[Paper Review] Having Your Cake and Eating It Too: Autonomy and Interaction in a Model of Sentence Processing

Kurt Eiselt, Kavi Mahesh+1 more|ArXiv.org|Aug 31, 1994

Natural Language Processing Techniques18 references6 citations

TL;DR

This paper proposes a hybrid model of sentence processing that unifies a single processor with modular knowledge sources, reconciling autonomy and integration in language comprehension. Using the COMPERE computational model, it demonstrates that this approach successfully explains both modular and integrated processing phenomena, offering a more human-like framework for natural language understanding.

ABSTRACT

Is the human language understander a collection of modular processes operating with relative autonomy, or is it a single integrated process? This ongoing debate has polarized the language processing community, with two fundamentally different types of model posited, and with each camp concluding that the other is wrong. One camp puts forth a model with separate processors and distinct knowledge sources to explain one body of data, and the other proposes a model with a single processor and a homogeneous, monolithic knowledge source to explain the other body of data. In this paper we argue that a hybrid approach which combines a unified processor with separate knowledge sources provides an explanation of both bodies of data, and we demonstrate the feasibility of this approach with the computational model called COMPERE. We believe that this approach brings the language processing community significantly closer to offering human-like language processing systems.

Motivation & Objective

• To resolve the long-standing debate in psycholinguistics between modular (autonomous) and integrated (monolithic) models of sentence processing.
• To address the limitations of purely modular or purely integrated models, which each explain only one set of empirical data.
• To propose and implement a hybrid architecture that combines a unified processor with separate knowledge sources to achieve both autonomy and interaction.
• To demonstrate the feasibility of this hybrid approach using the COMPERE computational model.
• To bring the field closer to developing human-like language processing systems by integrating the strengths of both theoretical camps.

Proposed method

• Designing a computational model, COMPERE, that uses a single central processor to coordinate multiple specialized knowledge sources.
• Structuring knowledge sources as independent modules that maintain relative autonomy while interacting through a shared processing engine.
• Implementing dynamic interaction between modules via shared representations and feedback mechanisms during sentence processing.
• Using a unified processing architecture to simulate real-time sentence comprehension, including syntactic and semantic integration.
• Calibrating the model to handle both garden-path and non-garden-path sentences, reflecting human processing patterns.
• Evaluating the model's behavior against empirical data from psycholinguistic experiments on sentence processing.

Experimental results

Research questions

• RQ1 Can a single processor model account for both autonomous and integrated aspects of human sentence processing?
• RQ2 How can modular knowledge sources maintain independence while contributing to a unified processing outcome?
• RQ3 Does a hybrid architecture better explain conflicting empirical findings in sentence processing than purely modular or monolithic models?
• RQ4 Can the COMPERE model simulate human-like processing behavior, including reanalysis and disambiguation?
• RQ5 What role does interaction between modules play in resolving syntactic ambiguities during real-time comprehension?

Key findings

• The hybrid model successfully explains both modular and integrated processing phenomena, resolving a key tension in language processing theory.
• COMPERE demonstrates that a single processor can coordinate multiple specialized knowledge sources without sacrificing modularity.
• The model accounts for reanalysis in ambiguous sentences, such as garden-path structures, by allowing interaction between modules.
• The architecture supports dynamic feedback and integration, enabling real-time processing that mirrors human comprehension patterns.
• The results show that autonomy and interaction are not mutually exclusive, and both can coexist in a single, coherent processing framework.
• The model provides a feasible path toward building human-like language processing systems by unifying theoretical strengths from both camps.