Language models (LMs) are pre-trained on raw text datasets to generate text sequences token-by-token. While this approach facilitates the learning of world knowledge and reasoning, it does not explicitly optimize for linguistic competence. To bridge this gap, we propose L2T, a pre-training framework integrating Language Learning Tasks alongside standard next-token prediction. Inspired by human language acquisition, L2T transforms raw text into structured input-output pairs to provide explicit linguistic stimulation. Pre-training LMs on a mixture of raw text and L2T data not only improves overall performance on linguistic competence benchmarks but accelerates its acquisition, while maintaining competitive performance on general reasoning tasks.
This study explores the application of linguistic theories to the analysis of the English language, with a specific focus on the interplay between syntactic structures and semantic interpretation. The primary objective is to investigate how formal grammatical rules govern sentence construction and how these rules, in turn, influence meaning and communicative effectiveness. To achieve this, the research employs a comparative analytical framework, examining a corpus of authentic written and spoken English data.Particular attention is given to complex phenomena such as syntactic ambiguity, where a single sentence can yield multiple interpretations based on its structure. For instance, the phrase "the shooting of the hunters" presents an immediate ambiguity, as it is unclear whether the hunters are the agents performing the action or the recipients of it. Similarly, the scope of quantifiers and negation, as in sentences like "Everyone didn't answer the question," is analyzed to elucidate how logical meaning is derived from grammatical form. The study also delves into the principles of information structure, exploring how concepts like topic, focus, and given versus new information are encoded through syntactic devices such as cleft constructions and passive voice. By systematically examining these areas, this research aims to contribute to a more nuanced model of English grammar that integrates formal syntax with pragmatic function, offering insights valuable to fields such as computational linguistics, language education, and translation studies.
While large language models (LLMs) have shown promise in translating extremely low-resource languages using resources like dictionaries, the effectiveness of grammar books remains debated. This paper investigates the role of grammar books in translating extremely low-resource languages by decomposing it into two key steps: grammar rule retrieval and application. To facilitate the study, we introduce ZhuangRules, a modularized dataset of grammar rules and their corresponding test sentences. Our analysis reveals that rule retrieval constitutes a primary bottleneck in grammar-based translation. Moreover, although LLMs can apply simple rules for translation when explicitly provided, they encounter difficulties in handling more complex rules. To address these challenges, we propose representing grammar rules as code functions, considering their similarities in structure and the benefit of code in facilitating LLM reasoning. Our experiments show that using code rules significantly boosts both rule retrieval and application, ultimately resulting in a 13.1% BLEU improvement in translation.
In this paper, we offer an overview of indigenous languages, identifying the causes of their devaluation and the need for legislation on language rights. We review the technologies used to revitalize these languages, finding that when they come from outside, they often have the opposite effect to what they seek; however, when developed from within communities, they become powerful instruments of expression. We propose that the inclusion of Indigenous knowledge in large language models (LLMs) will enrich the technological landscape, but must be done in a participatory environment that encourages the exchange of knowledge.
Mohammad Zbeeb, Hasan Abed Al Kader Hammoud, Sina Mukalled
et al.
We present AraLingBench: a fully human annotated benchmark for evaluating the Arabic linguistic competence of large language models (LLMs). The benchmark spans five core categories: grammar, morphology, spelling, reading comprehension, and syntax, through 150 expert-designed multiple choice questions that directly assess structural language understanding. Evaluating 35 Arabic and bilingual LLMs reveals that current models demonstrate strong surface level proficiency but struggle with deeper grammatical and syntactic reasoning. AraLingBench highlights a persistent gap between high scores on knowledge-based benchmarks and true linguistic mastery, showing that many models succeed through memorization or pattern recognition rather than authentic comprehension. By isolating and measuring fundamental linguistic skills, AraLingBench provides a diagnostic framework for developing Arabic LLMs. The full evaluation code is publicly available on GitHub.
In recent years, Artificial Intelligence (AI) has become an increasingly influential component of educational practice, particularly in language learning. AI-powered tools are now capable of offering immediate, personalized, and contextually sensitive support, helping learners develop writing accuracy, critical thinking, and revision strategies. In EFL contexts, AI enhances access to high-quality feedback and reduces teachers’ workload by automating lower-level tasks, allowing instructors to focus on formative guidance. This study investigates the expanding role of AI in EFL writing instruction, with a special focus on grammar feedback generated by ChatGPT, Gemini, and Perplexity. The case study involved 16–17-year-old learners from a military-academic lyceum, representing B1–B2 proficiency levels. Using researcher-designed rubrics, AI-generated feedback was evaluated across three key dimensions: technical accuracy, pedagogical clarity, and linguistic appropriacy. The results show that ChatGPT provides human-like, detailed, and supportive explanations that foster understanding, while Gemini delivers concise and structured feedback that suits more autonomous learners. Perplexity, in contrast, offers factual but brief comments, emphasizing accuracy over elaboration. These findings suggest that while AI tools can enhance feedback speed and consistency, their pedagogical value depends heavily on teacher mediation and careful prompt design. The study concludes that AI should not replace the teacher’s role but rather complement it acting as a co-pilot that supports learners through timely, personalized guidance. The research contributes to AI-informed pedagogy by linking feedback theory, rubric-based evaluation, and classroom practice.
As non-native users of English, multilinguals sometimes produce discourses that seemingly deviate from the English norm. One dispute on Facebook, for instance, tagged someone being linguistically incompetent for using double comparatives (DCs), seeing that its use (e.g. more taller) violates the English grammar. Hence this study directs its attention to DC’s emergence in online discourses to explore its acceptability and practicality. As a qualitative content inquiry informed by Chomsky’s (1961) generative grammar and theory of linguistic structure, this study found 9 DC structures, which at a glance, deviate the English norms, suggesting a variation in the application of comparative forms in English. Surprisingly, at a closer look, DCs are linguistic innovations characterizing users’ creativity and practical intent and their way to maximize the flexibility of English to serve a practical purpose. Understanding the purpose of Facebook discourses (e.g. to influence decisions) is crucial in this study to explain the use of DCs. As a result, promotional is the main purpose that justifies the use of DCs as it contains transactional and practical details to influence decisions (e.g. customers to grab discounted services). Significantly, this study introduces three types of DCs used in online discourses not yet explored in linguistics, namely: 1) intralingual (coreference) which coreferences with other lexical items composing units of linguistic references that denote meaning of an entire discourse in semantic level, 2) interlingual (intensifier) which intensifies the presentation of meaning according to how they are intended by users, and 3) extra-lingual (meaning marker) which adds related ideas not explicitly conveyed in a discourse. Although DCs appear to be acceptable in practical uses, particularly in online discourses, applying them in formal writing (where grammar accuracy is crucial) needs consensus among language experts because its usage may not be acceptable to convey technical information.
This study examines lexical and semantic variation in isiXhosa as spoken in the Gqeberha and Mthatha speech communities in South Africa's Eastern Cape Province. Although isiXhosa is predominantly spoken in this region, it exhibits substantial dialectal variation influenced by social, historical, and geographical factors. These influences have led to the emergence of distinct regional varieties. The study focuses on comparing lexical and semantic features across these two urban centres, with the aim of understanding the sociolinguistic dynamics shaping their linguistic identities. The research is anchored in the diglossia theory and adopts a qualitative methodology. Data was gathered through semi-structured interviews with isiXhosa speakers from both Gqeberha and Mthatha. The findings reveal significant differences in vocabulary, lexical semantics, grammar, and accent. The variety spoken in Mthatha aligns more closely with the standardised form of isiXhosa, while the Gqeberha variant reflects greater lexical innovation and contact influence. The Gqeberha variety exhibits stylistic innovations in its lexicon where those imaging words are of no semantic significance in speech. These differences have implications for indigenous language education and policy, particularly regarding the accommodation of regional variation. Furthermore, the study highlights how these linguistic differences have contributed to the formation of sub-hegemonies, emerging identities, and underlying tensions among speakers of the different isiXhosa varieties.
ABSTRACT Right from his earliest studies on the theories of Ferdinand de Saussure (1857–1913), E. F. K. Koerner consistently maintained that Saussure proposed ideas in agreement with, or in response to, the intellectual climate of his time. So much so, that for Koerner, Saussure’s real value lies more in his capacity to synthesise the ideas that were circulating at the end of the nineteenth and the beginning of the twentieth centuries, than in being the first to advance certain concepts, perceived as innovative by many of his later readers (cf. Koerner (1971) 1973; 1988; 2020). The present paper follows the same reasoning by asking what the first reference to Saussure, in 1919, made by the Brazilian philologist Manuel Said Ali Ida (1861–1953), meant within Brazilian philological and linguistic circles, at the turn of the nineteenth to the twentieth century. How could we interpret Said Ali’s reference to Saussure? In this paper, I argue that Said Ali’s gramatical writings synthesise clearly and consistently what was already ‘in the air’ in a fragmented and sometimes misunderstood way of applying the then-new scientific historical-comparative paradigm to Portuguese language history and description.
Nur Ifadloh, Ameen Saliman Abdullahi, Rani Aryanti Rukmana
This study investigates the pragmatic failures found in EFL learners’ academic email communication and evaluates the extent to which AI grammar tools can detect and address such failures. Drawing on theories of interlanguage pragmatics and politeness, the research identifies recurring issues in the realization of requests, apologies, and formal politeness—where learners often produce grammatically correct yet pragmatically inappropriate messages. These failures commonly stem from first-language pragmatic transfer and a lack of explicit instruction in target language norms. Adopting a mixed-methods approach, the study analyzed a corpus of 640 elicited emails from 80 EFL university students and assessed feedback from Grammarly, Quillbot, and ChatGPT using comparative qualitative and quantitative analysis. While the tools effectively corrected surface-level errors, they fell short in addressing context-sensitive pragmatic nuances such as indirectness, tone, and formality. The findings underscore the distinction between linguistic and pragmatic competence, highlight the limitations of current AI tools in fostering pragmatic awareness, and emphasize the need for explicit, context-rich instruction. This study contributes to a more integrated understanding of how human expertise and AI technologies can collaboratively support pragmatic development in digital language learning environments.
Social Media platforms have offered invaluable opportunities for linguistic research. The availability of up-to-date data, coming from any part in the world, and coming from natural contexts, has allowed researchers to study language in real time. One of the fields that has made great use of social media platforms is Corpus Linguistics. There is currently a wide range of projects which have been able to successfully create corpora from social media. In this paper, we present the development and deployment of a linguistic corpus from Twitter posts in English, coming from 26 news agencies and 27 individuals. The main goal was to create a fully annotated English corpus for linguistic analysis. We include information on morphology and syntax, as well as NLP features such as tokenization, lemmas, and n- grams. The information is presented through a range of powerful visualisations for users to explore linguistic patterns in the corpus. With this tool, we aim to contribute to the area of language technologies applied to linguistic research.
Garry Kuwanto, Chaitanya Agarwal, Genta Indra Winata
et al.
Code-switching, the phenomenon of alternating between two or more languages in a single conversation, presents unique challenges for Natural Language Processing (NLP). Most existing research focuses on either syntactic constraints or neural generation, with few efforts to integrate linguistic theory with large language models (LLMs) for generating natural code-switched text. In this paper, we introduce EZSwitch, a novel framework that combines Equivalence Constraint Theory (ECT) with LLMs to produce linguistically valid and fluent code-switched text. We evaluate our method using both human judgments and automatic metrics, demonstrating a significant improvement in the quality of generated code-switching sentences compared to baseline LLMs. To address the lack of suitable evaluation metrics, we conduct a comprehensive correlation study of various automatic metrics against human scores, revealing that current metrics often fail to capture the nuanced fluency of code-switched text. Additionally, we create CSPref, a human preference dataset based on human ratings and analyze model performance across ``hard`` and ``easy`` examples. Our findings indicate that incorporating linguistic constraints into LLMs leads to more robust and human-aligned generation, paving the way for scalable code-switching text generation across diverse language pairs.
While large language models exhibit certain cross-lingual generalization capabilities, they suffer from performance degradation (PD) on unseen closely-related languages (CRLs) and dialects relative to their high-resource language neighbour (HRLN). However, we currently lack a fundamental understanding of what kinds of linguistic distances contribute to PD, and to what extent. Furthermore, studies of cross-lingual generalization are confounded by unknown quantities of CRL language traces in the training data, and by the frequent lack of availability of evaluation data in lower-resource related languages and dialects. To address these issues, we model phonological, morphological, and lexical distance as Bayesian noise processes to synthesize artificial languages that are controllably distant from the HRLN. We analyse PD as a function of underlying noise parameters, offering insights on model robustness to isolated and composed linguistic phenomena, and the impact of task and HRL characteristics on PD. We calculate parameter posteriors on real CRL-HRLN pair data and show that they follow computed trends of artificial languages, demonstrating the viability of our noisers. Our framework offers a cheap solution for estimating task performance on an unseen CRL given HRLN performance using its posteriors, as well as for diagnosing observed PD on a CRL in terms of its linguistic distances from its HRLN, and opens doors to principled methods of mitigating performance degradation.
Computational modeling plays an essential role in the study of language emergence. It aims to simulate the conditions and learning processes that could trigger the emergence of a structured language within a simulated controlled environment. Several methods have been used to investigate the origin of our language, including agent-based systems, Bayesian agents, genetic algorithms, and rule-based systems. This chapter explores another class of computational models that have recently revolutionized the field of machine learning: deep learning models. The chapter introduces the basic concepts of deep and reinforcement learning methods and summarizes their helpfulness for simulating language emergence. It also discusses the key findings, limitations, and recent attempts to build realistic simulations. This chapter targets linguists and cognitive scientists seeking an introduction to deep learning as a tool to investigate language evolution.
Persian possesses a number of stative complex predicates with dâshtan ‘to have’ that express certain kinds of mental state. I propose that these possessive experiencer complex predicates be given a formal semantic treatment involving possession of a portion of an abstract quality by an individual, as in the analysis of property concept lexemes due to Francez and Koontz-Garboden (Language 91(3):533–563, 2015; Natural Language and Linguistic Theory 34:93–106, 2016; Semantics and morphosyntactic variation: Qualities and the grammar of property concepts, Oxford University Press, 2017). Augmented with an analysis of prepositional phrases introducing the target of the mental state and an approach to gradability in terms of measure functions (Wellwood in Measuring predicates, PhD dissertation, University of Maryland, College Park, 2014), the analysis explains various properties of possessive experiencer complex predicates, including the behavior of target phrases, the ability of the non-verbal element to be modified by a range of adjectives, the direct participation of the non-verbal element in comparative constructions, and the ability of degree expressions to modify both the non-verbal element and the VP containing the complex predicate. Theoretically, the analysis ties transitive mental state expressions to the grammar and semantics of property concept sentences, which are expressed via possessive morphosyntax cross-linguistically, and connects with syntactic proposals that independently argue for a universal underlyingly possessive morphosyntax for mental state predicates (Noonan in Case and syntactic geometry, PhD dissertation, McGill University, 1992; Hale and Keyser in Prolegomenon to a theory of argument structure, MIT Press, 2002). The work here also motivates modifications to Francez and Koontz-Garboden’s original proposal, and opens new questions in the original empirical domain of the analysis of possessive predicating strategies for the expression of property concept sentences.
Syftet med denna studie är att bidra med kunskap om läsförståelseundervisning i svenskämnet i mellanstadiet. Material från en observationsstudie i två klasser i årskurs 6 analyseras för att förstå och beskriva hur lärare undervisar i läsförståelse utifrån skönlitterära böcker som eleverna valt själva, så kallade tystläsningsböcker, respektive ett läromedel i svenskämnet. Genom innehållsanalys synliggörs vad som kan vara undervisning för att förstå ett specifikt textinnehåll respektive förståelsestrategier. Trots olika utgångspunkter vad gäller texter och läsaktiviteter ställer båda lärarna frågor och genererar diskussioner på hög nivå. De utmanar även eleverna att tänka om textens innehåll, form och budskap, men undervisar i låg grad om strategier för att förstå och komma ihåg textinnehåll. Ingen av lärarna påpekar explicit för eleverna att det sätt som de tar sig an läsaktiviteterna hjälper dem att förstå textinnehåll och att liknande strategier kan användas vid all läsning. En slutsats är att lärare behöver bli medvetna om att det inte räcker med att låta elever träna på att synliggöra textinnehåll eller att knyta textinnehåll till sig själva, utan det behövs också direkt undervisning om det. Potential att undervisa explicit om när och hur olika strategier kan användas för att stötta elever att bli självständiga läsare diskuteras.
English abstract
Teaching Reading Comprehension in Grade 6 Using Fiction and a Textbook in Swedish
This article aims to contribute to knowledge about the teaching of reading comprehension in the subject Swedish in middle school. Material from an observational study in two grade six classes is analysed to learn about and describe how teachers teach reading comprehension using fiction books chosen by students (so-called silent reading books) and a Swedish textbook, respectively. A content analysis makes visible teaching that supports students in understanding the content of a text, as well as teaching that provides students with reading strategies. Despite differences in their use of texts and reading activities, the teachers in both classrooms ask questions and initiate high-level discussions. They also challenge students to think about the content, form, and message of texts. However, strategies for understanding and remembering text content are taught to a low degree. None of the teachers explicitly point out to students that the way they approach reading activities can help them understand text content and that similar strategies can be used for all reading. One conclusion is that teachers need to become aware that it is not enough to let students practice making text content visible or to connect text content to themselves; explicit teaching about comprehension strategies is also needed. The article also discusses when and how different strategies can be used to support students to become independent readers.
Education (General), Language. Linguistic theory. Comparative grammar
Fixed lexical or syntactical expressions and formulae hallmark legal language. They serve both linguistic and legal purposes, and should be rendered accordingly in a target language and legal system. Most of the times, however, formulaic expressions are translated by resorting to calques, false cognates, or phrases that are uncommon in the target legal language (and legal system). This paper is aimed at exploring how and if corpus analysis can dispel doubts and help find acceptable translation candidates. As there are currently no publicly available legal corpora addressing corporate documents such as contracts and agreements, this paper wishes to bridge this gap by building and relying on an ad hoc corpus of authentic agreements written in English as a first language according to the laws of England and Wales. In this way, corpus evidence can help find equivalents and, possibly, address recurrent mistranslations from Italian into English. During the corpus analysis process, the paper shows and discusses search queries and how equivalents can be obtained. At the same time, it questions dictionary entries. The paper findings highlight that the consultation of the ad hoc corpus allows to find acceptable translations of Italian legal formulae and address recurrent mistranslations. English formulaic expressions, in fact, can be rendered satisfactorily thanks to the possibility of noticing word usages in context, keywords in contexts and collocations. Further research can encompass a wider variety of formulae and/or legal documents so that scholars and translators can be equipped with useful reference tools.
Language. Linguistic theory. Comparative grammar, Comparative law. International uniform law
The idea that until the beginning of the nineteenth century the sentence theory was considered in writings about the theory of eloquence has long been a locus communis in works on the history of Russian linguistics. Russian grammars contained «slovosochinenie» (i. e. syntax) as a syntactic section which stated the rules for connecting different parts of speech into combinations built according to the type of agreement and government. This «commonplace» of syntax historiography deserves a new addition. It is necessary to study the reverse influence issue and answer the question what role the first Russian grammars played in the development of rhetorical theory. The article analyses eighteenth- and nineteenth-century grammatical writings using the methods of extracting, commenting on, and interpreting fragments concerning rhetorical devices. As a result, the study has revealed a close connection between the Russian grammars by M. V. Lomonosov, A. A. Barsov, N. I. Grech, F. I. Buslaev and the theory of elocution. It is particularly true in terms of considering figures of speech, as well as in the sections on the rhetorical period types and the sequence order of the sentence parts. For instance, M. V. Lomonosov in his «Russian Grammar» analyses the examples of «transfer» and «omission», that is, inversion and ellipsis (omission) of sentence «parts». A. A. Barsov, who lists and defines nine «syntactic figures», developed this tradition further. N. I. Grech in his «Practical Russian Grammar» considers reverse word order, ellipsis, «tozhdeslovie» / tautology (intensifying lexical repetition) and «deviation» from strict grammatical agreement of names and verb forms (syllepsis). F. I. Buslaev, enlarging the syntactic section in his «Experience of Historical Grammar» with lexicological topics, gives close attention to tropes that demonstrate different ways of meaning transfer, i. e. metaphor and metonymy (for instance, synecdoche). The paper concludes that Russian linguists of the mid-tolate eighteenth and mid-nineteenth centuries aspired to add a section on the expressive use of grammatical language resources to the syntactic theory.
Teresa Kieseier, Dieter Thoma, M. Vogelbacher
et al.
Abstract Metalinguistic awareness (MLA) is a predictor of adult foreign language (FL) learning in instructed settings. Following Bialystok and Ryan (1985) two-component model of MLA, we distinguish ML analysis as the ability to compare and select language items from ML control as the ability to detect and manipulate rule-based linguistic patterns. So far, there is little evidence how components of MLA interact, and how they affect learning outcomes in early FL learning. In this paper, we investigate the effects of MLA on FL learning of English in primary school, where learning is mostly oral and proficiency in the target language is still rudimentary. We tested 200 students in German public primary schools on English vocabulary and grammar. Metalinguistic (ML) analysis was assessed in an interview, while ML control was operationalized in phoneme manipulation and letter fluency tasks. In addition, we assessed personal, cognitive, and social covariates. Results from linear mixed effects regression analyses indicate that (a) ML control predicts ML analysis skills, (b) ML analysis fosters FL vocabulary outcomes, and (c) ML control benefits grammar skills. We discuss theoretical and practical implications. Plain language summary In this paper, we studied how primary-school-age children’s knowledge about language helps them learning a foreign language. Specifically, we examined two aspects of what has been termed ‘metalinguistic awareness’. Theory assumes that metalinguistic awareness consists of (a) knowledge of how languages work and how they are similar or different (metalinguistic analysis) and (b) skills to manipulate or rearrange elements of language and play with them (metalinguistic control). For instance, speakers know how words are ordered in a sentence and how to add endings to make words rhyme. We were interested in how these two aspects of awareness are connected and how they help learners in different areas of foreign language learning. To answer these questions, we collected data from 200 4th-graders in Germany with different linguistic and social backgrounds learning English as a foreign language. We were particularly interested in the early stages of learning a foreign language, where proficiency is still limited. Results show that metalinguistic awareness indeed consists of an analysis and a control component affecting different parts of foreign language learning. Whereas the ability to compare linguistic structures (metalinguistic analysis) is important for English vocabulary learning, the manipulation of language elements (metalinguistic control) helps learning English grammar. Since different types of metalinguistic awareness are important factors during the learning process, teachers should incorporate activities that foster both in the classroom. Overall, comparing and reflecting on language(s) on a metalinguistic level can be a fruitful approach even at the beginning stages of foreign language learning.