A Computational Cognitive Model of Novel Word Generalization

Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 1795–1804, Lisbon, Portugal, 17-21 September 2015. c⃝2015 Association for Computational Linguistics. A Computational Cognitive Model of Novel Word Generalization Aida Nematzadeh, Erin Grant, and Suzanne Stevenson Department of Computer Science University of Toronto {aida,eringrant,suzanne}@cs.toronto.edu Abstract A key challenge in vocabulary acquisition is learning which of the many possible meanings is appropriate for a word. The word generalization problem refers to how children associate a word such as dog with a meaning at the appropriate category level in a taxonomy of objects, such as Dalma- tians, dogs, or animals. We present the first computational study of word general- ization integrated within a word-learning model. The model simulates child and adult patterns of word generalization in a word-learning task. These patterns arise due to the interaction of type and token frequencies in the input data, an influence often observed in people’s generalization of linguistic categories. 1 Introduction Learning word meanings is a challenging early step in child language acquisition. Imagine a child hears the word dax for the first time while observing a white rabbit jumping around – dax might mean WHITE RABBIT, RABBIT, ANIMAL, CUTE, LOOK, etc. (Quine, 1960). How does the child learn the correct meaning of a word from a large pool of potential meanings? A possi- ble explanation is that children infer a word’s meaning by identifying the commonalities across the situations in which the word occurs (Pinker, 1989). One mechanism for achieving this is cross- situational learning (e.g., Siskind, 1996; Frank et al., 2007; Fazly et al., 2010; Kachergis et al., 2012). Recent word learning experiments con- firm that both adults and children infer the cor- rect word-meaning mappings by keeping track of cross-situational statistics across individually am- biguous learning trials (Yu and Smith, 2007; Smith and Yu, 2008; Yurovsky et al., 2014). Although cross-situational learning is a general mechanism for narrowing down the meaning of a word, it does not explain how children overcome an interesting challenge in word learning: deter- mining the correct level of a hierarchical taxon- omy that a word refers to. For example, children learn that the word dog refers to all kinds of dogs, and not to a specific breed, such as Dalmatians, or to a more general category, such as animals – even though some of these choices (e.g., animals) are compatible with all the cross-situational evi- dence available for dog (because all dogs are also animals). We use the term “word generalization” to refer to this problem of associating a word with the meaning at an appropriate category level, given some sample of experiences with the word. Previous research has argued that children use a specific bias or constraint – the basic-level as- sumption – to focus their word generalizations appropriately (Markman, 1991; Golinkoff et al., 1994). According to this bias, children prefer to associate a word to a set of objects that form a basic-level category, such as dogs or trucks, and that share a …

Content truncated to 500 words