LitTransAI 1.0: A DeepSeek-Powered Literary Translation Platform with Integrated Machine Learning Algorithms

Abstract

Literary translation presents distinct challenges for machine translation systems, demanding preservation of stylistic, cultural, and emotional nuances that extend beyond basic linguistic accuracy. While large language models like ChatGPT-4 demonstrate considerable potential, they frequently encounter difficulties with cultural specificity and poetic elegance. We introduce LitTransAI 1.0, a novel DeepSeek-assisted platform that integrates a multidimensional scoring framework with a diagnostic feedback generation mechanism specifically designed for literary texts, addressing a gap in specialized, pedagogically viable tools. The system employs a hybrid architecture featuring a semantic-cohesion scorer for evaluating content fidelity and narrative coherence, a stylistic analyzer for capturing lexical diversity and syntactic complexity, and a cultural-context validator for identifying culture-specific items. A random-forest classifier, trained on a carefully curated corpus of human-translated literary works, categorizes translation errors and delivers prioritized, template-guided feedback. For practical deployment, we implement a lightweight P-Tuning v2 configuration that ensures low latency on classroom-grade hardware. On a balanced evaluation set comprising translated poems and prose excerpts from diverse literary periods, LitTransAI 1.0 achieves Quadratic Weighted Kappa scores of 0.82 for Content, 0.78 for Style, and 0.75 for Cultural Fidelity, closely matching the performance of a larger generative LLM baseline (DeepSeek) while operating 5.2 times faster at 135 tokens per second compared to 26 tokens per second. The feedback mechanism reduces false positives to 15%, substantially lower than the baseline's 58%. In an eight-week pedagogical intervention with translation students, platform users demonstrated significant improvements in handling metaphors, maintaining cultural references, and preserving authorial style. Instructor feedback indicated high levels of perceived usefulness and time efficiency, collectively demonstrating that LitTransAI 1.0 provides a reliable, scalable, and pedagogically viable solution for literary translation education and practice.

Ähnliche Arbeiten