Literature discovery is a critical component of scientific research. Modern discovery systems leveraging Large Language Models (LLMs) are increasingly adopted for their ability to process natural language queries (NLQs). To assess the robustness of such systems, we compile two NLQ datasets and submit them to nine widely used discovery platforms. Our findings reveal that LLM-based search engines struggle with precisely formulated queries, often producing numerous false positives. However, precision improves when LLMs are used not for direct retrieval but to convert NLQs into structured keyword-based queries. As a result, hybrid systems that integrate both LLM-driven and keyword-based approaches outperform purely keyword-based or purely LLM-based discovery methods.

Misinformation on social media poses significant risks, particularly when it concerns critical scientific issues such as climate change. One promising direction for mitigation is the development of automated fact-checking systems that verify claims against authoritative scientific sources. In this work, we present our solution to the ClimateCheck2025 shared task, which involves retrieving and classifying scientific abstracts as evidence for or against given claims. Our system is built around a multi-stage hybrid retrieval pipeline that integrates lexical, sparse neural, and dense neural retrievers, followed by cross-encoder and large language model (LLM)-based reranking stages. For stance classification, we employ prompting strategies with LLMs to determine whether a retrieved abstract supports, refutes, or provides no evidence for a given claim. Our approach achieves the second-highest overall score across both subtasks of the benchmark and significantly surpasses the official baseline by 53.79% on average across Recall@2, Recall@5, Recall@10, and B-Pref. Notably, we achieve state-of-the-art performance in Recall@2. These results highlight the effectiveness of combining structured retrieval architectures with the emergent reasoning capabilities of LLMs for scientific fact verification, especially in domains where reliable human annotation is scarce and timely intervention is essential.

Good scientific writing makes use of specific sentence and paragraph structures, providing a rich platform for discourse analysis and developing tools to enhance text readability. In this vein, we introduce SciPara, a novel dataset consisting of 981 scientific paragraphs annotated by experts in terms of sentence discourse types and topic information. On this dataset, we explored two tasks: 1) discourse category classification, which is to predict the discourse category of a sentence by using its paragraph and surrounding paragraphs as context, and 2) discourse sentence generation, which is to generate a sentence of a certain discourse category by using various contexts as input. We found that Pre-trained Language Models (PLMs) can accurately identify Topic Sentences in SciPara, but have difficulty distinguishing Concluding, Transition, and Supporting Sentences. The quality of the sentences generated by all investigated PLMs improved with amount of context, regardless of discourse category. However, not all contexts were equally influential. Contrary to common assumptions about well-crafted scientific paragraphs, our analysis revealed that paradoxically, paragraphs with complete discourse structures were less readable.

Scientific abstracts provide a concise summary of research findings, making them a valuable resource for extracting scientific arguments. In this study, we assess various unsupervised approaches for extracting arguments as aligned premise-conclusion pairs: semantic similarity, text perplexity, and mutual information. We aggregate structured abstracts from PubMed Central Open Access papers published in 2022 and evaluate the argument aligners in terms of the performance of language models that we fine-tune to generate the conclusions from the extracted premise given as input prompts. We find that mutual information outperforms the other measures on this task, suggesting that the reasoning process in scientific abstracts hinges mostly on linguistic constructs beyond simple textual similarity.

The abstracts of scientific papers typically contain both premises (e.g., background and observations) and conclusions. Although conclusion sentences are highlighted in structured abstracts, in non-structured abstracts the concluding information is not explicitly marked, which makes the automatic segmentation of conclusions from scientific abstracts a challenging task. In this work, we explore Normalized Mutual Information (NMI) as a means for abstract segmentation. We consider each abstract as a recurrent cycle of sentences and place two segmentation boundaries by greedily optimizing the NMI score between the two segments, assuming that conclusions are strongly semantically linked with preceding premises. On non-structured abstracts, our proposed unsupervised approach GreedyCAS achieves the best performance across all evaluation metrics; on structured abstracts, GreedyCAS outperforms all baseline methods measured by P_k. The strong correlation of NMI to our evaluation metrics reveals the effectiveness of NMI for abstract segmentation.

In scientific papers, arguments are essential for explaining authors’ findings. As substrates of the reasoning process, arguments are often decorated with discourse indicators such as “which shows that” or “suggesting that”. However, it remains understudied whether discourse indicators by themselves can be used as an effective marker of the local argument components (LACs) in the body text that support the main claim in the abstract, i.e., the global argument. In this work, we investigate whether discourse indicators reflect the global premise and conclusion. We construct a set of regular expressions for over 100 word- and phrase-level discourse indicators and measure the alignment of LACs extracted by discourse indicators with the global arguments. We find a positive correlation between the alignment of local premises and local conclusions. However, compared to a simple textual intersection baseline, discourse indicators achieve lower ROUGE recall and have limited capability of extracting LACs relevant to the global argument; thus their role in scientific reasoning is less salient as expected.