What have we learned from 15 years of research on cross-situational word learning? A focused review

Abstract

In 2007 and 2008, Yu and Smith published their seminal studies on cross-situational word learning (CSWL) in adults and infants, showing that word-object-mappings can be acquired from distributed statistics despite in-the-moment uncertainty. Since then, the CSWL paradigm has been used extensively to better understand (statistical) word learning in different language learners and under different learning conditions. The goal of this review is to provide an entry-level overview of findings and themes that have emerged in 15 years of research on CSWL across three topic areas (mechanisms of CSWL, CSWL across different learner and task characteristics) and to highlight the questions that remain to be answered.

Keywords: cross-situational statistical learning; cross-situational word learning; language acquisition; referential ambiguity; review; word learning.

Figure 1

Examples of the two most common variants of the cross-situational word learning paradigms. In Variant 1 (A), participants hear two novel words and see two novel objects on each trial in a passive learning phase (e.g., Yu and Smith, 2007). There is a separate testing phase where participants only hear one word and have to select the correct referent on each trial (similar to Variant 2). In Variant 2 (B), participants only hear one word and see two objects in a two-alternative forced choice trial (e.g., Roembke and McMurray, 2016). There is no separate testing phase. In both variants, it is unclear which of the two objects maps onto the blue object in Trial 1. In Trial 3, however, the word JEPLIN is presented with the target object and a different competitor than in Trial 1. At this point, a participant could know that JEPLIN maps onto the blue object. Many features can be manipulated in this paradigm (e.g., modality of the word, number of presented competitors; number of presented words).

Figure 2

Overview of the two most influential accounts of cross-situational word learning (CSWL). In a gradual associative account (A), associations are built with all possible referents (Trial 1). As the correct word-object-mapping co-occurs more frequently than any other word-competitor-pairing, their association becomes strongest (Trial 3). In contrast, in a propose-but-verify (hypothesis-testing) account, people form one hypothesis about each word-object-mapping (B). In this example, the wrong hypothesis is “proposed” on Trial 1, which is then verified and revised on subsequent trials (Trials 2–3). Images are retrieved from the MultiPic database (Duñabeitia et al., 2018).