Çağrı Toraman

Cagri Toraman, Izzet Emre Kucukkaya, Oguzhan Ozcelik et al.

The importance of social media is again exposed in the recent tragedy of the 2023 Turkey and Syria earthquake. Many victims who were trapped under the rubble called for help by posting messages in Twitter. We present an interactive tool to provide situational awareness for missing and trapped people, and disaster relief for rescue and donation efforts. The system (i) collects tweets, (ii) classifies the ones calling for help, (iii) extracts important entity tags, and (iv) visualizes them in an interactive map screen. Our initial experiments show that the performance in terms of the F1 score is up to 98.30 for tweet classification, and 84.32 for entity extraction. The demonstration, dataset, and other related files can be accessed at https://github.com/avaapm/deprem

Cagri Toraman, Eyup Halit Yilmaz, Furkan Şahinuç et al.

Tokenization is an important text preprocessing step to prepare input tokens for deep language models. WordPiece and BPE are de facto methods employed by important models, such as BERT and GPT. However, the impact of tokenization can be different for morphologically rich languages, such as Turkic languages, where many words can be generated by adding prefixes and suffixes. We compare five tokenizers at different granularity levels, i.e. their outputs vary from smallest pieces of characters to the surface form of words, including a Morphological-level tokenizer. We train these tokenizers and pretrain medium-sized language models using RoBERTa pretraining procedure on the Turkish split of the OSCAR corpus. We then fine-tune our models on six downstream tasks. Our experiments, supported by statistical tests, reveal that Morphological-level tokenizer has challenging performance with de facto tokenizers. Furthermore, we find that increasing the vocabulary size improves the performance of Morphological and Word-level tokenizers more than that of de facto tokenizers. The ratio of the number of vocabulary parameters to the total number of model parameters can be empirically chosen as 20% for de facto tokenizers and 40% for other tokenizers to obtain a reasonable trade-off between model size and performance.

Cagri Toraman, Oguzhan Ozcelik, Furkan Şahinuç et al.

The rapid dissemination of misinformation through online social networks poses a pressing issue with harmful consequences jeopardizing human health, public safety, democracy, and the economy; therefore, urgent action is required to address this problem. In this study, we construct a new human-annotated dataset, called MiDe22, having 5,284 English and 5,064 Turkish tweets with their misinformation labels for several recent events between 2020 and 2022, including the Russia-Ukraine war, COVID-19 pandemic, and Refugees. The dataset includes user engagements with the tweets in terms of likes, replies, retweets, and quotes. We also provide a detailed data analysis with descriptive statistics and the experimental results of a benchmark evaluation for misinformation detection.

Cagri Toraman, Furkan Şahinuç, Eyup Halit Yilmaz

The performance of hate speech detection models relies on the datasets on which the models are trained. Existing datasets are mostly prepared with a limited number of instances or hate domains that define hate topics. This hinders large-scale analysis and transfer learning with respect to hate domains. In this study, we construct large-scale tweet datasets for hate speech detection in English and a low-resource language, Turkish, consisting of human-labeled 100k tweets per each. Our datasets are designed to have equal number of tweets distributed over five domains. The experimental results supported by statistical tests show that Transformer-based language models outperform conventional bag-of-words and neural models by at least 5% in English and 10% in Turkish for large-scale hate speech detection. The performance is also scalable to different training sizes, such that 98% of performance in English, and 97% in Turkish, are recovered when 20% of training instances are used. We further examine the generalization ability of cross-domain transfer among hate domains. We show that 96% of the performance of a target domain in average is recovered by other domains for English, and 92% for Turkish. Gender and religion are more successful to generalize to other domains, while sports fail most.

Umitcan Sahin, Izzet Emre Kucukkaya, Oguzhan Ozcelik et al.

Text-embedded images can serve as a means of spreading hate speech, propaganda, and extremist beliefs. Throughout the Russia-Ukraine war, both opposing factions heavily relied on text-embedded images as a vehicle for spreading propaganda and hate speech. Ensuring the effective detection of hate speech and propaganda is of utmost importance to mitigate the negative effect of hate speech dissemination. In this paper, we outline our methodologies for two subtasks of Multimodal Hate Speech Event Detection 2023. For the first subtask, hate speech detection, we utilize multimodal deep learning models boosted by ensemble learning and syntactical text attributes. For the second subtask, target detection, we employ multimodal deep learning models boosted by named entity features. Through experimentation, we demonstrate the superior performance of our models compared to all textual, visual, and text-visual baselines employed in multimodal hate speech detection. Furthermore, our models achieve the first place in both subtasks on the final leaderboard of the shared task.

Izzet Emre Kucukkaya, Umitcan Sahin, Cagri Toraman

The task of multi-author writing style detection aims at finding any positions of writing style change in a given text document. We formulate the task as a natural language inference problem where two consecutive paragraphs are paired. Our approach focuses on transitions between paragraphs while truncating input tokens for the task. As backbone models, we employ different Transformer-based encoders with warmup phase during training. We submit the model version that outperforms baselines and other proposed model versions in our experiments. For the easy and medium setups, we submit transition-focused natural language inference based on DeBERTa with warmup training, and the same model without transition for the hard setup.

Umitcan Sahin, Izzet Emre Kucukkaya, Cagri Toraman

Fanfiction, a popular form of creative writing set within established fictional universes, has gained a substantial online following. However, ensuring the well-being and safety of participants has become a critical concern in this community. The detection of triggering content, material that may cause emotional distress or trauma to readers, poses a significant challenge. In this paper, we describe our approach for the Trigger Detection shared task at PAN CLEF 2023, where we want to detect multiple triggering content in a given Fanfiction document. For this, we build a hierarchical model that uses recurrence over Transformer-based language models. In our approach, we first split long documents into smaller sized segments and use them to fine-tune a Transformer model. Then, we extract feature embeddings from the fine-tuned Transformer model, which are used as input in the training of multiple LSTM models for trigger detection in a multi-label setting. Our model achieves an F1-macro score of 0.372 and F1-micro score of 0.736 on the validation set, which are higher than the baseline results shared at PAN CLEF 2023.

Burak Erinç Çetin, Yıldırım Özen, Elif Naz Demiryılmaz et al.

Generative large language models present significant potential but also raise critical ethical concerns. Most studies focus on narrow ethical dimensions, and also limited diversity of languages and models. To address these gaps, we conduct a broad ethical evaluation of 29 recent open-source large language models using a novel data collection including four ethical aspects: Robustness, reliability, safety, and fairness. We analyze model behavior in both a commonly used language, English, and a low-resource language, Turkish. Our aim is to provide a comprehensive ethical assessment and guide safer model development by filling existing gaps in evaluation breadth, language coverage, and model diversity. Our experimental results, based on LLM-as-a-Judge, reveal that optimization efforts for many open-source models appear to have prioritized safety and fairness, and demonstrated good robustness while reliability remains a concern. We demonstrate that ethical evaluation can be effectively conducted independently of the language used. In addition, models with larger parameter counts tend to exhibit better ethical performance, with Gemma and Qwen models demonstrating the most ethical behavior among those evaluated.

Cagri Toraman

Despite advancements in English-dominant generative large language models, further development is needed for low-resource languages to enhance global accessibility. The primary methods for representing these languages are monolingual and multilingual pretraining. Monolingual pretraining is expensive due to hardware requirements, and multilingual models often have uneven performance across languages. This study explores an alternative solution by adapting large language models, primarily trained on English, to low-resource languages. We assess various strategies, including continual training, instruction fine-tuning, task-specific fine-tuning, and vocabulary extension. The results show that continual training improves language comprehension, as reflected in perplexity scores, and task-specific tuning generally enhances performance of downstream tasks. However, extending the vocabulary shows no substantial benefits. Additionally, while larger models improve task performance with few-shot tuning, multilingual models perform worse than their monolingual counterparts when adapted.

Süha Kağan Köse, Mehmet Can Baytekin, Burak Aktaş et al.

Retrieval-Augmented Generation (RAG) enhances LLM factuality, yet design guidance remains English-centric, limiting insights for morphologically rich languages like Turkish. We address this by constructing a comprehensive Turkish RAG dataset derived from Turkish Wikipedia and CulturaX, comprising question-answer pairs and relevant passage chunks. We benchmark seven stages of the RAG pipeline, from query transformation and reranking to answer refinement, without task-specific fine-tuning. Our results show that complex methods like HyDE maximize accuracy (85%) that is considerably higher than the baseline (78.70%). Also a Pareto-optimal configuration using Cross-encoder Reranking and Context Augmentation achieves comparable performance (84.60%) with much lower cost. We further demonstrate that over-stacking generative modules can degrade performance by distorting morphological cues, whereas simple query clarification with robust reranking offers an effective solution.

Ayşe Aysu Cengiz, Ahmet Kaan Sever, Elif Ecem Ümütlü et al.

The reliance on translated or adapted datasets from English or multilingual resources introduces challenges regarding linguistic and cultural suitability. This study addresses the need for robust and culturally appropriate benchmarks by evaluating the quality of 17 commonly used Turkish benchmark datasets. Using a comprehensive framework that assesses six criteria, both human and LLM-judge annotators provide detailed evaluations to identify dataset strengths and shortcomings. Our results reveal that 70% of the benchmark datasets fail to meet our heuristic quality standards. The correctness of the usage of technical terms is the strongest criterion, but 85% of the criteria are not satisfied in the examined datasets. Although LLM judges demonstrate potential, they are less effective than human annotators, particularly in understanding cultural common sense knowledge and interpreting fluent, unambiguous text. GPT-4o has stronger labeling capabilities for grammatical and technical tasks, while Llama3.3-70B excels at correctness and cultural knowledge evaluation. Our findings emphasize the urgent need for more rigorous quality control in creating and adapting datasets for low-resource languages.

Deniz Yılmaz, Evren Ayberk Munis, Çağrı Toraman et al.

Document parsing is now widely used in applications, such as large-scale document digitization, retrieval-augmented generation, and domain-specific pipelines in healthcare and education. Benchmarking these models is crucial for assessing their reliability and practical robustness. Existing benchmarks mostly target high-resource languages and provide limited coverage for low-resource settings, such as Turkish. Moreover, existing studies on Turkish document parsing lack a standardized benchmark that reflects real-world scenarios and document diversity. To address this gap, we introduce OCRTurk, a Turkish document parsing benchmark covering multiple layout elements and document categories at three difficulty levels. OCRTurk consists of 180 Turkish documents drawn from academic articles, theses, slide decks, and non-academic articles. We evaluate seven OCR models on OCRTurk using element-wise metrics. Across difficulty levels, PaddleOCR achieves the strongest overall results, leading most element-wise metrics except figures and attaining high Normalized Edit Distance scores in easy, medium, and hard subsets. We also observe performance variation by document type. Models perform well on non-academic documents, while slideshows become the most challenging.

Burak Aktaş, Mehmet Can Baytekin, Süha Kağan Köse et al.

Text-to-SQL systems have achieved strong performance on English benchmarks, yet their behavior in morphologically rich, low-resource languages remains largely unexplored. We introduce BIRDTurk, the first Turkish adaptation of the BIRD benchmark, constructed through a controlled translation pipeline that adapts schema identifiers to Turkish while strictly preserving the logical structure and execution semantics of SQL queries and databases. Translation quality is validated on a sample size determined by the Central Limit Theorem to ensure 95% confidence, achieving 98.15% accuracy on human-evaluated samples. Using BIRDTurk, we evaluate inference-based prompting, agentic multi-stage reasoning, and supervised fine-tuning. Our results reveal that Turkish introduces consistent performance degradation, driven by both structural linguistic divergence and underrepresentation in LLM pretraining, while agentic reasoning demonstrates stronger cross-lingual robustness. Supervised fine-tuning remains challenging for standard multilingual baselines but scales effectively with modern instruction-tuned models. BIRDTurk provides a controlled testbed for cross-lingual Text-to-SQL evaluation under realistic database conditions. We release the training and development splits to support future research.

Eyup Halit Yilmaz, Cagri Toraman

Intent detection of spoken queries is a challenging task due to their noisy structure and short length. To provide additional information regarding the query and enhance the performance of intent detection, we propose a method for semantic expansion of spoken queries, called ConQX, which utilizes the text generation ability of an auto-regressive language model, GPT-2. To avoid off-topic text generation, we condition the input query to a structured context with prompt mining. We then apply zero-shot, one-shot, and few-shot learning. We lastly use the expanded queries to fine-tune BERT and RoBERTa for intent detection. The experimental results show that the performance of intent detection can be improved by our semantic expansion method.