Evaluating the Error Analysis Performance of the Deepseek AI Chatbot in Foreign Language Teaching

This study evaluates the error analysis performance of the DeepSeek artificial intelligence chatbot in foreign language teaching, focusing on its ability to detect, correct, and provide feedback on errors in German texts at the B1 proficiency level. Utilizing a structured methodology, 11 texts from the Menschen B1 textbook were modified to include 88 intentional errors, categorized into four types (omission, addition, selection, ordering) across grammatical and lexico-semantic levels. The chatbot’s performance was assessed through three stages: error detection, correction accuracy, and feedback quality. Results revealed that DeepSeek detected 85% of errors overall, with higher success rates for addition (91%) and ordering (91%) errors compared to omission (77%). Correction accuracy stood at 91%, with all selection errors corrected flawlessly, though lexico-semantic ordering errors showed lower correction accuracy (70%). Feedback for correctly corrected errors was 86% accurate, with lexico-semantic errors receiving near-perfect feedback (96%) compared to grammatical errors (75%). While DeepSeek demonstrated robust capabilities in error detection and correction, inconsistencies were observed, particularly in addressing omission errors and providing feedback for grammatical ordering errors. The findings highlight DeepSeek’s potential as a supplementary tool in foreign language education but underscore the need for refinement in handling specific error types and feedback clarity. This study contributes to the limited literature on DeepSeek’s educational applications and offers insights for optimizing artificial intelligence driven error analysis in language learning contexts.