Skip to content

Read-only version: For security reasons, the Research Information System (FIS) has been temporarily replaced by this read-only version. All public content and full texts remain available; login, editing, submission and statistics are not available. Questions: fis@uni-bamberg.de

Home › Publications › LLM meets BSLVC: Does ChatGPT share grammatical preferences with selected varieties of English?

Publication

LLM meets BSLVC: Does ChatGPT share grammatical preferences with selected varieties of English?

Deckert, Katharina; Vetter, Fabian (2026): LLM meets BSLVC: Does ChatGPT share grammatical preferences with selected varieties of English?, in: Bamberg: Otto-Friedrich-Universität, doi: 10.20378/irb-117467.

Details

Faculty/Chair
Faculty/Professorship
Author
Vetter, Fabian
Conference
21th Biennial International Conference on the Linguistics of Contemporary English (BICLCE), Klagenfurt (Austria)
Publisher Information
Otto-Friedrich-Universität : Bamberg
Year of publication
2026
Pages
1
Language
English
DOI
Licence
URN
Abstract

The effect Large Language Models (LLMs) have on (scientific) writing has been called “truly unprecedented” (Kobak et al. 2025: 6). For instance, they tend to shorten sentences (Rudnicka 2023), influence vocabulary usage (Kobak et al. 2025) and might cause “similarity in writing or content” (Liang et al. 2024) in academic writing. Combining the preference of machine learning models for frequent structures (Schneider 2022: 370) and the omnipresence of LLMs, it may very well be that AI-modified texts might be “homogenized along US norms”(Mair 2024: 19) while the status of “non-dominant minor and emerging national standards” might weaken (Mair 2024: 18). This contribution seeks to elucidate i) whether ChatGPT, unless prompted otherwise, indeed shares grammatical preferences with a specific variety of English, and ii) whether it is able to reproduce grammatical preferences of other varieties of English.

To this end, we prompt ChatGPT to fill in the Bamberg Survey of Language Variation and Change (BSLVC; Krug and Sell 2013) multiple times with five prompts: As a speaker of L1 varieties, i.e. (i) British English (BrE) and (ii) American English (AmE), of an L2 variety, i.e. (iii) Indian English, of a learner variety, i.e. (iv) German Learner English, and in the default mode, i.e. (v) without being given a specific variety of English it should adhere to. In a second step, we utilize the BSLVC dashboard (Vetter 2025) to compare the synthetic data to existing BSLVC data. ChatGPT’s default mode is expected to align most with AmE because of Mair’s (2024) assumption regarding the homogenization of texts. Based on his assumption concerning the weakening of non-dominant national standards in favor of the American standard, prompting ChatGPT to answer the questionnaire as an AmE speaker is presumed to align more with actual native speaker judgments compared to other variety specifications.

GND Keywords
Amerikanisches Englisch
Standardsprache
ChatGPT
Keywords
Large Language Models
grammatical variation
World Englishes
BSLVC
learner English
corpus linguistics
DDC Classification
420 English
RVK Classification
HF 614
Type
Conferenceobject
URI
Activation date
2026-11-03
Project(s)

Permalink: https://fis.uni-bamberg.de/handle/uniba/117467 · Data as JSON · Cite: BibTeX · RIS

Data as of: 2026-11-16

Restricted read-only access: login, editing and submission are temporarily disabled.