LeCoder: A large-scale automated coder for coding errors in word-production tasks

Area of Science:

• Psycholinguistics
• Computational Linguistics
• Neuropsychology

Background:

• Speech errors provide insights into language production mechanisms and disorders.
• Manual coding of speech error data is time-consuming, subjective, and requires large datasets.
• Existing methods lack objectivity and scalability for analyzing speech error patterns.

Purpose of the Study:

• To introduce LeCoder, the first open-source, automated error coder for English word and naming data.
• To develop a flexible, scalable, and generalizable tool for quantifying speech error relationships using a data-driven approach.
• To enhance the objectivity and replicability of speech error analysis in psycholinguistic and neuropsychological research.

Main Methods:

• Developed LeCoder using a data-driven approach based on large-scale English corpora.
• Quantified the target-response relationship in speech error data.
• Validated LeCoder's accuracy and generalizability on datasets coded by expert researchers and through machine learning.

Main Results:

• LeCoder demonstrates high accuracy compared to expert human coders.
• In some instances, LeCoder provides more logical categorizations than human coders.
• Machine learning approaches confirm LeCoder's generalizability to novel participants and items.

Conclusions:

• LeCoder offers an objective and efficient method for analyzing speech errors.
• The tool's high accuracy and generalizability encourage its adoption across research labs.
• Widespread use of LeCoder is expected to improve the replicability of findings in speech error research.