Criminal investigators and intelligence analysts have developed structured analytic techniques to evaluate competing hypotheses under incomplete information. This study examines whether such human expert investigative methodologies are also effective for narrative-based culprit inference in large language models (LLMs). Focusing on the task of analyzing evidence from complex narratives and identifying the perpetrator among suspects, we conducted experiments on 10 LLMs using the MuSR murder mystery benchmark. The PRISM framework, which applies investigative techniques, consistently outperformed existing general-purpose strategies across all models, with its effectiveness manifesting regardless of model scale. Ablation studies revealed that the hypothesis structuring stage is particularly crucial, accounting for 89% of the methodological improvement beyond information filtering. This suggests that domain-specific structures that specify “what to analyze” are more effective in LLM reasoning than simply increasing the number of reasoning paths.

If large language models (LLMs) are deployed to analyze evidence and evaluate suspects in criminal investigations, are they free from the very trap that has led countless human investigators to misjudgment—implicit bias swayed by information irrelevant to the essence of the case? To answer this question, this study systematically injected personas (gender, race, relationship) into neutralized murder mystery scenarios and examined the reasoning stability of LLMs. Experimental results revealed that implicit bias propagation was observed across all models. The phenomenon where models outwardly state “that information is irrelevant to the judgment” while their actual conclusions are already influenced by the injected persona was universally observed. Interestingly, model scale alone did not guarantee stability: while the largest model achieved the lowest instability, several smaller models outperformed much larger ones. The most notable finding concerns the differential vulnerability across persona types: while race and gender were processed relatively stably, relationship information—particularly hostile relationships—induced significantly higher reasoning contamination. More concerning is the fact that even when conclusions were correctly maintained, the reasoning process itself was extensively contaminated. These findings suggest that current alignment techniques have created a blind spot by focusing on identity-based bias while neglecting relationship-based bias, and propose that stability evaluation should encompass not only outputs but also reasoning processes.

Large Language Models (LLMs) demonstrate strong reasoning and self-correction abilities in high-resource languages like English, but their performance remains limited in low-resource languages such as Korean. In this study, we investigate whether reinforcement learning (RL) can enhance Korean reasoning abilities to a degree comparable to English. Our findings reveal that RL alone yields limited improvements when applied to models lacking inherent Korean reasoning capabilities. To address this, we explore several fine-tuning strategies and show that aligning the model’s internal reasoning processes with Korean inputs—particularly by tuning Korean-specific neurons in early layers—is key to unlocking RL’s effectiveness. We introduce a self-correction code-switching dataset to facilitate this alignment and observe significant performance gains in both mathematical reasoning and self-correction tasks. Ultimately, we conclude that the crucial factor in multilingual reasoning enhancement is not injecting new linguistic knowledge, but effectively eliciting and aligning existing reasoning capabilities. Our study provides a new perspective on how internal translation and neuron-level tuning contribute to multilingual reasoning alignment in LLMs.

This paper describes Jeonbuk National University (JBNU)’s system for the 2019 shared task on Cross-Framework Meaning Representation Parsing (MRP 2019) at the Conference on Computational Natural Language Learning. Of the five frameworks, we address only the DELPH-IN MRS Bi-Lexical Dependencies (DP), Prague Semantic Dependencies (PSD), and Universal Conceptual Cognitive Annotation (UCCA) frameworks. We propose a unified parsing model using biaffine attention (Dozat and Manning, 2017), consisting of 1) a BERT-BiLSTM encoder and 2) a biaffine attention decoder. First, the BERT-BiLSTM for sentence encoder uses BERT to compose a sentence’s wordpieces into word-level embeddings and subsequently applies BiLSTM to word-level representations. Second, the biaffine attention decoder determines the scores for an edge’s existence and its labels based on biaffine attention functions between roledependent representations. We also present multi-level biaffine attention models by combining all the role-dependent representations that appear at multiple intermediate layers.

Neural machine translation decoders are usually conditional language models to sequentially generate words for target sentences. This approach is limited to find the best word composition and requires help of explicit methods as beam search. To help learning correct compositional mechanisms in NMTs, we propose concept equalization using direct mapping distributed representations of source and target sentences. In a translation experiment from English to French, the concept equalization significantly improved translation quality by 3.00 BLEU points compared to a state-of-the-art NMT model.