JustiLM is a novel few-shot justification generation model designed for explainable fact-checking of real-world claims. The paper introduces a new benchmark dataset, ExClaim, derived from the WatClaimCheck dataset, containing 6,951 real-world claims with their corresponding veracity labels and human-written justifications. ExClaim also includes a large searchable corpus of 957,949 chunk-level documents for fine-grained evidence retrieval. The proposed JustiLM model is based on a retrieval-augmented language model (RAG) that uses fact-check articles as auxiliary information during training. It leverages a pre-trained RAG model, Atlas, and incorporates distillation techniques to enhance the model's ability to generate justifications based on retrieved evidence. The model is trained using both article-level and chunk-level distillation techniques, which help in aligning the retrieved documents with the fact-check articles and improving the quality of generated justifications. Experiments show that JustiLM outperforms several baselines, including In-Context Learning (ICL) enabled models like Flan-T5, Llama2, and the state-of-the-art few-shot RAG model Atlas. It also performs competitively with the latest GPT-4 model. A straightforward extension of JustiLM for joint veracity prediction and justification generation improves the veracity prediction task with large margins. The paper also discusses the limitations of previous approaches that rely on fact-check articles for justification generation, highlighting the need for a more realistic approach that uses retrieved evidence instead. The proposed method addresses these limitations by using a retrieval-augmented language model and distillation techniques to generate justifications based on retrieved evidence. The results demonstrate that JustiLM can generate high-quality justifications that are more similar to those written by fact-checkers. The model's performance is evaluated using various metrics, including ROUGE, MAUVE, and SummaCC, which measure the quality of generated justifications in terms of factual accuracy, fluency, and consistency. The experiments also show that JustiLM generalizes well to new claims and performs consistently across different test sets. The paper concludes that the proposed approach provides a realistic and effective solution for explainable fact-checking of real-world claims, with the potential to enhance the credibility and trustworthiness of fact-checking systems. Future work includes exploring the adaptation of various LLM-based reasoning methods into JustiLM to further improve the task of justification generation.JustiLM is a novel few-shot justification generation model designed for explainable fact-checking of real-world claims. The paper introduces a new benchmark dataset, ExClaim, derived from the WatClaimCheck dataset, containing 6,951 real-world claims with their corresponding veracity labels and human-written justifications. ExClaim also includes a large searchable corpus of 957,949 chunk-level documents for fine-grained evidence retrieval. The proposed JustiLM model is based on a retrieval-augmented language model (RAG) that uses fact-check articles as auxiliary information during training. It leverages a pre-trained RAG model, Atlas, and incorporates distillation techniques to enhance the model's ability to generate justifications based on retrieved evidence. The model is trained using both article-level and chunk-level distillation techniques, which help in aligning the retrieved documents with the fact-check articles and improving the quality of generated justifications. Experiments show that JustiLM outperforms several baselines, including In-Context Learning (ICL) enabled models like Flan-T5, Llama2, and the state-of-the-art few-shot RAG model Atlas. It also performs competitively with the latest GPT-4 model. A straightforward extension of JustiLM for joint veracity prediction and justification generation improves the veracity prediction task with large margins. The paper also discusses the limitations of previous approaches that rely on fact-check articles for justification generation, highlighting the need for a more realistic approach that uses retrieved evidence instead. The proposed method addresses these limitations by using a retrieval-augmented language model and distillation techniques to generate justifications based on retrieved evidence. The results demonstrate that JustiLM can generate high-quality justifications that are more similar to those written by fact-checkers. The model's performance is evaluated using various metrics, including ROUGE, MAUVE, and SummaCC, which measure the quality of generated justifications in terms of factual accuracy, fluency, and consistency. The experiments also show that JustiLM generalizes well to new claims and performs consistently across different test sets. The paper concludes that the proposed approach provides a realistic and effective solution for explainable fact-checking of real-world claims, with the potential to enhance the credibility and trustworthiness of fact-checking systems. Future work includes exploring the adaptation of various LLM-based reasoning methods into JustiLM to further improve the task of justification generation.