Prior works have shown that fine-tuning on new knowledge can induce factual hallucinations in large language models (LLMs), leading to incorrect outputs when evaluated on previously known information. However, the specific manifestations of such hallucination and its underlying mechanisms remain insufficiently understood. Our work addresses this gap by designing a controlled dataset Biography-Reasoning, and conducting a fine-grained analysis across multiple knowledge types and two task types, including knowledge question answering (QA) and knowledge reasoning tasks. We find that hallucinations not only severely affect tasks involving newly introduced knowledge, but also propagate to other evaluation tasks. Moreover, when fine-tuning on a dataset in which a specific knowledge type consists entirely of new knowledge, LLMs exhibit elevated hallucination tendencies. This suggests that the degree of unfamiliarity within a particular knowledge type, rather than the overall proportion of new knowledge, is a stronger driver of hallucinations. Through interpretability analysis, we show that learning new knowledge weakens the model’s attention to key entities in the input question, leading to an over-reliance on surrounding context and a higher risk of hallucination. Conversely, reintroducing a small amount of known knowledge during the later stages of training restores attention to key entities and substantially mitigates hallucination behavior. Finally, we demonstrate that disrupted attention patterns can propagate across lexically similar contexts, facilitating the spread of hallucinations beyond the original task.

Large Language Models have demonstrated impressive reasoning capabilities across multiple languages. However, the relationship between capabilities in different languages is less explored. In this work, we decompose the process of reasoning tasks into two separated components: knowledge retrieval and knowledge-free reasoning, and analyze the relationship between cross-lingual transferability and these two components. With adapted commonsense reasoning datasets and constructed knowledge-free reasoning datasets, we show that the knowledge-free reasoning capability can be nearly perfectly transferred across various source-target language directions despite the secondary impact of resource in some specific target languages, while cross-lingual knowledge retrieval significantly hinders the transfer. Moreover, by analyzing the hidden states and feed-forward network neuron activation during the reasoning, we show that higher similarity of hidden representations and larger overlap of activated neurons could explain the better cross-lingual transferability of knowledge-free reasoning than knowledge retrieval. Thus, we hypothesize that knowledge-free reasoning shares similar neurons in different languages for reasoning, while knowledge is stored separately in different languages.

Despite their strong ability to retrieve knowledge in English, current large language models show imbalance abilities in different languages. Two approaches are proposed to address this, i.e., multilingual pretraining and multilingual instruction tuning. However, whether and how do such methods contribute to the cross-lingual knowledge alignment inside the models is unknown. In this paper, we propose CLiKA, a systematic framework to assess the cross-lingual knowledge alignment of LLMs in the Performance, Consistency and Conductivity levels, and explored the effect of multilingual pretraining and instruction tuning on the degree of alignment. Results show that: while both multilingual pretraining and instruction tuning are beneficial for cross-lingual knowledge alignment, the training strategy needs to be carefully designed. Namely, continued pretraining improves the alignment of the target language at the cost of other languages, while mixed pretraining affect other languages less. Also, the overall cross-lingual knowledge alignment, especially in the conductivity level, is unsatisfactory for all tested LLMs, and neither multilingual pretraining nor instruction tuning can substantially improve the cross-lingual knowledge conductivity.

Large Language Models (LLMs) possess extensive knowledge and strong capabilities in performing in-context reasoning. However, previous work challenges their out-of-context reasoning ability, i.e., the ability to infer information from their training data, instead of from the context or prompt. This paper focuses on a significant aspect of out-of-context reasoning: Out-of-Context Knowledge Reasoning (OCKR), which is to combine multiple knowledge to infer new knowledge. We designed a synthetic dataset with seven representative OCKR tasks to systematically assess the OCKR capabilities of LLMs. Using this dataset, we evaluated several LLMs and discovered that their proficiency in this aspect is limited, regardless of whether the knowledge is trained in a separate or adjacent training settings. Moreover, training the model to reason with reasoning examples does not result in significant improvement, while training the model to perform explicit knowledge retrieval helps for retrieving attribute knowledge but not the relation knowledge, indicating that the model’s limited OCKR capabilities are due to difficulties in knowledge retrieval. Furthermore, we treat cross-lingual knowledge transfer as a distinct form of OCKR, and evaluate this ability. Our results show that the evaluated model also exhibits limited ability in transferring knowledge across languages.