Lossy-Context Surprisal: An Information-Theoretic Model of Memory Effects in Sentence Processing

Area of Science:

• Cognitive Science
• Computational Linguistics
• Psycholinguistics

Background:

• Human sentence processing research aims to predict word comprehension difficulty in context.
• Existing models are typically expectation-based or memory-based.
• A gap exists in unifying these approaches to explain complex memory-expectation interactions.

Purpose of the Study:

• To present a novel, unified model of incremental sentence processing difficulty.
• To explain how incomplete memory representations influence word processing.
• To account for phenomena like language-dependent structural forgetting and dependency locality effects.

Main Methods:

• Developed the 'lossy-context surprisal' model.
• The model posits processing difficulty is proportional to surprisal given a lossy memory context.
• Utilized computational modeling to analyze interactions between memory and expectation in sentence comprehension.

Main Results:

• The lossy-context surprisal model offers an intuitive explanation for language-dependent structural forgetting.
• Demonstrated that dependency locality effects are a special case of the broader 'information locality' principle.
• Showed the model's ability to unify key features of prior expectation-based and memory-based theories.

Conclusions:

• The proposed model provides a more comprehensive framework for understanding sentence processing difficulty.
• Lossy memory representations are crucial for explaining complex psycholinguistic phenomena.
• The concept of information locality offers a generalized principle for memory-based effects in language.