Frequency, Neighborhood Density, and Phonological Similarity Effects in Picture Naming: An Artificial Lexicon Study

Frequency, Neighborhood Density, and Phonological Similarity Effects in Picture Naming: An Artificial Lexicon Study Austin F. Frank (afrank@bcs.rochester.edu), Michael K. Tanenhaus, Richard N. Aslin, Anne Pier Salverda Department of Brain and Cognitive Sciences, University of Rochester Abstract ognized more slowly than words from sparse neighborhoods (Vitevitch & Luce, 1998). The effects of neighborhood density on production are in- consistent. Vitevitch and colleagues have shown a facilita- tory effect of neighborhood density on the production of En- glish words (Vitevitch, 2002; Vitevitch, Armbruster, & Chu, 2004; Vitevitch & Sommers, 2003). The same researchers, however, have found the opposite effect in Spanish. Span- ish speakers produced words from dense neighborhoods more slowly than words from sparse neighborhoods (Vitevitch & Stamer, 2006). Additionally, the results of specific experi- ments have sometimes been difficult to replicate. Vitevitch & Sommers, 2003 show fewer tip-of-the-tongue states for words from dense neighborhoods, but fail to find any effect of neigh- borhood density in picture naming. This stands in contrast to Vitevitch, 2002 and Vitevitch et al., 2004, where naming la- tencies were shorter for words from dense neighborhoods. In comprehension, the inhibitory influence of neigh- borhood density on recognition has been attributed to phonologically-based competition during lexical access (Mc- clelland & Elman, 1986). A similar process could be involved in production. On the other hand, an interactive activation model of production has been used to demonstrate that shared phonology among neighbors should lead to faster responses and fewer errors for words from dense neighborhoods (Dell & Gordon, 2003). At this point, the results from production studies cannot fully explain the role that phonological simi- larity plays in lexical access for single word production. One possible confound in studies of neighborhood effects is that neighborhood density is correlated with other distribu- tional properties in the lexicon (e.g. phonotactic probabilities; see Vitevitch, Luce, Pisoni, & Auer, 1999). When control- ling for correlated factors and trying to sample the lexicon at different levels of other lexical properties (neighborhood den- sity, lexical frequency), it can be difficult to gather an ideally- balanced set of stimuli for studying the interactions among lexical properties in language processing. Furthermore, while current methods of estimating frequency and neighborhood density are useful, these estimates fail to account for impor- tant factors like individual experience with language. Cit- ing these reasons and others, Magnuson et al. conducted a series of experiments based on an artificial lexicon (2003). Over the course of several days, subjects learned to recognize novel shapes by their “names”. Drawing the names of the shapes from a specially designed artificial lexicon allowed the researchers to achieve precise control over theoretically in- teresting factors such as lexical frequency and neighborhood Subjects learned to name novel shapes using words from an artificial lexicon. Use of an artificial lexicon allowed for tight control over properties of the words in the lexicon, includ- ing lexical frequency, neighborhood density, and phonologi- cal similarity. After training, subjects’ naming latencies and error rates displayed some of the same performance patterns as would be expected in natural language studies of picture naming. These encouraging results argue for the consideration of artificial lexicons as research tools in studies of production, and offer exciting possibilities for future work in language pro- cessing. Keywords: picture naming, word production, word learning, artificial lexicon, lexical representation, phonological neigh- borhoods, frequency effects In both comprehension and production, certain properties of words are known to affect the way those words are pro- cessed. Some of these properties include lexical frequency, phonological similarity to recently used words, and phono- logical neighborhood density. Experimental evidence shows that frequency of use and phonological structure play an important role in how a word is processed, in both comprehension and production. Lis- teners recognize high-frequency words more quickly than low-frequency words (Marslen-Wilson, 1987). Speakers pro- duce high-frequency words more proficiently, as evidenced by shorter naming latencies and fewer errors (Jescheniak & Levelt, 1994; Levelt et al., 1991; Dell, 1986). While experience with a word consistently facilitates its processing, the impact of a word’s phonology on its use is more varied. Exposure to an auditorily-presented phonologi- cally related word during lexical access leads to faster naming performance (Meyer & van der Meulen, 2000; Levelt et al., 1991); but producing a phonologically-related word prior to a naming a picture slows speakers down (Wheeldon, 2003). Phonologically related words need not even be explicitly pre- sented as primes in order to affect lexical access. Words that are phonologically similar are referred to as neighbors 1 (Luce & Pisoni, 1998). Words with many neighbors are said to come from dense neighborhoods, and words with few neighbors from sparse neighborhoods. The density of a word’s phonological neighborhood affects its processing, even when its neighbors are not explicitly in- volved in the language task. In comprehension, for exam- ple, words from dense neighborhoods are systematically rec- 1 In this study and most of the studies presented below, phonolog- ical similarity is heuristically defined as a pair of words that differ by no more than one added, altered, or deleted phoneme. See Luce & Pisoni, 1998 for more on this “shortcut rule”.