Analysis Errors in the Image-to-Text Conversion and Translation of Students' Handwriting Using Google Application

Abstract

In recent decades, advances in artificial intelligence have driven digital transformation and led to the development of digital applications such as Google Lens. Google Lens utilizes smartphone cameras and machine learning technology to recognize objects and perform functions such as translation and scanning; however, its accuracy remains a subject of debate. Handwritten text presents particular challenges due to variations in writing styles, inconsistencies in letter formation, and contextual influences, all of which may lead to inaccuracies in image-to-text conversion and translation processes. This study aims to analyze the types and frequencies of errors in the results of image-to-text conversion and the translation of students' handwritten texts using Google Lens. A qualitative descriptive approach with document analysis was employed, focusing on 15 handwritten essays by English education students at the University of Mataram. The findings reveal that the most dominant type of error is character recognition errors (55% of conversion errors), followed by omitted text errors (27%) and word segmentation errors (18%). In addition, translation-stage errors were identified, including syntactic errors (75% of translation errors) and untranslated text (25%). These findings indicate that Google Lens still has limitations in recognizing and translating handwritten text. The accuracy of the results is strongly influenced by the clarity of the handwriting and the quality of the image. While this study identifies specific error patterns, it does not systematically compare Google Lens output with human translations using a standardized rubric; therefore, further research is needed to validate claims regarding superiority over human translation.