This work contributes a preliminary cross-system comparison for an underexplored language pair and lays the groundwork for larger-scale evaluations of gender-inclusive AT, indicating that all four models struggle to faithfully convey gender information from source to target.
Machine translation (MT) systems often fail to correctly translate gender, especially when converting from a gender-neutral language like English to a gendered target language such as Romanian. This bias results in translations that default to masculine forms or reinforce gender stereotypes. We propose a hybrid pipeline to mitigate this issue by combining large language model (LLM)-based gender classification with neural machine translation (NMT). Our system uses a fine-tuned LLM to detect the intended gender of target words in English sentences and insert inline gender hint tags. These tagged sentences are then passed to a Transformer model fine-tuned to generate morphologically correct Romanian translations. To support this, we introduce three novel datasets for gender disambiguation and translation. Our approach improves gender accuracy on the WinoMT and WinoGender benchmarks by over 40 percentage points compared to a baseline MT system. This is the first method to explicitly address and evaluate gender bias in English-Romanian MT using both LLM inference and tag-aware translation.
I. Grigore, Sergiu Nisioi· International Conference on...· 0 citations
This study investigates the translation of gender from English into Arabic, focusing on the performance of senior-level translation students in a literary context. Building on previous research on verbal and adjectival translation, the article adopts a mixed-methods approach combining quantitative frequency analysis with qualitative examination of error types. The corpus consists of selected passages from Naguib Mahfouz’s Midaq Alley, translated into Modern Standard Arabic by student translators. Gender-related renderings are classified as similar, different or unattempted in order to assess both accuracy and omission. The findings indicate that while gender is generally handled correctly at the lexical level, significant difficulties arise in maintaining agreement within phrases and clauses, particularly in cases involving inanimate or abstract nouns and structurally complex constructions. The study highlights the impact of grammatical asymmetry between English and Arabic on translation performance and underscores the need for more systematic instruction in contrastive grammar and contextual analysis. It contributes to a more nuanced understanding of gender as a key factor in English–Arabic literary translation and provides implications for translator training.
K. Mansoor, Daniel Dejica· Scientific Bulletin of the P...· 0 citations
Assessment of how GT and GPT-4o translate English VPE into Arabic focuses on the accuracy of ellipsis reconstruction and the translation strategies employed, highlighting the importance of better-quality training data for NMT and LLM tools for both discourse-level processing and context-dependent data.
Eassa Ali, Abbas Brashi, Dana Awad et al.· The International Journal of...· 0 citations
The Cross-Lingual Comprehension Gap (CLCG) is defined as the reduction in response quality when the same content and question are presented in a target language rather than in English.
J-PragEval-v0 is introduced, a minimal-pair benchmark isolating four such phenomena from surface fluency, and Pragmatic Representation Steering is specified, a parameter-free inference-time method that edits residual-stream activations along the class-mean-difference directions probing identifies.
This paper analyses various recent state-of-the-art variants of large language models (LLMs) and neural machine translation (NMT) for Indian languages in comparison to statistical machine translation (SMT) and tackles key questions, such as idiomatic expressions, morphologically complex grammar or the scarceness of parallel corpora.
Jayanand A. Kamble, S. Jadhav, V. J. Kadam· International Journal of Inf...· 0 citations
