The paper investigates the quality of machine translation (MT) and traces its development through two main approaches – Statistical Machine Translation (SMT) and Neural Machine Translation (NMT) – by comparing English-to-Slovak outputs produced by Google Translate. The aim of the paper is to evaluate the quality of MT outputs from the point of view of two typologically different languages – English, a predominantly analytic language, and Slovak, a primarily synthetic language – using a sample of newspaper texts, which are often translated by machine due to their wide vocabulary and varied subject matter. The research results indicate that NMT (obtained in 2023), compared to its predecessor SMT (obtained in 2017), has significantly improved in almost all framework categories. The NMT output is much more fluent, sounding more natural and comprehensible. In contrast, shortcomings can be found in the category of syntactic-semantic correlativeness and lexical semantics. In such cases, neural MT may struggle to select the appropriate fit-in-context meaning; moreover, these lexemes can further shift the meaning of the entire sentence, clause, or even utterance.
The text analyses the verbal means by which communicators, especially members of the younger generation, express their evaluative attitudes towards various types of denotations in the electronic communication environment. These linguistic means are divided according to the types of evaluation into ambiguous in meaning, allowing different interpretations depending on the context, and further into means expressing positive and negative evaluation. The author’s primary focus is on the predominant English-derived elements, encompassing both single words and phrases, as well as English sentences in their original form. The majority of these words are currently used slang neologisms, which are characterised by expressiveness, economy of expression, linguistic creativity, and a striving for originality. It is also important to consider motivation and the methods used in relation to the specific characteristics of communication in the online environment.
This study investigates the adaptation of the Mistral 7B large language model for the Slovak language, addressing the limited availability of high-quality open-source models for low-resource languages. While commercial models like GPT-4 and Claude exhibit strong Slovak proficiency, their proprietary nature restricts transparency and customization. To overcome this, we fine-tuned the open-weight Mistral 7B model using the Araneum Slovacum VII Maximum corpus (5.3 billion tokens), creating a specialized Slovak variant, Mistral-SK-7b. The training, conducted on the Leonardo supercomputer (10,000 GPU hours), yielded significant improvements: the fine-tuned model generates grammatically correct and contextually coherent Slovak text, eliminating the errors (code-switching, repetition loops, and lexical interference from other languages) observed in the original Mistral-7B-v0.1. The resulting model, released under the Apache 2.0 license, provides a publicly accessible resource for Slovak NLP applications while preserving the base model’s multilingual capabilities. Our work demonstrates the feasibility of adapting state-of-the-art LLMs for linguistically underrepresented languages and underscores the role of open models in promoting digital language preservation.
The objective of the article is twofold: first, to employ the knowledge of the recurrence of low-frequency words in authorial texts; and second, to prevent the misuse of this knowledge. Contrary to the prevailing authorship attribution theory and practice (Evert et al. 2017, Juola 2008), our research has revealed that the personal linguistic profile is not primarily composed of frequent words with grammatical functions. Instead, we have identified that a distinct set of full-meaning words defines an individual’s linguistic profile (Faltýnek 2020, Faltýnek – Matlach 2021). An examination of these meanings reveals an individual’s unconscious language habits and, consequently, their personality settings. Such personal profiling is referred to as “deep content” and “deep sentiment analysis”. The innovation in question has the potential to facilitate a novel form of linguistic personalization in digital communication, one that has not been previously observed or utilized. The main aim of this article is to describe the algorithm to conduct single-person linguistic deep content and deep sentiment profiling and personalization. We will describe technical steps to provide such a form of digital communication processing and to facilitate the adjustment of a text targeted at an individual, described as a System and method for adapting text-based data structures to text samples (Patent No.: US11797753B2, Faltýnek et al. 2023). This algorithm can be used to (a) produce a personal linguistic profile (analogically to psychometrics instruments such as NEO-FFI Big Five, Minnesota Multiphasic Personality Inventory (MMPI)), (b) target digital communication to an individual by “translating” a text to their language (i.e. linguistic habits) and stimulate desired feelings to a predetermined content. The algorithm is, however, also designed (c) to be used to avoid procedures (a) and (b) using any kind of digital communication platform by an individual. This algorithm is implemented in the software Cloakspeech (Faltýnek – Benešová – Kučera 2025), which provides personalization of AI-generated texts: AI speaks like a particular person.
Creating comprehensible documents for asylum seekers from Arabic-speaking regions presents a challenge due to differences in legal cultures, institutionally established norms, and vastly different social and cultural backgrounds. Such materials should consider the recipients’ level of understanding within their social context, cultural conventions, and the diverse connotations of terms used. This article examines selected features of original Slovak legal documentation on migration and asylum seekers translated into Arabic, alongside its updated and revised version developed within the framework of the RESPLAIN project. It also explores ways to simplify Arabic legal language for recipients from diglossic Arabic-speaking environments arriving in Slovakia, where understanding complex legal texts in Modern Standard Arabic (MSA) cannot be taken for granted.
