We analyzed 133 linguistic features for automated proficiency classification under the Common European Framework of Reference (CEFR) in a two-tier architecture: deterministic Tier 1 (lexical, morphological, and syntactic measures) and machine-learning-based Tier 2 (semantic coherence, topic structure, cohesion, and error-analysis signals). Experiments were conducted on a corpus of 3,205 learner texts from mixed sources, with triangulated validation against expert-verified Cambridge examination subsets. The materials were collected in 2022--2025 and included a substantial institutional corpus of over 3,000 essays and other writing texts produced by students of Moscow Institute of Physics and Technology (MIPT) studying English as a foreign language and regularly assessed by our AI-powered testing system ISTOK (Intelligent System for Testing General Language Competencies). Feature matrices were standardized after missing-value handling (fold-local median imputation for cross-validation and zero-fill for held-out reporting). In supervised evaluation, the best Tier 1+2 model reaches 66.72% exact accuracy (macro F1 = 0.69) and 94.53% adjacent accuracy (within one CEFR level) on a 3,198-sample CEFR-labeled benchmark split; an extended comparison including preliminary Tier 3 features achieves 67.50%. Unsupervised analyses show strong structure for extreme levels (A1 99.5% purity; C2 82.4% purity) and moderate alignment with professional Cambridge labels (Adjusted Rand Index = 0.303). We report block ablations and compact subset searches, with strongest signals from morphological complexity and lexical sophistication, and consistent incremental gains from error-based features. The results provide a validated, interpretable feature inventory and practical guidance for feature selection in automated language assessment systems.

The article examines the application of deep reinforcement learning methods to solve the problem of automated scheduling of driver shifts for urban passenger transport. The Crew Scheduling Problem belongs to the class of NP-hard combinatorial optimization problems and is characterized by a multitude of complex constraints related to labor legislation and the operational specifics of the transport network. The problem formulation, considering route changes, is described. The problem is formalized as a Markov Decision Process, taking into account specific constraints of the transport industry: maximum working hours, lunch breaks, and minimum rest time between trips. The state space is formulated, including features of control stops, the current trip, and candidate shifts. A mechanism for prioritized selection of candidate shifts is described to reduce the dimensionality of the action space. A multi-component reward function is described, considering the number of shifts utilized, deadhead travel time, and driver utilization. The agent's architecture is implemented based on the Actor-Critic method with the Proximal Policy Optimization algorithm. An experimental study was conducted on real data from the transport network of the city of Yaroslavl, including 6 routes and 974 trips. A comparative analysis was conducted with alternative methods: DQN, REINFORCE, and a heuristic approach represented by a greedy algorithm. The comparative analysis of the results demonstrated the superiority of the PPO algorithm over the other approaches. As a result of the study, it was concluded that reinforcement learning methods can be used to solve transportation optimization problems.

This paper investigates the task of automatic word alignment in parallel texts, a fundamental step for training machine translation systems, conducting comparative linguistic studies, and creating linguistic resources. Given the scarcity of annotated data for many language pairs, the applicability of Large Language Models (LLMs) becomes particularly relevant due to their high generalization capabilities and ability to solve tasks without extensive fine-tuning on target datasets. This study presents a comparative analysis of the effectiveness of modern general-purpose LLMs versus specialized alignment algorithms using Russian-English parallel data. The research involved testing ten state-of-the-art models (including Gemini 3 Pro, GPT-5.2, and Claude Sonnet 4.5) using various prompting strategies (zero-shot, few-shot), alongside five baseline approaches ranging from statistical methods (fast-align, eflomal) to neural network architectures (AwesomeAlign, AccAlign, BinaryAlign). Performance was evaluated based on Precision, Recall, F-measure, and Alignment Error Rate (AER) metrics using annotated data from the Russian National Corpus. Experimental results indicated that the specialized BinaryAlign algorithm maintains the lead in overall alignment quality (F-measure 0.883, AER 0.113). However, leading LLMs, specifically Gemini 3 Pro Preview and GPT-5.2, demonstrated results surpassing those of most classic and early neural network baselines. Notably, for the most effective models, including in-context examples often reduced performance compared to the zero-shot setting. Thus, modern LLMs can serve as a reliable tool for high-quality alignment in the absence of training data, opening new perspectives for processing low-resource language pairs.

The paper investigates the impact of training set size on the quality of automatic short answers grading, formulated as a classification task. The impact was evaluated using a method based on measuring the similarity between the assessed answer and a given reference answer, calculated via embedding vectors, in combination with a logistic regression classifier. Experiments were conducted on corpora of answers to questions in computer science, history, and software development using Qt framework. The sizes of the corpora were 547, 522, and 931 answers, respectively. Two experiments were conducted during the study. In the first experiment, the change in classification quality was assessed as the training set size was reduced. It showed that when the binary classification is utilized (an answer can be either correct or incorrect), reducing the size of the training set leads to a smaller decline in quality compared to ternary classification (which includes a class of partially correct answers). In the second experiment, the possibility of improving classification quality by expanding small‑sized training sets through data augmentation was investigated. It demonstrated that augmentation performed using the DeepSeek generative model can significantly improve results in several cases, which is important for practical applications under data scarcity conditions. Additionally, the experiments revealed that when different language models are used to generate embeddings, the magnitude of change in classification quality with varying training set sizes can differ significantly. Specifically, using certain models — such as rubert‑tiny2 and MiniLM‑L12‑v2 — to produce embeddings yields more stable results than using other models.

We consider the problem of feasible vertex coloring with the minimum number of colors for connected undirected graphs that contain no self-loops or multiple edges. For every given $k \geq 3$, the problem of checking the existence of a feasible vertex coloring of the graph with k colors is NP-complete. Therefore, studying graph-scaling processes while preserving or limiting their chromatic numbers is of interest.

In this paper, we study the nature of changes in the chromatic number of graphs with an increase in the number of vertices and edges using gluing operations by identifying their isomorphic subgraphs. $G = (G_{1} \circ G_{2}) \tilde{G}$ — is the resulting graph of the gluing operation of graphs $G_1$ and $G_2$; $\tilde{G} \subseteq G$ is the subgraph obtained as a result of identifying isomorphic subgraphs $G_1' \subseteq G_1$ and $G_2' \subseteq G_2$; $|V(G)| = |V(G_1)| + |V(G_2)| - |V(\tilde{G})|, |E(G)| = |E(G_1)| + |E(G_2)| - |E(\tilde{G})|$. Gluing operations in which one of the graphs $G_1$ or $G_2$ is isomorphic to another graph or its subgraph and the identification of subgraphs $G_1^{'}\subset G_1$ and $G_2^{'} \subset G_2$ is carried out in accordance with the isomorphism $G_1' \cong G_2'$, are called cloning operations.
A constructive description of a class of 2-chromatic graphs is obtained based on the gluing and cloning operations. Constraints on the gluing and cloning operations that ensure the preservation of the chromatic number of scalable graphs are formulated. It is established that when performing cloning operations, $\chi(G) =\max{\chi(G_1),\chi(G_2)}$. Examples of assembling 2-chromatic graphs using operations satisfying these constraints are given. For an arbitrary gluing operation $\chi(G) \leqslant \max {\chi(G_1),\chi(G_2)} + |V(\tilde{G})| - |V(\tilde{G'})|$, where $\tilde{G'}$ is the maximal complete subgraph of $\tilde{G}$. The possible growth of the chromatic number of graphs is estimated when scaling with various restrictions on the superposition of gluing operations.

The large states pace of programs makes their direct verification by model checking difficult or impossible. The presence of symmetry in a program often allows simplifying the model and reducing its state space, leading to significant decrease of verification time. The classical approach consists in detecting a symmetry group and constructing a quotient model based on it — a simplified model for verification purposes. However, not all tools provide support for symmetry, and those that do may still struggle because finding an appropriate symmetry group is computationally complex problem.

This work proposes an approach to program development based on explicit symmetry exploitation, which is an alternative to the classical one. In the program, a core is extracted — a coordination center working under consideration of symmetry and responsible for ensuring temporal properties. The core coordinates computations outside itself — those placed in the wrapper surrounding the core. As a result, the core has a small state space, replace the quotient model and allows verification using a model checker without symmetry support. The wrapper cannot interfere in the operation of the verified core and violate its properties. The approach is demonstrated by the example of the development and verification of the Mars rover resource arbiter. The arbiter coordinates access of n processes to m resources where both n and m are natural numbers. Programming languages C/C++ and the Spin model checker tool are used. The behavioral model of the core is automatically extracted by the Spin tool from the C code. Temporal properties expressed via Linear Temporal Logic (LTL) are subject to verification.