Relational labels can improve relational retrieval

Relational labels can improve relational retrieval Anja Jamrozik (a.jamrozik@u.northwestern.edu) Dedre Gentner (gentner@northwestern.edu) Department of Psychology, Northwestern University 2029 Sheridan Road, Evanston, IL 60208 USA Abstract Even though relational retrieval is typically rare, it is more likely for experts in a domain. For example, when solving challenging science problems, experts often retrieve problems that share common relational structure (e.g., Clement, 1988). Likewise, the likelihood of relational retrieval is better for students with greater mathematical expertise than for novices (Novick, 1988), and mathematical expertise can predict the likelihood of transferring a solution strategy to analogous math problems (Novick & Holyoak, What contributes to experts’ improvement in relational retrieval? Two factors that might be involved are having experienced many opportunities to compare examples and acquiring a technical vocabulary. There is abundant evidence that comparison of examples can improve the likelihood of relational retrieval. When learners compare two instances of the same relational structure, the process of alignment renders their common structure more salient. This process of schema abstraction increases the likelihood of retaining this common relational structure and transferring it to other instances (e.g., Gentner Loewenstein, Thompson, & Forbus, 2009; Gentner & Markman, 1997; Gick & Holyoak, 1983; Markman & Gentner, 1993; Reeves & Weisberg, 1994; Ross & Kennedy, 1990). In the current research, we focus on the second factor and ask how acquiring a vocabulary may impact relational retrieval 1 . As someone develops expertise in a domain, they may acquire terms that name common relations or relational patterns in that domain. We ask whether the use of such relational language can improve relational retrieval. Retrieval that is based on common relational structure, such as an underlying principle or pattern, is useful but typically rare. Based on evidence that comparison-derived schema abstraction can improve relational retrieval, we asked whether the use of relational labels can also promote abstraction and improve relational retrieval. Using a cued-recall paradigm, we varied the presence of relational labels at encoding and test. As compared to a no-label baseline condition, relational retrieval improved when relational labels were given at encoding and at test and also when relational labels were given only at encoding. The findings demonstrate that one way to improve relational retrieval is through the use of labels that name relational structure. Keywords: relational retrieval; relational language; inert knowledge Introduction When encountering a new example or problem, we sometimes retrieve examples from memory that share relational structure with the current example. This can be very useful as it allows us to transfer existing knowledge to the new example. For instance, if a social psychology student learns about the classic findings that a person’s attitude can become resistant against very persuasive arguments after the person has argued against weak versions of such arguments (e.g., McGuire, 1961; McGuire & Papageorgis, 1961), then this might remind the student of how someone can become immune to a disease after being exposed to a weakened form of that disease. Based on this connection, they may be able to draw some conclusions about the new situation, such as why the attitude becomes resistant to change, or why the initial arguments against the attitude have to be weak. As useful as it can be, relational retrieval—retrieval based only on common relational structure—is typically rare (e.g., Gentner, Rattermann, & Forbus, 1993; Gick & Holyoak, 1980, 1983; Holyoak & Koh, 1987; Ross, 1987, 1989). Instead, memory retrieval is likely to be based either on overall similarity or on surface commonalities, such as matching entities (e.g., Brooks, Norman, & Allen, 1991; Gentner et al., 1993; Holyoak & Koh, 1987; Ross, 1987, 1989). This is an instance of the inert knowledge problem (Whitehead, 1929) – that people are often unable to retrieve knowledge and apply it to new situations even when that information has been stored in memory (e.g., Barnett & Ceci, 2002; Bransford, 1979). Relational Language In the current research, we ask whether an abstraction process like the one that operates during comparison also applies when relational language is used. Specifically, we ask whether using known relational terms like reciprocity or inoculation to label examples promotes the abstraction of their relational structure and leads to improved relational retrieval. The idea is that using a relational term to label a situation can promote the abstraction of relational structure The value of relational language and comparison might in some cases be related because using the same term for different examples can invite comparison of these examples (e.g., Gentner, 2003, 2010; Gentner & Medina, 1998; Gentner & Namy, 1999). However, our focus here will be on the individual effects of relational language and comparison.