The English language is changing faster than before, partly due to the influence of the Internet. Digital language includes a large number of discourse markers (DMs), many of which can be considered innovative. Acronymization, pragmatic specialisation, and compensatory lexical innovation are the most common lexical processes that can be witnessed in the DMs used in computer-mediated communication (CMC). The following novel DMs were identified in recent Twitter chats: lol, tbh, omg, meh, and idk. These DMs perform several functions, such as showing emotions, signalling uncertainty, hesitation, or mitigation. Interpreting these functions may not be an easy or obvious task for AI. The primary aim of the study is to evaluate the pragmatic competence of an LLM, Gemini 3 Pro, regarding the interpretation of these novel DMs. A mixed-method research process was employed: LLM-generated outputs were compared with the findings of the relevant literature, quantitative corpus analysis, and our qualitative human interpretation to assess the model’s analytical usefulness. Gemini 3 Pro was found to show a high level of pragmatic competence in terms of interpreting the functions of DMs, but sometimes tended to overgeneralise, or failed to understand the tone of the text and the intention of the speaker to use a DM.

Lexical innovation refers to the process of creating new lexical items, enabling languages to adapt to evolving socio-cultural and material realities. The domains of business, economics, and finance are among the most productive ones of lexical innovation. The present research study lies at the intersection of lexical innovation, idiomaticity, and large language model (henceforth, LLM) research and investigates lexical productivity, semantic shift, and globalization (Anglocentric changes) in business-related colour idioms by comparing human translation and annotation with the output of LLMs. The current experiment involves an initial study carried out for five languages: English (the pivotal one), Albanian (AL), Hebrew (HE), Hungarian (HU), Lithuanian (LT), and Standard European Portuguese (PT). The research results reveal that LLMs show high mutual agreement, but the agreement with humans is lower. The internal consistency of LLMs reflects shared Anglocentric metaphor encoding rather than convergence toward human idiomatic usage. It demonstrates that human expertise remains essential for high-quality idiomatic translation, particularly for culture-specific expressions.

Potentially idiomatic expressions (PIEs) carry meanings inherently tied to the everyday experience of a given language community. As such, they constitute an interesting challenge for assessing the linguistic (and to some extent cultural) capabilities of NLP systems. In this paper, we present XMPIE, a parallel multilingual and multimodal dataset of potentially idiomatic expressions. The dataset, containing 34 languages and over ten thousand items, allows comparative analyses of idiomatic patterns among language-specific realisations and preferences in order to gather insights about shared cultural aspects. This parallel dataset allows evaluation of language model performance for a given PIE in different languages and whether idiomatic understanding in one language can be transferred to another. Moreover, the dataset supports the study of PIEs across textual and visual modalities, to measure to what extent PIE understanding in one modality transfers or implies in understanding in another modality (text vs. image). The data was created by language experts, with both textual and visual components crafted under multilingual guidelines, and each PIE is accompanied by five images representing a spectrum from idiomatic to literal meanings, including semantically related and random distractors. The result is a high-quality benchmark for evaluating multilingual and multimodal idiomatic language understanding.