<?xml version="1.0" standalone="yes"?>
<Paper uid="C90-3012">
  <Title>Language Without A Central Pushdown Stack</Title>
  <Section position="2" start_page="0" end_page="64" type="intro">
    <SectionTitle>
1. Introduction
</SectionTitle>
    <Paragraph position="0"> The long-range goal of our research project is to develop and implement in computer simulation a unified, psycholinguistically realistic model of language behaviour--specifically, language production, language comprehension, and language development. Our model is constructed in the framework of relational network linguistics (also known as cognitive-stratificational grammar (Copeland &amp; Davis 1980)), while also incorporating features of spreading activation (Collins &amp; Loftus 1975), competition (Bates &amp; MacWhinney 1987), and other models. In this paradigm, linguistic information is represented in the form of an asynchronous, massively parallel network (in the sense of Schnelle et al. 1988) whose nature can be seen as intermediate between that of mainstream connectionist networks and traditional generative grammars, incorporating aspects of both formalisms. The basics of relational network grammars will be set forth in SS3.</Paragraph>
    <Paragraph position="1"> Our specific goal here is to describe our attempts to simulate the psychological facts about the production of syntactic embedding in a non-ad hoc way within the above framework. We will show that adequately accounting for human performance requires an examination of at least two other representational levels (or strata) in addition to the syntax, namely the event level and a level we call the lexotacticso In the process we also hope to demonstrate the unified nature of our theory across linguistic levels.</Paragraph>
    <Paragraph position="2"> Particular consideration will be given to the cognitive data structures and processing mechanisms required. Our claim in this regard is that temporary memory in the form of a centralized data store, whether it be a pushdown stack, a queue, a collection of stacks, or whatever, is an inappropriate component for modelling human linguistic processing. We will argue instead for a large collection of very simple, localized processing units which possess small amounts of storage by virtue of the possible states which they may be in; in short, a network of finite-state devices. These processors will be general, in the sense that they are useful for cognitive processes outside the domain of language.</Paragraph>
  </Section>
class="xml-element"></Paper>