A constraint-based lexicalist account of the subject/object attachment preference

Area of Science:

• Psycholinguistics
• Computational Linguistics
• Cognitive Science

Background:

• Readers exhibit a strong preference for parsing ambiguous noun phrases as direct objects of preceding verbs.
• This 'object bias' can create processing difficulties, known as garden-path sentences, even when grammatically implausible.
• Previous theories proposed lexically blind initial parsing followed by lexical filtering.

Purpose of the Study:

• To demonstrate that the observed object bias and processing difficulties arise naturally from constraint-based lexicalist models.
• To investigate the role of verb frequency in sentence processing within these models.
• To compare model behavior with established psycholinguistic findings.

Main Methods:

• Developed a computational model using a simple recurrent network (SRN).
• Trained the SRN to predict upcoming verb complements using a corpus of verbs from the Penn Treebank.
• Analyzed the model's parsing preferences and sensitivity to verb frequency.

Main Results:

• The model demonstrated a significant object bias, mirroring human reading preferences.
• Verb frequency influenced the model's predictions, consistent with psycholinguistic data.
• The results support constraint-based lexicalist approaches to sentence processing.

Conclusions:

• Constraint-based lexicalist models can naturally account for the object bias in sentence processing.
• Lexical information, including verb frequency, plays a crucial role from the initial stages of parsing.
• The findings align computational modeling with empirical psycholinguistic evidence.