoy & Nelson, 1982). Both these processes can introduce a bias in the resulting norms. Chaining occurs when participants respond with associations based on a previous response rather than responding to the cue. To some extent, this effect can be reduced through instructions. Retrieval inhibition occurs when participants repeatedly sample responses for one cue. Research on retrieval inhibition in memory has shown that when people have to produce subsequent responses to the same cue, additional responses become hindered by a retrieval blockage (e.g., Raaijmakers & Shiffrin, 1981; Roediger, 1973). Again, some measures can limit the effect of response inhibition. First, in contrast to tasks in which participants give exhaustive associations (until they cannot gen -eratnymospe),thakcnbrsitedolya limited number of responses. Retrieval inhibition is expected to be of minimal influence whenever a limited number of associations per participant are gathered, in comparison with studies in which th continuous association task was time delimited. Second, lexpected is tthatthe effect of retrieval inhibition will be reduced if the stimuli are more diverse. In this study, we present a Dutch data set of associations for 1,424 words using a continuous association task. Despite a long tradition of studying word associations, recent Dutch word association norms are hard to find. A recent extensive study for English words was conducted b (...truncated)