The determine the reading order of the text extracted from a searchable PDF produced by an OCR software from an old newspaper is the first task in the process of preparation of corpora of old newspapers. In the paper we present an algorithm for generation of reading order of black selected from the corresponding PDF. Also we performed a tuning of the parameters of the algorithm. The optimization provides 10 % improvement.

This study investigates whether human–LLM interaction in academic writing exhibits cross-cultural variation. Using NLP-informed corpus methods, we analyze nine semi-structured student interviews from three national contexts (Romania, Bulgaria, Switzerland) to examine how AI use is linguistically constructed across three dimensions of epistemic positioning: agency strength, authority dynamics, and discourse-level stance. Results show a strong predominance of distancing and hedging strategies, with AI consistently framed as a functional writing support tool rather than an epistemic authority. At the same time, modest but systematic cross-country differences indicate culturally embedded variation in how students discursively negotiate epistemic responsibility and evaluation in AI-assisted writing practices.

In the paper we present a set of small LLM-based models for solving the basic NLP tasks for Bulgarian - POS tagging, Lemmatization, Dependency parsing, Named Entity Recognition, Named Entity Linking, Event Annotation, among others. In order to create fine-tuned models for these tasks, we first pre-train models using architectures like BERT, Modern-BERT, and T5 with different sizes, over Bulgarian data only. For each of the tasks we report our approach towards the fine-tuning, the results from the experiments and also the evaluation. Then we define a way to visualize the results over HTML documents which contain the analyzed texts. Our rationale are as follows: most, if not all SSH research scenarios, need a reliable processing chains that can be customized with respect to the specific needs. These scenarios also would need proper visualization for human observation. We aim to provide such a basic LLM-based toolkit.

The paper presents the results from the experiments with two large language models (LLMs) - T5 and Llama – for extracting events from a Bulgarian event corpus. The two models were pretrained by us on 35 Billion Token Bulgarian Corpus. The extraction was performed within the context of one sentence. Our approach aims at balancing the ACE-oriented approach that uses triggers in event detection, and the MUC-oriented one that uses more general event types. The evaluation relies on the IoU (Intersection over Union) of token spans and is twofold. The first one refers to the predicted event token span. Here if the span is correct, the semantic roles within the event are further checked. The second one refers to the triple of an event type, its semantic roles and participants. The results are promising. A qualitative evaluation is provided as well.