Frequency Effects in the Production of Noun Phrases

In this study we explore some of the properties involved in the production of multi-word English utterances. In order to produce such utterances, the speech production system must retrieve several lexical units and organize them in a well-formed utterance, which follows the rules of the language being spoken. According to current models, the linguistic mechanisms involved in this process can be divided into two main stages: grammatical encoding and phonological encoding (see, e.g., Bock and Levelt 1994; Levelt, 1989). The first stage involves the selection of the various words that belong to the NP along with their grammatical properties (the so-called grammatical encoding). During the second stage, the phonological information of the lexical nodes that have been selected is retrieved, and the phonological adjustments needed to produce those words in the proper phonological context are computed (the so-called phonological encoding). After these stages the processes of articulation programming and execution start (see, e.g. Caramazza 1997; Dell 1986; Levelt, Roelofs, and Meyer, 1999). In the present study we will focus on the characteristics of phonological encoding. More specifically, we will concentrate on evaluating the amount of phonological information that is concurrently activated in the system during speech production. The study of speech error corpora has been a major source of information for our understanding of the speech production system (including the properties we have briefly described. See, e.g., Garrett 1975, 1992). The kinds of errors that are observed and their distributional properties constrain possible models of speech production. Consider for example the error that arises when a speaker misplaces two words in an utterance, by exchanging their intended positions (e.g., "she donated a library to the book", instead of " she donated a book to the library", from Garrett 1975). Very often, these errors involve members of the same grammatical class (e.g. two nouns or two verbs) and they are interpreted as arising at a level in which lexical selection takes place. Moreover, the lexical items that interact in such errors occupy relatively