Lost in Multilinguality: Dissecting Cross-lingual Factual Inconsistency in Transformer Language Models

Multilingual language models (MLMs) store factual knowledge across languages but often struggle to provide consistent responses to semantically equivalent prompts in different languages. While previous studies point out this cross-lingual inconsistency issue, the underlying causes remain unexplored. In this work, we use mechanistic interpretability methods to investigate cross-lingual inconsistencies in MLMs. We find that MLMs encode knowledge in a language-independent concept space through most layers, and only transition to language-specific spaces in the final layers. Failures during the language transition often result in incorrect predictions in the target language, even when the answers are correct in other languages. To mitigate this inconsistency issue, we propose a linear shortcut method that bypasses computations in the final layers, enhancing both prediction accuracy and cross-lingual consistency. Our findings shed light on the internal mechanisms of MLMs and provide a lightweight, effective strategy for producing more consistent factual outputs.

Through the LLM Looking Glass: A Socratic Self-Assessment of Donkeys, Elephants, and Markets

While detecting and avoiding bias in LLM-generated text is becoming increasingly important, media bias often remains subtle and subjective, making it particularly difficult to identify and mitigate. In this study, we assess media bias in LLM-generated content and LLMs' ability to detect subtle ideological bias. We conduct this evaluation using two datasets, PoliGen and EconoLex, covering political and economic discourse, respectively. We evaluate eight widely used LLMs by prompting them to generate articles and analyze their ideological preferences via self-assessment. By using self-assessment, the study aims to directly measure the models' biases rather than relying on external interpretations, thereby minimizing subjective judgments about media bias. Our results reveal a consistent preference of Democratic over Republican positions across all models. Conversely, in economic topics, biases vary among Western LLMs, while those developed in China lean more strongly toward socialism.

XCOMPS: A Multilingual Benchmark of Conceptual Minimal Pairs

We introduce XCOMPS in this work, a multilingual conceptual minimal pair dataset covering 17 languages. Using this dataset, we evaluate LLMs' multilingual conceptual understanding through metalinguistic prompting, direct probability measurement, and neurolinguistic probing. By comparing base, instruction-tuned, and knowledge-distilled models, we find that: 1) LLMs exhibit weaker conceptual understanding for low-resource languages, and accuracy varies across languages despite being tested on the same concept sets. 2) LLMs excel at distinguishing concept-property pairs that are visibly different but exhibit a marked performance drop when negative pairs share subtle semantic similarities. 3) Instruction tuning improves performance in concept understanding but does not enhance internal competence; knowledge distillation can enhance internal competence in conceptual understanding for low-resource languages with limited gains in explicit task performance. 4) More morphologically complex languages yield lower concept understanding scores and require deeper layers for conceptual reasoning.

On Relation-Specific Neurons in Large Language Models

In large language models (LLMs), certain neurons can store distinct pieces of knowledge learned during pretraining. While knowledge typically appears as a combination of relations and entities, it remains unclear whether some neurons focus on a relation itself -- independent of any entity. We hypothesize such neurons detect a relation in the input text and guide generation involving such a relation. To investigate this, we study the Llama-2 family on a chosen set of relations with a statistics-based method. Our experiments demonstrate the existence of relation-specific neurons. We measure the effect of selectively deactivating candidate neurons specific to relation $r$ on the LLM's ability to handle (1) facts whose relation is $r$ and (2) facts whose relation is a different relation $r' \neq r$. With respect to their capacity for encoding relation information, we give evidence for the following three properties of relation-specific neurons. $\textbf{(i) Neuron cumulativity.}$ The neurons for $r$ present a cumulative effect so that deactivating a larger portion of them results in the degradation of more facts in $r$. $\textbf{(ii) Neuron versatility.}$ Neurons can be shared across multiple closely related as well as less related relations. Some relation neurons transfer across languages. $\textbf{(iii) Neuron interference.}$ Deactivating neurons specific to one relation can improve LLM generation performance for facts of other relations. We will make our code publicly available at https://github.com/cisnlp/relation-specific-neurons.

Bring Your Own Knowledge: A Survey of Methods for LLM Knowledge Expansion

Adapting large language models (LLMs) to new and diverse knowledge is essential for their lasting effectiveness in real-world applications. This survey provides an overview of state-of-the-art methods for expanding the knowledge of LLMs, focusing on integrating various knowledge types, including factual information, domain expertise, language proficiency, and user preferences. We explore techniques, such as continual learning, model editing, and retrieval-based explicit adaptation, while discussing challenges like knowledge consistency and scalability. Designed as a guide for researchers and practitioners, this survey sheds light on opportunities for advancing LLMs as adaptable and robust knowledge systems.

Understanding In-Context Machine Translation for Low-Resource Languages: A Case Study on Manchu

In-context machine translation (MT) with large language models (LLMs) is a promising approach for low-resource MT, as it can readily take advantage of linguistic resources such as grammar books and dictionaries. Such resources are usually selectively integrated into the prompt so that LLMs can directly perform translation without any specific training, via their in-context learning capability (ICL). However, the relative importance of each type of resource e.g., dictionary, grammar book, and retrieved parallel examples, is not entirely clear. To address this gap, this study systematically investigates how each resource and its quality affects the translation performance, with the Manchu language as our case study. To remove any prior knowledge of Manchu encoded in the LLM parameters and single out the effect of ICL, we also experiment with an encrypted version of Manchu texts. Our results indicate that high-quality dictionaries and good parallel examples are very helpful, while grammars hardly help. In a follow-up study, we showcase a promising application of in-context MT: parallel data augmentation as a way to bootstrap the conventional MT model. When monolingual data abound, generating synthetic parallel data through in-context MT offers a pathway to mitigate data scarcity and build effective and efficient low-resource neural MT systems.

NoLiMa: Long-Context Evaluation Beyond Literal Matching

Recent large language models (LLMs) support long contexts ranging from 128K to 1M tokens. A popular method for evaluating these capabilities is the needle-in-a-haystack (NIAH) test, which involves retrieving a "needle" (relevant information) from a "haystack" (long irrelevant context). Extensions of this approach include increasing distractors, fact chaining, and in-context reasoning. However, in these benchmarks, models can exploit existing literal matches between the needle and haystack to simplify the task. To address this, we introduce NoLiMa, a benchmark extending NIAH with a carefully designed needle set, where questions and needles have minimal lexical overlap, requiring models to infer latent associations to locate the needle within the haystack. We evaluate 12 popular LLMs that claim to support contexts of at least 128K tokens. While they perform well in short contexts (<1K), performance degrades significantly as context length increases. At 32K, for instance, 10 models drop below 50% of their strong short-length baselines. Even GPT-4o, one of the top-performing exceptions, experiences a reduction from an almost-perfect baseline of 99.3% to 69.7%. Our analysis suggests these declines stem from the increased difficulty the attention mechanism faces in longer contexts when literal matches are absent, making it harder to retrieve relevant information.

A Federated Approach to Few-Shot Hate Speech Detection for Marginalized Communities

Hate speech online remains an understudied issue for marginalized communities, and has seen rising relevance, especially in the Global South, which includes developing societies with increasing internet penetration. In this paper, we aim to provide marginalized communities living in societies where the dominant language is low-resource with a privacy-preserving tool to protect themselves from hate speech on the internet by filtering offensive content in their native languages. Our contribution in this paper is twofold: 1) we release REACT (REsponsive hate speech datasets Across ConTexts), a collection of high-quality, culture-specific hate speech detection datasets comprising seven distinct target groups in eight low-resource languages, curated by experienced data collectors; 2) we propose a solution to few-shot hate speech detection utilizing federated learning (FL), a privacy-preserving and collaborative learning approach, to continuously improve a central model that exhibits robustness when tackling different target groups and languages. By keeping the training local to the users' devices, we ensure the privacy of the users' data while benefitting from the efficiency of federated learning. Furthermore, we personalize client models to target-specific training data and evaluate their performance. Our results indicate the effectiveness of FL across different target groups, whereas the benefits of personalization on few-shot learning are not clear.

How far can bias go? -- Tracing bias from pretraining data to alignment

As LLMs are increasingly integrated into user-facing applications, addressing biases that perpetuate societal inequalities is crucial. While much work has gone into measuring or mitigating biases in these models, fewer studies have investigated their origins. Therefore, this study examines the correlation between gender-occupation bias in pre-training data and their manifestation in LLMs, focusing on the Dolma dataset and the OLMo model. Using zero-shot prompting and token co-occurrence analyses, we explore how biases in training data influence model outputs. Our findings reveal that biases present in pre-training data are amplified in model outputs. The study also examines the effects of prompt types, hyperparameters, and instruction-tuning on bias expression, finding instruction-tuning partially alleviating representational bias while still maintaining overall stereotypical gender associations, whereas hyperparameters and prompting variation have a lesser effect on bias expression. Our research traces bias throughout the LLM development pipeline and underscores the importance of mitigating bias at the pretraining stage.

Derivational Morphology Reveals Analogical Generalization in Large Language Models

What mechanisms underlie linguistic generalization in large language models (LLMs)? This question has attracted considerable attention, with most studies analyzing the extent to which the language skills of LLMs resemble rules. As of yet, it is not known whether linguistic generalization in LLMs could equally well be explained as the result of analogical processes, which can be formalized as similarity operations on stored exemplars. A key shortcoming of prior research is its focus on linguistic phenomena with a high degree of regularity, for which rule-based and analogical approaches make the same predictions. Here, we instead examine derivational morphology, specifically English adjective nominalization, which displays notable variability. We introduce a new method for investigating linguistic generalization in LLMs: focusing on GPT-J, we fit cognitive models that instantiate rule-based and analogical learning to the LLM training data and compare their predictions on a set of nonce adjectives with those of the LLM, allowing us to draw direct conclusions regarding underlying mechanisms. As expected, rule-based and analogical models explain the predictions of GPT-J equally well for adjectives with regular nominalization patterns. However, for adjectives with variable nominalization patterns, the analogical model provides a much better match. Furthermore, GPT-J's behavior is sensitive to the individual word frequencies, even for regular forms, a behavior that is consistent with an analogical account of regular forms but not a rule-based one. These findings refute the hypothesis that GPT-J's linguistic generalization on adjective nominalization involves rules, suggesting similarity operations on stored exemplars as the underlying mechanism. Overall, our study suggests that analogical processes play a bigger role in the linguistic generalization of LLMs than previously thought.