Normalizace nestandardizovaných variet ranné angličtiny pomocí AI jazykových modelů

Titlbachová, Magdalena

(in English): This bachelor's thesis investigates the potential of large language models (LLMs) for normalizing non-standardized varieties of Early Modern English (EModE) into Present-Day English. The aim was to determine how effectively "mainstream", non-specialized models (ChatGPT, Gemini, Copilot, Llama, and Claude) can standardize texts from various periods and genres of EModE - how their outputs compare to manually standardized versions. The analysis covered 15 texts from five genres (literary texts, religious texts, administrative texts, educational texts, correspondence) and three chronological stages (1500-1550, 1550-1600, 1600-1650). The texts were manually standardized and shortened to approximately 500 words. Model performance was evaluated using edit distance (ED), precision, and recall. ChatGPT achieved the best results (average ED 3.1; precision 0.998; recall 0.991), followed by Claude (4.9) and Gemini (5.9). Based on ED, models were more effective in later EModE periods, however, precision and recall remained relatively stable across periods. By genre, the models performed best on religious texts (ED 3.1) and worst on educational texts (12.3). The findings suggest that modern LLMs, particularly ChatGPT, can outperform traditional tools (compared to standardized texts from the Early...

guest :: login Digital Repository
		Search		Submit		Help		About