Papers reviewed and determined not to be word norm studies. Use the flag icon to report errors or suggest re-inclusion.
16504 papers
Despite recent advances in digital resources for other Coptic dialects, especially Sahidic, Bohairic Coptic, the main Coptic dialect for pre-Mamluk, late Byzantine Egypt, and the contemporary language of the Coptic Church, remains critically under-resourced. This paper presents and evaluates the first syntactically annotated corpus of Bohairic Coptic, sampling data from a range of works, including Biblical text, saints' lives and Christian ascetic writing. We also explore some of the main differences we observe compared to the existing UD treebank of Sahidic Coptic, the classical dialect of the language, and conduct joint and cross-dialect parsing experiments, revealing the unique nature of Bohairic as a related, but distinct variety from the more often studied Sahidic.
Age and language experience both shape the speed of visual word recognition for children and adults. There is a considerable debate in the literature regarding whether these effects are primarily facilitating or impeding and whether the influences of age and language experience can be distinct and delineated. In order to address these questions, we collected data from Hungarian participants, analyzing data from 80 children (ages 9-17) and 387 adults (ages 18-90), on 250 words in an online visual lexical decision task. We used a pre-calibrated word list, based on prior familiarity ratings, to assess the participants' vocabulary size and compared the effects of vocabulary size, age, and years spent in education on response speed in correct lexical decision trials over real words (as opposed to filler non-words). We found that vocabulary size and education facilitate, while age impedes word recognition speed in the task, and that vocabulary size effects are mediated by both age and education. These age-related trends were observed across a broad age range, although conclusions regarding the oldest participants (70+) must remain tentative due to their limited representation in our sample.
The drive for internationalization in higher education has accelerated over the past two decades, reshaping how universities teach, collaborate, and move knowledge across borders. Central to this transformation are not only institutional partnerships and student exchange programs, but the more subtle and powerful roles of language, literature, and culture. These elements shape how students understand the world and how universities construct global learning spaces. English has taken on the role of academic lingua franca, simplifying international communication. Yet, its dominance raises complex questions about equity, inclusion, and cultural diversity. At the same time, we’re witnessing rapid developments in artificial intelligence (AI) that promise to radically alter how we think about access, communication, and language learning in higher education. Against this backdrop, the European Higher Education Area (EHEA) has set a clear goal: by 2030, at least 20% of graduates should have studied or trained abroad (European University Association, 2023). To reach this target, we must reconsider whether English alone can carry the weight of internationalization, or whether a more multilingual, AI-supported approach is needed.
Mood, an individual’s emotional state, fundamentally shapes how the brain interprets sensory input by providing a continuous affective context for prediction and evaluation. In language processing, mood may bias the interpretation of emotionally valenced words, amplifying or dampening their perceived affect. Yet, the temporal dynamics of these mood-valence interactions remain poorly understood. To clarify inconsistent evidence on the timing and nature of mood-valence interactions, we examined how induced mood influences early stages of emotional word processing using EEG. Participants performed a valence-rating task for positive, negative, and neutral words in a baseline condition and following positive or negative mood induction. Event-related potentials were analysed across early processing windows (N1, P2, EPN) using cluster-based permutation statistics. Positive mood selectively attenuated N1 amplitudes for highly valenced words, consistent with reduced prediction error under mood-congruent expectations. Later components (P2, EPN) showed decreased amplitudes for both high and neutral valence, suggesting reduced model updating under mood-congruent expectations. Negative mood, in contrast, produced weaker and temporally delayed modulations. Behaviourally, participants responded more quickly to valenced words under induced mood conditions, supporting the neural findings. Interpreted within a predictive coding framework, these results support the theoretical view that mood functions as a hyperprior, tuning the precision of predictive models during language comprehension. Positive mood appears to enhance predictive flexibility and facilitate the processing of affectively congruent words, whereas induced negative mood reduces positive affect. Taken together, the findings highlight how affective states dynamically modulate early predictive mechanisms in emotional language processing.
State-Space Models (SSMs) have emerged as efficient alternatives to computationally intensive architectures like Transformers, particularly for sequence modeling. However, a fundamental challenge in their training is the reliance on static loss functions, which may not be optimal across all learning stages. To address this issue, in this paper a hybrid model integrating the Hyena architecture with a Dynamic Loss Network (DLN) is proposed which is guided by a Learn-to-Teach (L2T) approach (L2T-DLN). In this framework, the Hyena model is a student, and its loss function is optimized adaptively. A teacher model, leveraging a memory of the student's past performance, guides the DLN in dynamically balancing the primary cross-entropy loss and a regularization term. Experiments on the Penn Treebank (PTB) dataset show that our approach significantly improves language modeling performance. Our proposed model achieved a validation Perplexity of 102.6, a notable improvement over the 110.4 achieved by a baseline Hyena model using a static loss function. This research indicates that combining SSMs with adaptive loss function markedly enhances the quality and efficiency of deep learning models for sequential data, showing potential for applications in Natural Language Processing (NLP), time-series analysis, and biological signal processing.
Large language models (LLMs) have reignited debate about whether machines without minds or intentions can genuinely participate in linguistic practice. Critics portray them as ‘stochastic parrots’ that manipulate form without meaning, whereas defenders emphasize their impressive functional capacities. This paper argues that these disputes conflate distinct dimensions of meaning and agency. I extend Huw Price’s distinction between i-representation and e-representation (roughly, inferential versus environment-tracking types of representation) by differentiating physical e-representation—such as a fuel gauge, grounded in causal coupling—from symbolic e-representation, exemplified in language and mediated by agents. This refinement clarifies what is at issue: LLMs clearly display i-representational competence through their participation in inferentially structured discourse. Whether their outputs possess symbolic e-representational content, however, is contested and framework-relative. It depends on whether agent-mediated uptake is taken to suffice, or whether additional grounding conditions—such as intentions, causal connections, or proper functions—are required. I further distinguish norm-sensitivity—the capacity to track and adapt to linguistic norms, which grounds their i-representational competence—from norm-responsibility, the reflexive capacity to own commitments and bear accountability. Technical analysis of LLM architectures shows that they exhibit advanced norm-sensitivity through statistical learning but entirely lack norm-responsibility. LLMs thus occupy a distinctive position: they are genuine functional participants in linguistic practices, yet fall short of the reflexive agency characteristic of responsible speakers.
Abstract This paper presents a novel framework for modeling role and task allocation in cooperative wheeled soccer robot systems by leveraging latent knowledge extracted from past collaborative interactions. Inspired by recent advances in heterogeneous multi-robot collaboration, the proposed method encodes a soccer team as a set of Multidimensional Relational Structures (MDRSs), capturing both temporal and spatial relations among robot roles, actions, and stimuli. A structured dataset, termed the Soccer Robot Collaboration Treebank (SRCT), is introduced to represent play-by-play histories of robot behaviors, parsed through a formal grammar to support structured learning. Probabilistic modeling and Non-Negative Tensor Decomposition (NTD) are applied to the resulting tensors, enabling robust inference and latent knowledge estimation even in scenarios with sparse data or communication loss. Simulated experiments using a team of wheeled soccer robots in the Webots environment demonstrate the system’s ability to dynamically reassign roles, reason over incomplete histories, and predict collaborative behaviors such as passing, defending, or role-switching. The results show that the proposed framework enhances both strategic flexibility and robustness, providing a foundation for real-time decision-making in robotic soccer under uncertainty.
The article is devoted to studying socially determined linguistic processes, which are traditionally associated with the broad problem of social variation of communication (discourse) and linguistic variability of the German language. It presents the results of a study conducted within the framework of cognitive sociolinguistics - the linguistics of social meanings. The author observes continuity in the development of scientific thought, explores the problem of linguistic variability in two modes - theoretical and applied. The subject of the research is the terminological apparatus and specific linguistic facts that are based on the cognitive, communicative and social functions of language, that is, to be the reality of the thought of individual and/or collective knowledge of representatives of a certain society and the means of their communication. The purpose of the article is to analyze theoretical propositions, linguistic terms and existing specific language forms that convey social meanings, marked by a social feature in their terminological interpretations with a focus on the linguistic picture of the Germany new lands. The novelty of the research lies in cognitive-semantic analysis, systematization and modeling this phenomenon in the social aspect. As a result, the author comes to her own conclusions, which lead to understanding and rethinking the old views on the problems of dialectology in the context of modern realities of the German language society and the data of modern linguistics. The methodology of the research and the description of its results are determined by the principle of interdependence of the three most important didactic and linguistic strata - the study of modern language from the standpoint of linguistic norms, the study of linguistic variability and the analysis of language change in the framework of its historical development. General scientific methods and special methods of cognitive linguistics are used to analyze theoretical material and linguistic facts, including explanatory description, interpretation, cognitive modeling, cognitive dominance, and focusing.
This paper examines how social media discourse affects the learning of the English language on the tertiary level in Lahore and how informal online communication influences the academic English of students in Lahore. The qualitative design was employed to gather data on the basis of semi-structured interviews with English language learners and teachers. Braun and Clarke’s (2006) framework was used to conduct thematic analysis on the transcribed data. The results indicate that though social media provides appropriate exposure to vocabulary, pronunciation, and language use in real life, the casualness of linguistic norms has a very strong impact on the students’ academic writing and their communicative accuracy. The participants claimed to use short forms, abbreviations, slang and mixed-language texting on a regular basis, and these transferred to essays and presentations. Other themes were distraction and a lack of studying discipline, the inability to stick to the formal register, and the misunderstanding of words acquired in unconfirmed online situations. Teachers also reported on the same lines, as they observed poor writing standards in academic writing and increased dependence on social-media-driven language patterns. Pedagogical mechanisms to counteract these effects were also determined in the study with the focus being on register awareness, purposeful digital task integration, and curriculum modernization. On the whole, the study has determined that the power of social media is twofold, both positive when moderated and negative when uncontrolled and recommends that informed teaching and learning methods are needed to help students balance between informal online communication and formal academic language.
The article explores irony as a complex linguistic and pragmatic phenomenon that plays a significant role in creating satirical effect within the sketch genre. The object of analysis is the sketch «Les Flics» by French comedian Coluche, in which irony functions as a tool of social critique, particularly through the ridicule of flaws within the institution of police. The study identifies the main linguistic, stylistic, and pragmatic means used to construct irony. It demonstrates that the central communicative strategy shaping the critical attitude toward the police system is the deliberate provocation of the audience and the emphasis on the absurdity of the depicted social conditions. The comic effect arises from the contrast between socially expected norms of behavior (including verbal conduct) and the reality represented by the narrator-character. Stylistic devices such as antiphrasis, metaphor, metonymy, grotesque, and sarcasm, along with linguistic features such as colloquial register, slang, and violations of linguistic norms, are shown to contribute to the portrayal of deep linguistic and social deviation in the police character. The article draws on contemporary linguistic theories, including pragmatics, speech act theory, polyphony theory, and relevance theory. It concludes that irony in the sketch is both staged and situational, with the comic effect emerging from the conflict between expectation and reality. Special attention is given to implicature, polyphonic structure of the utterance, and the satirical function of the narrator. The study shows that irony not only enhances the aesthetic dimension of the text but also performs a socially critical function by shaping a discourse of resistance. Future research directions include the analysis of other Coluche sketches in the context of linguistic critique of society.
Lexical Semantic Change (LSC) provides insight into cultural and social dynamics. Yet, the validity of methods for measuring different kinds of LSC remains unestablished due to the absence of historical benchmark datasets. To address this gap, we propose LSC-Eval, a novel three-stage general-purpose evaluation framework to: (1) develop a scalable methodology for generating synthetic datasets that simulate theory-driven LSC using In-Context Learning and a lexical database; (2) use these datasets to evaluate the sensitivity of computational methods to synthetic change; and (3) assess their suitability for detecting change in specific dimensions and domains. We apply LSC-Eval to simulate changes along the Sentiment, Intensity, and Breadth (SIB) dimensions, as defined in the SIBling framework, using examples from psychology. We then evaluate the ability of selected methods to detect these controlled interventions. Our findings validate the use of synthetic benchmarks, demonstrate that tailored methods effectively detect changes along SIB dimensions, and reveal that a state-of-the-art LSC model faces challenges in detecting affective dimensions of LSC. LSC-Eval offers a valuable tool for dimension- and domain-specific benchmarking of LSC methods, with particular relevance to the social sciences.
Introduction Researchers working in the field of cognitive aging frequently encounter highly motivated yet nervous older participants during data collection in the laboratory. Such anecdotal experiences raise the question of whether the affective or physiological response of older participants to psychological laboratory experiments differs to that of young adults, who might be less motivated but also less nervous, as they may be more used to the environment and to learning and memory tests. Methods In the present study, we collected saliva samples and subjective affective ratings during an EEG experiment on memory, and at home, in young and older adults, while also taking into account sex effects. Results There was no significant interaction involving time point (laboratory vs. at home) and age group. However, across both time points older males showed significantly higher cortisol-levels than older females, while there was no difference for younger males and females. The trajectories in cortisol levels throughout the session, especially around the memory task, differed by age: While there was a decrease in cortisol levels for younger adults from before to after the memory task, we did not observe such a decrease in older participants. There were few age differences in alpha-amylase or negative affect. However, older adults showed higher ratings of positive affect than younger participants. Importantly, lower cortisol levels before the memory task were associated with higher associative memory performance for older adults. Discussion Affective reactions to psychological laboratory tasks may hence be an important factor to consider in psychological experiments in the field of cognitive aging.
This study reviews the English language test of Singapore’s Primary School Leaving Examination, a high-stakes national assessment taken annually by nearly all primary six students for secondary school placement. Given the test’s importance in shaping students’ academic pathways and recent format changes, it is crucial to evaluate its validity, specifically its ability to provide accurate and fair assessments of students’ English language proficiency and academic readiness. The review outlines the test’s educational and policy context, followed by a description of the latest formats for both the English language and foundation English language versions. The analysis focuses on core dimensions of test validity, including content representativeness, construct validity, criterion-related validity (concurrent and predictive), and reliability (inter-rater reliability and internal consistency). Drawing on official documents and limited empirical studies, the review finds moderate improvements in content representativeness and construct validity. However, both longstanding and emerging concerns (e.g., the exclusion of local linguistic norms and genre scope) indicate that key limitations remain. While predictive validity, inter-rater reliability, and internal consistency appear supported, empirical research remains sparse across all reviewed test qualities, particularly in concurrent validity. The review integrates identified research gaps and proposes inquiry directions to inform future test development and policy adaptation. Strengthening the evidence base is essential for ensuring a valid, reliable, and equitable assessment system in Singapore’s primary education landscape.
This paper presents the structure and principal components of the linguistic resources required for sentiment analysis in the Uzbek language. The research aims to identify and develop effective approaches for constructing a linguistic database - referred to as SentiUzNet - and to establish a foundational sentiment lexicon tailored specifically to the characteristics of the Uzbek language. In particular, the paper discusses key principles for annotating words with sentiment polarity and subjectivity scores, as well as methodological foundations for building a lexicographic database to support automated emotional analysis of texts. A significant part of the research focuses on experimenting with large-scale user-generated content, specifically social media comments written in Uzbek. These datasets were used to train and evaluate sentiment analysis models, thereby allowing an assessment of their performance and practical applicability. The results of this research represent one of the first comprehensive attempts to facilitate automatic sentiment detection in the Uzbek language and are expected to contribute substantially to the advancement of natural language processing technologies in under-resourced linguistic settings.
Deep neural networks employ specialized architectures for vision, sequential and language tasks, yet this proliferation obscures their underlying commonalities. We introduce a unified matrix-order framework that casts convolutional, recurrent and self-attention operations as sparse matrix multiplications. Convolution is realized via an upper-triangular weight matrix performing first-order transformations; recurrence emerges from a lower-triangular matrix encoding stepwise updates; attention arises naturally as a third-order tensor factorization. We prove algebraic isomorphism with standard CNN, RNN and Transformer layers under mild assumptions. Empirical evaluations on image classification (MNIST, CIFAR-10/100, Tiny ImageNet), time-series forecasting (ETTh1, Electricity Load Diagrams) and language modeling/classification (AG News, WikiText-2, Penn Treebank) confirm that sparse-matrix formulations match or exceed native model performance while converging in comparable or fewer epochs. By reducing architecture design to sparse pattern selection, our matrix perspective aligns with GPU parallelism and leverages mature algebraic optimization tools. This work establishes a mathematically rigorous substrate for diverse neural architectures and opens avenues for principled, hardware-aware network design.
Knowledge distillation (KD) is a widely adopted technique for compressing large models into smaller, more efficient student models that can be deployed on devices with limited computational resources. Among various KD methods, Relational Knowledge Distillation (RKD) improves student performance by aligning relational structures in the feature space, such as pairwise distances and angles. In this work, we propose Quantum Relational Knowledge Distillation (QRKD), which extends RKD by incorporating quantum relational information. Specifically, we map classical features into a Hilbert space, interpret them as quantum states, and compute quantum kernel values to capture richer inter-sample relationships. These quantum-informed relations are then used to guide the distillation process. We evaluate QRKD on both vision and language tasks, including CNNs on MNIST and CIFAR-10, and GPT-2 on WikiText-2, Penn Treebank, and IMDB. Across all benchmarks, QRKD consistently improves student model performance compared to classical RKD. Importantly, both teacher and student models remain classical and deployable on standard hardware, with quantum computation required only during training. This work presents the first demonstration of quantum-enhanced knowledge distillation in a fully classical deployment setting.
<span lang="EN-US">The illicit act of appropriating programming code has long been an appealing notion due to the immediate time and effort savings it affords perpetrators. However, it is universally acknowledged that concerted efforts are imperative to identify and rectify such transgressions. This is particularly crucial as academic institutions, including universities, may inadvertently confer degrees for work tainted by this form of plagiarism. Consequently, the primary objective of this research is to scrutinize the feasibility of identifying plagiarism within pairs of Verilog algorithms and texts. this study aims to detect plagiarism in textual content and Verilog code by leveraging diverse linguistic characteristics from the WordNet lexical database. The primary objective is to achieve optimal accuracy in identifying instances of plagiarism, incorporating features such as modifications to text structure, synonym substitution, and simultaneous application of these strategies. The system's architecture is intricately designed to unveil instances of plagiarism in both textual content and Verilog code by extracting nuanced characteristics. The systematic process includes preprocessing, detailed analysis, and post-processing, supported by a feature-rich database. Each entry in the database represents a distinctive similarity case, contributing to a thorough and comprehensive approach to plagiarism detection.</span>
Multilingual Large Language Models (LLMs) have shown remarkable performance across various languages; however, they often include significantly less data for low-resource languages such as Urdu compared to high-resource languages like English. To assess the linguistic knowledge of LLMs in Urdu, we present the Urdu Benchmark of Linguistic Minimal Pairs (UrBLiMP) i.e. pairs of minimally different sentences that contrast in grammatical acceptability. UrBLiMP comprises 5,696 minimal pairs targeting ten core syntactic phenomena, carefully curated using the Urdu Treebank and diverse Urdu text corpora. A human evaluation of UrBLiMP annotations yielded a 96.10% inter-annotator agreement, confirming the reliability of the dataset. We evaluate twenty multilingual LLMs on UrBLiMP, revealing significant variation in performance across linguistic phenomena. While LLaMA-3-70B achieves the highest average accuracy (94.73%), its performance is statistically comparable to other top models such as Gemma-3-27B-PT. These findings highlight both the potential and the limitations of current multilingual LLMs in capturing fine-grained syntactic knowledge in low-resource languages.
Emojis are widely used in digital communication to convey emotional cues alongside text, yet their impact on word-level reading within sentence contexts remains unclear. We conducted an eye-tracking experiment to examine how positive (e.g., 🤩) versus neutral (e.g., 🧑🦳) face emojis embedded mid-sentence in otherwise neutral sentences affect the processing of the preceding and following words (e.g., positive “Did you change your hair 🤩 something is different” vs. neutral “Did you change your hair 🧑🦳 something is different”). We observed robust parafoveal-on-foveal (PoF) effects on the n–1 word, with longer fixations in first-fixation, gaze duration, and single-fixation measures when the parafoveal emoji was positive rather than neutral. This valence effect persisted even after accounting for mislocated fixations, suggesting that positive emotional content genuinely modulates foveal word processing. In contrast, the n+1 word showed no valence-based facilitation, implying that the influence of a positive mid-sentence emoji does not extend to subsequent words in continuous reading. At the sentence level, positive emojis were associated with faster overall reading times and higher valence ratings, although dashed (no-emoji) sentences in the pre-test were rated more positively than emojified versions in the experiment. These findings reinforce models of eye movement control that allow parallel processing of foveal and parafoveal information, highlighting how affective face emojis can shape real-time reading dynamics.
Word Sense Disambiguation (WSD) is a fundamental task in Natural Language Processing (NLP), addressing the challenge of identifying correct word meanings in context. This task is particularly complex for morphologically rich and resource-limited languages like Hindi, which exhibit significant lexical ambiguity compounded by limited availability of annotated corpora. To address these challenges, we propose a supervised approach combining the multilingual BERT model (mBERT) with Hindi WordNet as a structured lexical resource. Using few-shot learning, we fine-tune mBERT on a dataset constructed from Hindi WordNet to disambiguate contextually ambiguous words across four parts of speech (POS): nouns, verbs, adjectives, and adverbs. Experiments on standard Hindi WSD benchmarks demonstrate that our method significantly outperforms traditional rule-based and embedding-based approaches, achieving 96.48% accuracy—an approximate 3% improvement over the strongest baseline. These results validate the effectiveness of integrating contextualized embeddings from pre-trained language models with structured lexical databases, highlighting the promise of hybrid techniques for advancing WSD in low-resource languages and providing a framework applicable to other morphologically complex languages with similar resource constraints.
The linguistic features of the Uzbek language - complex agglutinative morphology, free word order, and limited resources - necessitate a specialized approach and thorough research in the application of morphological and syntactic methods. Within the framework of the study, morphological analysis methods and syntactic analysis methods are reviewed based on scientific sources. Each section presents the existing advantages and disadvantages, experience of their use in the Uzbek language, as well as a comparative analysis with foreign languages. Rule-based methods, statistical models (HMM, CRF, etc.), Neural network-based approaches (BiLSTM-CRF, seq2seq) of morphological analysis in the Uzbek language are discussed, and the results are given in examples and percentages. It is shown that syntactic parsing is implemented using dependency and constituency parsing analysis methods. The issue of building a UD treebank for the Uzbek language with SOV order is considered. The impact of complex morphological structure and free word order in sentences on the construction of parsers is highlighted. As a result of the studied approaches, the issue of building hybrid parsers, integrating them with morphological analysis and assigning grammatical categories of words to the parser is raised. Also, the development of neural constituency parsers based on neural networks and the effectiveness of the results obtained from them are analyzed.
The obligatory use of third-person honorifics is a distinctive feature of several South Asian languages, encoding nuanced socio-pragmatic cues such as power, age, gender, fame, and social distance. In this work, (i) We present the first large-scale study of third-person honorific pronoun and verb usage across 10,000 Hindi and Bengali Wikipedia articles with annotations linked to key socio-demographic attributes of the subjects, including gender, age group, fame, and cultural origin. (ii) Our analysis uncovers systematic intra-language regularities but notable cross-linguistic differences: honorifics are more prevalent in Bengali than in Hindi, while non-honorifics dominate while referring to infamous, juvenile, and culturally exotic entities. Notably, in both languages, and more prominently in Hindi, men are more frequently addressed with honorifics than women. (iii) To examine whether large language models (LLMs) internalize similar socio-pragmatic norms, we probe six LLMs using controlled generation and translation tasks over 1,000 culturally balanced entities. We find that LLMs diverge from Wikipedia usage, exhibiting alternative preferences in honorific selection across tasks, languages, and socio-demographic attributes. These discrepancies highlight gaps in the socio-cultural alignment of LLMs and open new directions for studying how LLMs acquire, adapt, or distort social-linguistic norms. Our code and data are publicly available at https://github.com/souro/honorific-wiki-llm
Abstract The age-related positivity bias refers to the finding that older adults recount events more positively (or less negatively) as compared to younger adults (i.e., a main effect of age on memory valence). This bias is closely related to the positivity effect, which reflects an interaction between age and valence of information to be remembered. We examined the age-related positivity bias and positivity effect using a one-year longitudinal design with a sample that spanned adulthood (N = 374; age range 19-90; M= 47.41; SD= 16.75). Participants answered questions regarding their memories of learning about the outcome of the 2020 U.S. presidential election. Analyses examined the association between age and valence ratings (positive, negative) and ratings of feelings (happy, elated, upset, and shaken) at Time 1, as well as the association with age between change scores for each of those variables, while controlling for who the participant voted for in the election. Results indicate that increased age was associated with reporting feeling less negative at the time of the event, and also remembering feeling more positive (elated and happy) when reconstructing the event one year later, thereby providing evidence of the positivity bias. There was no evidence of an age by valence interaction in a 2 (Valence) x 3 (Age) mixed ANCOVA on the positive and negative change scores, indicating there was not a positivity effect. Depressive symptoms partially mediated the relationship between age and valence variables, indicating that depressive symptoms may be one mechanism for explaining the age-related positivity bias.
Background: Many previous studies highlighting a relationship between depression and emotional face recognition have relied on measures of classification accuracy to determine recognition deficits. However, the perception of emotions is also related arousal levels and valence, and more research is needed to determine how depression impacts these dimensions.Aims: To compare performance on both an objective forced choice emotional recognition task and subjective emotional face valence rating task in participants with self-reported high depression.Methods: Based on screening using the depression sub-scale of the DASS-42, 46 participants (23 males, 23 female) were in the high depression group (mean DASS-42 34±5) and 50 participants in the control groups (25 males, 25 females) with DASS-42 scores of either 0 or 1. All participants completed both a performance-based task (objective) as well as a rating task (subjective) of emotional facial expressions. Results: The data indicate that difference in performance exist in classification accuracy between the groups, with depressed participants demonstrating reduced accuracy for anger, sadness and neutral facial expressions. Additionally differences in subjective ratings exist in the depressed group, but with the important caveat that these only relate to faces display positive emotional expressions.Discussion: The limitations of relying solely on objective tasks where recognition accuracy is the main outcome measure are discussed as well as the data quantitatively demonstrating a reduced response in the depression group to positive stimuli. This study justifies the need for future studies using both objective and subjective measures to assess emotion classification deficits in depression.
BACKGROUND AND OBJECTIVES: It is well documented that the fear of specific stimuli and situations can be acquired through the social observation of the actions of another person. In contrast, it is still a matter of debate, whether processes related to fear attenuation, extinction, and extinction-retrieval can equally be achieved through social observation after de novo fear conditioning. METHODS: Here, we used a differential fear conditioning procedure and investigated whether the variation of the context of video-based vicarious extinction learning (VEL) will affect subsequent extinction learning and extinction-retrieval. Conditioned fear acquisition, extinction, and extinction-retrieval was measured using psychophysiological (skin conductance responses) and subjective measures (CS-UCS contingency ratings and CS-valence ratings). RESULTS: Participants showed enhanced fear extinction learning after VEL as compared to controls. VEL improved extinction learning relative to controls but appeared to be highly context-dependent. The beneficial effect of VEL on subsequent extinction learning was abolished when the context in which the model was performing in the video was different from the context in which the observer performed all stages of the experiment. LIMITATIONS: Data were obtained in a non-clinical sample which does not permit the extrapolation of findings to clinical populations. CONCLUSION: Our results suggests that safety information derived from VEL promotes fear extinction when model and observer perform the experiment in the same context. Given that fear extinction is considered as an experimental proxy of exposure therapy, our findings might be instructive for the development of novel clinical interventions to promote exposure treatment efficacy.
Emojis are widely used in digital communication to convey emotional cues alongside text, yet their impact on word-level reading within sentence contexts remains unclear. We conducted an eye-tracking experiment to examine how positive (e.g., 🤩) versus neutral (e.g., 🧑🦳) face emojis embedded mid-sentence in otherwise neutral sentences affect the processing of the preceding and following words (e.g., positive “Did you change your hair 🤩 something is different” vs. neutral “Did you change your hair 🧑🦳 something is different”). We observed robust parafoveal-on-foveal (PoF) effects on the n–1 word, with longer fixations in first-fixation, gaze duration, and single-fixation measures when the parafoveal emoji was positive rather than neutral. This valence effect persisted even after accounting for mislocated fixations, suggesting that positive emotional content genuinely modulates foveal word processing. In contrast, the n+1 word showed no valence-based facilitation, implying that the influence of a positive mid-sentence emoji does not extend to subsequent words in continuous reading. At the sentence level, positive emojis were associated with faster overall reading times and higher valence ratings, although dashed (no-emoji) sentences in the pre-test were rated more positively than emojified versions in the experiment. These findings reinforce models of eye movement control that allow parallel processing of foveal and parafoveal information, highlighting how affective face emojis can shape real-time reading dynamics.
The article examines speech culture as a key component of language competence among higher education students. The author emphasizes that mastering the norms of the literary language, adhering to ethical and stylistic standards in communication, is an indicator not only of a person’s general education but also of their readiness for professional and social interaction. The main components of speech culture are analyzed, including accuracy, clarity, logic, appropriateness, purity, expressiveness, and aesthetic quality of speech. Particular attention is paid to common violations of linguistic norms observed in the student environment: the use of colloquial, slang, and foreign words without necessity, unjustified calques, bureaucratic expressions, as well as syntactic and orthoepic errors. The article outlines the main causes of linguistic carelessness, such as low reading culture, the influence of social media, and the decline in linguistic standards in everyday and educational communication. The author proposes a number of pedagogical and methodological strategies aimed at cultivating a high level of speech culture among students. These include the integration of communicative training into the educational process, regular involvement of students in stylistic text analysis, and the activation of creative language practices. Examples of typical speech situations are provided to demonstrate the contrast between cultured and uncultured language use, highlighting the importance of speaker selfreflection in improving overall language competence. The relevance of the study is due to the growing importance of speech culture in the modern educational environment, where effective communication is a key component of a specialist’s professional training.
This article attempts to investigate the complicated relationship of form of language, narrative vagueness, and cultural interpretation within Frank Stockton’s “The Lady, or the Tiger?” Applying text linguistic methods to the analysis of this classic short story, the researcher tackles two research questions, specifically, how does the linguistic structure of the narrative reinforce the uncertain ending of “The Lady, or the Tiger?” What are the effects of cultural and language influences on how the reader interprets the important characters and themes of “The Lady, or the Tiger?” Grounded in theoretical frameworks from Halliday and Hasan’s cohesion and coherence, and more recent cognitive linguistics, the paper addresses how the deployment of ambiguity within language functions as a reader engagement and theme exploration strategy. The research concludes that the narrative’s linguistic construction—defined by its deliberate plot of unresolvable conflict, indirect characterization, and sophisticated temporalities—is a significant factor in the development of the legendary vagueness of the story. The research further contends that cultural contexts and linguistic norms governing the reader’s interpretation heavily influence the apparently moral, motivational, and destined lives of characters and thus the broader themes of justice, choice, and human nature. Finally, the paper concludes that “The Lady, or the Tiger?” illustrates the outstanding power of language in creating narrative doubt and highlights the importance of cultural spectacles in the reception and interpretation of literature. In this sense, the story illustrates the general implications of text linguistics in the understanding of narrative ambiguity and cultural impact.
The nouns of our language refer to either concrete entities (like a table) or abstract concepts (like justice or love), and cognitive psychology has established that concreteness influences how words are processed. Accordingly, understanding how concreteness is represented in our mind and brain is a central question in psychology, neuroscience, and computational linguistics. While the advent of powerful language models has allowed for quantitative inquiries into the nature of semantic representations, it remains largely underexplored how they represent concreteness. Here, we used behavioral judgments to estimate semantic distances implicitly used by humans, for a set of carefully selected abstract and concrete nouns. Using Representational Similarity Analysis, we find that the implicit representational space of participants and the semantic representations of language models are significantly aligned. We also find that both representational spaces are implicitly aligned to an explicit representation of concreteness, which was obtained from our participants using an additional concreteness rating task. Importantly, using ablation experiments, we demonstrate that the human-to-model alignment is substantially driven by concreteness, but not by other important word characteristics established in psycholinguistics. These results indicate that humans and language models converge on the concreteness dimension, but not on other dimensions.
This study explores the syntactic network characteristics of English e-commerce live-streaming discourse by employing a syntactic treebank and syntactic complex network analysis. The main findings are: (1) The syntactic network of English e-commerce live-streaming discourse exhibits small-world and scale-free properties, which are hallmark traits of complex networks. (2) The central nodes of the network are be, I, and the, with be serving as the most central node, while I and the act as local central nodes. (3) The central node be demonstrates both strong centrifugal and centripetal forces. Its centrifugal force is most frequently associated with subject relations and adjective complements, while its centripetal force is characterized by auxiliary and clausal complements. These findings indicate that the syntactic structure of English e-commerce live-streaming discourse is highly robust. This robustness underscores the discourse’s functional purpose: to convey information clearly while engaging users through personalization and specificity. Furthermore, the study highlights the critical role of be in attributive and descriptive constructions. Overall, this research provides insights into the syntactic organization of e-commerce discourse and demonstrates the effectiveness of complex network analysis in linguistic studies.
This paper examines the phonetic value of the semivowel in Gavrilo Stefanović Venclović’s Služabna knjiga. The work was written between 1711 and 1716 in the Old Church Slavonic language using old Cyrillic script. The script is of the semi-uncial type with elements of cursive writing. The analysis is based on photographs of the manuscript (РГБ, Собрание Н. П. Румянцева Ф. 256 № 401). The text was transcribed using the Transkribus software platform, while most of the excerpted examples were processed using the AntConc software. The main findings of the research can be summarized as follows: (a) semivowel signs and the apostrophe at the end of a word have a purely orthographic function (e.g. даръ, хотѣщимь, готов); (b) as in the vernacular, in the Serbian Slavonic language epoch, the semivowel in a strong position generates the reflex a (e.g. вѣнацъ, диванъ, четвртакъ); (c) the secondary semivowel is vocalized as a (e.g. оганъ, петарь, жизан), or not recorder at all (e.g. жизнъ, огнъ, помыслъ); (d) the semivowel in a weak position, according to the rules of the Serbian Slavonic linguistic norm, is pronounced as a within the prepositions въ (e.g. въ бꙑтїе), съ (e.g. съ нами), and къ (e.g. къ г҃ꙋ), prefixes въ- (e.g. въходиⷮ), въз- (e.g. въⷥдиханїе), and съ- (e.g. съдѣлавъ), and the root въс- (e.g. въсакаа), as evidenced by examples where the letter а appears in place of the former weak position semivowel (e.g. васака, множаствѣ, саблюденїе)
The study examines how children, parents and staff in a kindergarten create a social space in the kindergarten’s cloakroom through (linguistic) actions and language choices. Based on one year of ethnographic fieldwork, which includes participant observation, documentation of the kindergarten’s semiotic landscape, field conversations and interviews with staff, the study shows how the cloakroom becomes a site for multilingual practices, while the kindergarten otherwise is dominated by a Norwegian language norm. The study demonstrates how children and adults take ownership of this space and create a multilingual environment through their actions. The analysis is grounded in Lefebvre’s theory of the production of social space – through spatial practice, representations of space and lived space – Gadamer’s perspectives on play and Pascual-de-Sans’ concept of idiotopy. The findings reveal that language choices and the construction of the cloakroom as a social space are influenced by complex processes related to place, time and agency. When the cloakroom is less in focus for the staff, it can become an important site for children’s play, where they negotiate both linguistic norms and other rules. The study argues for the significance of such social spaces as part of the kindergarten’s linguistic environment, where children can take ownership and make language choices that deviate from established majority language norms. At the same time, the study highlights the importance of time in research on language and place, both theoretically and methodologically, as well as the roles of researchers in gaining access to such spaces through invitations from the children.
Despite the growing use of NLP in second language (L2) research, model accuracy in L2 settings remains underexplored. This study addresses this gap by evaluating and fine-tuning a Korean language model to extract morphosyntactic features (i.e., morpheme tokenization/tagging and dependency parsing) from L2-Korean texts. We begin by evaluating a domain-general Korean language model on a gold-annotated L2-Korean treebank. We then fine-tune the model on L2-Korean data and quantify the resulting gains across diverse L1- and L2- datasets. Finally, we examine how model reliability varies with learner proficiency scores. Three key findings emerge: while the domain-general model excels at morpheme tokenization, it underperforms on morpheme tagging and dependency parsing; fine-tuning substantially improves adaptability to L2 morphosyntax; and proficiency has minimal effect on morpheme-level tasks but significantly affects dependency-parsing reliability. These results highlight the importance of incorporating L2 training data to improve morphosyntactic analysis in L2 settings and caution against uncritical reliance on automated dependency annotations, especially when performance varies across proficiency levels.
Large language models (LLMs) have gained a lot of attention and achievements recently because of their significant comprehension and generative abilities. However, the large-scale parameters of LLMs require considerable computational resources in the training and inference process, which restricts their wide application. To overcome this challenge, we propose an efficient mixed precision weight quantization (EMWQ) method for LLMs in this article. Specifically, we introduce a new outlier detection method by analyzing the weight distribution instead of the conventional weight magnitude. Then, we propose a dual-quantization strategy that quantizes both the outlier critical columns and the residual matrices with different precision. Besides, we introduce two effective EMWQ-based application frameworks, the EMWQ-R and EMWQ-O in our study. Comprehensive experiments are conducted on the Penn Treebank (PTB), C4, ARC-Easy datasets, and MMLU benchmark across various tasks. The comparison results demonstrate that the proposed EMWQ achieves state-of-the-art performance in mixed precision quantization and further reduces computational memory cost. Besides, it has higher generalizability compared with conventional methods.
The article discusses multi-senses connectives and their annotation in text corpora.The author clarifies the concept, which is usually applied to three different phenomena: i) the uncertainty of annotators in choosing one of the meanings of a polysemic connector; ii) the possibility of establishing more than one relation, both explicit and implicit, between two fragments of text; iii) the "combination" of several meanings by a connective.The author then considers how these cases are annotated in the Penn Discourse Treebank and in the Supracorpora Database of Connectives, which use a multi-label annotation for discourse relations.The study also raises a number of theoretical questions: what information can we obtain from double labels, which relations are distributionally close, i.e. can be established in the same contexts, how to separate the contribution of the connective and of the context to the overall meaning of a sequence of sentences, how legitimate is it to talk about the combination of meanings by the connective, how to annotate the features of the context of the connectives so that this data can be used for Natural Language Processing?The solution to these questions is important for both theoretical cognitive and applied research (in particular, for machine learning and machine translation).
This paper presents a real-time American Sign Language (ASL) recognition system utilizing a hybrid deep learning architecture combining 3D Convolutional Neural Networks (3D CNN) with Long Short-Term Memory (LSTM) networks. The system processes webcam video streams to recognize word-level ASL signs, addressing communication barriers for over 70 million deaf and hard-of-hearing individuals worldwide. Our architecture leverages 3D convolutions to capture spatial-temporal features from video frames, followed by LSTM layers that model sequential dependencies inherent in sign language gestures. Trained on the WLASL dataset (2,000 common words), ASL-LEX lexical database (~2,700 signs), and a curated set of 100 expert-annotated ASL signs, the system achieves F1-scores ranging from 0.71 to 0.99 across sign classes. The model is deployed on AWS infrastructure with edge deployment capability on OAK-D cameras for real-time inference. We discuss the architecture design, training methodology, evaluation metrics, and deployment considerations for practical accessibility applications.
The article explores orthographic interference in the process of improving the normative tools of national writing. Orthographic interference is a linguistic phenomenon that arises from the interaction of different language systems, manifesting as deviations or violations of writing norms. The study analyzes changes and adaptations of linguistic norms and their impact on the writing skills of language learners. It focuses on identifying linguistic mechanisms to improve the national writing system by preventing and reducing orthographic interference. The study employs content analysis, comparative methods, and qualitative analysis. Over 60 students’ written works were examined to determine the frequency, typology, and causes of orthographic errors. The analysis identified common types of interlingual and intralingual interference. Interlingual interference results from the influence of Russian and English graphic, phonological, and morphological features, while intralingual interference arises from inadequate understanding of the phonetic and phonological foundations of the language. Content analysis identified the most frequent orthographic errors, with primary causes including insufficient mastery of spelling rules, differences between native and target language writing systems, and teaching method shortcomings. These errors serve as indicators of students’writing experience and language proficiency. The identified types of interference contribute to improving normative tools, assessing the effectiveness of educational programs and methodologies, standardizing orthographic norms, and enhancing writing culture in multilingual societies. The findings support the codification of Kazakh orthographic norms and the development of a scientific and methodological foundation for reducing linguistic interference.
a Real close relationship exists between transferring and public media, imposing a noticeable light on the importance of language especially translation in casting discourse media since translation goes on an essential role in representing the diversity between cultures and perspectives in media, however translation may distorting the culture disparities and context unless not managed cautiously,in this sense, translators have to create a balance between loyalty to the original texts and the target ones taking into consideration the diversity in linguistic and cultural norms, this balance can be molded by applying certain approaches and strategies, this article aims to present theoretical and analytical paths to explore the diverse techniques and mechanisms used in transferring media discourse which are important to be adopted in the process of translation given that translation in the media can play a really malicious role in spreading misinformation and tendentious narratives, faulty or biased translation effects the public perception dramatically and fabricates stereotypes of misconceptions, accordingly, this study deals with real world media discourse and examines the strategies adopted by the translators, the findings show that the process of transferring from one language to another comprises modifying the linguistic norms and cultural backgrounds,spotting the light on power and responsibility that emerge during the process of translation since societies become increasingly interconnected so it is an urgent matter for translators and journalists to maintain ethical, cultural and linguistic standards to preserve accuracy and solidity of translated content and to protect the public perception from distorted narratives and misinformation as much as possible.
This study examines the linguistic landscape (LL) of Chikan Old Street, a historic district in Zhanjiang, Guangdong, through the lens of the SPEAKING model. As one of the most well-preserved historical districts in southern China, Chikan Old Street embodies a rich maritime heritage and commercial traditions, making it a compelling site for LL research. The study investigates the interaction between official language policies, regional linguistic identity, and globalization, providing insights into how language hierarchies are constructed in heritage sites. A mixed-methods approach is employed, integrating quantitative corpus analysis, qualitative semiotic interpretation, and public perception surveys. The findings reveal a clear stratification of language use: Chinese dominates official signage, with pinyin and English in subordinate positions, reflecting state-imposed linguistic norms. Private signage, however, demonstrates greater linguistic flexibility, incorporating Cantonese expressions, traditional Chinese characters, and creative bilingual adaptations. By highlighting the negotiation between top-down language standardization and bottom-up linguistic agency, this study contributes to broader discussions on language policy, cultural heritage preservation, and multilingual accessibility in historical districts. The findings underscore the need for improved linguistic planning, standardized translation policies, and greater public engagement in signage design to ensure that linguistic landscapes in heritage sites are both culturally authentic and globally navigable.
Abstract Written culture in medieval Scandinavia featured the concurrent use of different languages–Latin and the local vernaculars–and two scripts–the Roman script and the runic script. Although Latin carried substantial cultural capital as the cosmopolitan language of religious and juridical authorities, diplomacy, and high literate culture, the vernacular runic written tradition had already been established for many centuries and continued to be used alongside Latin and the Roman script. This coexistence was particularly evident in the epigraphic landscape and resulted in several inscriptions that mixed both languages and scripts. This article explores the choices of language and script made by two medieval inscribers who, despite being primarily trained within a vernacular runic tradition, used Latin and/or the Roman script, particularly in their signatures. The study argues that these inscribers leveraged the symbolic value of Latin and the Roman script to project a distinctive professional identity in response to a changing linguistic market in which these linguistic resources were gaining value. The analysis also illustrates how these inscribers’ highly individual practices resulted from a negotiation of different linguistic norms and highlights a tension between their level of literacy and their efforts to display cultural capital. Finally, the article argues that the use of Latin and the Roman script was motivated more by their symbolic value than by purely communicative necessity, offering parallels to modern instances of language commodification.
This research was conducted within the framework of gender linguistics, the main task of which is to study how grammatical categories related to biological sex are reflected in language and affect the perception of men and women in the minds of speakers of a particular language. The research paper focuses on the use of feminatives in Russian and Greek among bilinguals and monolinguals. The aim of the study was to identify trends in the use of feminine correlates of the names of professions and job positions in legal and political fields, as well as to determine the influence of the linguistic environment on the formation of linguistic norms regarding gender terminology. The authors employed an experimental approach based on a comparative analysis of the use of feminatives among Russian-Greek natural bilinguals aged 16-25 living in Cyprus and learning English as a foreign language, and artificial bilinguals, native speakers of Russian/Greek who also speak English. The experiment participants were given English sentences containing professional designations. The task set for them was to translate the sentences in such a way that the agentive subject or addressee was indicated as a female; after that a systematic analysis of the use of feminatives and their derivational models was conducted. This analysis revealed gender asymmetry in both linguistic contexts, reflecting the unequal representation of male and female genders in the linguistic consciousness of the speakers. The main conclusions emphasize the importance of the linguistic environment and cultural factors in shaping gender-specific lexicon used in everyday communication and media, and indicate the presence of interference in the speech of bilinguals.
Previous studies suggested that pitch characteristics of lexical tones in Standard Chinese influence various sensory perceptions, but whether they iconically bias emotional experience remained unclear. We analyzed the arousal and valence ratings of bi-syllabic words in two corpora (Study 1) and conducted an affect rating experiment using a carefully designed corpus of bi-syllabic words (Study 2). Two-alternative forced-choice tasks further tested the robustness of lexical tones' affective iconicity in an auditory nonce word context (Study 3). Hierarchical linear models, generalized linear mixed models, and cross-validation were employed to understand the relationship between lexical tones and the emotional responses of tone-carrying words. Results consistently indicated that words with a falling-falling tonal sequence, both real and nonce words, received higher arousal ratings than those with rising-rising and rising-low tones. Only in nonce words, the high-high sequence was more likely to be associated with the low-arousal option; the falling-falling tone sequence was more often linked to negative-valence choice, while high-high and rising-rising tones with positive-valence. These findings, though subtle, suggest that the use of pitch in lexical tones influences emotional responses during the processing of tone-carrying words, pointing to an inherent iconic quality in lexical tones that may subtly shape speakers' emotional experiences.
This essay deals with two colour-related adjectives, badius and baietus, in the Medieval Latin documentary sources of Catalonia studied by the Glossarium Mediae Latinitatis Cataloniae (GMLC). The collection of documental testimonies has been conducted through the lexical database Corpus Documentale Latinum Cataloniae (CODOLCAT), a digital corpus of the Latin texts of this linguistic domain. These sources of notarial and juridical nature contain a considerable amount of colour adjectives, which usually serves to identify and differentiate lexical elements within the same referential class. One of the most attested colour adjectives is badius “bay, brown”, frequently documented with its variant baius and always referring to equines. There are also few occurrences of baietus “brownish”, a lexical innovation derived from badius. To refine their precise definitions, this study explores the forms and uses of both adjectives, taking into consideration the contexts in which they appear. Additionally, this study highlights the importance of the integration of digital tools into lexicographical research and emphasizes the need to incorporate insights from other linguistic domains to achieve a more comprehensive understanding of the words under analysis.
This repository contains data accompanying the publication "Auditory localization and subjective assessment of autonomous cleaning robot sounds: A VR experiment on speed, operating mode and alerting signals", submitted for review to the Acta Acustica. The dataset contains: Audio and video material Stimuli consisting of robot recordings under all evaluated conditions (0.3 m/s and 0.8 m/s speed, with and without cleaning, with and without noise AVAS or multi-tone AVAS, both with and without added amplitude modulation). All sounds were exported as 32-bit float wav files; i.e., reading the files into Matlab with audioread results in calibrated Pa values. The files uploaded here were used as source signals in the auralization, assuming a distance of 1 m. The final binaural stimuli were rendered by TASCAR and include an attenuation corresponding to the simulated 7 m distance. 30cms_cleaning_noAVAS.wav 30cms_noCleaning_multiTone.wav 30cms_noCleaning_multiToneAM.wav 30cms_noCleaning_noAVAS.wav 30cms_noCleaning_noise.wav 30cms_noCleaning_noiseAM.wav 80cms_cleaning_noAVAS.wav 80cms_noCleaning_multiTone.wav 80cms_noCleaning_multiToneAM.wav 80cms_noCleaning_noAVAS.wav 80cms_noCleaning_noise.wav 80cms_noCleaning_noiseAM.wav ambienceNoise.wav Excerpt of background noise played back during the experiment. localizationTaskDemo.mp4 Participant POV recording of localization task. This recording was done with a fixed head position, in the actual experiment participants were turning their heads freely. Experiment results and analysis localizationData.csv Table containing the mean and standard deviation of absolute localization error, aggregated for each participant and stimulus. subjectiveData.csv Table containing mean and z-scored annoyance, arousal, trust, and valence ratings for each participant and stimulus. stimuliAnalysis.csv Table containing results of level, loudness, sharpness, roughness, tonality, fluctuation strength, and impulsiveness analysis for all stimuli.
Word order difference between source and target languages is a major obstacle to cross-lingual transfer, especially in the dependency parsing task. Current works are mostly based on order-agnostic models or word reordering to mitigate this problem. However, such methods either do not leverage grammatical information naturally contained in word order or are computationally expensive as the permutation space grows exponentially with the sentence length. Moreover, the reordered source sentence with an unnatural word order may be a form of noising that harms the model learning. To this end, we propose an Implicit Word Reordering framework with Knowledge Distillation (IWR-KD). This framework is inspired by that deep networks are good at learning feature linearization corresponding to meaningful data transformation, e.g. word reordering. To realize this idea, we introduce a knowledge distillation framework composed of a word-reordering teacher model and a dependency parsing student model. We verify our proposed method on Universal Dependency Treebanks across 31 different languages and show it outperforms a series of competitors, together with experimental analysis to illustrate how our method works towards training a robust parser.
Using masculine forms for mixed-gender groups or individuals of unknown gender leads people to think of men. In grammatically gendered languages, using feminine and paired forms (gender-inclusive language, GIL) is a common and effective strategy to increase the visibility of women. Although GIL benefits women as a group, its adoption may encounter resistance, especially among employed women. They may refrain from using feminine forms to refer to themselves due to apprehension about potential backlash for deviating from professional and linguistic norms. Additionally, they may hesitate, so as not to evoke societal stereotypes that associate femininity with lower competence and status in professional settings. In two representative samples of Polish self-identified women, we examined the prevalence of GIL forms in professional self-reference. In both studies, approximately half of the participants used feminine forms. The tendency to use GIL was less pronounced among employed participants, with only a third of currently employed women using feminine job titles. Gender identification moderates this effect, with working women who strongly identify with other women being more likely to use feminine forms. These findings shed light on the potential social and professional factors influencing the adoption of GIL by documenting who uses these forms in what contexts. By identifying potential barriers to broader adoption this study underscores the need to address these challenges with professional and policy-oriented interventions.