Subba Reddy Oota, Vijay Rowtula, Satya Sai Srinath Namburi
et al.
Recent work has shown that scaling large language models (LLMs) improves their alignment with human brain activity, yet it remains unclear what drives these gains and which representational properties are responsible. Although larger models often yield better task performance and brain alignment, they are increasingly difficult to analyze mechanistically. This raises a fundamental question: what is the minimal model capacity required to capture brain-relevant representations? To address this question, we systematically investigate how constraining model scale and numerical precision affects brain alignment. We compare full-precision LLMs, small language models (SLMs), and compressed variants (quantized and pruned) by predicting fMRI responses during naturalistic language comprehension. Across model families up to 14B parameters, we find that 3B SLMs achieve brain predictivity indistinguishable from larger LLMs, whereas 1B models degrade substantially, particularly in semantic language regions. Brain alignment is remarkably robust to compression: most quantization and pruning methods preserve neural predictivity, with GPTQ as a consistent exception. Linguistic probing reveals a dissociation between task performance and brain predictivity: compression degrades discourse, syntax, and morphology, yet brain predictivity remains largely unchanged. Overall, brain alignment saturates at modest model scales and is resilient to compression, challenging common assumptions about neural scaling and motivating compact models for brain-aligned language modeling.
The fusion of language and vision in large vision-language models (LVLMs) has revolutionized deep learning-based object detection by enhancing adaptability, contextual reasoning, and generalization beyond traditional architectures. This in-depth review presents a structured exploration of the state-of-the-art in LVLMs, systematically organized through a three-step research review process. First, we discuss the functioning of vision language models (VLMs) for object detection, describing how these models harness natural language processing (NLP) and computer vision (CV) techniques to revolutionize object detection and localization. We then explain the architectural innovations, training paradigms, and output flexibility of recent LVLMs for object detection, highlighting how they achieve advanced contextual understanding for object detection. The review thoroughly examines the approaches used in integration of visual and textual information, demonstrating the progress made in object detection using VLMs that facilitate more sophisticated object detection and localization strategies. This review presents comprehensive visualizations demonstrating LVLMs' effectiveness in diverse scenarios including localization and segmentation, and then compares their real-time performance, adaptability, and complexity to traditional deep learning systems. Based on the review, its is expected that LVLMs will soon meet or surpass the performance of conventional methods in object detection. The review also identifies a few major limitations of the current LVLM modes, proposes solutions to address those challenges, and presents a clear roadmap for the future advancement in this field. We conclude, based on this study, that the recent advancement in LVLMs have made and will continue to make a transformative impact on object detection and robotic applications in the future.
In this work we introduce the concepts of linguistic transformation, linguistic loop and semantic deficit. By exploiting Lie group theoretical and geometric techniques, we define invariants that capture the structural properties of a whole linguistic loop. This result introduces new line of research, employing tools from Lie theory and higher-dimensional geometry within language studies. But, even more intriguingly, our study hints to a mathematical characterization of the meta-linguistic or pre-verbal thought, namely of those cognitive structures that precede the language.
Syntactic discontinuity is a grammatical phenomenon in which a constituent is split into more than one part because of the insertion of an element which is not part of the constituent. This is observed in many languages across the world such as Turkish, Russian, Japanese, Warlpiri, Navajo, Hopi, Dyirbal, Yidiny etc. Different formalisms/frameworks in current linguistic theory approach the problem of discontinuous structures in different ways. Each framework/formalism has widely been viewed as an independent and non-converging system of analysis. In this paper, we propose a unified system of representation for both continuity and discontinuity in structures of natural languages by taking into account three formalisms, in particular, Phrase Structure Grammar (PSG) for its widely used notion of constituency, Dependency Grammar (DG) for its head-dependent relations, and Categorial Grammar (CG) for its focus on functor-argument relations. We attempt to show that discontinuous expressions as well as continuous structures can be analysed through a unified mathematical derivation incorporating the representations of linguistic structure in these three grammar formalisms.
Stefano Epifani, Giuliano Castigliego, Laura Kecskemeti
et al.
Background: Mentalization integrates cognitive, affective, and intersubjective components. Large Language Models (LLMs) display an increasing ability to generate reflective texts, raising questions regarding the relationship between linguistic form and mental representation. This study assesses the extent to which a single LLM can reproduce the linguistic structure of mentalization according to the parameters of Mentalization-Based Treatment (MBT). Methods: Fifty dialogues were generated between human participants and an LLM configured in standard mode. Five psychiatrists trained in MBT, working under blinded conditions, evaluated the mentalization profiles produced by the model along the four MBT axes, assigning Likert-scale scores for evaluative coherence, argumentative coherence, and global quality. Inter-rater agreement was estimated using ICC(3,1). Results: Mean scores (3.63-3.98) and moderate standard deviations indicate a high level of structural coherence in the generated profiles. ICC values (0.60-0.84) show substantial-to-high agreement among raters. The model proved more stable in the Implicit-Explicit and Self-Other dimensions, while presenting limitations in the integration of internal states and external contexts. The profiles were coherent and clinically interpretable yet characterized by affective neutrality.
To date there have not been many studies that examine how Mandarin is acquired in an immersion setting. In this study we examine how early-stage immersion learners of Mandarin acquire a grammatical structure— the ba construction. This is a frequently used structure in Mandarin that has a non-canonical SOV word order, an order for which English has no counterpart. Taking a qualitative approach, we collected learner utterances over a 4-month period. It was found that the learners did not like to deviate from the canonical SVO word order and it was difficult for them to produce the ba construction. However, when they did use the construction, their utterances satisfied a complex predicate constraint that is imposed on the construction, suggesting that the learners have knowledge of the constraint. Above all, immersion young learners have a grammar of their own. Their language development offers a new window into bilingual language acquisition.
Education (General), Language. Linguistic theory. Comparative grammar
Melusi Malinga, Isaac Lupanda, Mike Wa Nkongolo
et al.
South Africa and the Democratic Republic of Congo (DRC) present a complex linguistic landscape with languages such as Zulu, Sepedi, Afrikaans, French, English, and Tshiluba (Ciluba), which creates unique challenges for AI-driven translation and sentiment analysis systems due to a lack of accurately labeled data. This study seeks to address these challenges by developing a multilingual lexicon designed for French and Tshiluba, now expanded to include translations in English, Afrikaans, Sepedi, and Zulu. The lexicon enhances cultural relevance in sentiment classification by integrating language-specific sentiment scores. A comprehensive testing corpus is created to support translation and sentiment analysis tasks, with machine learning models such as Random Forest, Support Vector Machine (SVM), Decision Trees, and Gaussian Naive Bayes (GNB) trained to predict sentiment across low resource languages (LRLs). Among them, the Random Forest model performed particularly well, capturing sentiment polarity and handling language-specific nuances effectively. Furthermore, Bidirectional Encoder Representations from Transformers (BERT), a Large Language Model (LLM), is applied to predict context-based sentiment with high accuracy, achieving 99% accuracy and 98% precision, outperforming other models. The BERT predictions were clarified using Explainable AI (XAI), improving transparency and fostering confidence in sentiment classification. Overall, findings demonstrate that the proposed lexicon and machine learning models significantly enhance translation and sentiment analysis for LRLs in South Africa and the DRC, laying a foundation for future AI models that support underrepresented languages, with applications across education, governance, and business in multilingual contexts.
Projecting visual features into word embedding space has become a significant fusion strategy adopted by Multimodal Large Language Models (MLLMs). However, its internal mechanisms have yet to be explored. Inspired by multilingual research, we identify domain-specific neurons in multimodal large language models. Specifically, we investigate the distribution of domain-specific neurons and the mechanism of how MLLMs process features from diverse domains. Furthermore, we propose a three-stage mechanism for language model modules in MLLMs when handling projected image features, and verify this hypothesis using logit lens. Extensive experiments indicate that while current MLLMs exhibit Visual Question Answering (VQA) capability, they may not fully utilize domain-specific information. Manipulating domain-specific neurons properly will result in a 10% change of accuracy at most, shedding light on the development of cross-domain, all-encompassing MLLMs in the future. The source code is available at https://github.com/Z1zs/MMNeuron.
AbstractSpanish has two forms to introduce comparative standards: que ‘that’ and de ‘of.’ The comparative morpheme is always the same más ‘-er/more.’ While que-comparatives show no variation in their syntactic properties, there is significant variation within de-comparatives regarding extraposition, scope, ACD resolution and the syntax of comparative numerals. Despite this variation, I argue that a uniform account is possible. I propose that más has the same syntax across the board (i.e. it takes the late-merged standard as complement, Bhatt and Pancheva 2004) and semantically it is a generalized quantifier over degrees (Heim 2001). The analysis (i) ensures that más and the standard form a constituent, (ii) allows for inverse scope, ACD resolution inside the standard of comparison and extraposition.
This work explores the degree to which grammar acquisition is driven by language `simplicity' and the source modality (speech vs. text) of data. Using BabyBERTa as a probe, we find that grammar acquisition is largely driven by exposure to speech data, and in particular through exposure to two of the BabyLM training corpora: AO-Childes and Open Subtitles. We arrive at this finding by examining various ways of presenting input data to our model. First, we assess the impact of various sequence-level complexity based curricula. We then examine the impact of learning over `blocks' -- covering spans of text that are balanced for the number of tokens in each of the source corpora (rather than number of lines). Finally, we explore curricula that vary the degree to which the model is exposed to different corpora. In all cases, we find that over-exposure to AO-Childes and Open Subtitles significantly drives performance. We verify these findings through a comparable control dataset in which exposure to these corpora, and speech more generally, is limited by design. Our findings indicate that it is not the proportion of tokens occupied by high-utility data that aids acquisition, but rather the proportion of training steps assigned to such data. We hope this encourages future research into the use of more developmentally plausible linguistic data (which tends to be more scarce) to augment general purpose pre-training regimes.
Max Müller-Eberstein, Rob van der Goot, Barbara Plank
et al.
Representational spaces learned via language modeling are fundamental to Natural Language Processing (NLP), however there has been limited understanding regarding how and when during training various types of linguistic information emerge and interact. Leveraging a novel information theoretic probing suite, which enables direct comparisons of not just task performance, but their representational subspaces, we analyze nine tasks covering syntax, semantics and reasoning, across 2M pre-training steps and five seeds. We identify critical learning phases across tasks and time, during which subspaces emerge, share information, and later disentangle to specialize. Across these phases, syntactic knowledge is acquired rapidly after 0.5% of full training. Continued performance improvements primarily stem from the acquisition of open-domain knowledge, while semantics and reasoning tasks benefit from later boosts to long-range contextualization and higher specialization. Measuring cross-task similarity further reveals that linguistically related tasks share information throughout training, and do so more during the critical phase of learning than before or after. Our findings have implications for model interpretability, multi-task learning, and learning from limited data.
Mokhtar Ebrahimi, Nasrollah Emami, Ghodrat Ghasemipour
et al.
Aghraz Al-Seyasah Fi A'araz Al-Reyasah, authored by Zahiry Samarghandy has become one of the excellent and outstanding old prose texts, because of containing technical, political, social, philosophical, didactic, and ethical teachings and addressing governance practices. These practical and miscellaneous teachings have led to frequent transcriptions and insertion of significant distortions and falsifications into the text of the mentioned book. Jaafar Sheảr by having in hand four manuscripts of Aghraz Al-Seyasah but without benefitting all of them (specially Meshkảt᾿s transcript) has corrected and printed it in many cases mentioned the manuscript᾿s differences. Furthermore, negligence in trusteeship towards manuscripts, especially the base manuscript, hastiness in correction, and the weakness and indifference in typesetting have inserted some other distortions, falsifications, and errors in the aforesaid book. This study has tried to countercheck the printed copy of Aghraz Al-Seyasah with the base and Meshkảt᾿s manuscript, and then by comparing some phrases of it with the other works of Zahiry Samarghandy (Sandbadnameh and Ghorrat–al-Alfaz wa Nozhat-al-Alhaz) and writer᾿s stylistic norms analyze and correct some distortions falsifications and errors in the printed text of Aghraz Al-Seyasah.
Ad hoc parsers are everywhere: they appear any time a string is split, looped over, interpreted, transformed, or otherwise processed. Every ad hoc parser gives rise to a language: the possibly infinite set of input strings that the program accepts without going wrong. Any language can be described by a formal grammar: a finite set of rules that can generate all strings of that language. But programmers do not write grammars for ad hoc parsers -- even though they would be eminently useful. Grammars can serve as documentation, aid program comprehension, generate test inputs, and allow reasoning about language-theoretic security. We propose an automatic grammar inference system for ad hoc parsers that would enable all of these use cases, in addition to opening up new possibilities in mining software repositories and bi-directional parser synthesis.
This paper describes the University of Maryland's submission to the Special Task on Formality Control for Spoken Language Translation at \iwslt, which evaluates translation from English into 6 languages with diverse grammatical formality markers. We investigate to what extent this problem can be addressed with a \textit{single multilingual model}, simultaneously controlling its output for target language and formality. Results show that this strategy can approach the translation quality and formality control achieved by dedicated translation models. However, the nature of the underlying pre-trained language model and of the finetuning samples greatly impact results.
We define a general mathematical framework for linguistics based on the theory of fibrations, called FibLang. We start by modelling the interaction between linguistics and cognition in the most general way possible, with a heavy focus on conceptually motivating any assumption we make. The advantage is that FibLang remains agnostic with respect to any particular axiomatization of grammar one may choose. As such, it is compatible with already existing categorical models of language (such as for example, DisCoCat), providing a formally sound framework to apply mathematical tools developed in the context of category theory, mainly categorical logic, to the study of language
Since the beginning of 2020, the COVID-19 pandemic has forced universities around the world to engage quickly and efficiently with online teaching platforms. Yet even before the pandemic hit, many programmes had been addressing challenges related to globalisation and technologisation within the teaching and learning context by moving to online or blended teaching and learning modes. In both the immediate pre- and post-pandemic context, movement towards the use of innovative technologies such as virtual reality (VR) to enhance the student experience have occurred in disciplines that heavily rely on practice-based learning (such as the health sciences and psychology). This paper describes an innovative approach to community interpreter training, which is in high demand in Australia. The VR project under examination here aims to provide evidence-based, pedagogically-sound, authentic, situated learning scenarios in a safe, virtual environment so that students are better prepared to deal with the complexities of the role of an interpreter in family violence (FV) settings. Using the VR platform, trainees will be given the opportunity to engage in simulated interpreting tasks working with victims of FV, social workers, police and other field-specific protagonists. In this article, we outline the methodology applied to the provision of interpreter training in this specific VR context. This methodology will serve as a blueprint for other institutions — particularly those offering specialised interpreter training — looking to minimise the threat to face-to-face contexts introduced by the pandemic, but also eager to expand into more experiential teaching offerings that reach beyond traditional modes used for interpreter training.
Ronivaldo Moreira de Souza, Mauricio Ribeiro da Silva
Ao longo do processo de formação sócio-histórica do Brasil, os colonizadores construíram narrativas com imagens estereotipadas dos africanos escravizados e de sua religiosidade a partir de uma visão europeia e cristã. Mesmo depois da abolição da escravidão esses estereótipos permaneceram associados à religiosidade afro-brasileira e foram apropriadas pelas narrativas jornalísticas, contribuindo para reverberar essa estereotipia até o tempo presente. O objetivo deste artigo é analisar os estereótipos associados à religiosidade afro-brasileira na década de 1920 nas narrativas jornalísticas da imprensa carioca. Tendo como base os pressupostos teórico-metodológicos da Análise do Discurso de Escola Francesa, compomos o corpus da pesquisa com 24 textos, dos quais, dois são tomados para análise no artigo e considerados como representativos da seleção. Constatamos que nas narrativas analisadas a religiosidade afro-brasileira é demonizada, configurando a estereotipia.
Numerous works have analyzed biases in vision and pre-trained language models individually - however, less attention has been paid to how these biases interact in multimodal settings. This work extends text-based bias analysis methods to investigate multimodal language models, and analyzes intra- and inter-modality associations and biases learned by these models. Specifically, we demonstrate that VL-BERT (Su et al., 2020) exhibits gender biases, often preferring to reinforce a stereotype over faithfully describing the visual scene. We demonstrate these findings on a controlled case-study and extend them for a larger set of stereotypically gendered entities.
Offensive content is pervasive in social media and a reason for concern to companies and government organizations. Several studies have been recently published investigating methods to detect the various forms of such content (e.g. hate speech, cyberbullying, and cyberaggression). The clear majority of these studies deal with English partially because most annotated datasets available contain English data. In this paper, we take advantage of available English datasets by applying cross-lingual contextual word embeddings and transfer learning to make predictions in low-resource languages. We project predictions on comparable data in Arabic, Bengali, Danish, Greek, Hindi, Spanish, and Turkish. We report results of 0.8415 F1 macro for Bengali in TRAC-2 shared task, 0.8532 F1 macro for Danish and 0.8701 F1 macro for Greek in OffensEval 2020, 0.8568 F1 macro for Hindi in HASOC 2019 shared task and 0.7513 F1 macro for Spanish in in SemEval-2019 Task 5 (HatEval) showing that our approach compares favourably to the best systems submitted to recent shared tasks on these three languages. Additionally, we report competitive performance on Arabic, and Turkish using the training and development sets of OffensEval 2020 shared task. The results for all languages confirm the robustness of cross-lingual contextual embeddings and transfer learning for this task.
Semantic types of causative verbs of the Khakas language, which are reflected in the texts of the Khakas heroic tales Altyn Aryg and Ah Chibek Aryg. Recorded from the narrator P. V. Kurbizhekov (1910–1966), Khara Khuskhun, recorded from the narrator P. V. Todanov. The relevance of this study is due to the fact that causative verbs are active in use and are of great interest to linguistic researchers. The need to study the semantic types of causative verbs in the texts of Khakas heroic tales is caused by the lack of development of this issue today, which determines the novelty of the study. The purpose of the article is to identify the semantics and functioning of causative verbs in the texts of Khakas heroic tales. The tasks of the research include identifying all the causative verbs found in the texts of heroic tales; determination and analysis of causative affixes involved in the formation of causative verbs; identification of word-formation models of causative verbs, presented in the texts of heroic tales. The analysis of the factual material was carried out using a complex of linguistic methods and techniques. The most promising method for solving the tasks was the method of continuous sampling of examples, which provides for the selection of examples for analysis and writing out of the texts of heroic legends in a row all the examples of the analyzed type found in it, a descriptive method for identifying causative verbs and their consistent description from the point of view of their structure and functioning. The method of distributive analysis made it possible to reveal the valence of the causative verb. As a result of the research, the semantic types of causative verbs were identified and analyzed, namely, factitive, permissive and factual-reflexive, permissive-reflexive types; the derivational models of causative verbs that are relevant for the language of the heroic epic are determined. For verbs with factual semantics, five derivational models have been identified; for permissive – three models; factual-reflexive and permissive-reflexive according to the same model. The study also showed that in the texts of Khakas heroic tales, a double causative is often used, which expresses both factual and permissive meaning. The results obtained can find application in the reading of lecture courses on the Khakas language at the philological faculties of universities, in the compilation of educational and teaching aids, dictionaries. In the future, interesting results can be obtained from studies of causative constructions characteristic of the texts of heroic legends, in a comparative and comparative typological light. References Kildibekova T. A. Action verbs in the modern Russian language. Saratov, Saratov University Publ. House, 1985, 157 p. (In Russ.) Dadueva E. A. Causative verbs with permissive meaning in Buryat and Russian languages. Oriental Studies. 2019, no. 1 (41), pp. 99–107. DOI: 10.22162/2619-0990-2019-41-1-99-107. (In Russ.) Dadueva E. A. Functional and semantic category of causativity in Russian and Buryat languages. Ulan-Ude, Publ. House of the Buryat State University, 2011, 128 p. (In Russ.) Tarieva L. U. On the question of regular means of education of causative verbs in the ergative language. Vestnik VSU. Series: Linguistics and intercultural communication. 2018, no. 2, pp. 119–122. (In Russ.) Khachemizova M. A. Causative in the Adyghe language: its essence, marking, functioning. Abstract of the dissertation of Candidate of Philological Sciences. Maykop, 2003, 19 p. (In Russ.) Khalilova Z. M. Causatives in the Bezhtinsky language. Vestnik VSU. Series: Linguistics and intercultural communication. 2017, no. 4, pp. 11–115. (In Russ.) Moldanova I. M. Semantic types of causative verbs of the Khanty language (on the material of the Kazym dialect). Native language: linguistic journal. 2018, no. 2, pp. 81– (In Russ.) Kormushin I. V. On grammatical and lexical in verbal causatives. In: Tyurkologicheskiy sbornik. Moscow, Nauka Publ., 1966, pp. 64–73. (In Russ.) Yuldashev A. A. Ponuditelnyi zalog. In: Comparative-historical grammar of the Turkic languages. Morphology. Moscow, Nauka Publ., 1988, pp. 284– (In Russ.) Guzev V. G. Essays on the theory of Turkic inflection. Verb. Leningrad, LSU Publ. House, 1990, 164 p. (In Russ.) Danilova N. I. Category of causativity in the Yakut language (based on the material of the novel by N. E. Mordinov – Amma Achchygyia “Spring Time”). In: O. N. Betlingk and questions of Turkic philology: materials of the All-Russian scientific conference dedicated to the 160th anniversary of the publication of the work of Academician O. N. Betlingk “On the Yakut language” (Yakutsk, June 23–24, 2011). Yakutsk, IHRISN SB RAS Publ., 2013, pp. 189–193. (In Russ.) Danilova N. I., Efremov N. N., Samsonova E. M. Functional and semantic categories in the Yakut language. Causativity. Evidentiality. Iteratively. Novosibirsk, Nauka Publ., 2013, 240 pp. (In Russ.) Almadakova N. D. Causative verbs and causative voice. In: Grammar of the modern Altai language. Morphology. Editorial board: I. A. Nevskaya (ed.), N. D. Almadakova, A. N. Maizina, etc. Gorno-Altaisk, S. S. Surazakov Research Institute of Altaistics, 2017, pp. 229–230. (In Russ.) Telyakova V. M. One-act causative and passive constructions of the Shor language in comparative coverage. Siberian Philological Journal. 2009, no. 3, pp. 140– (In Russ.) Telyakova V. M. Causative constructions with predicate actants in the Shor language in comparative lighting. Bulletin of the Novosibirsk State University. Series: History, Philology. 2009, vol. 8, no. 2, pp. 84–89. (In Russ.) Grammar of the Khakas language. Ed. N. A. Baskakov. Moscow, Nauka Publ., 1975, 417 p. (In Russ.) Chugunekova A. N. Morphological means of expression of the causative verb in Khakas language. In: Modern pedagogy: problems and prospects: materials of the International scientific-practical conference dedicated to the Year of the Bashkir language, the 85th anniversary since the birth of Z. G. Uraksin and the 85th birthday of M. V. Zaynullin (Ufa, November 18–21, 2020). Ufa, Samrau Publ. House, 2020, pp. 431–435. (In Russ.) Chugunekova A. N. Models of elementary simple sentences of causal object displacement in space (on the material of the Khakas language). Oriental Studies. 2020, vol. 13, no. 1, pp. 208–224. DOI: 10.22162/2619-0990-2020-47-1-208-223. (In Russ.) Kholodovich A. A. Typology of causative constructions. Morphological causative. Leningrad, Nauka Publ., 1969, 311 p. (In Russ.) Altyn Aryg: Heroic tales recorded from S. P. Kadyshev. Ed. T. G. Tachieva. Abakan, Khakas House, 1987, 232 p. (In Khakas) Akh Chibek Aryg. The heroic legend. Recorded and prepared for publication by T. G. Tachiev. Abakan, Khakas Branch of the Krasnoyarsk Book Publ. House, 1968, 180 p. (In Khakas) Khara Khuskhun. Heroic legend in the Khakas language. Prepared for publication by V. E. Mainagasheva and A. A. Todanova. Abakan, Khakas Branch of the Krasnoyarsk Book Publ. House, 1977, 195 p. (In Khakas) Alyptyg nymakh. Prepared for publication Y. I. Chaptykova and N. S. Chistobaeva. Abakan, Khakas Book Publ. House, 2018, 126 p. (In Khakas) Ai Huuchin: A heroic tale written by P. V. Kurbizhekov. Ed. by V. E. Mainagasheva. Abakan, Khakas Publ. House, 1991, 302 p. (In Khakas) Dadueva E. A. Semantic features of constructions with double-causation verbs in the Buryat language. Tomsk Journal of Linguistic and Anthropological Research. Series Linguistics. 2019, iss. 4, pp. 9–16. (In Russ.)