George Kachergis: George Kachergis Doctoral Dissertation

Language acquisition is a ubiquitous, challenging problem involving fundamental cognitive abilities of attention, learning, and memory. From infants to adult travelers, language learners are faced with figuring out which words refer to which referents from situations that contain many words and referents. By remembering the words and referents (e.g., objects) that co-occur most frequently over time, learners may acquire correct noun meanings—an idea known as cross-situational learning (Gleitman, 1990). However, given that in most situations there are many possible word-object pairings but time and attention are limited, learners likely use strategies to restrict the number of meanings they consider. Building on associative learning research on attention (e.g., Kruschke, 2001), my thesis models noun acquisition using a simple learning mechanism with biases. I present and model a number of word-learning experiments that investigate both theorized learning constraints (e.g., mutual exclusivity: the assumption that each word maps to one object, and vice-versa) and factors such as word frequency, contextual diversity, and active learning, all of which occur naturally. Besides being of interest to psychologists and linguists, the word-object mapping problem is related to machine translation (language-language mappings) in computer science, and more generally to inference from contingency tables (for an overview, see Agresti, 1992). Models of word learning beg application in education, where they may be used to predict optimal training sequences and even identify individual differences in terms of underlying psychological abilities. In adult cross-situational learning studies (e.g., Yu & Smith, 2007; Kachergis, Yu, & Shiffrin, 2009a, 2009b; Suanda & Namy, 2012), participants are asked to learn the referent of novel words by watching a series of training trials. On each trial learners see an array of unfamiliar objects (e.g., four sculptures) and hear pseudowords (e.g., stigson, bosa). The referent of each pseudoword is ambiguous on a given trial, because although each word refers to a single onscreen object, the intended referent is not indicated. In a typical learning scenario, participants attempt to learn 18 word-object pairings from 27 trials, with four words and four objects given per trial. In this design, each word-referent pair is presented six times over the five-minute training period. Learning a correct word-object pairing requires accumulating word-object co-occurrences in some fashion. Cross-situational learning has also been observed in infants (Smith & Yu, 2008; Vouloumanos & Werker, 2009) and toddlers (Akhtar & Montague, 1999). Research shows that even simple cross-situational learning models can learn large lexicons in a reasonable amount of time, both in simulated corpora (Blythe, Smith, & Smith, 2010) and from a corpus of parent-child interactions (Frank, Goodman, & Tenenbaum, 2009). How might people accomplish cross-situational learning? One approach views word learning as a problem of induction with an enormous hypothesis space, and proposes a number of language-specific constraints to restrict the space (Markman, 1992). In this view, infants generate hypotheses that are consistent with this set of constraints and principles. For example, the global principle (or bias) of mutual exclusivity (ME) assumes that every object has only one name (Markman & Wachtel, 1988). At a lower level, Clark (1987) proposed the fill-the-lexical gap bias, which causes children to want to find a name for an object with no known name,