Vol 30, No 3 (2026)

By 2026, automated sign language recognition has moved from research laboratories into applied domains, including mobile translators, educational software and inclusive services. For Russian Sign Language (RSL), however, this field remains fragmented and is rarely addressed from a linguistic perspective. The aim of the article is threefold: to systematise the state of research on AI-based recognition of RSL with an explicitly linguistic frame; to identify the principal gap, namely the absence of a continuous parallel RSL-spoken Russian corpus; and to propose a concrete component-based architecture for an RSL-to-Russian translation system grounded in published evidence. The review focuses on seven blocks: the phonology, morphology and prosody of RSL; the gap between Russian and international corpora (BSL Corpus, DGS-Korpus, How2Sign, RWTH-PHOENIX-Weather versus the NSTU corpus and TheRuSLan); the evolution of AI architectures from HMM-based continuous recognition to transformer-based translation and large language models; the Russian landscape of datasets and systems built around Slovo, Bukva and Logos; the linguistic validity criteria for synthetic sign data; the mapping between Stokoe-style sign phonology and the outputs of real-time pose estimators; and the choice of a language model for the spoken-Russian decoder. The paper argues that a successful RSL-to-Russian (Sign-to-Text) system must couple linguistic knowledge - Stokoe phonology, predicate typology, mouthings and information structure - with engineering components in a principled way. In particular, it proposes a pipeline that combines RTMPose Whole-body as a real-time front-end, a Logos-pretrained vision encoder, a Vector Quantization sign tokenizer in the spirit of SignLLM and a language decoder from the GigaChat family. The contribution of the study is a literature-grounded synthesis of the state of the art with explicit identification of research lacunae and a justified component architecture for a target system.

Rhetorical move analysis has gained popularity in research aimed at describing the rhetorical structure of academic genres, but traditional methods involving humans are sometimes subjective, time-intensive, and inconsistent. This study explores the viability of using ChatGPT for move analysis of grant proposal abstracts (GPAs). It aims to evaluate the accuracy and reliability of ChatGPT in identifying rhetorical moves in grant proposal abstracts compared to human annotation. To achieve this purpose, the research seeks answers to the questions how ChatGPT can be instructed to assist in identifying and annotating rhetorical moves in grant proposal abstracts, what types of rhetorical moves are more accurately identified by ChatGPT, and which types present challenges for it, and to what extent ChatGPT’s annotation differs from that of a human annotator. Using a corpus of Russian-language GPAs submitted to the Russian Scientific Fund, the study developed a prompt intended to guide ChatGPT in identifying rhetorical moves typical of this genre. The accuracy of AI identification was assessed in comparison with the annotation performed by a human. The results show that a carefully created prompt with detailed examples and clear definitions yields high levels of agreement with a human annotator. However, the study showed that while ChatGPT demonstrated good performance in identifying rhetorical moves, its annotations differed from the annotations performed by a human. The human annotator frequently outperformed ChatGPT, particularly in recall (the completeness of detecting all instances of a given move) for certain categories. Nevertheless, the study showed that ChatGPT can be a useful tool not just for applied linguistics researchers, but also for improving pedagogical approaches to grant writing teaching. The AI-based identification of rhetorical moves can be used to provide EAP students with examples of efficient rhetorical methods in successful grant applications, allowing them to get a better understanding of genre conventions and improve their grant writing skills.

In conflicts marked by asymmetrical power and contested legitimacy, news headlines are not neutral labels of events; they are condensed discursive sites through which media institutions can shape public understanding, sympathy and accountability. The study examines Western media reporting on the 2023 Palestinian-Israeli conflict (RPIC). It aims to reveal the linguistic and narrative strategies that are employed in media discourse to influence public understanding of the conflict. It analyzes 97 news headlines published between October 7, 2023, and February 2025 in seven major Western outlets (The Guardian, Los Angeles Times, The New York Times, The New York Post, The Washington Post, CNN, and Reuters). Two theoretical frameworks were adopted for the purpose of this study: Critical Discourse Analysis (CDA) and Narrative Framing Theory. The data analysis revealed that the examined Western media frequently employ certain discursive and narrative strategies, such as the obscuring of agency, lexical choices, narrative depersonalization, sentimentalization through the emphasis on familial roles, and emotional framing. It was also found that the use of such strategies may align with certain dominant political narratives by highlighting Israeli suffering and minimizing Palestinian suffering. The study concludes that some Western news headlines do not always describe the conflict objectively; rather, they construct certain ideological and political narratives that directly affect international understanding of the conflict. Thus, the study recommends that media discourse be critically scrutinized to identify and challenge potentially nonobjective journalistic practices. The findings contribute to CDA and narrative-framing research by showing how headline-level linguistic choices can produce unequal hierarchies of visibility, empathy, and accountability in conflict reporting.

In contemporary urban environments, public signage reflects not only language use but also broader processes of globalization, identity construction, and language commodification. This study investigates the linguistic landscape of Margonda Street, Depok, Indonesia, focusing on the representation of Indonesian, English, and other foreign languages in commercial signage. It aims to analyze how linguistic resources are displayed in urban commercial space and to identify the symbolic meanings and commercial values associated with their use. This study employed a mixed-methods case study design. Data consisted of 601 commercial signs documented through photographic observation and 40 semi-structured interviews with business owners or managers. Visual data were analyzed using the Linguistic Landscape framework, while interview data and field notes were examined through thematic analysis. The findings reveal that monolingual Indonesian signage remains the dominant form of commercial communication, followed by English, bilingual, multilingual, and hybrid signs. English was particularly visible in businesses associated with urban lifestyles, education, and services, whereas Japanese, Korean, and Arabic appeared in more specialized sectors. Interview findings indicate that language choice is influenced not only by communicative needs but also by symbolic and economic considerations. English is commonly associated with modernity, prestige, and global connectedness; Japanese with authenticity; Korean with popular culture; and Arabic with religious trust and legitimacy. The coexistence of these linguistic resources reflects a process of glocalization in which global and local influences are negotiated within a shared commercial environment. The study concludes that languages in commercial signage function simultaneously as communicative resources, symbolic capital, and economic assets. The linguistic landscape of Margonda Street demonstrates how foreign languages acquire commercial value while Indonesian continues to serve as a central marker of accessibility and local identity.

Digital book retailers often provide prospective purchasers with only a brief synopsis prior to purchase. The means by which brief texts, most often under 250 words, can simultaneously summarize, brand, promote, and persuasively present an 80.000-word novel remains under-described in the existing literature. The aim of the study is to examine how mystery/thriller synopses use semantic compression (maximising meaning per word) to meet both narrative and marketing goals. The study focuses on three main aspects: (1) the use of semantic compression to achieve persuasive effects; (2) the lexical and grammatical patterns that are characteristic of this microgenre; (3) the linguistic mechanisms behind the compression process. The research is based on a corpus of 300 (56.220 words) English Goodreads synopses (2022-2025). Phase 1 used AntConc and CLAWS-C7 tagging for length, sentence profile, lexical density, keywords, and voice/mood. Phase 2 applied double-coded discourse analysis grounded in Systemic Functional Linguistics, small-format text theory, and paratext theory, charting compression devices at macro-, meso-, and micro-levels. Synopses average 187 words, show high lexical density (52%), and prefer interrogatives and active voice. 86% follow a seven-move functional-pragmatic structure: hook, micro-portrait, temporal leap, thematic bundle, suspense question, comparative tag, sell-line. Compression is further executed via paratactic stacks, em-dash grafts, cardinality lists, modifier clusters, and systematic omission of subplots and endings. Recycled evaluatives and genre labels act as semantic shortcuts that activate readers’ genre schemata. Semantic compression is theorized as an important cognitive-discursive phenomenon functioning at the junction of literature and commerce. Book synopses are hybrid literary-marketing micro-genres whose layered compression turns narrative surplus into reader engagement. The study offers a replicable template that writers, publishers, copywriters, and automated-copy systems can use to craft more persuasive small-format texts.