Abstract Large language models (LLMs) are increasingly used for lexical retrieval tasks, yet it remains unclear how they operationalize lexical ambiguity, particularly regarding the distinction between homonymy and polysemy. While diachronic lexicography usually distinguishes homonymy as the convergence of etymologically unrelated lexical forms, and polysemy as the divergence of meanings within a single lexical item, LLMs may rely primarily on synchronic associations when generating examples of ambiguous words. This study investigates how three publicly available LLMs (ChatGPT, Claude, and Gemini) retrieve lexically ambiguous items under three prompt conditions: a neutral ambiguity prompt, a prompt requesting homonyms without etymological constraints, and a prompt explicitly requiring etymologically unrelated homonyms. Retrieved items were manually classified as either homonymous or polysemous through verification against established etymological sources. The results show that neutral prompts predominantly yielded polysemous items, while increasingly restrictive prompts produced progressively higher proportions of homonyms. However, substantial differences emerged across models, and even under explicit diachronic constraints some outputs reflected semantic extension, unsupported etymological reasoning, or plausible but historically unattested forms. The findings suggest that ambiguity retrieval in LLMs is highly sensitive to prompt formulation and that models tend to operationalize lexical ambiguity primarily in synchronic terms unless explicitly instructed otherwise. The study demonstrates that prompt design plays a crucial role in the retrieval of homonyms in lexicographic applications of LLMs.
Abstract Recognizing the increasing significance of cross-cultural dictionaries in a world marked by conflict, this article addresses issues in the scholarship and practice of cross-cultural lexicography. It begins by clarifying key concepts, arguing that cross-cultural lexicography is a purpose-driven activity designed to foster understanding and meaningful engagement across cultures. Building on Bergenholtz and Tarp’s distinction between lexicographic purposes and functions, the article then proposes a preliminary theoretical framework for cross-cultural lexicography that integrates lexicographic purposes and functions. This framework is illustrated through a case study of A Historical Dictionary of Chinese Cultural Terminologies, demonstrating how lexicographic functions are applied to fulfill lexicographic purposes. Ultimately, this study seeks to improve the development of cross-cultural dictionaries, thereby promoting more effective and meaningful cultural engagement.
Abstract In this paper, we advance relational lexicography as a framework for understanding contemporary community-centred dictionary work for Indigenous, endangered, and minoritized languages. Reviewing the results of a global survey of 144 practitioners working across 126 languages, we demonstrate that dictionary-making is increasingly collaborative, ethical and reflexive, and embedded in wider networks of language revitalization, education, archives, and place-based knowledge. The responses show that dictionaries are no longer understood primarily as static reference products, but as dynamic social processes shaped by relationships among speakers, learners, researchers, dialects, technologies, and communities. These findings are situated within broader shifts from colonial and authoritative models of lexicography toward critical, decolonial, and participatory approaches. We argue, therefore, that relational lexicography provides a useful conceptual framework for conceptualizing dictionary-making as a socially situated, culturally accountable, and future-oriented practice, while exploring its practical, ethical, institutional, and theoretical implications for contemporary language documentation and revitalization in community-centred and collaborative lexicographic projects worldwide today.
Abstract This contribution starts and ends with the work of Patrick Hanks (1940-2024), the forerunner of GenAI in lexicography, as well as the most impactful practical and theoretical lexicographer going into the future. Patrick Hanks, in other words, has most literally been ‘citing the future’ for many decades preceding that future. Following a preamble in which I look back to look forward, I take his Festschrift as a starting point, presented to him when he turned 70, and move on to his voluminous International Handbook of Modern Lexis and Lexicography, completed halfway upon his death. Next, and as a case study, I employ a sample of 100 articles from the field of ‘GenAI in lexicography’ to first chart the sample’s evaluations of earlier work. I then proceed by looking at (1) the languages studied in the sample, (2) the related academic disciplines that come into play, (3) the topics covered, (4) the perceived usefulness of ‘GenAI in lexicography’, (5) the AI tools used or discussed, and (6) the dictionary slots given attention. Based on all of this, the article concludes with the true impact of Patrick Hanks on GenAI in lexicography, both in academic terms (spanning a period of 40 years: from the COBUILD project in 1983 to the widespread utilisation of ChatGPT as of 2023) and personal terms (spanning a period of 35 years: from Ken Church in 1989 to G-M de Schryver in 2024).
Abstract This study explores the potential of Generative AI (GenAI), specifically ChatGPT, for the acquisition and semantic organization of collocations in support of collocation-resource construction. We examine ChatGPT’s ability to generate English collocation candidates for eleven semantic collocation categories formalized as Lexical Functions (LFs). LFs are treated as a fine-grained intermediate analytical layer from which broader, user-oriented semantic groupings can be derived. Two prompting strategies are compared: a few-shot approach, in which the model receives twenty illustrative examples of a given LF and generates fifty further candidates, and an iterative profile-refinement approach, in which a lexicographer guides the model in constructing a semantic and syntactic LF profile subsequently used for sample generation. The outputs were manually evaluated by specialists in Explanatory Combinatorial Lexicology. The results show that the few-shot strategy captures some central semantic properties of several LFs, while profile refinement generally yields higher accuracy and greater semantic coherence, especially for semantically complex LFs. The acquired LF profiles were then used to prompt ChatGPT to construct a collocation-dictionary entry for Fear, modelled on the format of the Oxford Collocations Dictionary (OCD). The comparison shows that LF-guided prompting favours semantic differentiation and explicit structural organization by sense and semantic function, whereas the OCD entry displays greater idiomatic selectivity and lexicographic compactness. Overall, LF-guided interaction with GenAI appears to be a promising semi-automatic instrument for compiling semantically structured collocation resources that can support dictionary construction, although expert validation, generalization, and post-editing remain necessary.
Abstract This study presents a comparative review of 484 empirical studies on the use of English learner’s dictionaries (ELDs) and Chinese learner’s dictionaries (CLDs) published between 2001 and 2024. Drawing from the Web of Science and CNKI databases, the analysis categorises the literature into eight research themes and seven methodological approaches to delineate the developmental trajectories of these two fields. Results indicate that ELD use research demonstrates relatively broader thematic exploration and increasing adoption of objective, process–oriented methodologies. Concurrently, CLD use research remains in an exploratory phase, predominantly relying on self–reported data to examine user needs and skills, while lacking in–depth inquiry into users’ cognitive processing and the typological features specific to the Chinese language. Despite these developmental differences, both fields exhibit shared gaps regarding how dictionaries are integrated into instructional and autonomous learning contexts, as well as in the exploration of the interplay between dictionaries and other elements within the language learning resource ecosystem. The paper concludes by proposing a research agenda that incorporates frameworks from educational technology and the learning sciences to advance this ecological perspective. It calls for further research into the use of AI–enhanced learner’s dictionaries (AILDs) in response to the evolving technological landscape. Furthermore, the specificities of the Chinese language and its acquisition warrant greater international scholarly attention, constituting a crucial yet under–researched dimension in current CLD use studies.
This study investigates the metabolism of 329 English loanwords listed in the Modern Chinese Dictionary (MCD1-7) from an ecolinguistic perspective, focusing on their distribution, vitality, and endangerment. First, the synchronic ecology of these loanwords was analyzed using the MCD as a static corpus, which provides a comprehensive overview of their current status. Second, their diachronic ecology was evaluated through niche breadth measurements based on the Media Language Corpus (MLC), a dynamic corpus. To further validate the dynamic vitality of these loanwords, a Questionnaire of Lexical Acknowledgement was designed and implemented. The results reveal significant variation in the ecology of these loanwords, influenced by both external and internal factors, such as word length, frequency, semantic domain, part of speech, and translation method. Notably, only 113 loanwords exhibit medium ecological niches, indicating that approximately one third retain meaningful vitality. Among the 96 endangered loanwords, 32 are extinct, 39 are critically endangered, 19 are endangered, and 6 are vulnerable. The study underscores that the metabolism of English loanwords in Chinese is influenced by external factors (ecological environment) and internal factors (linguistic system).
This study compared the effectiveness of a widely used AI chatbot (ChatGPT) and an online dictionary (Cambridge Dictionary) in enhancing L2 vocabulary learning during word lookup. Ninety-three Saudi university students from various academic disciplines and institutions nationwide were assigned to three groups: ChatGPT, dictionary, and control. The ChatGPT and dictionary groups participated in an intervention where they looked up twenty lower-frequency words, while the control group received no treatment. Immediately afterward, all groups took posttests, and delayed posttests were administered one week later. Quantitative results showed no significant differences between the ChatGPT and dictionary groups on either outcome at either posttest. Although previous research has reported advantages for ChatGPT in lexical task performance, much of that work has focused on immediate outcomes rather than longer-term retention. Th findings of the present study suggest that both tools may provide sufficiently rich lexical input to support vocabulary learning. Overall, well-designed learner dictionaries remain pedagogically robust resources and continue to serve as a useful benchmark for evaluating emerging AI technologies.
For the ongoing creation of a database of Spanish verbs, we developed a methodological proposal for the automatic tagging of semantic types in running text according to Hanks' Corpus Pattern Analysis (CPA) guidelines. In this task, a text document is the input and the output is the tagging of each noun, noun phrase or proper noun with one of the semantic types in the CPA Ontology. The present proposal is based on a combination of algorithms for automatic ontology population, named entity recognition and, most importantly, word sense disambiguation, to assign the appropriate type to a noun according to the context. The paper includes an evaluation of the method tagging a random sample of 200 Wikipedia pages in Spanish and English. Evaluation figures by a panel of three experts show 84% precision and 88% recall in Spanish and 83% precision and 93% recall in English. These are competitive results considering the simplicity and computational efficiency of the algorithm.
The Termini Online Hungarian Dictionary (TOHD) - cf. Ben & odblac; et al., 2025 - documents the lexicon of Hungarian as spoken in countries neighbouring Hungary. The dictionary includes two types of lexical items from Hungarian language varieties used beyond the national borders: on the one hand, items that are wholly or partly of foreign origin; on the other hand, lemmas that also occur in standard Hungarian but display different meanings. The recognition of Hungarian as a pluricentric language - widely acknowledged since the late 1990s - provides an important theoretical linguistic framework for documenting these regionally differentiated lexical items (see Kontra, 2022, p. 43; Fenyvesi, 2005) and elevates its importance to a higher level. Each entry is illustrated with authentic example sentences, which allow for the analysis of a word's specific grammatical and pragmatic usage. The present study examines the novelties and distinctive features of the dictionary in detail, with particular attention to its structure, editorial challenges, and educational use. Since dictionary use forms part of the Hungarian National Core Curriculum, the paper also discusses pedagogical approaches that help students become familiar with language varieties and with the Termini Online Hungarian Dictionary, while simultaneously developing their dictionary-use skills. With the help of the dictionary, students gain a thorough understanding of the pluricentric nature of the Hungarian language, which enhances sensitivity to regional language varieties and fosters a broader, more inclusive perspective on language use and linguistic identity in the digital age.
The aim of this article is to discuss the Polish philologist Albert (Wojciech) de Biberstein Kazimirski (1808-1887), and one of his works - the frequently reissued, though now largely forgotten Dictionnaire fran & ccedil;ais-polonais (Paris 1839), a plagiarized version of which caused a great controversy in the nineteenth century among Polish elites. After a brief presentation of the author's biography, the publishing history of the dictionary is examined in detail, with particular emphasis on the first edition. This is followed by an in-depth analysis of its macrostructure and microstructure. To better understand the dictionary's success among Poles living in Paris at the time, it is essential to understand the circumstances in which it was created. Discussing it against the background of the era and highlighting the author's other achievements provides a clearer picture of this successful work.
With the growing integration of mobile technology into language learning, dictionary apps have become useful tools for English as a Foreign Language (EFL) learners. This study investigated Chinese user feedback on the Longman Dictionary of Contemporary English (LDOCE) app to uncover user needs, perceptions, and sentiments. Based on 4,643 user comments collected between August 2016 and December 2024, the research employed word frequency analysis, semantic network construction, Latent Dirichlet Allocation (LDA) topic modelling, and sentiment analysis. Five key themes were identified: technical issues, version updates, lexical content, dictionary perceptions, and pronunciation. The findings showed that while users valued the app's authoritative content and learning features, they also raised concerns about app crashes, monetization, and insufficient pronunciation options. Furthermore, although the app's functions have evolved rapidly over the past decade, the dictionary content itself has only been updated once, raising questions about content maintenance in dictionary apps. This study demonstrates that user feedback provides valuable insights that can guide both the functional design and lexicographic priorities of future dictionary apps.
Corpus Pattern Analysis (CPA) is a technique for mapping meaning onto words in text. It was first proposed by Patrick Hanks in 2004 and has since been applied to detect and analyse recurrent syntagmatic patterns centred around verbs across various languages. In this paper, following a suggestion by Hanks himself (Hanks 2004a, 2004b, 2012, 2013), we explore the possibility of applying it to patterns centred around nouns. We conduct an exploratory study of three Spanish nouns with the goal of identifying their most recurrent patterns. Results show that: i. the current CPA apparatus can be successfully used to identify noun patterns, but it requires adjustments and extensions, particularly, the construction of a new ontology for verbs and adjectives; ii. in contrast to verbs, nouns can have more than one pattern per meaning, especially in the case of literal senses, and their meaning may be assigned by collocates that are outside their pattern; iii. metaphorical patterns show more syntactico-semantic restrictions, which may be useful for establishing links between metaphors and language.
In this paper, we explore the relationship between language and ontology. We present the results of a manual linking exercise we performed between the CPA (Hanks 2004) and the DOLCE (Masolo et al. 2003) ontologies, with the goal of identifying the distinctions and similarities between the two systems from a cognitive and applied perspective. The motivation for our research stems from the differing nature of the two resources. The CPA ontology is a bottom-up system in which categories are identified through manual clustering of the noun fillers of argument positions of verbs gathered from large corpora. In contrast, DOLCE is a top-down ontology, where categories are not based on extensive linguistic evidence but are instead stipulated on formal grounds. We are interested in verifying whether the two methodologies (bottom-up and top-down) lead to differences in the resulting ontologies. The preliminary results reveal that the most general types in T-PAS can be mapped fairly well into DOLCE's upper level. The experiment also shows that the bottom-up system is finer-grained and anthropocentric, in the sense that its categories reflect human experience, as conveyed through language use. On the other hand, the DOLCE taxonomy is ontologically more robust than the CPA hierarchy. Two substantial issues remain open: the mapping of the Abstract category, which is interpreted differently in the two ontologies, and the treatment of systematic polysemy.
As a long-term scientific project on the history of the Italian language, the Lessico Etimologico Italiano (LEI) represents one of the most important and ambitious historical and etymological dictionary projects ever undertaken. The LEI, which started in 1968, documents and analyzes every single word of the Italian language and all Italian dialects from their beginnings to today. Until 2018, the editorial process used the traditional lexicographical method of creating annotated index cards to collect information. Each index card contains text areas, handwritten annotations, and/or textual stamps. Of particular interest are the etymon, context (the source text), and textual stamp, which contains the abbreviated title of the book or source from which it was copied or taken. In this paper, we present a novel approach for efficiently processing a large number of scanned index cards to accelerate the philological work required for producing the LEI. For this purpose, a deep learning workflow for automatic detection, alignment, and recognition of textual stamps on digitized index cards is proposed. To support large-scale indexing under an initially incomplete stamp inventory, we introduce an embedding-based retrieval workflow that enables the identification and integration of previously unseen stamps during operation. Our experimental evaluations show excellent results for stamp detection and stamp recognition, with a mean average precision of 98.80% and an accuracy of 97.02%, respectively. In addition, we compare our stamp recognition approach with two recent Large Language Model (LLM) approaches for OCR. Our best approach achieves 97.02% accuracy, which surpasses the OCR performance of the two LLMs (91.60% and 89.47%, respectively). Furthermore, we compare our approach to a recent multimodal large language model (MM-LLM) on a benchmark dataset and observe substantially lower accuracy for the LLM (up to 84.58%) compared to our approach (98.61%).
When ChatGPT became publicly available in November 2022, artificial intelligence (AI) transformed our daily lives. In the paper, the use of an AI tool, namely the Copilot and Kraj & scaron;avar algorithms, are compared with respect to recognizing abbreviations and expansions in Italian texts, notably scientific papers in the field of tourism. An overview of Italian dictionaries of abbreviations is first provided, before an introduction to Kraj & scaron;avar-an algorithm designed to automatically recognize abbreviations and expansions in electronic texts. A brief presentation of the procedure for obtaining texts in the field of tourism follows. The texts were selected in line with the text typology classification developed by Mikoli & ccaron; (Mikoli & ccaron;, 2007). After the classification, scientific papers from the field of tourism were selected before being filtered with Kraj & scaron;avar and Copilot. The findings are presented and discussed with a view to including rules so as improve the recognition of abbreviations and expansions in Italian texts in a future compilation of Italian dictionaries of abbreviations.
