Recent work has examined language models from a linguistic perspective to better understand how they acquire language. Most existing benchmarks focus on judging grammatical acceptability, whereas the ability to interpret meanings conveyed by grammatical forms has received much less attention. We introduce the Linguistic Minimal-Pair Benchmark for Evaluating Constructional Understanding in Language Models (CxMP), a benchmark grounded in Construction Grammar that treats form-meaning pairings, or constructions, as fundamental linguistic units. CxMP evaluates whether models can interpret the semantic relations implied by constructions, using a controlled minimal-pair design across nine construction types, including the let-alone, caused motion, and ditransitive constructions. Our results show that while syntactic competence emerges early, constructional understanding develops more gradually and remains limited even in large language models (LLMs). CxMP thus reveals persistent gaps in how language models integrate form and meaning, providing a framework for studying constructional understanding and learning trajectories in language models.
Gianluca Barmina, Nathalie Carmen Hau Norman, Peter Schneider-Kamp
et al.
We present an enhanced benchmark for evaluating linguistic acceptability in Danish. We first analyze the most common errors found in written Danish. Based on this analysis, we introduce a set of fourteen corruption functions that generate incorrect sentences by systematically introducing errors into existing correct Danish sentences. To ensure the accuracy of these corruptions, we assess their validity using both manual and automatic methods. The results are then used as a benchmark for evaluating Large Language Models on a linguistic acceptability judgement task. Our findings demonstrate that this extension is both broader and more comprehensive than the current state of the art. By incorporating a greater variety of corruption types, our benchmark provides a more rigorous assessment of linguistic acceptability, increasing task difficulty, as evidenced by the lower performance of LLMs on our benchmark compared to existing ones. Our results also suggest that our benchmark has a higher discriminatory power which allows to better distinguish well-performing models from low-performing ones.
Hamdan Al Ahbabi, Gautier Marti, Saeed AlMarri
et al.
Self-supervised learning models for speech processing, such as wav2vec2, HuBERT, WavLM, and Whisper, generate embeddings that capture both linguistic and paralinguistic information, making it challenging to analyze tone independently of spoken content. In this work, we introduce a method for disentangling paralinguistic features from linguistic content by regressing speech embeddings onto their corresponding text embeddings and using the residuals as a representation of vocal tone. We evaluate this approach across multiple self-supervised speech embeddings, demonstrating that residual embeddings significantly improve tone classification performance compared to raw speech embeddings. Our results show that this method enhances linear separability, enabling improved classification even with simple models such as logistic regression. Visualization of the residual embeddings further confirms the successful removal of linguistic information while preserving tone-related features. These findings highlight the potential of residual embeddings for applications in sentiment analysis, speaker characterization, and paralinguistic speech processing.
Logical reasoning with large language models (LLMs) has received growing attention. One mainstream approach translates natural language into formal logic and then applies symbolic solvers for deduction. While effective in many tasks, these LLM-based translators often fail to generate consistent symbolic representations when the same concept appears in different linguistic forms. Such inconsistencies break logical coherence and lead to solver errors. However, most existing benchmarks lack this type of linguistic variation, which frequently occurs in real-world text, leaving the problem underexplored. To address this gap, we present SoLT, a benchmark that systematically rewrites reasoning datasets into diverse yet logically equivalent forms across multiple levels. Beyond evaluation, SoLT also provides a general method to enrich any dataset with linguistic diversity while preserving both meaning and logic. To further enhance the stability of LLM-based reasoning, we propose MenTaL, which explicitly guides models to build a concept-symbol mapping table during translation. By linking equivalent expressions to shared symbols, MenTaL maintains consistency and mitigates symbol drift. Experiments on SoLT demonstrate that LLMs indeed suffer from inconsistent symbol mapping under linguistic variation, leading to significant drops in reasoning accuracy. Meanwhile, applying MenTaL brings clear and stable performance improvements across diverse inputs. Overall, our findings reveal that overlooking linguistic diversity hides key weaknesses in LLM-based translators, and our work offers a step toward more reliable logical reasoning in varied real-world scenarios. Our code is available at https://github.com/wufeiwuwoshihua/LinguDiver.
Large language models (LLMs) have emerged as powerful tools for supporting second language acquisition, particularly in simulating interactive dialogues for speaking practice. However, adapting the language difficulty of LLM-generated responses to match learners' proficiency levels remains a challenge. This work addresses this issue by proposing a framework for controlling language proficiency in educational dialogue systems. Our approach leverages three categories of linguistic features, readability features (e.g., Flesch-Kincaid Grade Level), syntactic features (e.g., syntactic tree depth), and lexical features (e.g., simple word ratio), to quantify and regulate text complexity. We demonstrate that training LLMs on linguistically annotated dialogue data enables precise modulation of language proficiency, outperforming prompt-based methods in both flexibility and stability. To evaluate this, we introduce Dilaprix, a novel metric integrating the aforementioned features, which shows strong correlation with expert judgments of language difficulty. Empirical results reveal that our approach achieves superior controllability of language proficiency while maintaining high dialogue quality.
The work aims to systematize and identify the main areas of research on scientific and academic discourse in Russian linguistics. This review-analytical study is based on dissertations for the degree of Doctor of Philology and Candidate of Philology over the past ten years, which constitutes its scientific novelty. The author focuses on the issues of the cognitive-discursive paradigm of linguistic knowledge, which has become firmly established in the science of language in recent years. The work examines the main trends in the study of various genres of academic (scientific) communication. Particular attention is paid to current approaches to the study of the terminological system of scientific discourse. The review revealed the active development of the theory of scientific discourse under the influence of the cognitive-discursive approach in the study of scientific text. It can be concluded that the anthropocentrism of modern linguistic research has allowed for a new perspective on some traditionally distinguished problems, rethinking and deepening many concepts and categories typically discussed in the stylistics of scientific speech.
Metaphor Components Identification (MCI) contributes to enhancing machine understanding of metaphors, thereby advancing downstream natural language processing tasks. However, the complexity, diversity, and dependency on context and background knowledge pose significant challenges for MCI. Large language models (LLMs) offer new avenues for accurate comprehension of complex natural language texts due to their strong semantic analysis and extensive commonsense knowledge. In this research, a new LLM-based framework is proposed, named Linguistics-aware In-context Learning with Data Augmentation (LaiDA). Specifically, ChatGPT and supervised fine-tuning are utilized to tailor a high-quality dataset. LaiDA incorporates a simile dataset for pre-training. A graph attention network encoder generates linguistically rich feature representations to retrieve similar examples. Subsequently, LLM is fine-tuned with prompts that integrate linguistically similar examples. LaiDA ranked 2nd in Subtask 2 of NLPCC2024 Shared Task 9, demonstrating its effectiveness. Code and data are available at https://github.com/WXLJZ/LaiDA.
Large Language Models (LLMs) have demonstrated promising capabilities for code generation. While existing benchmarks evaluate the correctness and efficiency of LLM-generated code, the potential linguistic bias - where code quality varies based on the natural language used to describe programming tasks - remains underexplored. In this paper, we aim to investigate this linguistic bias through the lens of English and Chinese. To facilitate our investigation, we present a unified evaluation framework comprising a curated dataset of 52 Python programming questions with parallel bilingual task descriptions, automated correctness verification, and efficiency quantification tools based on runtime complexity estimation. Based on this framework, we conduct the first empirical study towards the linguistic bias in LLM-generated code on eight popular LCGMs, as well as GPT-3.5-Turbo and GPT-4. We observe that these LCGM-generated code show different correctness on an average of 12% bilingual programming tasks, where 39% also exhibits diverse efficiency. Our findings indicate that LLMs commonly exhibit linguistic bias for code generation.
Analyses of transformer-based models have shown that they encode a variety of linguistic information from their textual input. While these analyses have shed a light on the relation between linguistic information on one side, and internal architecture and parameters on the other, a question remains unanswered: how is this linguistic information reflected in sentence embeddings? Using datasets consisting of sentences with known structure, we test to what degree information about chunks (in particular noun, verb or prepositional phrases), such as grammatical number, or semantic role, can be localized in sentence embeddings. Our results show that such information is not distributed over the entire sentence embedding, but rather it is encoded in specific regions. Understanding how the information from an input text is compressed into sentence embeddings helps understand current transformer models and help build future explainable neural models.
URIEL is a knowledge base offering geographical, phylogenetic, and typological vector representations for 7970 languages. It includes distance measures between these vectors for 4005 languages, which are accessible via the lang2vec tool. Despite being frequently cited, URIEL is limited in terms of linguistic inclusion and overall usability. To tackle these challenges, we introduce URIEL+, an enhanced version of URIEL and lang2vec that addresses these limitations. In addition to expanding typological feature coverage for 2898 languages, URIEL+ improves the user experience with robust, customizable distance calculations to better suit the needs of users. These upgrades also offer competitive performance on downstream tasks and provide distances that better align with linguistic distance studies.
Evaluating the cognitive capacities of large language models (LLMs) requires overcoming not only anthropomorphic but also anthropocentric biases. This article identifies two types of anthropocentric bias that have been neglected: overlooking how auxiliary factors can impede LLM performance despite competence ("auxiliary oversight"), and dismissing LLM mechanistic strategies that differ from those of humans as not genuinely competent ("mechanistic chauvinism"). Mitigating these biases necessitates an empirically-driven, iterative approach to mapping cognitive tasks to LLM-specific capacities and mechanisms, which can be done by supplementing carefully designed behavioral experiments with mechanistic studies.
Abstract Plant diseases significantly threaten global agriculture, impacting crop yield and food security. Nearly 30% of the crop yield is lost due to plant diseases. Efficient identification and classification of plant diseases through computer vision techniques have become imperative for timely intervention. However, popular plant disease datasets often suffer from data imbalance, with certain classes underrepresented, hindering the performance of machine learning models. Traditional data augmentation methods, such as rotation and flipping, are limited in their effectiveness, especially when faced with imbalanced datasets. To address this limitation, we explore advanced data augmentation techniques, including Generative Adversarial Networks (GANs) such as CycleGAN and LeafGAN, which have shown promise in generating synthetic images. However, we propose an innovative approach of Object-based single Style Transfer on a single neural network for augmenting the plant disease dataset. This technique focuses on mitigating data imbalance issues within datasets, which can adversely affect the model’s ability to generalize across diverse classes. The proposed method is compared with state-of-the-art data augmentation techniques, highlighting its superiority in addressing data imbalance issues. Our approach aims to produce more realistic and diverse synthetic images, leading to improved model generalization and accuracy in plant disease classification tasks validated using different classifiers. The efficiency of our approach is validated through extensive experimentation and benchmarking against existing methods.
Computational linguistics. Natural language processing, Electronic computers. Computer science
Gelek mîrektiyên kurdan hene ku gelek bi hêz temendar bûne. Ji wan yek Mîrektiya Bitlîsê ye ku 760 salî desthilatdar bûye. Mirektiyan Cizîrê û Hekariyê jî bi nav û deng bûne. Bi navê beglik û mîrek wekî Miks, Hîzan Mehmûdî û wekî din jî hebûne. Lêbelê Mîrektiya Kêsanê heta niha nehatiye naskirin. Xebat li ser vê Mîrektiyê nehatine kirin. Me jî xebatek li ser vê mijarê hem di qadê da hem jî di çavkaniyan da kir. Bi alîkriya rêberên herêmê em çûne Sunbanê. Me yek ji endamên wê mîrmalbatê ku Cemîl Begê ye tesbît kir û agahiyên pêwîst jê wergirtin. Çi ku li ser erdnigariya vê binemala mîrektiyê bi hûrgilî yan xebat nehatine kirin, yan jî ji aliyê çand, huner, dîrok û erdkolojiyê ve kêm xebatên zanistî çêbûne. Lewra hê jî gelek cih hene ku li ser herêmeke din tên hesibandin, yan jî nayêne naskirin; lê di nav gel da bi navekî cuda têne binavkirin. Ew herêm yan jî şaxên malmîrekan tam nehatine tesbîtkirin. Zemanê berê, serdema navîn rêveberiya herêma Kurdistanê ji aliyê axa, mîr û begên kurdan ve hatiye îdarekirin. Yek ji wan jî Mîrektiya/Beglika Kêsanê ye. Ev Mîretiya han, ji dema şerê Ulama û Şeref Xanê Bitlîsî bi şûn ve ava bûye û heta dema Murad Begê Kêsanî dewam kiriye. Desthilatdariya Osmanî ew rûxandiye. Mîrekên Kêsanê herçiqas nêzê 250 û 300 sal ev herêm îdare kirine jî dîroka medreseya wê digihîje heta 383 salan berê. Ev herêma ku em ê li serê bixebitin, navenda Mîrektiyê Kêsanê û Keleha Sunbanê derdora 60-70ê gundî nav xwe da dihewîne. Em ê behsa Keleha Mîran, Medreseya ‘Ebdulah Beg, dêr û camiyên wan bikin. Herweha dê jiyana Fileh û Musulmanên ku wê hingê çawa bi awayekî biedalet bi hev ra mi‘amele kirine bibe mijara lêkolînê. Dê şopa çîrok û serpêhatiya vê mîrektiya ku wekî sancaqekê bûye cihê gerandin û nivîstekên medrese û kêlikên têkildarî Begen Kêsanê bihên vekolan.
Indo-Iranian languages and literature, Language. Linguistic theory. Comparative grammar
Статья посвящена проблеме образов прошлого в медиакоммуникации, которая рассматривается в качестве важнейшего проводника информационной социализации молодежи. Ее роль возрастает в условиях общественной нестабильности, распространении социальной тревожности, когда требуется формирование новых моделей самоидентификации человека. Креативные форматы в социальных медиа – сторителлинг, меминг, историческая сетевая литература, юмористические сюжеты в рекламе служат своеобразными защитными психологическими механизмами, способствуют преодолению конфликтности исторического сознания. В статье подчеркивается, что концепт репрезентации исторической памяти в журналистике, социологии и маркетинге является дискуссионным. Предлагается при характеристике медийного пространства использовать понятие «социальные представления о прошлом». С целью осмысления особенностей восприятия рекламы, использующей исторических личностей и события, обобщаются данные социологического исследования. С помощью методов вторичного анализа данных и формализованного интервью решаются задачи выявления степени вовлеченности в рекламную коммуникацию с исторической ретроспекцией, установления интересующего аудиторию исторического периода, коммуникативной и коммерческой эффективности, приемлемости использования юмора. Интервью проводилось на основе репрезентативной выборки по признакам пола и образования представителей молодежи г. Воронежа (N=148). В заключении сделаны выводы о воздействии исторических сюжетов как в социальной, так и коммерческой рекламе на молодежную аудиторию. Главным стоппером является именно исторический контекст. Запомнились люди-символы прошлого – А. С. Пушкин, Л. Н. Толстой, Петр I, Иван Грозный. Позитивные эмоции при просмотре рекламной информации с историческими включениями преобладают. В целом влияние на поведение современного молодого потребителя в ходе информационной социализации довольно значительно, а коммуникативная эффективность и привлекательность контента высоки.
We study feature interactions in the context of feature attribution methods for post-hoc interpretability. In interpretability research, getting to grips with feature interactions is increasingly recognised as an important challenge, because interacting features are key to the success of neural networks. Feature interactions allow a model to build up hierarchical representations for its input, and might provide an ideal starting point for the investigation into linguistic structure in language models. However, uncovering the exact role that these interactions play is also difficult, and a diverse range of interaction attribution methods has been proposed. In this paper, we focus on the question which of these methods most faithfully reflects the inner workings of the target models. We work out a grey box methodology, in which we train models to perfection on a formal language classification task, using PCFGs. We show that under specific configurations, some methods are indeed able to uncover the grammatical rules acquired by a model. Based on these findings we extend our evaluation to a case study on language models, providing novel insights into the linguistic structure that these models have acquired.
It is well known that AI-based language technology -- large language models, machine translation systems, multilingual dictionaries, and corpora -- is currently limited to 2 to 3 percent of the world's most widely spoken and/or financially and politically best supported languages. In response, recent research efforts have sought to extend the reach of AI technology to ``underserved languages.'' In this paper, we show that many of these attempts produce flawed solutions that adhere to a hard-wired representational preference for certain languages, which we call techno-linguistic bias. Techno-linguistic bias is distinct from the well-established phenomenon of linguistic bias as it does not concern the languages represented but rather the design of the technologies. As we show through the paper, techno-linguistic bias can result in systems that can only express concepts that are part of the language and culture of dominant powers, unable to correctly represent concepts from other communities. We argue that at the root of this problem lies a systematic tendency of technology developer communities to apply a simplistic understanding of diversity which does not do justice to the more profound differences that languages, and ultimately the communities that speak them, embody. Drawing on the concept of epistemic injustice, we point to the broader sociopolitical consequences of the bias we identify and show how it can lead not only to a disregard for valuable aspects of diversity but also to an under-representation of the needs and diverse worldviews of marginalized language communities.
We introduce a novel data generation method for contradiction detection, which leverages the generative power of large language models as well as linguistic rules. Our vision is to provide a condensed corpus of prototypical contradictions, allowing for in-depth linguistic analysis as well as efficient language model fine-tuning. To this end, we instruct the generative models to create contradicting statements with respect to descriptions of specific contradiction types. In addition, the model is also instructed to come up with completely new contradiction typologies. As an auxiliary approach, we use linguistic rules to construct simple contradictions such as those arising from negation, antonymy and numeric mismatch. We find that our methods yield promising results in terms of coherence and variety of the data. Further studies, as well as manual refinement are necessary to make use of this data in a machine learning setup.
An important layer of this research is dedicated to critical studies, which are directed at the strategies
of creating a peculiarly Indian literary theory and practice. We have a desire to separate the indigenous
tradition from the broad American, in particular, Anglo-American and recognize Indian writing as a
component of the multicultural paradigm. Currently, there is a noticeable confrontation between two
camps of literary critics: one of them is oriented to European literary theories and believes that they should
form the basis of literary interpretations of indigenous writers’ works; another wing is determined by the
need to clarify the peculiarities of the literary paradigm of “Indian realism” in the context of a globalized
society taking into account new literary models of the perception of ethnic minorities. The need to write
the article is caused by the lack of a comprehensive understanding of the problem in Ukrainian literary
studies and the growing objective interest in the works of Native American writers, in particular Susan
Power. The article proposes a conceptual and methodological determination of the study of a literary text
written in the style of Indian realism, which makes it possible to reveal the raised scientific problem at
many levels.
The article examines how the drama of loss, search and a new acquisition of national identity by
the Sioux people was artistically and aesthetically reflected in the text of Susan Power`s novel “The Grass
Dancer”. The presentation and consideration of the problem of national and cultural identity provides an
opportunity to see the artistic diversity in the understanding of the personal destiny of a person and the
people in general, literary ideas about the Sioux people beliefs peculiarities, their aesthetic component
and place in the national cultural canon. The main thing is to avoid the trap of a politicized and ideologized
theory of multiculturalism, in which modern critics increasingly see an opportunity to interpret the
texts of indigenous writers, which is actually the ideology of colonial domination hidden behind political
correctness.
The main purpose of this article is to outline a coherent theoretical and empirical model of multi-level
functioning of Dakota national identity concept in the novel “The Grass Dancer” by Susan Power. Also the aim is to substantiate the leading concept of Sioux national identity literary manifestations interpretation
as a unique code, epistemological, socio-cultural and artistic-aesthetic factor that plays a significant role in
the modern worldview formation process and myth-making of Dakota society representatives.
The following article involves historical-cultural and structural approaches, which correspond to the
purpose and tasks of the research; methods of cultural-semantic analysis and phenomenological methods
were also used.
The persistent deconstruction of the Eurocentric canon of world literature, not only at the level
of academic discussions, but also in the system of university teaching of world literature, demands new
texts such as “The Grass Dancer”. The reformatting of canons is, of course, a permanent process, but the
globalization of the literary canon today acquires a qualitatively new scale and breadth proposed by Susan
Power. Multiculturalism with its influence on cultural dynamics and the idea of national and cultural identity
can’t be considered the driving cultural stimulus of changes in all its ambiguity. To an even greater extent,
transculturalism, proposed by Power, aimed at defining common interests and common values across
cultural and national borders for non-native readers. That is her main contribution to the construction of
a more globalized literary canon. Susan Power as a Native American writer has repeatedly addressed the
specified range of the Indigenous problems, which constantly tested the agreement prevalent in the nonnative science with the most urgent problems of Native literary studies.
Abstract Speaking constitutes one of the main goals of learning a second language (L2). Despite the increasing attention on the role of planning and language transfer in L2 learning, the combined effect of using different languages and pre-task planning on language production remains unclear. This study investigated whether the use of different languages in planning affects speaking performance and whether the effect differs by language proficiency. A total of 84 students in Chinese universities learning English as a foreign language participated in several speaking tasks after planning using their first language (L1) Chinese or L2 English. Findings showed that using L1 in planning results in significantly higher syntactic complexity, accuracy, and fluency in speaking performance than using L2 in planning, while the difference in lexical diversity were not statistically significant. Further analysis shows that for speech accuracy, the facilitative effect of L1 was stronger among low-proficient than high-proficient learners. Findings from this study support the use of L2 learners’ entire linguistic repertoire in speaking activities and provides implications on speech production theories as well as translanguaging pedagogies.
Special aspects of education, Language acquisition
The article is devoted to the analysis of the structure and functioning of multi-component compound sentences in Valerii Shevchuk's speech. The novel "On the Humble Field" will serve as the research material. Since in modern linguistics there are still no scientific investigations, the subject of which was the analysis of multi-component complex sentences in Valerii Shevchuk's artistic texts, our research is relevant.
The goal of the article is to characterize the structural organization of multi-component compound sentences revealed in the novel "On the Humble Field" by Val. Shevchuk, to carry out their classification, to consider the peculiarities of the use of these constructions in the context of the work. In order to realize the specified goal, the following tasks must be solved: 1) identify the multi-component compound sentences recorded in the novel "On the Humble Field"; 2) analyze the structure of these sentences; 3) to classify the detected multi-component compound sentences; 4) trace the functioning of these sentences in the analyzed work.
As a result of examining the syntactic parameters of Valery Shevchuk's language production, there was revealed the use of complex constructions, in particular multi-component compound sentences. There was carried out the classification of these constructions by the method of modeling subordinating parts and their connections, three main varieties were distinguished: complex subordinate clauses with sequential subordination, complex subordinate clauses with subordination: homogeneous and heterogeneous, the most - with sequential subordination. In the novel "On the Humble Field" written by Val. Shevchuk multi-component complex sentences have different stylistic load. Often they characterize various circumstances under which events take place, depict nature. It was noticed that multi-component complex sentences often express the psychological state of the characters.
As a result of the analysis, it was found that in his works Val. Shevchuk actively uses multi-component compound sentences, among which compound sentences with sequential subordination are the most common. It was noticed that the writer most often uses these multi-component complex constructions with pictorial and emotional-expressive load.