A model of language processing as hierarchic sequential prediction.

Area of Science:

• Computational linguistics
• Cognitive science
• Natural language processing

Background:

• Hierarchic sequence models in memory research often have shallow hierarchies, leading to conflated memories.
• Existing models struggle with deep hierarchical structures in natural language processing.

Purpose of the Study:

• To introduce a broad-coverage probabilistic sentence processing model.
• To flatten sentence processing operations into a shallow hierarchy of learned sequences using a modified left-corner parsing strategy.

Main Methods:

• Utilized a variant of a left-corner parsing strategy.
• Developed a broad-coverage probabilistic model with constraints on hierarchy depth.
• Tested the model on large newspaper corpora.

Main Results:

• The proposed model successfully processed large newspaper corpora.
• Achieved accuracy comparable to state-of-the-art parsers.
• Demonstrated effective handling of sentence structures without relying on deep memory hierarchies.

Conclusions:

• A shallow hierarchy constraint in computational memory models can enhance sentence processing.
• This model offers an effective alternative to complex hierarchical approaches in natural language processing.
• The findings have implications for understanding memory and developing advanced NLP tools.