Linda Zeng

Linda Zeng, Rithwik Gupta, Divij Motwani et al.

Retrieval-augmented generation (RAG) has shown impressive capabilities in mitigating hallucinations in large language models (LLMs). However, LLMs struggle to maintain consistent reasoning when exposed to misleading or conflicting evidence, especially in real-world domains such as politics, where information is polarized or selectively framed. Mainstream RAG benchmarks evaluate models under clean retrieval settings, where systems generate answers from gold-standard documents, or under synthetically perturbed settings, where documents are artificially injected with noise. These assumptions fail to reflect real-world conditions, often leading to an overestimation of RAG system performance. To address this gap, we introduce RAGuard, the first benchmark to evaluate the robustness of RAG systems against misleading retrievals. Unlike prior benchmarks that rely on synthetic noise, our fact-checking dataset captures naturally occurring misinformation by constructing its retrieval corpus from Reddit discussions. It categorizes retrieved evidence into three types: supporting, misleading, and unrelated, providing a realistic and challenging testbed for assessing how well RAG systems navigate different types of evidence. Our experiments reveal that, when exposed to potentially misleading retrievals, all tested LLM-powered RAG systems perform worse than their zero-shot baselines (i.e., no retrieval at all), while human annotators consistently perform better, highlighting LLMs' susceptibility to noisy environments. To our knowledge, RAGuard is the first benchmark to systematically assess the robustness of the RAG against misleading evidence. We expect this benchmark to drive future research toward improving RAG systems beyond idealized datasets, making them more reliable for real-world applications. The dataset is available at https://huggingface.co/datasets/UCSC-IRKM/RAGuard.

Linda Zeng

Neural Machine Translation (NMT) systems struggle when translating to and from low-resource languages, which lack large-scale data corpora for models to use for training. As manual data curation is expensive and time-consuming, we propose utilizing a generative-adversarial network (GAN) to augment low-resource language data. When training on a very small amount of language data (under 20,000 sentences) in a simulated low-resource setting, our model shows potential at data augmentation, generating monolingual language data with sentences such as "ask me that healthy lunch im cooking up," and "my grandfather work harder than your grandfather before." Our novel data augmentation approach takes the first step in investigating the capability of GANs in low-resource NMT, and our results suggest that there is promise for future extension of GANs to low-resource NMT.

Diji Yang, Linda Zeng, Kezhen Chen et al.

Large Language Models (LLMs) exhibit great potential in complex multi-step reasoning through inference-time thinking but still struggle with deciding when to stop thinking due to limited self-awareness about their knowledge boundaries. While human preference alignment has shown extraordinary opportunities, expensive labeling challenges adherence to scaling law. Language model self-critique, as an alternative to using human-labeled reasoning data, is questioned with its inherited biases. This work addresses these challenges by distilling the LLM's own reasoning processes into synthetic behavioral data, eliminating the need for manual labeling of intermediate steps. Building on this concept, we propose Distillation-Reinforcement-Reasoning (DRR), a three-step framework that leverages the LLM's inherent behaviors as external feedback by first generating behavioral data using the Reasoner (LLM) to reflect its reasoning capabilities, then training a lightweight discriminative reward model (DM) on behavioral data, and finally deploying the DM at inference time to assist the Reasoner's decision-making. Experiments on multiple benchmarks show that the DRR framework outperforms self-critique approaches without relying on additional complex data annotation. Benefiting from lightweight design, ease of replication, and adaptability, DRR is applicable to a wide range of LLM-centric tasks.

Diji Yang, Linda Zeng, Jinmeng Rao et al.

Retrieval Augmented Generation (RAG) has shown strong capability in enhancing language models' knowledge and reducing AI generative hallucinations, driving its widespread use. However, complex tasks requiring multi-round retrieval remain challenging, and early attempts tend to be overly optimistic without a good sense of self-skepticism. Current multi-round RAG systems may continue searching even when enough information has already been retrieved, or they may provide incorrect answers without having sufficient information or knowledge. Existing solutions either require large amounts of expensive human-labeled process supervision data or lead to subpar performance. This paper aims to address these limitations by introducing a new framework, SIM-RAG, to explicitly enhance RAG systems' self-awareness and multi-round retrieval capabilities. To train SIM-RAG, we first let a RAG system self-practice multi-round retrieval, augmenting existing question-answer pairs with intermediate inner monologue reasoning steps to generate synthetic training data. For each pair, the system may explore multiple retrieval paths, which are labeled as successful if they reach the correct answer and unsuccessful otherwise. Using this data, we train a lightweight information sufficiency Critic. At inference time, the Critic evaluates whether the RAG system has retrieved sufficient information at each round, guiding retrieval decisions and improving system-level self-awareness through in-context reinforcement learning. Experiments across multiple prominent RAG benchmarks show that SIM-RAG is an effective multi-round RAG solution. Furthermore, this framework is system-efficient, adding a lightweight component to RAG without requiring modifications to existing LLMs or search engines, and data-efficient, eliminating the need for costly human-annotated mid-step retrieval process supervision data.

Linda Zeng

Code-mixing (CM), where speakers blend languages within a single expression, is prevalent in multilingual societies but poses challenges for natural language processing due to its complexity and limited data. We propose using a large language model to generate synthetic CM data, which is then used to enhance the performance of task-specific models for CM sentiment analysis. Our results show that in Spanish-English, synthetic data improved the F1 score by 9.32%, outperforming previous augmentation techniques. However, in Malayalam-English, synthetic data only helped when the baseline was low; with strong natural data, additional synthetic data offered little benefit. Human evaluation confirmed that this approach is a simple, cost-effective way to generate natural-sounding CM sentences, particularly beneficial for low baselines. Our findings suggest that few-shot prompting of large language models is a promising method for CM data augmentation and has significant impact on improving sentiment analysis, an important element in the development of social influence systems.

Linda Zeng, Steven Y. Feng, Michael C. Frank

Multilingualism is incredibly common around the world, leading to many important theoretical and practical questions about how children learn multiple languages at once. For example, does multilingual acquisition lead to delays in learning? Are there better and worse ways to structure multilingual input? Many correlational studies address these questions, but it is surprisingly difficult to get definitive answers because children cannot be randomly assigned to be multilingual and data are typically not matched between languages. We use language model training as a method for simulating a variety of highly controlled exposure conditions, and create matched 100M-word mono- and bilingual datasets using synthetic data and machine translation. We train GPT-2 models on monolingual and bilingual data organized to reflect a range of exposure regimes, and evaluate their performance on perplexity, grammaticality, and semantic knowledge. Across model scales and measures, bilingual models perform similarly to monolingual models in one language, but show strong performance in the second language as well. These results suggest that there are no strong differences between different bilingual exposure regimes, and that bilingual input poses no in-principle challenges for agnostic statistical learners.