Neural Network Interference in EFL Students' Academic Writing: A Longitudinal Corpus Study

Szerzők és előadók

Olga G. Gorina · Svetlana N. Kucherenko

Szekció és helyszín

AI, Writing, Knowledge and Academic Practice
2026. november 25., szerda · 13:00–14:15
Díszterem, Szentkirályi utca 28.

Teljes, beküldött absztrakt

This study examines how students' academic writing in English changes under the influence of generative AI. We analyse a longitudinal corpus of Research Proposals written by fourth-year Bachelor students as part of an English for Academic Purposes course between 2021 and 2026. The corpus covers three periods: before AI became widely available (2021–2022), a transitional year (2023), and the phase of regular AI use (2024–2026). We introduce the term neural network interference to describe how AI-generated language patterns enter students' writing and reshape it. Three types are distinguished. Conscious interference occurs when students deliberately use AI suggestions and evaluate them critically. Unconscious interference means copying AI patterns without reflection. Zero interference refers to cases where students use a construction correctly without understanding its grammar or pragmatic function. Alongside interference we observe acceleration: lexical sophistication grows faster than grammatical control, creating an uneven skill profile consistent with the notion of the learner's "muddle" (Lewis, 1993). Corpus data reveal changes across multiple linguistic levels, including hedging patterns, syntactic variety, sentence length distribution, and stylistic consistency within individual texts. We argue that these findings call for a reconsideration of writing assessment criteria and EAP pedagogy. We also propose diagnostic metrics that help instructors distinguish genuine skill development from surface-level AI fluency.

Forrás