| This study investigates whether the linguistic structures obtained in experimentally-elicited data can be explained within a usage-based framework. An underlying assumption of this work is that frequency of use drives conventionalization of structure and representation in memory, where usage factors further impact storage properties of linguistic units. It therefore follows that as speakers access their grammars in an elicitation task, their experimental responses should reflect the storage properties of their linguistic mental representations.; In order to establish speakers' most probable linguistic representations, twelve words were analyzed for two types of structures in the Switchboard Corpus of Recorded Telephone Conversations: (1) the most frequent collocation for a word, defined here as the word's most frequently-occurring, fully-specified lexical string (e.g., big bucks for bucks), and; (2) the most frequent construction for a word, defined here as the most frequently-occurring, morpho-syntactic pattern for a word that has some morphological material expressed alongside open categories (e.g., [ NUMERAL bucks]).; Fifty-four native speakers of English were then provided with the twelve words investigated in the corpus data and asked to provide an example utterance for each. Elicited responses were coded as either an instance of the corpus collocation, corpus construction, or some other corpus pattern. The distribution of elicited data across these three categories was analyzed using a chi-square goodness-of-fit test.; Two results of the study support the hypothesis that elicited data reflect the storage properties of frequent conversational structures. First, it was found that the most frequent collocations in corpora were rarely reproduced in elicitation. This is consistent with a usage-based model in which collocations, having separate lexical storage, are susceptible to token frequency effects, leading to autonomous storage but hampering access in elicitation. Second, the results show that a word's most frequent construction in corpora tends to be reproduced in elicitation. This, too, is consistent with a usage-based grammar that contains localized abstractions whose representations typically maintain lexical connections, facilitating access of the construction in elicitation. In light of these findings, elicited data promise additional support for usage-based theoretical constructs. |
