PASS-FC: Progressive and Adaptive Search Scheme for Fact Checking of Comprehensive Claims

TL;DR

PASS-FC addresses time-sensitive and entity-ambiguous claims by grounding atomic facts in explicit time spans and unique entity descriptors, then employing an adaptive, multilingual search loop with credible-source filtering and a reflection mechanism. Across six benchmarks and ten languages, it outperforms strong baselines, including larger LLMs, highlighting the value of temporal grounding and cross-lingual evidence. Ablation studies confirm the critical role of temporal grounding and the adaptive search, while cross-lingual retrieval provides genuinely new support. The work advances automated fact-checking toward robust, time-aware, multilingual evidence gathering and veracity labeling, with code and results to enable further research.

Abstract

Automated fact-checking (AFC) still falters on claims that are time-sensitive, entity-ambiguous, or buried beneath noisy search-engine results. We present PASS-FC, a Progressive and Adaptive Search Scheme for Fact Checking. Each atomic claim is first grounded with a precise time span and disambiguated entity descriptors. An adaptive search loop then issues structured queries, filters domains through credible-source selection, and expands queries cross-lingually; when necessary, a lightweight reflection routine restarts the loop. Experiments on six benchmark--covering general knowledge, scientific literature, real-world events, and ten languages--show that PASS-FC consistently outperforms prior systems, even those powered by larger backbone LLMs. On the multilingual X-FACT set, performance of different languages partially correlates with typological closeness to English, and forcing the model to reason in low-resource languages degrades accuracy. Ablations highlight the importance of temporal grounding and the adaptive search scheme, while detailed analysis shows that cross-lingual retrieval contributes genuinely new evidence. Code and full results will be released to facilitate further research.

Figures (12)

• Figure 1: Workflow comparison between a traditional fact-checking pipeline (left) and PASS-FC (right). PASS-FC enriches atomic claims with temporal and entity information, then employs advanced and multilingual search to collect relevant and sufficient evidence. Text inside rectangles shows retrieved snippets.
• Figure 2: Overview of the PASS-FC framework. We claim all of the procedures highlighted in bold orange as our initial contributions to the automatic fact-check (AFC) task.
• Figure 3: Temporal grounding example for the claim "Universal Studios features a Madagascar zone."
• Figure 4: Hyperparameter analysis. (a) and (b): Impact of evidence number and reflection trigger labels on performance, using 100 randomly sampled AVeriTeC training examples. (c) Performance gains from iterations across all datasets using GPT-4omini, and on Factbench using GPT-4o. Evaluation metrics include evidence recall (where applicable) and macro-F1 score.
• Figure 5: Analysis of XLE. (a): Distribution of languages chosen by XLE in the force-XLE setting from the ablation study. (b) The distribution of reasons that choose every language.