Problem
Research questions and friction points this paper is trying to address.
cross-lingual transfer
script unification
Turkic languages
multilingual models
surface overlap
Universal or Language-Family-Specific Script Unification for Cross-Lingual Transfer? A Case Study on Turkic Languages
🤖 AI Summary
This study addresses the challenge of limited cross-lingual transfer performance among closely related languages—such as those in the Turkic family—due to divergent writing systems. It presents the first systematic comparison between the general-purpose romanization tool Uroman and the Turkic-specific Common Turkic Script (CTS) for script unification. Using fastText word embeddings trained on Wikipedia corpora from 11 Turkic languages, the authors evaluate named entity recognition (NER) and part-of-speech (POS) tagging performance on WikiANN and Universal Dependencies datasets, also comparing against the CANINE-c model. Results show that CTS and Uroman achieve comparable NER performance, both significantly outperforming monolingual baselines, while neither script unification method consistently dominates in POS tagging, with fastText remaining competitive across multiple treebanks. The findings highlight that script unification efficacy is jointly influenced by linguistic properties, subword overlap, and the availability of supervised signals.
📝 Abstract
Closely related languages written in different scripts expose little surface overlap to multilingual models, limiting cross-lingual transfer. We compare two approaches to script unification: the general-purpose uroman romanizer and the family-specific Common Turkic Script (CTS). We train matched fastText models on transliterated Wikipedia corpora from 11 Turkic languages and evaluate them on WikiANN named entity recognition and Universal Dependencies part-of-speech tagging. CTS and uroman show no significant difference on NER, while both substantially outperform the official monolingual fastText baselines. POS results reveal no universal winner: language-specific differences are associated with the cross-lingual character n-gram coverage induced by each representation, while within-language coverage becomes more important when target-language supervision is available. Although CANINE-c achieves higher overall POS averages, the substantially simpler fastText-based systems remain competitive on several treebanks. Overall, the effectiveness of script unification depends on the language, the induced subword overlap, and the available supervision.
Innovation
Methods, ideas, or system contributions that make the work stand out.
script unification
cross-lingual transfer
Turkic languages
fastText
character n-gram coverage
🔎 Similar Papers
No similar papers found.
