Lexical is as lexical does: computational approaches to lexical representation.

Area of Science:

• Neuroscience
• Cognitive Psychology
• Computational Linguistics

Background:

• The concept of 'lexical representation' in neuroimaging and neuropsychology lacks precise definition.
• Existing computational models of word recognition vary in how they represent lexical knowledge.

Purpose of the Study:

• To critically examine the construct of lexical representation.
• To evaluate the neural plausibility of different representational schemes in computational models.
• To synthesize current evidence regarding the nature of lexical representations.

Main Methods:

• Review of computational models (localist vs. connectionist).
• Analysis of behavioral, neuropsychological, and neuroimaging evidence.
• Discussion of neural plausibility and model architectures.

Main Results:

• Structural representations lack neural plausibility; distributed representations are favored.
• Connectionist models with recurrent connections can form functional lexical representations.
• Evidence supports a role for semantic information and modality/task-specific representations.

Conclusions:

• Lexical representations are likely distributed and functionally specific.
• Connectionist architectures offer potential implementations.
• Further interdisciplinary research combining modeling and neuroscience is essential.