Sound Similarity Relations in the Mental Lexicon

The standard definition of neighborhood density defines two words to be neighbors if they differ by one and only one segment (Luce & Pisoni, 1998). This definition assumes that the length of the shared part is irrelevant to sound similarity. However, confusability of non-linguistic sound sequences (Fallon Coble & Robinson, 1992; Kidd & Watson, 1992) as well as judgments of sound similarity between spoken pseudowords (Kapatsinski, in press, b) depend on the proportion of total duration of the word or sound sequence that is mismatched, and not on the absolute duration of the mismatch. To bring the definition of neighborhood into alignment with these results, this paper defines words to be neighbors if they share at least two thirds of their total duration, measured in segments. This simple change reduces the proportion of words with no neighbors to all words in the lexicon from over 58 percent to just 7 percent, increasing the applicability of the Neighborhood Activation Model (Luce & Pisoni, 1998). Lexical decision, naming, and familiarity judgment data indicate that speaker/hearers are sensitive to the more distant neighbors brought in by the new neighborhood definition. The large-scale properties of the network are discussed and future directions are indicated.