Talk · International Conference on Computer Science and Engineering · October 26, 2024
Aligning Sentences for Kazakh-Turkish Parallel Corpora
Diana Rakhimova, Eşref Adalı, Aidana Karibayeva
Abstract
This paper presents a hybrid approach to sentence alignment for the Kazakh-Turkish parallel corpus, addressing the challenges posed by linguistic and structural differences between the two languages. The system is divided into two modules: the Hunaling module and the Rule-Based module. The Hunaling module first performs an initial alignment using sentence length to establish a rough correspondence between sentences. This is followed by a refinement stage that leverages lexical matching, focusing on Turkic cognates and shared roots. The Rule-Based module then applies syntactic and contextual rules to fine-tune the alignment, using a small annotated corpus to guide adjustments. The proposed approach achieves a precision of 90%, recall of 88%, and an F1-score of 88.5%, demonstrating significant improvements over traditional methods. This methodology offers a robust solution for aligning sentences in Kazakh-Turkish parallel corpora, with potential applications in machine translation and multilingual NLP.
More by Aidana Karibayeva
- Speech-to-Speech Translation from Kazakh into Turkic Languages: Pretrained Speech Encoders, Turkic Speech Units, and Lessons from Negative ResultsComputers · 2026
- Toward an Adaptive Speech-to-Summary Pipeline for Turkic Languages: Language-Specific ASR, Conditional Morphology-Aware Correction, Pivot Translation, and SummarizationComputers · 2026
- RESTORING PUNCTUATION IN AUTOMATIC SPEECH RECOGNITION FOR KAZAKH LANGUAGEEKTU Journal of Information and Communication Sciences · 2026
- From Speech to Readable Text: Automatic Punctuation in ASR Systems of the Kazakh LanguageCommunications in computer and information science · 2026
