SindBERT, the Sailor: Charting the Seas of Turkish NLP

TL;DR

This work addresses the underrepresentation of morphologically rich Turkish in large-scale pretraining by introducing SindBERT, the first large-scale RoBERTa-style Turkish encoder trained from scratch on 312 GB of Turkish text. The authors provide base and large configurations, a Turkish-optimized 52k BPE tokenizer, and pretraining on TPUv4 with a standard RoBERTa schedule, followed by fine-tuning on PoS tagging, NER, offensive language detection, and TurBLiMP. Across four benchmarks, SindBERTbase is competitive with existing Turkish encoders, while SindBERTlarge achieves top results in PoS and OffensEval, though scaling yields selective gains and TurBLiMP shows limited advantages for the large model. The results highlight that corpus quality and language-specific tokenization can outperform sheer data volume, suggesting that future Turkish NLP progress will hinge on curated data, discriminative evaluation, and domain adaptation rather than mere scale. SindBERT is released openly under MIT and serves as a foundation for further Turkish NLP development and analysis of scaling limits in morphologically rich languages.

Abstract

Transformer models have revolutionized NLP, yet many morphologically rich languages remain underrepresented in large-scale pre-training efforts. With SindBERT, we set out to chart the seas of Turkish NLP, providing the first large-scale RoBERTa-based encoder for Turkish. Trained from scratch on 312 GB of Turkish text (mC4, OSCAR23, Wikipedia), SindBERT is released in both base and large configurations, representing the first large-scale encoder-only language model available for Turkish. We evaluate SindBERT on part-of-speech tagging, named entity recognition, offensive language detection, and the TurBLiMP linguistic acceptability benchmark. Our results show that SindBERT performs competitively with existing Turkish and multilingual models, with the large variant achieving the best scores in two of four tasks but showing no consistent scaling advantage overall. This flat scaling trend, also observed for XLM-R and EuroBERT, suggests that current Turkish benchmarks may already be saturated. At the same time, comparisons with smaller but more curated models such as BERTurk highlight that corpus quality and diversity can outweigh sheer data volume. Taken together, SindBERT contributes both as an openly released resource for Turkish NLP and as an empirical case study on the limits of scaling and the central role of corpus composition in morphologically rich languages. The SindBERT models are released under the MIT license and made available in both fairseq and Huggingface formats.

Figures (1)

• Figure 1: Perplexity of the SindBERT models. Top: validation perplexity measured at checkpoints. Bottom: training perplexity measured at each optimization step.