In Japanese–English machine-translation,especially for creative and localized content, standard MT metrics typically summarize translation quality as scalar scores, but provide little diagnostic information about how style changes in translation. This is limiting because politeness, directness, interpersonal stance, and character voice are encoded differently across Japanese and English, and stylistic change may reflect localization-oriented adaptation rather than error. We propose the Style Diagnostic Profile (SDP) framework: SDP assigns each utterance a seven-dimensional bipolar style vector, and 𝛥sdp, the difference between target and source SDP vectors, represents translation-induced stylistic change as a multi-axis shift vector. To evaluate the framework, we first validate an LLM-based SDP scorer against bilingual human annotations on professional JA→EN subtitles and show that it approximates human style judgments. Using Style25, a controlled JA↔EN style-perturbation benchmark, we demonstrate that different perturbation categories occupy distinct regions in 𝛥sdp space. As a case study, we analyze professional subtitles and observe stylistic adaptation rather than simple style copying. These results position the SDP framework as a diagnostic layer complementary to adequacy-oriented MT evaluation.

Warning: This paper contains examples of stereotypes and social bias. LLMs are increasingly used in interactive settings by the general public, making the evaluation of model behavior in multi-turn conversational scenarios important for safety, including stereotyping-related harms. However, existing multi-turn social bias evaluations often rely on pre-specified or template-based user inputs that do not adapt to model responses and typically assume a fixed dialogue length in advance. In this paper, we study social bias dynamics in response-conditioned multi-turn interactions using a controlled evaluation protocol that generates follow-up user queries from the evolving dialogue history and allows evaluation over variable numbers of turns. Experimental results show that LLMs exhibit social bias even in coherent, response-conditioned multi-turn interactions, revealing late-emerging bias, non-monotonic bias patterns, and bias re-emergence. These results motivate evaluations that extend beyond fixed-turn, pre-scripted protocols. Our findings highlight the importance of analyzing social bias as a turn-level dynamic phenomenon.

LLM-based machine translation systems can now do more than translate a source segment in isolation. At test time, they can translate directly, prepend previous context, or refine a context-aware draft. However, these interventions are not uniformly beneficial. Context can resolve discourse-sensitive choices, but irrelevant context and extra refinement increases cost and may perturb adequate translations. We introduce ContextRouter, a lightweight policy router that learns which intervention to apply for each source segment. During training, we first generate candidates for all contextual interventions and score them with a routing objective combining overall translation quality with incurred token costs. We then train our ContextRouter with group-relative supervision, so it learns within-example preferences rather than absolute global action scores. At inference, our router uses only test-time-available signals from the source, previous source-side context, and a no-context draft to determine and execute only the selected context intervention. On a filtered context-aware subset of TED, QED, and OpenSubtitles covering multiple languages and domains, our ContextRouter improves the quality-cost tradeoff over fixed no-context, always-context, always-refinement, QE-reranking, and score-regression policies, and human evaluation confirms that annotators prefer its chosen interventions. These results support treating context and refinement actions in LLM-MT not as always present defaults but conditional test-time interventions instead.

Warning: This paper contains examples of stereotypes and biases. Large Language Models (LLMs) exhibit considerable social biases, and various studies have tried to evaluate and mitigate these biases accurately. Previous studies use downstream tasks to examine the degree of social biases for evaluation and mitigation. While the output of LLMs highly depends on prompts, prior works evaluating and mitigating bias have often relied on a limited variety of prompts. In this paper, we investigate the sensitivity of LLMs when changing prompt variations (task instruction, few-shot examples, debias-prompt) by analyzing task performance and social bias of LLMs. Our experimental results reveal that LLM rankings fluctuate across prompts for both task performance and social bias. We also confirmed that the impact of format changes can differ for each bias category. Performance improvement from prompt settings may not result in reduced bias. Moreover, the ambiguity of instances is a common factor in LLM sensitivity to prompts across advanced LLMs. We recommend using diverse prompts, as in this study, to compare the effects of prompts on social bias in LLMs.

While large language models (LLMs) excel at machine translation (MT), the impact of how LLMs utilize different forms of contextual information on discourse-level phenomena remains underexplored. We systematically investigate how different forms of context such as prior source sentences, models’ generated hypotheses, and reference translations influence standard MT metrics and specific discourse phenomena (formality, pronoun selection, and lexical cohesion). Evaluating multiple LLMs across multiple domains and language pairs, our findings consistently show that context boosts both translation and discourse-specific performance. Notably, the context strategy of combining source text with the model’s own prior hypotheses effectively improves discourse consistency without gold references, demonstrating effective use of model’s own imperfect generations as diverse contextual cues.

For extracting meaningful topics from texts, their structures should be considered properly. In this paper, we aim to analyze structured time-series documents such as a collection of news articles and a series of scientific papers, wherein topics evolve along time depending on multiple topics in the past and are also related to each other at each time. To this end, we propose a dynamic and static topic model, which simultaneously considers the dynamic structures of the temporal topic evolution and the static structures of the topic hierarchy at each time. We show the results of experiments on collections of scientific papers, in which the proposed method outperformed conventional models. Moreover, we show an example of extracted topic structures, which we found helpful for analyzing research activities.