A neurocomputational approach to prepositional phrase attachment ambiguity resolution.

Area of Science:

• Computational Neuroscience
• Natural Language Processing
• Cognitive Science

Background:

• Prepositional Phrase (PP) attachment ambiguity is a significant challenge in Natural Language Processing (NLP).
• Resolving this ambiguity often requires semantic understanding to determine correct syntactic structure.
• Existing computational models vary in their ability to handle this complex linguistic phenomenon.

Purpose of the Study:

• To develop and evaluate a novel neurocomputational model for resolving PP attachment ambiguity.
• To investigate the emergence of neural cell assemblies (CAs) representing semantic similarities.
• To assess the model's performance against established machine learning algorithms.

Main Methods:

• A large-scale network of biologically plausible fatiguing leaky integrate-and-fire neurons was employed.
• The network was trained using semantic hierarchies from WordNet on ambiguous sentences from the Penn Treebank corpus.
• Emergent, massively overlapping neural cell assemblies (CAs) were utilized for semantic categorization.

Main Results:

• The neurocomputational model demonstrated the emergence of overlapping CAs that effectively captured semantic similarities.
• The model achieved an average resolution accuracy of 84.56% for PP attachment ambiguity.
• Performance was found to be on par with existing state-of-the-art machine learning algorithms.

Conclusions:

• The proposed neurocomputational model offers a biologically plausible approach to solving PP attachment ambiguity.
• Emergent neural cell assemblies can effectively represent semantic information for syntactic disambiguation.
• This model presents a promising alternative or complement to traditional NLP techniques for semantic parsing.