Language Use in Normal Speakers and its Disorders 1 . The Architecture of Normal Spoken Language Use

The normal language user's production and understanding of speech involves the highly skilled coordination of myriad processes. When a speaker conceives of some communicative intention, he will select and order information whose expression may realize that intention. He will also formulate that information, i. e., give it linguistic shape. This includes retrieving the appropriate words from memory and assigning them their proper grammatical roles and syntactic positions. The speaker will further compute a phonetic specification for the developing utterance, and use it to guide the articulatory execution that produces overt speech. The addressee will normally try to reconstruct the speaker's communicative intentions. She will perform an acoustic-phonetic analysis on the continuous speech signal in order to segment it into recognizable words and phrases. She will retrieve the syntactic properties and meanings of successive words and parse the string into meaningful phrases and sentences. And she will interpret this information in terms of the context of interaction, the purpose of the exchange, the presuppositions about the speaker's intentions, etc. This complex system operates largely unawares. A speaker concentrates his attention on what is to be said, not on how it is to be done. The rate of fluent speech, some 2 to 3 words per second, is too high for a speaker to ponder over each and every word or syntactic construction. And there is no way for a speaker to consciously prepare some fifteen speech sounds per second. Similarly, the listener normally attends to the content of what is said, not to the phonetic shape of words or the syntactic complexities of the utterance. Like in all other skills, the lower level processes are automatic, They don't use central resources; they come for free. In the following I will outline the architecture of this skill. This involves the dissection of the speech processing system into component subsystems. It also involves a characterization of the representations computed by the processing components as well as the manner in which these representations are computed. And it requires a specification of how the processing components cooperate in producing their joint product. These matters will be discussed in reference to Figure 1.1. It is a 'blueprint' of the language user, depicting the main processing components and their connections. We will proceed anticlockwise through the system, beginning at the highest conceptual level at which the speaker generates information to be expressed. We will follow the flow of information down to articulation, the generation of overt speech. We will then turn to the perception and parsing of speech, back to the conceptual interpretative level. After this grand tour some remarks will be made about the modes of cooperation between the various components, in particular about incrementality and autonomy of processing.