Error Analysis in the Age of AI: A Psycholinguistic Study of L2 Writer Dependency on Automated Writing Assistants

Abstract

Although the pedagogical utility of Automated Writing Assistants (AWAs) has received increasing attention, relatively little is known about their long-term psycholinguistic effects on second language (L2) English writers. This study examines the impact of sustained AWA use among intermediate L2 learners enrolled in a 16-week academic writing course. A mixed-methods longitudinal design was employed, involving 48 participants randomly assigned to either an AWA-supported group or a control group. Data were derived from corpus-based error analysis focusing on grammatical, mechanical, and lexical categories alongside stimulated recall interviews and a post-intervention writing task conducted without AWA support. The findings suggest that the AWA group demonstrated more rapid surface-level error reduction, particularly in article and preposition usage; this improvement in error reduction diminished once the tool was withdrawn. Participants in this group exhibited increased error recurrence and reported heightened metalinguistic uncertainty, with revisions closely resembling prior AWA outputs rather than evidencing internalised control. This pattern was interpreted as indicative of procedural dependency, wherein AWAs appeared to function as external monitors that constrained the development of autonomous self-regulation. The study underscored the need for pedagogical strategies that temper short-term gains in accuracy with sustained metacognitive engagement, thereby supporting the cultivation of linguistic autonomy over time.