Investigating variability in morphological processing with Bayesian distributional models

Area of Science:

• Psycholinguistics
• Cognitive Science
• Computational Linguistics

Background:

• Processing morphologically complex words is crucial for language comprehension.
• Non-native speakers exhibit significant variability in language performance.
• Traditional analyses often focus on average effects, potentially obscuring important performance differences.

Purpose of the Study:

• To investigate masked morphological priming effects for derived and inflected English words in non-native speakers.
• To explore how different word forms influence reaction time variability.
• To demonstrate the value of analyzing performance variability beyond mean effects in language processing research.

Main Methods:

• Utilized masked morphological priming with derived ('printer') and inflected ('printed') words priming their stems ('print').
• Employed Bayesian distributional models to analyze reaction times using a shifted-lognormal distribution.
• Assessed the impact of priming on both the mean (mu) and standard deviation (sigma) of response times.

Main Results:

• Both derived and inflected primes showed similar effects on mean reaction times.
• Inflectional priming significantly increased response time variability (sigma) compared to derivational priming.
• Results align with prior research indicating greater variability in non-native processing of inflected forms.

Conclusions:

• Performance variability is a critical factor in understanding language processing, particularly in non-native speakers.
• Analyzing distributional parameters beyond the mean (e.g., standard deviation) can reveal distinct effects of linguistic manipulations.
• This approach enhances psycholinguistic models by disentangling subtle processing differences and informing theories of word recognition.