Volume 16, Issue 2

The development of generative AI presents both opportunities and challenges for language teaching. Understanding the linguistic features of AI-generated texts is essential, as it supports users in engaging with AI critically and appropriately for writing tasks. While existing studies have predominantly focused on English writing, the present study examines German argumentative essays produced by ChatGPT, DeepSeek, L1 speakers, and L2 learners, with a focus on linguistic complexity. The results reveal that AI-generated essays generally exhibit higher linguistic complexity. Specifically, DeepSeek essays demonstrate greater lexical complexity, whereas ChatGPT essays are characterized by more complex syntax. In comparison to human-authored essays, AI-generated essays tend to be more formal, marked by frequent nominalizations and a more extensive use of conjunctions. The findings are further interpreted in light of prior research and the underlying mechanisms of generative AI. Based on these results, pedagogical implications for foreign language writing instruction are proposed.

The paper compares texts which are part of Czech language didactic tests, the examinations used as the acceptance prerequisite for secondary schools (eight-year [E8], six-year [E6], and four-year [E4] ones) or as the secondary school final check (called “maturita” [M] in Czech) of the level of knowledge of the pupils’ mother tongue. The corpus comprises 334 texts in 56 tests (more than 66,000 tokens) and covers the timespan of 2019–2023. The indexes employed in the comparison are average token length (ATL), activity (Q), moving-average type–token ratio (MATTR), moving-average morphological richness (MAMR), and verb distances (VD). The statistically significant differences are search for using Kruskal–Wallis test and Dwass–Steel–Critchlow–Fligner test. It has been found out that the texts differ as to ATL, Q, and VD, but further statistical testing declared mostly the E8–M difference as statistically significant. The correlation analysis has confirmed that the three indexes are correlated, this implying that the difference is a product of one, verbalisation-to-nominalization tendency. Finally, we performed a k-means cluster analysis (preceded by t-SNE, used to reduce the number of dimensions), which divided the texts into two groups – “A” (easy) and “B” (difficult). These two groups are distributed rather equally in E8, E6, E4, but in M, the B-texts considerably prevail.

Metadiscourse markers are linguistic tools that help authors or writers, and speakers, to organise their discourse and communicate their ideas to the reader effectively. These are lexical tokens employed by the writers to engage readers in their writings and to convince them of their stance or opinion about the targeted domain. Metadiscourse markers are generally exploited in academic writings, and a multitude of literary articles focus on studying their usage in scientific discourse. In the present research, an endeavour is made to study the usage, application, and impact of metadiscourse markers over fiction, non-fiction, and academic discourses. To thoroughly examine the distribution and role of these linguistic elements, a diverse corpus is assembled, consisting of sixty literary works from fiction, non-fiction, and academic genres. The distributions of various metadiscourse markers are observed, and the concentration of their usage is then analysed and thoroughly discussed. The empirical results reveal that the usage of metadiscourse markers varies across different genres. Academic authors rely more on the interactive features of metadiscourse markers, while writers of fiction and non-fiction literature tend to exercise the interpretative traits of them. Authors or writers aptly plan the usage of metadiscourse markers and corresponding strategies to hit the aimed audience with the desired impact.

Zusammenfassung Die vorliegende Studie arbeitet aktuelle lexikalische Transfer- bzw. Entlehnungsvorgänge (im Aufsatz Transferneologismen genannt) im Zusammenhang mit dem russisch-ukrainischen Krieg am Material der ukrainisch-, der russisch- und der deutschsprachigen Medienkommunikation mittels einer datenbasierten Untersuchung mit primär qualitativem Design heraus. Die themenzentrierte Analyse hat deutlich gemacht, dass tiefgreifende politische Veränderungen kommunikative Folgen für den Sprachgebrauch haben: Die interlingualen Wechselwirkungen erreichen selbst bei Sprachen mit vergleichsweise unterschiedlichen Strukturtypen und Systemen – wie Ukrainisch und Russisch einerseits und Deutsch andererseits – eine bemerkenswerte Intensität. Hinsichtlich des Umgangs mit den Transferneologismen konnten vier zentrale Praktiken festgestellt werden: (a) die Originalform ohne Kommentar, (b) die Originalform + ihre Transkription, (c) nur ihre Transkription und (d) die Originalform + ihre Übersetzung oder Umschreibung. Es wurde dabei nachgewiesen, dass die Entstehung von Okkasionalismen auch im zwischensprachlichen Bereich durchaus ein produktiver Vorgang ist. Die behandelten Okkasionalismen als Neologismen im weiteren Sinne stellen zudem die hochgradige Variabilität des lexikalischen Teilsystems von Sprache unter Beweis.

The article deals with the developing of the German-Ukrainian online specialised dictionary portal that draws on the strengths of academic dictionary production and Internet lexicographic technology. Methodically, this is done by compiling German-Ukrainian specialized dictionaries, the digitisation of which is achieved within the framework of XML-based data modelling for Internet dictionaries in order to guarantee the software independence and longevity of the data. Such data modelling also allows the differentiation between subject areas, the retrieval of collocations, the search for word types as well as the statistical processing of frequencies with regard to their significance. The creation of dictionaries for the portal was realised in five steps. In the first step, a procedure for corpus conception was worked out that enables ad hoc generation of internet-based corpora. In the second step, language corpora were specifically prepared to implement corpus retrieval of single-word and multi-word terms and to isolate specialised vocabulary within common lexis. After an initial compilation of the list of technical terms, the coverage of terms relevant to the subject area was checked using textbooks and selected texts in the subject area. Then the lexicographic processing of the compiled list of technical terms took place. Subsequently, specialised dictionaries were evaluated, i.e. the compiled specialised vocabulary was checked on the basis of the degree of terminological coverage. To ensure an optimal design of the online dictionaries portal, we use a modern standard dictionary compiling programme TshwaneLex , with which any lexicographer can create reference works for any language without basic IT knowledge.