Аңдатпа
The Kazakh language is hard for learners and machine translation because of its complicated morphology, regional differences, and subtle differences in meaning. This research presents a crowdsourcing-driven adaptive translation system that amalgamates automated algorithms with user feedback to improve translation quality. The system has three parts: automated translation, collecting user feedback, and retraining the model over and over again. This lets it adapt to idiomatic expressions and morphological complexity in real time. A total of 142 people from five regions of Kazakhstan, both native and non-native speakers, looked at translations of Kazakh text snippets and gave suggestions for better translations or comments when needed. We used mean ratings, standard deviation, and an accuracy metric (the percentage of translations that scored 8 or higher) to measure the quality of the translations. We also used comments from users to measure their engagement. Participants evaluated translations of Kazakh text snippets into English and provided corrections or contextual comments when necessary. The results showed that the average translation score was$8.33(\text{SD}=0.73)$and the accuracy rate was 85.9%, which shows that users were consistently happy with the service. Qualitative feedback captured important contextual nuances, like changing “Тұсында” from “In the house” to “near, beside, at.” Moderate correlation was found between engagement and translation quality$(r=0.62, p<0.01)$, which shows that working together makes the system work better. These results show that using automated translation with crowdsourced feedback works well for hard languages like Kazakh.